# Tea

Teas I have drunk, with reviews and future purchases (personal, food, statistics)
created: 13 Apr 2011; modified: 22 Apr 2017; status: in progress; confidence: log;

Tea is one of my favorite drinks: a remarkable variety of flavors, caffeinated but cheaper than coffee and healthier than soda, easily prepared, socially acceptable, with a long & rich history intertwined with geopolitics. I particularly favor greens, oolongs, & ku-ki chas; blacks tend to be too bitter and unpleasant for me, pu-erh tastes both strange and bitter, while whites strike me as subtle to the point of tastelessness. I sometimes enjoy herbal-teas/tisanes but they are not a focus.

I order mostly1 from the Massachusetts mail-order loose-leaf Upton Tea Imports because they specialize in loose tea (while not all loose leaf teas are good, almost all good teas are sold loose-leaf), I’m a sucker for browsing their catalogue & reading excerpts from books about the history of tea like the Ancient Tea Horse Road, their prices seem pretty reasonable, their selection is broad (they boast of having something like 400 kinds of tea at any time, which I can believe), and they offer small samples of almost all their teas (which is a fantastic way to taste scores of teas without having to buy $20 of each tea which one may not like). In contrast I haven’t been terribly happy with other retailers like the selection of loose teas on Amazon (surprisingly sparse) or Teavana (physically convenient to buy from but quite limited selection & overpriced). Upton allows reviews if you’ve bought at least a certain quantity, but otherwise your notes are private. This strikes me as a little unfair (a sampler of 10g is more than enough to judge a tea!) and my reviews are a valuable guide to me in ordering, so I keep local copies of my reviews & notes. (Note that Upton’s recycles some listings so a hyperlink may not point where it should.) I drink primarily using a Colonial Williamsburg Beware The Fox stoneware mug (3-inch diameter), putting the tea in to steep using a Finum medium brewing basket and using ~1.5g of tea for two steeps if possible. (My Finum lasted around 5 years before the mesh starts getting clogged enough that it takes an unreasonable time to drain into the cup, especially with the finer green teas; cleaning helps but doesn’t fully restore circulation.) # Tea ## Oolong When I was young, I was a great fan of hot chocolate, but hot chocolate is troublesome to make if you are making real hot chocolate (with milk & everything). I tried coffee once or twice, but it was even more disgusting than beer. Herbal teas were drinkable, though, and I slowly graduated to green tea. Then one day a my mother bought a Bigelow box set of teas which happened to include an oolong tea. I instantly fell in love with oolong - not quite as raw and grassy as green tea but not so bitter & disgusting as black tea. (Not that green tea is bad; I still liked it, and all my favorite oolongs tend towards the green side of the oolong spectrum. I just prefer oolongs.) In roughly chronological order: • Tie-Guan-Yin Oolong First Grade (★★★★☆ / ★★★☆☆) A very nice tieguanyin (which is one of my favorite kinds of oolong). The flavor is straight oolong: in between green and black, with a tiny bit of sweetness. One of the best I’ve had. Handles re-steeping well. (It is largely the same as the second-grade, but the second had a sort of woody taste to it that the first doesn’t.) On the strength of this tasting from 2009, I ordered 400g of it in 2012 to be my standard tea when I ran out of samplers; to my great disappointment, it does not taste as good as I remember it. I don’t know whether my palate has become more demanding or whether the quality has fallen. An online acquaintance happened to order some at the same time, and was very satisfied with it, suggesting the former. • Tindharia Estate Oolong (★★★☆☆) Nothing memorable. • Bao Jun (★★☆☆☆) Like the Tindharia, nothing memorable. In fact, this was pretty weak in flavor. • Far too bitter and dark and burnt tasting! • Formosa Jade Oolong Imperial (★★★★★) The finest Jade Oolong that we have ever sampled. Those who are looking for the best of what Taiwan can produce will want to try at least the sample size (12 grams). The first time I ordered a sample, I thought the Imperial was extremely good - one of the, if not the best, oolongs I’ve ever had. But also expensive, so I did not order it again for several years. Again I was struck by the wonderful complex fragrance one inhales as one opens the bag. I wasn’t quite so impressed the second time, having had many more oolongs since then; it is indeed excellent, but I have to stand by my original appraisal that it is too expensive. Formosa Jade Oolong Imperial • Tie-Guan-Yin Oolong Second Grade (★★★☆☆/★★★★☆) Just slightly woody. Otherwise, a solid good oolong. Downside is that it does not resteep so the price advantage is less than appears. • China Oolong Buddha’s Palm (★★☆☆☆) Too smoky. • Osmanthus Oolong Se Chung (★★★★☆) It’s a solid oolong, but the floral taste (I don’t know how to describe the osmanthus flavor) really makes this for me. I like to mix a little of it into some of my other oolongs, though it’s not the best re-steeper I’ve ever had. This was my default oolong for a long time because 500g was just$18. One of the downsides of buying in such bulk is that the osmanthus fragments exhibited a Brazil nut effect and the last hundred cups were more osmanthus than tea.
• Fen Huan Dan Cong (★★☆☆☆)

The description promises a strong flavor, but perhaps I prepared it poorly because the flavor struck me as weak, nor did I particularly notice any peach. I was disappointed; I’d’ve been better off buying some more of the Osmanthus or 1st-grade Imperial.
• A solid oolong somewhere between the Second and First Grade oolongs
• Fancy Oolong Imperial (★★★★☆)

Very good; similar to the First Grade Imperial oolong.
• Benshan (★★★★☆)

I bought this and the roasted barley tea (see later) from the Rainbow Grocery Cooperative when I was visiting my sister in San Francisco. Benshan is a fairly green oolong and right up my alley, although it struck me as lacking the slight sweetness and floral overtones I expect from the best oolongs. But regardless, it was pretty tasty, and adding a little bit of the barley made the benshan oolong even better.
• Iron Buddha from Teavana (★★★☆☆)

Standard oolong; nothing memorable.
• Oolong Fine Grade (★★★☆☆); standard oolong
• Formosa Amber Oolong (★★☆☆☆); too black-tea-like
• Formosa Jade Oolong (★★★☆☆ / ★★★★☆); quite tasty, in the same vein as the First and Second Grade oolongs (although not as good)
• China Oolong Se Chung (★★☆☆☆); just as described - too woody for me
• Ruan Zhi Thai (★★★☆☆)

I didn’t expect much of a Thai tea, since I’ve never heard of oolongs from Thailand before. To a little surprise, I found it to be a completely normal oolong. Nothing floral to the taste, just a plain ordinary oolong. I would not have suspected you of lying if you had told me it was a Formosan oolong.
• Very good oolong. Comparable to the First and Second grade Imperial oolongs, without doubt.
• A disappointment; nothing special - the subtle notes are too subtle for me.
• Tie-Guan-Yin Special Tribute (★★★☆☆)

Rolled leaf-balls. Similar to the Oolong Fine Grade; but has a somewhat mysterious floral taste I can’t really compare to anything. Doesn’t seem to re-steep very well.
• Wuyi Golden Guan Yin (★★☆☆☆)

Loosely rolled long leaves; weak flavor with nothing of interest about it. (I’ll agree with the Upton’s description that it’s not bitter, but calling it sweet or having a raisin-like flavor is just hyperbole.) Disappointing.
• Floral Jinxuan (★★★☆☆)

At first, I thought this was ordinary, but upon resteeping I noticed the promised floral notes - they reminded me strongly of the osmanthus oolong.
• Formosa Oolong Spring Dragon (★★★☆☆)

Like the Special Tribute, but weaker in flavor, I think.
• Tea at the Empress (★★☆☆☆); I picked up this dark blue cylindrical tin of teabags somewhere or other. It doesn’t even specify what kind of tea it is, but apparently it has something to do with a hotel, and claims to be from The Fairmont Store (although no item is listed similar to the tin).

It’s not very good oolong. It starts off fairly bitter and doesn’t improve, but at least it doesn’t get too horrible as it resteeps. Regardless, I don’t know where I would get more and I would not get more if I knew.
• Empress Guei-Fei Oolong (★★★☆☆)

At 5 minutes of steeping, a pretty ordinary oolong; by 10 minutes, a strong floral taste had developed. Continued steeping made the flavor weaker and bitterer (as one would expect), but no other changes. It reminded me of the osmanthus oolong. During the second tasting, the floral flavor was not as overpowering; I was careful to use the same tsp amount of tea for each of the 9 teas, which suggests that perhaps last time I used too much of the Empress. Not bad at all, I may order it again.

At 5 minutes, another ordinary oolong, but by 10 minutes, the flavor has not become bitter but rather continued to develop into a very oolong flavor. Little change with re-steepings. In the second tasting, I noted that it was a sharper blacker flavor than Anxi and Empress. A good oolong, might be a candidate for my standard tea (but would need to check prices of the others).
• Formosa Oolong Choicest (★★☆☆☆)

The 5 minute steeping tasted both woody and floral, an odd combination which bothered me (I had expected more - it cost twice what the Oolong Choice Grade did). The 10 minute steeping wasn’t much better: it was sweeter tasting, but the stem/wood flavor was even stronger, and it didn’t improve or change very much at any subsequent steeping. It’s possible I prepared it wrong or picked a pinch of stems, but it seems unlikely I will pay the premium for this tea when I am not sure I can even describe it as good. (In the second tasting, I noted only that it was slightly sour.)
• Anxi tikwanyin (★★★☆☆)

Another gift from my sister. This is a mild medium oolong with relatively little floral taste compared to everything else I’ve been testing. As expected from the Anxi county tea region, their Tie-Guan-Yin is perfectly acceptable.
• Momo Oolong Super Grade, Lupicia Fresh Tea (★★★☆☆ / ★★★★☆)

As the name indicates, this is a peach-flavored oolong. I bought a bag of 10 teabags during Sakura Matsuri 2012. I wondered if $12 was too much to pay, but the bag seemed oddly heavy and the back said each bag had 2g of tea in them! 2 grams is a lot, and 20g is more reasonable for$12 - similar to Upton’s samples when S&H is included. (When I checked online, I saw the loose tea was $13 for 50g. Oh well.) The bags were the first I’ve seen made with a plastic mesh, and when I brewed the first one, the taste was far too strong. It was without doubt peach-flavored. For the next batches, I cut open the bag and used a fourth of the contents. This made a much more reasonable flavor, which holds up well under resteeping, and the peach-flavor is not as artificial-tasting as the other peach tea I have now. One thing I’ve learned after drinking many mugs is that this tea quickly becomes flavorless - it doesn’t hold up under resteeping; this may be because it was designed for quick release as tea bags - but hopefully the loose tea is unshredded leaves and this would be less of a problem. When I run out of tea, I may order a batch of Lupicia since besides the Momo Oolong, they have some oolongs I haven’t tried before. • Floral, but oddly it also tasted sour. Not recommended, to say the least, but perhaps the first tasting was simply an aberrant cup. On later tastings, I didn’t notice further sourness, and it seemed more acceptable. Dosing is difficult because the large whole leaves are very tightly wrapped but sometimes are just stems, so it is easy to add too few or too many. • Neither left a strong enough impression to review although the Floral Superior lived up to at least the first part of the name; they were both similar to the Special Grade. At times during this tasting, I wondered if Upton had screwed up & they were the same teas (but they couldn’t’ve been because the tea leaves were visibly different). The Floral Superior does not handle resteeping well, quickly losing flavor. • Super Fancy Oolong (★☆☆☆☆) Indescribable taste, but whatever it is, makes it bad. • Roasted Oolong (★★★☆☆) Pretty much as expected: a standard oolong taste with a smoky aftertaste. Smoky oolongs are not my cup of tea, but I had to try. The upside is that it turns out to resteep very well, and the smoky aftertaste slowly changes to a sweeter honey-like aftertaste. • Magnolia Blossom Oolong (★★★☆☆) The magnolia flavor is strong with this one. I was surprised to instantly recognize the flavor, because as far as I knew I had never had anything magnolia-flavored before. The flavor itself leaves me mixed - I sort of like but also sort of don’t. This may be one of the teas best consumed only at intervals or mixed in with another. It doesn’t resteep well, almost immediately losing any flavor. • Pre-Chingming Da Hong Pao (★★☆☆☆) Floral and weak. More green-white than oolong. • Organic Da Hong Pao Oolong (★★☆☆☆) A stronger Pre-Chingming Da Hong Pao, which then undercuts the improvement by tacking on an aftertaste which is not smoky but burnt. In general, this batch of oolongs was a disappointment: either boring or bad. I may finally have exhausted Upton’s oolong catalog. • A Christmas gift, this flavored oolong comes in the nice little plastic mesh bags that non-loose-tea products seem to be moving towards these days. The Se Chung and Shui Xian blend is heavily flavored with safflower, peach, and apricot for a somewhat overwhelmingly floral taste which makes it hard to judge the underlying oolong (it seems OK, but not great). Seems to handle a few resteeps well. • Discover Tea’s Ti Kuan Yin (★★★☆☆) A perfectly ordinary and satisfactory oolong; it handles steeping well and delivers a cup medium between green and black. While I was at their Williamsburg shop, I had a cup of their Glenburn Moonshine Oolong; it’s hard to judge from one cup you didn’t make, but while the leaves have a lovely silver fuzz and the brew was pretty good, I didn’t like it sufficiently to justify the 2-3x premium over the tie kuan yin. • Bao Zhong oolong with coconut extract. I am not a fan of coconut flavor and bought it out of curiosity when I wandered into their Williamsburg shop before Christmas 2013 (I also bought an ounce of their genmai-cha). It was better than I expected: the coconut is a light overlay and not overpowering, and the base Bao Zhong seems to be fine. • Solid oolong, much like a tieguanyin with the floral after-taste I love so much in oolongs. Resteeps normally without becoming too bitter. • Tao of Tea, Black Dragon Oolong Tea (★★★☆☆) A black tea in all but name; very similar to the Amali African Queen. Steeps perhaps twice. Didn’t much enjoy, but not as bitter & unpleasant as most black teas. • Tea’s Etc, Ginseng Loose Leaf Oolong Tea (★★★☆☆/★★★★☆) I hadn’t tried a ginseng tea before, and when this one popped up on Amazon, I thought I’d give it a try. While I strongly suspect the health benefits of ginseng have been overblown2, the flavor might still be nice. The tea comes in coated pellets, with some wisps of straw-colored plant matter which I assume are ginseng itself. The ginseng flavor is sweet, mild, fruity & difficult for me to compare to anything (I guess I should just describe it as ginseng-like!). I think I like it, although like the coconut oolong I wouldn’t want to drink too many cups in a row of it. • Daniel Clough, Golden Lily Wulong (★★★☆☆) 1 of 4 oolongs gifted me by Clough after his travels in China. Interesting and not what I expected, since the tea looked more like a tieguanyin. The Golden Lily almost doesn’t taste like an oolong at all: it tastes sweet, perhaps like honey?, and something harder to describe - googling, it seems the usual description is milky, which on further reflection seems like it’s a good analogy. • Clough, Lan Gui Ren ginseng (★★☆☆☆) A ginseng oolong like the previous Tea’s Etc; there’s no straw in it, and the coated pellets are much smaller, although unlike the other, the pellets do open up into tea leaves. Weakly ginseng, sweeter, and almost completely tasteless after the first steep. This one was a disappointment; I hope the other ginseng turns out to be better. • Clough, unknown ginseng (★★★☆☆) A normal foil baggy of little ginseng pellets; no straw, small more irregular pellets, green color. Similar the Tea’s Etc one. • Clough, unknown oolong (★★★☆☆/★★★★☆) A vacuum-sealed small sample (10g?) of Chinese tea; I didn’t recognize any of the names or characters on it (I took a photo just in case it turned out to matter). The first steep is a fairly tasty quasi-tieguanyin, but subsequent steeps are absolutely tasteless, which meant I used it up quickly. • Tao of Tea, Wu Yi Oolong Tea (★★★☆☆) Very similar to Tao of Tea’s Black Dragon Oolong Tea, which I didn’t much like either, but is better than the usual black. • Summit Tea Company, Tie Guan Yin Oolong Tea (★★★☆☆) Medium oolong, somewhat floral, survives only one steep, not terrible but fairly weak flavor. Tie Guan Yin on a budget: I’m not sure if one can do better for cheaper, but one could easily do better. • Art of Tea, Iron Goddess of Mercy Oolong Tea (★★★★☆) Reasonable Tie Guan Yin, very green, nice floral aftertaste; sensitive to temperature, though, and easily prepared too hot. Probably can do better quality vs price-wise. Container is a bit flimsy and if it falls to the ground, will spill contents all over (as I found out the hard way). • Tao of Tea’s Royal Phoenix Oolong Tea (★★★☆☆/★★★★☆) Toasty texture, fragrant aroma and sweet, high-bounce taste similar to nectarines and peaches. Origin is Guangdong Province, China No reviews, but I thought the description sounded promising and Tao of Tea has earned a little bit of trust, so I took a gamble with it. The leaves are long stringy black leaves. It resteeps well. My initial impression was that the flavor is indeed somewhat sweet and, grandiose name notwithstanding, it tastes like a middle of the road oolong with no particular additional flavors or aftertaste - just sort of oolong-y. I was disappointed: OK, not good I think I must have prepared it badly the first few times (perhaps too hot or steeped too long) because as I drink the rest of it, I’m enjoying it more and the flavor seems closer to the floral sort of Tie Guan Yin flavor I like most. • Huang Jin Gui Oolong (★★★★☆) This premium Oolong is produced in Anxi county of Fujian province, with a light oxidation level of less than 20%. The name Huang Jin Gui translates to golden osmanthus, referring to the cup’s light gold hue and the osmanthus-like aroma and flavor. (This and the next 3 are all Chinese teas.) Description is entirely accurate for once. I liked it. • China Tie-Guan-Yin Organic (★★★☆☆/★★★★☆) This organic selection has a sweet aroma with hints of tropical flowers and a suggestion of toasted coconut. The cup has interesting notes of stone fruit, golden raisins, and walnuts. The finish has a fruity/floral quality, which is balanced by a light mineral note. Regular TGY. I don’t find the complexities that Upton’s described, but it’s fine. • Wu Yi Water Fairy Oolong (★★★☆☆) While not a true Wu Yi Mountain tea, this Fujian province Oolong is a flavorful and affordable alternative. The dark, chocolate-brown leaves produce a dusky ecru liquor with a harmonious flavor profile, accented with a sweet, lingering finish. Some who have enjoyed this selection have commented about nuances of honeysuckle, citrus and peach. On the black end of the spectrum; it’s not as bitter as the previous Wu Yi I tried, which I am grateful for and makes it reasonably drinkable, but this one settles it: Wu Yis just aren’t for me. Time to give up on them, and probably time to start avoiding any oolong which is sufficiently oxidized to be described as black or chocolate-colored. • Zhang Ping Shui Hsian Oolong (★☆☆☆☆) This loosely-rolled Fujian province tea is tightly packed into paper-wrapped bricks. Infusing reveals bold, skillfully crafted leaves with a fresh aroma and a hearty cup with a lilac/hyacinth fragrance. The sweet finish has a delicate suggestion of cardamom. I thought this sounded cute - paper-wrapped bricks of tea, a throwback to the traditional methods of packaging and storing tea in China. And it sounded quite good too, a greener oolong right up my alley. But this one was a serious disappointment! The bricks turn out to be a lot of small bricks, and they are a pain to work with; you cannot simply reach in and get some tea, you have to break off compacted chunks of tea, which are hard to measure right and scatter debris (if you do it outside the bag, it’s a mess to clean up, and if you do it inside, the dust will fall to the bottom). I could put up with this format except to my perplexity, the tea seems almost tasteless, not hearty; I tried steeping at a variety of water temperatures (though Upton’s calls for 190 degrees, which is not exotic or unusual), convinced I was simply preparing it wrong, but none of them did the trick. (To avoid death by a thousand cleanups, I wound up crushing all the bricks by hand in a big bowl and then pouring them back in.) Expensive, messy, and tasteless. • Tranquil Tuesday: Phoenix Honey Orchid Oolong (★★★☆☆) • Golden Dragon (?): Tie Guan Yin Oolong Tea - Iron Goddess of Mercy (WuLong) (★★☆☆☆) Tasteless in much the same way as the Zhang Ping Shui Hsian Oolong was - not a bad taste, but hardly there. Last perhaps one steep and then even the weak flavor is gone. Mine came in a mostly-unlabeled foil bag (so I have no idea where it’s really from), and the reviews for the tins are more positive, suggesting I was sent a lower-quality alternate tea. • The Tao of Tea, Osmanthus Oolong Tea (★★★☆☆) Doesn’t compare well with Upton’s osmanthus oolong. Same problem as the Tao of Tea genmai-cha: the added flavor (osmanthus) is almost untastable and the base tea is nothing to write home about. • Yamamotoyama, Oolong Tea (★★★☆☆) Conveniently available in grocery stores, and not as bad as one might expect of bag tea. On the black end of things, without any of the green or floral tastes but more of a robust kukicha-like flavor. • [Pre-Chingming New Style Fairy Oolong](http://www.uptontea.com/shopcart/item.asp?itemID=ZOP3 “2g,$1.80) (★★★☆☆)

From Hunan province, this 2015 special production Pre-Chingming Oolong is notable for its outstanding aroma and cup. The flavor notes are intense, with a pronounced orchid/lilac quality as well as a light mineral hint. The finish lingers pleasantly and sweetly.

A strong aroma whose floral qualities reminds me of the even more intense scent of the Jade Imperial; I find the flavor more akin to jasmine than orchid. Beyond that, the flavor is mild and meek, and green. Overall, I don’t think the floral aroma makes up for the lack of other distinction in its flavor.
• [Pre-Chingming Fenghuang Dan Cong](http://www.uptontea.com/shopcart/item.asp?itemID=ZOP2 “2g, $1.80) (★★★☆☆) From Guangdong province, this venerable-style Oolong tea is made from ancient single trunk Camellia sinensis trees. Notable for its peach-like flavor and a pronounced sweet character, this 2015 harvest is suitable for multiple steepings, as with the Gong-fu method. Remarkably sweet, this takes oolongs to a place I did not expect them to go. It is even milder than the New Style and there is a definite fruity flavor to it which I can’t pin down beyond citrus-y, although I don’t think peach-like captures it. Here too, while distinct, the flavor does not capture my heart. • Select Tie-Guan-Yin Oolong (★★★★☆) The leaves of this selection display a range of color, with tans, dark olive greens and browns. Rolled in a semi-loose fashion, this tea is processed in the Muzha style (i.e., with a finishing light roast). The smooth liquor is sweet, with both fruity and flowery notes. The finish is clean and lightly sweet. Produced in Anxi, Fujian province. A solid, standard TGY: dark green liquor, resteeps well, and has the virtue of a good TGY in combining the characteristic floral overtones with a robuster main flavor. There are better oolongs but not terribly many. • Jasmine Oolong (★★★★☆) The base tea of this offering is a quality Tie-Guan-Yin Oolong with light oxidation, expertly scented with jasmine blossoms. Most of the blossoms are removed after scenting, which results in a smoother cup. The liquor is a fine marriage of orchid, jasmine, and other floral notes. A strong jasmine you smell as soon as you open the package, which overwhelms the oolong without being bitter or grassy. I’m developing a fondness for jasmine, and this hits the spot. • Tie-Guan-Yin Standard Oolong (★★☆☆☆/★★★☆☆) This Oolong tea has a neat appearance, with evenly rolled leaves ranging in color from dark olive green to a lighter lime green. The clear infusion is light, with a green-yellow hue. The pleasant flavor has notes of almond milk and light floral/citrus hints. Produced in the Fujian province of China. Thoroughly mediocre. It’s not cheap enough to offset how it’s not much of a TGY and doesn’t resteep. It makes a good TGY cup if you use twice as much as usual but that further destroys the cost advantage. • Formosa GABA Oolong (★★★☆☆/★★★★☆) Japanese researchers created GABA tea serendipitously in the 1980’s. Wanting effective methods to preserve tea, not fully fermented leaf was exposed to nitrogen. The glutamic acid inherent in tea was transformed to Gamma-Aminobutyric Acid, or GABA for short. This selection is notable for its broad and interesting flavor profile, with notes of mango, peach and guava. The cup has a creamy body with a pronounced, lingering sweetness. It’s unusual to see a tea with an origin story fit to rival a superhero’s, and GABA itself is an interesting chemical with the potential to augment tea’s theanine as an anxiolytic (although the much greater efficacy of phenibut suggests GABA on its own may be impotent); as soon as I read the description, I knew I had to try it. This one of the two GABA teas Upton’s carries (the other is Japanese Green GABA (Gabaron)); I regret not ordering a sample of the other as well, so I could compare them to each other and to other green/oolongs and get an idea of what part of the taste or effect may be attributable to the nitrogen/GABA process and what is part of the underlying tea which happened to get processed that way. (I have noticed that when it comes to additives or different processes like genmai-cha, the base tea determines how much I like it as much as the additives and that they often are of lower quality; so when I dislike something, it may be the additive, or it may be the base tea. So I need to try at least two, or dislike intensely what is clearly the additive, before I can be reasonably certain and ignore that category henceforth.) The flavor itself is as described, with a fruity rather than floral overtone. Eventually I ordered the Gabaron to compare side by side. The Gabaron tasted like a normal enough green tea, and I couldn’t detect any similarities. • Se Chung Oolong Classic (★★★☆☆) This is a great everyday Oolong at an attractive price. The neatly rolled leaves yield a golden-yellow liquor with a light floral aroma. The medium-bodied cup has a sweet vegetal quality with woody hints and fruity undertones. A pleasant astringency lingers in the finish. Regular oolong, not much flavor • Formosa Jade Oolong Supreme (★★★☆☆/★★★★☆) Those who appreciate the finer grades of Tung-Ting style Oolong will find this one an exceptional value. The flavor is surprisingly more refined than item TT86. Honey-like, floral. • A classic restaurant grade tea, with a smooth character, and classic Formosa Oolong flavors. Priced for everyday use, this selection is an excellent choice for its value and quality. Woody with some smoke. • Formosa Amber Oolong Select (★★★★☆) This grade of Amber Oolong (Wu-Long) has more complex flavor and finer leaf style than our TT55. Highly recommended. Floral, sweet aftertaste, handles long steeping well • Reminiscent of a roasted Tie-Guan-Yin, this fannings grade Oolong infuses in one to two minutes. The smooth cup has an earthy molasses character and a light red apple note, with hints of honey and pecans. The lingering finish has a slight minty note. (Fannings are very fine, almost dust-like tea.) Steeps almost instantly, dark brown liquor. Bitter with a coffee-like aroma, it tastes exactly like black tea and not at all like an oolong, even one of the darker and smokier ones. • Thurbo Estate FTGFOP1 Darjeeling Oolong (DJ-300) (★★☆☆☆/★★★☆☆) This tea is an attractive mix of well-twisted, wiry leaves, decorated with downy silver buds. The light amber cup has a pronounced sweet aroma with light fruity notes. A smooth, creamy mouth feel introduces flavor notes of stone fruit and nuts, which some have likened to pecan. A highly unusual Indian oolong. Medium brown liquor, not particularly black-tasting. Some unidentifiable funky overtone for me that triggers association with mold and wet dogs and puts me off despite the admittedly nice white-tea-esque visual appearance of the leaves. • This is a rare production of Oolong style tea from Assam. The leaves are beautifully made, with a color range of multi-hued browns and some silver and gold tips. The liquor has a light sweet note, which deepens into a complex flavor profile with caramel hints. This tea is produced by hand using old-time methods, including drying over a charcoal fire. Similar to the Darjeeling in unusual origin, appearance, and liquor color. It lacks the offputting overtone and has a sweeter and pleasant taste. • Tao of Tea: Frozen Summit (★★★★☆) • Tao of Tea: Oriental Beauty (★★☆☆☆) • China Black Tea Tie-Guan-Yin (★★☆☆☆) Produced from a cultivar used for Tie-Guan-Yin Oolong, this black tea selection has an interesting, complex aroma and flavor. The large, midnight brown leaves yield an amber cup with an aroma of chestnuts and wood. A refreshing minty suggestion complements the nutty/woody flavor notes. Oolong lovers will delight in this unique offering. Black leaves yielding a standard bitter and unpleasant black tea, with no discernible connections to normal TGY oolongs, really emphasizing the difference that the processing makes between green/oolong/black. • Formosa Amber Oolong Select (★★★☆☆) This grade of Amber Oolong (Wu-Long) has more complex flavor and finer leaf style than our TT55. Highly recommended. Darker yellow liquor with extremely sweet milk-like taste. Resteeps well. This offering is a 2016 special production Pre-Chingming Oolong from Hunan province. The striking, dark-olive leaf is quite bold and fully intact. Intense orchid/lilac notes are pronounced in both the aroma and the smooth, buttery flavor. Subtle nuances of stone fruit and vegetal hints round out the flavor. Long green leaves producing equally green liquor reminiscent of gyokuro's 'grassy' overtones but with definite floral aftertaste. Resteeps not too well. • Japanese Oolong Organic (★★★☆☆) This unusual tea has dry leaves of differing shapes and colors. The end result is an outstanding cup with a complex flavor profile that is unmistakably Oolong in character, with woody hints, delicate floral notes, and a sweet lingering finish. Like the Standard Grade Formosa Oolong, this is a basic oolong heavy on the woody flavor and somewhat stem-y. A decent cheap accompaniment to a meal but not a great oolong. • Eastern Beauty Oolong (★★★☆☆) The bold leaves of this limited production Eastern Beauty Oolong yield an amber-gold cup with a sweet, rich character. Heady notes of ripe fruit and honey are present in both the aroma and the complex flavor, which finishes with a light suggestion of spice. Meh. Not nearly as flavorful as other Eastern Beauties. • Clipper Ship Tea Company: Oriental Beauty Oolong (★★★☆☆) • Ginger’s Oolong (★★★☆☆/★★★★☆) When we added ginger to our peachy and delicious Formosa Oolong, we created Ginger’s Oolong, a fun and flavorful spin on our popular Peaches & Ginger tea. Kosher. Details: This is an old blend done over ten years ago. Our Peaches & Ginger is a popular black tea blend. So we thought, Formosa Oolong has peach notes, lets add some ginger root. Thus Ginger’s oolong was born. Dry Leaves: Dark brown leaves. Liquor: The ginger in this tea makes the liquor slightly darker, a medium brown. Aroma: The oolong provides the dark peach notes and the ginger gives spicy aromas. Caffeine Level: Caffeinated Body: The oolong is medium bodied. Flavors: The subtle flavors of peaches and toasted nuts are nicely contrasted by the ginger. Gingery but not overpoweringly so. • Rou Gui Oolong (★★★☆☆/★★★★☆) It is a pleasure to offer again Rou Gui Oolong. We love its roasted fruit flavors. It is made in the same area as Da Hong Pao: the Wuyi Mountains of Fujian in China. Dry Leaves: Twisted, dark brown leaves. Liquor: Amber. Aroma: Roasted apricots. Caffeine Level: Caffeinated. Body: Medium. Flavors: Apricots & Peaches. • Pomegranate Oolong (★★★☆☆) We infuse full leaves of premium Ti Quan Yin Oolong with tangy pomegranate to create this complex full-bodied blend that brews into a sweetly fragrant and silky textured cup of tea. The Ti Quan Yin Oolong we use is named after the Chinese Goddess of Mercy. Please note: Pomegranate Oolong in our box of 50 tea bags and our Historic Royal Palaces tin have been discontinued. Please continue to enjoy this tea in our other tea sachet collections, loose tea, or Fresh Brew Iced Tea pouches. Details: We wanted to offer a flavored oolong blend. This would make oolongs more approachable to some tea lovers. So we chose a good oolong and added the pomegranate. Dry Leaves: Rolled green leaves. Liquor: A very light clear green-yellow. Aroma: On top of the floral and citrus flavors of the tea lies the sweet citrus aromas of pomegranates. Caffeine Level: Caffeinated. Body: Medium body. Flavors: A lovely tasting oolong that is light and refreshing with strong flavors of pomegranate. ## Green • Pinhead Gunpowder is a green Chinese tea. Pale straw colored, the brew is light and refreshing in flavor. Each leaf is hand-rolled into a pellet-shaped ball. Because the tightly rolled shape helps the tea retain its freshness, it was one of the first teas to be exported from China. 1/4 lb. Loose Tea, No.42316 I originally bought a packet on a trip to Colonial Williamsburg around 2005 or so. It struck me as rather grassy and the tightly-rolled leaves seemed to easily oversteep and become bitter. Having received another batch in Christmas 2013 and preparing it with more respect for being a green tea (sub-boiling water, much shorter steep time), I find it more palatable and so I think my early impressions may have been more my fault than the tea’s fault. • Xian Shan Pouchong (★★★☆☆) Rolled green tea; strongly reminiscent of oolongs and definitely on the border. Fairly good considered as a green/oolong cross, but nothing memorable about the flavor - similar to Oolong Fine Grade. • Green Tea Pomegranate, English Tea Shop (★☆☆☆☆) • Satori Tea Company’s Sencha Klaus (★★☆☆☆) Gift from sister; a tin of variegated green (long thin leaves, stems, broken leaves) mixed with flakes of thin orange peel or skin. As the name indicates, it’s a Christmas-style tea which makes it taste like potpourri. The flavor is interesting; after the first few minutes, it struck me as a sweeter kind of green but I can’t figure out the flavor - minty? Floral? Some sort of citrus orange? After another 5 minutes, it’s much stronger and I feel confident identifying it as an orange flavor. It’s strong enough that I don’t think I want to drink it on its own, but perhaps I could mix in the Dae-Jak. (Satori’s description identifies the contents: almond bits, cinnamon bits, natural flavor and orange blossoms. Makes sense.) I ultimately wound up picking out all the orange peel to make it more palatable. • TeaAndAbsinthe’s sun dew apricot mango mix (★★★☆☆) Purchased at ICON 2012; after losing the bidding war for the tea item, I resolved to go find the original vendor in the dealers’ room, which I succeeded in doing. To my surprise, they were primarily a steampunk clothing vendor who happened to have one shelf-unit of tea mixes. Mostly blacks and rooibos, but there was one green that smelled nice and I was piqued that I had lost the bidding war Saturday for the awesome original ICON artwork and then the bidding war Sunday for the 3 teas, and it was just$3 an ounce. I had a nice chat with the guy, and bought an ounce of the mango green tea.

It has a pleasant green flavor with no real negatives, and the mango/apricot is not overwhelming. It degrades gracefully under resteeping. Overall, it’s quite good: better than most floral flavorings, above the peach tea (but below osmanthus oolong) in my estimation. Unfortunately, when I checked their website, they seem to offer no online shopping or long-distance ordering capability. I guess I will have to wait for ICON 2013 to buy some more.
• Stash Premium, Mangosteen Green Tea (★☆☆☆☆)

A disappointment. Not a good green, and the mangosteen just tasted too sweet. I didn’t bother with a second steep.
• Davids Tea, Daydreamer (★★★☆☆)

Small sample packet - a sencha green with mango & mangosteen. Much better than the Stash Premium. It started off well, and handled resteeping admirably. Competitive with TeaAndAbsinthe’s sun dew apricot mango mix, although a simpler overall flavor.
• Korean greens:

• Dae-Jak (★★☆☆☆)

After 5 minutes, struck me as rather grassy, akin to gyokuro, but with a weaker flavor. By 10 minutes, it was still grassy but a certain unpleasant edge had crept in, which was still there after the resteep. Not impressed. During the second-tasting, the unpleasant edge was weaker than I remembered, but otherwise both the Dae-Jak and Jung-Jak tasted the same.
• Jung-Jak (★★☆☆☆)

Very similar to the Dae-Jak, but less sweet (when tasting them side by side); the sweetness passed Dae-Jak at 10 minutes, and at 15 minutes, I wasn’t noticing the unpleasant edge. Better than the Dae-Jak, but I still doubt I’ll be ordering it again.
• Gyokuro Kenjyo (★★★☆☆)

At 1 minute, it’s a sharp tasting green which reminds me of a previous green tea I’ve had, but maddeningly, I can’t seem to place the specific aftertaste. At 5 minutes, the taste is stronger (but not more bitter or worse).
• Pre-Chingming Snow Dragon (★☆☆☆☆)

At 1 and 5 minutes, this is almost tasteless. I’d liken it to a white tea, which it may well be better classified as. I’d call it bad, but that implied there was any real flavor to dislike.
• Kagoshima Kabuse Sencha (★★★☆☆)

A ordinary sencha, the only thing I’d note is the slight floral note. Handles resteeping well.
• Yamamotoyama’s Genmai-cha Green Tea with Roasted Brown Rice 16-pack (★★★☆☆)

Picked up at my grocery store for $2 out of curiosity. As the instructions warn you, this green doesn’t handle resteeping very well and turns bitter after a few minutes. The roasted brown rice flavor is very strong and one can smell it upon opening a teabag packet. The green tea itself is acceptable. The combination is not bad, but I think the rice is over-toasted and comes off as a bit too burnt. The lesson here may be to find my own source of more lightly toasted brown rice. • Spice & Tea Exchange, Genmaicha (★★★☆☆) This improves on the Yamamotoyama. The rice is toasted much more lightly. I liked it, especially for drinking in the morning, although it doesn’t handle resteeps well and tastes a bit burned. I think genmai-cha can probably be even better, though. • Tao of Tea, Genmaicha Green Tea And Toasted Rice (★★☆☆☆) Devoid of the toasted-rice flavor - there’s grains of rice, yes, but it’s hard to believe they were ever toasted. It doesn’t taste nearly as good as the other two genmai-chas, and was a waste of money since it’s not that good a green tea on its own. • Organic China Gen-mai Cha (★★★★☆) A traditional combination of organic green tea with toasted brown rice produces a mild and smooth cup with nutty nuance and sweet, lingering aftertaste. A sweet and mild green, with an equally mild toasted-rice flavor. Definitely a good gen-mai cha. • Gen-mai Cha (Japan) (★★★☆☆) Literally, Gen-mai Cha means brown rice tea. Toasted and partially puffed rice is blended with large-leaf Sencha. Not as mild as the China gen-mai, with more of a green edge. The toasted-rice taste isn’t there, though. • Koto Genmaicha (★★★★☆) Genmaicha is a blend of Japanese green tea and roasted rice. Our Koto Genmaicha is a custom blend of Uji green tea and premium roasted Niigata brown rice. Enjoy its deep aroma and complex nutty flavor. Yields a vibrant yellow green tea. This tea is grown exclusively in the Uji region of Japan. The rice is grown exclusively in the Niigata, Japan. • Sato Matcha Genmaicha (★★★☆☆) Matcha Genmaicha is a blend of Japanese green tea and roasted rice coated with Matcha powder. Our Sato Matcha Genmaicha is a custom blend of Uji green tea and premium roasted Niigata brown rice coated with our Uji Matcha. Enjoy its complex aroma and deep nutty and leafy flavor. Yields a deep green colored tea. This tea is grown exclusively in the Uji region of Japan. The rice is grown exclusively in the Niigata of Japan. I’m not sure what the point of the matcha powder is, aside from coloring. It doesn’t taste much different. • Gyokuro (★★★★☆/★★★★★) The finest Japanese green tea, only shade-grown tips are used for Gyokuro. Prized for its delicate flavor and natural sweetness. Delicious. It’s described as sweet, and it really is! The taste is entirely different from your more usual green teas. It’s a pity it’s so much more expensive (3x$/g).
• The Tao of Tea, Handrolled Jasmine Pearls Green Tea (★★★★☆)

Marking a distinct departure from their gen-mai cha, which was seriously weak tea, here the jasmine is, if anything, too strong, overwhelming the green to the point where I’m not sure what it is. Since I like jasmine, this is good, and it resteeps well. (The only downside is that I was unable to drink the whole thing since about halfway through, I discovered a corner of the tin had become enveloped in a fuzzy mold; I keep my teas next to my sink & dish-drying rack, which might’ve caused that, but on the other hand, I can’t recall any of my other teas ever developing a fuzzy mold.)
• The Tao of Tea, Pearl Green Tea (★★★☆☆/★★★★☆)

A solid green tea, a bit sharp & oversteeps easily. Like a gunpowder.
• Produced for the Japanese domestic market, this special Sencha has beautiful, evenly shaped, deep-green leaves with yellow highlights. The vegetal aroma is layered with lively nutty tones. The cup has medium body and well-balanced flavor. Certified organic.

Like the other senchas. A pretty yellow-green liquor and even flavor (not heavy on the vegetable nor sweet nor bitter). Doesn’t impress me as much as the Sencha Fukamushi or Tamaryokucha, and bitter on resteeps.
• This offering is grown on the Nishi farm in the Kagoshima prefecture using the Yabukita cultivar, the most popular variety in Japan. Both the leaf and cup have a deep, dark green color and heady, rich aroma. The exquisite liquor has sweet, vegetal notes, a brothy mouth feel and pleasing, light astringency. The leaves are typical of the Fukamushi style, with both larger and small leaf pieces.

Powdery dark-green; the Upton description is dead on: a sweet vegetal taste with just the right amount of astringency. (No idea what is meant by brothy.)
• Organic Tamaryokucha (★★★★☆)

Tamaryokucha, (coiled tea) is a superb, handcrafted green tea that has a full, complex flavor. Vegetal notes and a pleasing pungency are accentuated by hints of dark berries and a sweet aftertaste. Produced in Takachiho, Miyazaki from the Yabukita cultivar.

Tastes much like the gyokuro: the vegetal notes leavened by a sweet distinctive aftertaste. But much cheaper.
• Japanese Ho-ji Cha (★★★★☆)

Bancha green tea is roasted evenly until it is brown, imparting a unique flavor. The liquor is golden brown; the taste is mellow.

A followup from Japanese Ku-Ki Ho-Ji Cha: since I liked the ku-ki when it was roasted, how would I like green tea similarly roasted? The answer is: quite well. It tastes much like the Ku-Ki Ho-Ji Cha, with perhaps a fuller flavor and a bit less bitter. I think I like it better as far as the roasted flavor goes (there’s no point in comparing with the Ku-Ki Cha Green Kamakura; that’s a green ku-ki, and this is roasted green) but I will probably order some of both later to directly compare.
• Kakegawa Matcha Organic (★★★☆☆)

A quality Matcha produced near Kakegawa city in Shizuoka prefecture. Packed in a silver matte tin.

This is the first tea I’ve ever bought where it came canned with a pull-tab. I suppose they are serious about preservation. The matcha is a bright green powder so fine and so consistent I briefly had the conviction that it was makeup or some industrial powder. The powder is so fine, in fact, that I immediately gave up the idea of using my usual Finum brewing basket, as the matcha would clearly clog the mesh and take forever to drain; instead, following descriptions of the tea ceremony and whipping up a dense liquid, I put spoonfuls directly into my tea mug with the water, and stirred vigorously. The water immediately takes on a totally opaque and somewhat-disturbing bright green appearance. The taste is strong and consistent and somewhat gyokuro-like, with almost no complexity or change in flavor over time that I noticed. It is certainly a matcha tea. Varying how much I put in did not seem to make much of a difference. 1g a mug (so 30 servings) works fine. Overall, while I’m impressed by the vividness of the green and the consistency of the powder and flavor, the flavor itself does not impress me enough to justify the price.
• This extraordinary grade of Matcha Gen-mai Cha is seldom seen outside of Japan. The components are of an exceptional quality, and produce a harmonious, well-balanced infusion that is eminently smooth. Toasted rice adds a soothing and mellow quality to the complex vegetal flavor. Highly recommended. JAS certified organic.

This one was odd: it didn’t taste much like the regular matcha did. What it tasted exactly like was the Organic China Gen-mai Cha (but more expensive). I’ll be sticking with the other gen-mai chas.
• Huang Ya Imperial Yellow Tea (★★★☆☆)

The leaves are long, twisted and threaded with fawn-colored tip. The aroma has nuances of pear and light citrus. The delicate liquor is smooth, with sweet notes of peach and steamed vegetables.

I had never heard of yellow tea before this. WP and various tea sellers make it sound like it’s most akin to green. Mild, inoffensive, entirely unremarkable flavor like a weak black, perhaps.
• 88th Night Shincha (★★★★☆)

This Shincha (first flush) tea is harvested on the 88th day of spring, and is manufactured using the traditional lightly-steamed process. The tea is made from the unique Okumidori varietal, known for yielding a very sweet and aromatic cup. The liquor has a mild aroma, with pleasant floral notes, accented by a delicate hint of freshly cut hay.

A very pleasant green tea in the same vein as the Tamaryokucha & Fukamushi.
• Tamaryokucha Koga (★★★★☆)

Tamaryokucha is a Japanese green tea that is processed uniquely to achieve its coiled shape. Our Tamaryokucha Koga is a custom blend of high Sencha grade Japanese tea. Enjoy its subtle sweetness with a strong floral fragrance and flavor. Yields a bright yellow green tea. This tea is grown exclusively in the Ureshino region of Saga, Japan.

• Uji Shibano Tea (★★★★☆)

Uji tea is a high grade Japanese tea grown exclusively in the Uji region of Japan. Our Uji Shibano tea is custom blended for the high quality. Enjoy a slightly sweet and gentle flavor and vibrant aroma. Yields a deep yellow green tea. This tea is grown exclusively in the Uji region of Japan

Similar to the Sencha Fukamushi.
• Chun Mee (Moon Palace) (★★★☆☆)

Literally translated, Chun Mee means precious eyebrows. One of our most popular organic China green teas. Steep about 2 minutes.

Cheap but astringent green. Oversteeps at the drop of a hat and you can only use it once. Treated carefully, it is barely adequate.
• Sencha Special Grade Yamato (★★★☆☆/★★★★☆)

A superior grade of Sencha, with a brighter flavor and smoother finish than basic Sencha. Highly recommended.

Difficult for me to distinguish this sencha from the 88th Night Shincha/Tamaryokucha/Fukamushi, but I think it is somewhat milder, and I am not as impressed.
• Pi Lo Chun Imperial (★☆☆☆☆)

Also known as Green Snail Spring, this exemplary China tea has a sweet, delicate aroma with mineral/seagrass notes. The flavor is very smooth with a buttery mouth feel. A light vegetal nuance and hint of melon complement a clean finish.

Tasteless.
• Bancha First Grade Organic (★★★☆☆)

First grade Bancha, produced during peak season, is a tea with herbaceous aroma, smooth character, and pale jade liquor. This is a tea with a full mouth feel, bright cup, and clean finish.

• Foucha Imperial Organic (★★★☆☆)

This outstanding selection is also known as Buddha’s Tea, a reference to its former cultivation at Buddhist monasteries. The sweet cup has a sublimely rich flavor, with a buttery mouth feel and pleasing finish. A connoisseur’s selection.

Both the Bancha and Foucha were better than the disappointment of the Pi Lo Chun Imperial, but neither lived up to the advertising.

• Pan Long Yin Hao (★★★☆☆)

Originally produced in the early 1980s, Pan Long Yin Hao (Curled Dragon Silver Tips) has won numerous awards in competitions held by China’s Ministry of Agriculture. This fine plucking is processed into a tighter roll than is typical for this style of tea. The cup is smooth and flavorful with predominantly sweet vegetal notes. The liquor has a round, brothy quality and pleasant finish.

• China Green Gyokuro Organic (★★★☆☆/★★★★☆)

A fine, Japanese style green tea, produced with traditional shade-grown techniques. The liquor is delicate and possesses a natural sweetness.

An attempt to clone the Japanese gyokuro plants & processing; it sort of succeeds in that you can tell from more than just the coloring/leaf-shape that it’s intended to be gyokuro, but the flavor shows there’s still work to be done.
• Japanese Sencha (★★★☆☆)

Delicate but brisk, this splendid green tea has a clean vegetal flavor. Rapidly gaining the recognition it deserves, this tea refreshes the palate with a hint of sweetness.

Delicate here again turns out to mean neutral and not particularly strongly flavored.
• Fujian Green Needle (★★★☆☆)

This tea is produced from a fine plucking, with the leaf sets expertly crafted so that the first and second leaf envelop the downy white bud. The mild liquor has an enticing aroma, with a light sweetness and refined character. A very gentle pungency refreshes the palate.

Classified as a green by Upton’s, I would call it a white; the softness and fuzziness of the blades are common among whites. The description is accurate enough, but being a white, I don’t like it.
• This unique selection is created using nitrogen during the production process. The resulting leaf is rich in the substance GABA (Gamma-Aminobutyric Acid), which some purport to have salubrious properties. We like this tea because it has a rich taste with intense buttery notes in the aroma and flavor, and oceanic hints in the liquor. A truly satisfying cup.

The GABA effects aside, this was unremarkable green tea.
• Gunpowder Green Peppermint (★★★☆☆)

Gunpowder green tea from China, scented with peppermint oil. (Formerly item TF45)

Peppermint oil is not overwhelmingly strong, but it does mask most of the green tea flavor. I would rather just drink the spearmint.
• Kyo Bancha Organic (★★★☆☆)

This small production, Uji region Bancha tea is entirely handmade, from plucking to roasting. The bold leaves are produced from larger, older leaves and naturally contain much less caffeine than an average green tea. The liquor has a full yet smooth character with a pleasant toasty note in the flavor. This offering is a fairly rare treat for the Japanese tea enthusiast.

Unusually for Upton’s, the largest quantity carried is 30g. The leaves themselves are even more unusual: they are large, on par in size with cooking bay leaves but dark and almost shiny, reminding me of oak & maple leaves which have decomposed over a long winter. The Upton’s description mentions roasting the bancha, so this is a ho-ji cha. The other ho-ji chas left more of an impression. Select Dragon Well Lung Ching 4
• Crafted from carefully selected leaves, this tea has a distinctively sweet, light-toasty flavor with hints of chestnut, and a pleasing vegetal finish. Highly recommended.

Like gyokuro without the bite.
• Yunnan Green Mao Feng (★★★☆☆)

The production of this classic tea begins with a fine plucking of select buds. The superb cup has a pale liquor with a light vegetal quality and a buttery mouth feel. The initial flavor notes, with hints of peach, are followed by a delicate menthol nuance. The finish is clean and refreshing.

Brothy sweet undertone, with definite peach.
• This tea is produced on Jiu Hua Mountain, one of the four sacred Buddhist peaks in China. Crafted from a classic two leaves and a bud fine plucking, this leaf produces a delightful cup with a surprisingly full mouth feel and sweet floral nuances. The finish has a light, cleansing astringency. 3/4

• Lung Jing Te Ji/China Green Tea Top Lung Ching (★★★☆☆/★★★★☆)

This Dragon Well is made from a fine plucking, which is then processed in a traditional manner and fired in charcoal pans for finishing. The infusion has a light jade-yellow color. The aroma of the liquor is delightful, and portends the sweet corn nuances and chestnut hints teased from the cup.

• Season’s Pick Young Hyson (★★★☆☆)

The aroma of this selection is lightly floral. The mild cup has sweet tones of honey and a light smoky undertone. The finish has a cranberry note and gentle tartness.

Bonus sample; thick chunky leaves. Gunpowder green-like, bit floral, slight bitter, oversteeps quickly. Might still be worth it since 800g is only $26. • The cup has an aroma of sweet peas and sea grass, with delicate floral notes. The first sip reveals a mild, sweet, and buttery smooth flavor. Light vegetal notes of sweet peas and corn silk are complemented by a savory undertone in the clean finish. Wiry; sweet and gyokuro-like. I could see myself drinking this as a cheaper green. (Not as cheap as the Young Hyson though - 800g costs$51.)
• A quality selection with a bright and full-flavored cup, accented by sweet nuances and subtle nutty overtones. The aroma and liquor have a fruity aspect which follows in the finish.

Wiry; similar to the Mao Jian Wu Lu in being semi-gyokuro like but much less so.
• An organic silver-tip green tea was scented with fresh, organic jasmine blossoms, creating this delightful tea. The cup aroma is candy sweet and redolent with the sublime scent of jasmine. The liquor is smooth and well balanced, with a pleasing level of jasmine flavor that lingers into the finish.

Heavily jasmined, hard to distinguish anything beyond that; oversteeps fairly quickly.
• Sencha Yabukita Organic #2 (★★★★☆)

This selection from Kirishima, in Kagoshima prefecture, is crafted from the Yabukita cultivar and usually reserved for the domestic market. The mouth feel has a decidedly brothy character, and the flavor has a classic umami aspect. A superb specimen of this style and grade.

• This selection was created using time-honored, hand-processing techniques that go back centuries. The cup has a soft and buttery complexion with a range of complex flavor notes. The liquor is extremely well-balanced and finishes on a clean, sweet note. The twenty-gram packets are factory packed and nitrogen flushed. Samples are repackaged at our facility. Produced in Shizuoka prefecture from the Saemidori cultivar.

Very fine sticky slices of dark green which cling to plastic surfaces (static electricity?) and expand surprisingly much when steeped. Simultaneously sweet and savory with a strange seafood-like aftertaste.
• Pre-Chingming Green Gu Zhang Mao Jian (★★☆☆☆/★★★☆☆)

This 2016 Pre-Chingming offering has wiry, well-twisted leaves with a profusion of delicate, downy buds. The pale jade liquor is fragrant with light floral hints. The sweet cup is smooth and buttery, with a brothy mouth feel and clean finish.

• The flat, spring green leaves of this 2016 Pre-Chingming selection produce a yellow jade liquor with a fresh vegetal aroma. The light cup is smooth and savory with hints of melon and chestnut. A crisp tang lingers in the finish.

• The dark green leaf of this 2015 offering is twisted and curled. The cup has a vegetal aroma with hints of mint and sweet flowers. The liquor has a light vegetal flavor with a buttery mouth feel and hints of spice.

• Pre-Chingming Hunan Hairpoint (★★☆☆☆/★★★☆☆)

The attractive, dark green leaves are well-twisted and interspersed with many silvery-white tips. The dry leaf aroma has a smooth, fresh hay quality, with a hint of what some have likened to raw dough. The infused leaf has vegetal notes with a very subtle smoky suggestion. The light liquor has multiple layers of flavor. This 2015 selection is from Hunan province.

• Capital Teas, Gyokuro (★★★☆☆/★★★★☆)

Japan’s finest green tea, made from single buds picked only in April or May; full-flavored with umami characteristics.

For gyokuro, this is excellently priced, and while I did not have any of Upton’s more expensive regular gyokuro at hand to compare side by side, I think the quality was only somewhat less. Sadly, hardly halfway through the tin, I realized I had made a major mistake in not transferring the bulk of it into a sealed container & using a smaller amount in a small container to protect against the extremely humid section of my kitchen where I store my teas, as I noticed that a mold had started to grow in one corner and the bottom half had compacted into a single mass & discolored. (This is similar to what happened with my Tao of Tea Handrolled Jasmine Pearls Green Tea balls.) I had to throw out the rest. I immediately transferred the Capital Tea Himalayan Golden Monkey, which was in a similar tin, into separate containers with some desiccant bags - but tragically it was too late for the gyokuro.
• Clipper Ship Tea Company: Imperial Green (★★★☆☆)
• Pi Lo Chun (★★☆☆☆/★★★☆☆)

This Pi Lo Chun selection has beautifully crafted leaves with silver downy buds. The pale gold infusion has a light floral fragrance. The satisfying cup is sweet and smooth with fruity notes and gentle floral hints.

smoky aftertaste?
• China Green Sencha (★★★☆☆)

A good green tea for everyday consumption. In the style of Japanese Sencha, this China tea represents an affordable alternative to the more costly Japanese varieties.

• Organic Green Ceylon OP (★★☆☆☆)

This green tea offering has large, well-twisted leaves that produce a light golden liquor with a delicate floral aroma. The smooth cup is well-balanced with a full, buttery mouth feel and gentle pungency. The flavor has sweet tropical fruit notes and light citrus hints. A suggestion of spice may be found in the clean, lingering finish. bleh

• Organic Gyokuro (★★★★☆/★★★★★)

Limited lots of organic Gyokuro tea are produced each year. This lot has smooth umami notes, a rich cup and a clean, sweet finish. It was grown in Miyazaki prefecture using the Yabukita cultivar. We are happy to offer this tea while our current supply lasts.

Tastes as good as the other gyokuro.
• Formosa Pouchong (★★★☆☆/★★★★☆)

The large, twisted leaves of this Pouchong selection yield a pale golden cup with a light floral essence in both the aroma and the flavor. A sweet honey note enhances the mild vegetal quality of the liquor. The finish is clean and refreshing.

• Pre-Chingming Pi Lo Chun (★★★☆☆)

One of China’s most well-known green teas, Pi Lo Chun, or Green Snail Spring, is named for its unique appearance. This 2016 Pre-Chingming selection has a mix of olive green leaves and silver tips rolled into the classic spiral shapes. The pale yellow-jade infusion is smooth and buttery with a vegetal aroma, hinting of fresh corn. Some have noted a hint of spice, suggestive of anise, in the gently pungent finish.

• Vietnam Green Mao Feng Organic (★★☆☆☆/★★★★☆)

Reminiscent of a Yunnan green tea, this organic offering from Vietnam has bold, olive-green leaves with a crepe-y appearance. A fruity note is present in both the aroma and the cup. The bright golden liquor has a note of sweet uncured tobacco with peach hints in the finish.

(I would describe the sweet uncured tobacco as more of a smoky-puerh-like flavor.)
• This 2016 Pre-Chingming offering is composed of bright-green, well-twisted leaves with a sprinkling of downy silver tips. A lovely floral note is present in both the aroma and the velvety smooth cup. The light jade liquor is flavorful and sweet with a full, buttery mouth feel and clean aftertaste.

• Korea Green Tea Sejak Organic (★★★☆☆/★★★★☆)

The dark emerald green leaves of this offering produce a light jade liquor with a lively, vegetal aroma and a clean, savory quality. The cup has a slightly brothy mouth feel with sweet, buttery hints and a crisp finish. This tea has been expertly handcrafted using time honored methods and is good for multiple infusions.

Remarkably sencha-like.
• The Tao of Tea, Sencha Green Tea, Loose Leaf (★★★★☆)
• The Tao of Tea, Dragonwell Green Tea, Loose Leaf (★★★☆☆)
• The Tao of Tea, Tea Forest Green Tea, Loose Leaf (★★★☆☆)
• Weishan Mao Feng (★★★☆☆)

During our most recent trip to China, we journeyed to the Weishan area. There, in back of the Buddhist temple, is an organic garden where we found a spring tea. It turned out to be a delightful discovery. Light in the cup, Weishan Mao Feng features just the right roast flavors. Details: Weishan is well known inside Hunan Province for its Buddhist temple honoring Guan Yin, and like many such centers there was tea made around the temple. Originally, it was probably monks that made the tea and then passed their knowledge down the local peasants. Most of the teas are very rustic with much use of charcoal firing that makes for a very smoky (and unpleasant for most Westerners) flavor. In fact they make a blend of tough leaves and dried vegetables that is made into a soup for the hard working local peasants. Although this region is not well known outside of Hunan, we like this example of the lighter, more elegant teas that are starting to be made. Dry Leaves: Mao Feng means downy tip in Chinese, meaning the bud is young and covered with white down like a bird. These are the tricomes or small outgrowths that protect the very young tea tip (also called a bud) from insects and hostile weather conditions. As the leaves matures, the tricomes disappear. So when one sees the downy tricomes it is a sign of a very young tip with all the desired qualities of sweetness and elevated levels of antioxidants. These leaves are definitely small with a silvery tips. Liquor: The color of the this tea is very light green with a slight khaki tinge that comes from the final firing over charcoal. Aroma: Light with vegetal roasts. This is the result of a light firing over charcoal in the traditional manner. Caffeine Level: Caffeinated Body: As mentioned, the leaves are very small and there are many tips, so the body is very light. Flavors: The mixture of light vegetal flavors of steamed green beans or artichokes is accented by the definite roast flavors from the final charcoal firing.

### Ku-ki

An obscure niche of tea is kukicha/ku-ki/kuki-cha/kuki hojicha/twig/stick tea: a de-mono tea made from, as it sounds, the byproduct stems of the tea leaves proper. Despite having few or no tea leaves, this near-oxymoron turns out to yield a tasty tea. (And since leaves secrete caffeine as an insecticide while tougher stems don’t need as much protection, ku-ki will usually be low in caffeine.) They seem to usually be considered a kind of green tea but that might be due to the connection to sencha/gyokuro/matcha manufacturing since some come roasted and overall ku-is reminds me more of oolongs than greens. I discovered ku-kis entirely serendipitously, which is too bad; their obscurity means many people who might like them will probably never hear of or give them a try.

• Choice Organic Tea’s Twig Ku-ki cha (★★★☆☆ / ★★★★☆)

This was a random try of a tea bag, and I was a little dubious - twig kukicha doesn’t sound very promising, and twiggy is usually a bad adjective coming from me. But the first steep turned out to be fairly good, as did the second steep. The Wikipedia description of it as mildly nutty and slightly sweet turns out to be on the money; it also reminded me of genmai-cha. There was only one tea bag, so my first impression will remain limited, but I think I will try some kukichas in the future. (Upton’s stocks 3 Japanese kukichas and 1 Chinese.)
• Organic China Ku-ki Cha (★★★★☆)

To my sorrow, this was the only ku-ki tea Upton’s had in stock when I ordered this batch, and not the one I was most interested in (the roasted ku-ki cha). This may be a continuing effect from the Fukushima incident which cut off many rarer Japanese teas.

Regardless, I like it. It has a sort of hybrid green-oolong taste, but with a nutty or roasted-barley overtone. (The only downside was that I drink my teas without a strainer or tea ball, and the stems & twigs all float!) This suggests that the one packet I tried before was not an aberration; if Upton’s doesn’t have any more when next I order, I’ll probably look for another retailer which does have some.
• Japanese Ku-Ki Ho-Ji Cha (Roasted Ku-Ki Cha) (★★★☆☆ / ★★★★☆)

This carefully roasted Ku-Ki Cha (twig tea) produces a golden brown cup with a sweet vegetal flavor. A mild tea that is naturally low in caffeine. The distinct toasty notes linger pleasingly on the palate. Not to be confused with regular Ho-Ji Cha.

Fortunately, they did have two available when I ordered in June 2015 (the two Japanese ones, but not the Chinese). The roasted ku-ki cha is all-twig and dark brown; it tastes strong and like heavily-toasted genmai-cha, nutty and bordering on smoky. (Reminds me of the Choice Organic Tea one.) An interesting flavor worth trying. - Organic Ku-Ki Cha Green Kamakura (★★★★☆)

> This is an organic green tea, produced from Camellia sinensis twigs. The mellow cup has delicate, fresh hay flavor notes. Naturally low in caffeine.

Not pure (unroasted green) twigs, but perhaps one-thirds green-tea leaves. A sweet grassy flavor combined with the twiggy flavor which I liked.
• Browsing Whole Foods for the first time in years aside from noting that the clientele is surprisingly tall compared to at Walmart and that many New Age fads like homeopathy and probiotics have yet to run their course, I lamented that their selections remained entirely teabag-oriented (with the honorable exception of some surprisingly reasonably priced matcha powder) when I spotted a most unexpected box of kukicha teabags. Having exhausted Upton’s selection, I was curious, and it’s always good to support the more obscurer kinds when they show up in a mainstream retailer so I gave it a try.

Eden takes a similar tack as Choice Organic Tea’s teabags: it’s a mild flavor, neither bitter nor sweet, with a sort of nutty or coffee vibe. Nothing special but a good my-first-kukicha..
• Hojicha Kurenai (★★★☆☆ / ★★★★☆)

Our Hojicha Kurenai is exclusively made from tea leaf stems. The stems provide a longer lasting flavor and more infusions. Enjoy a mild roasted flavor with a deep and pleasant aroma. Yields a deep red colored tea. This tea is grown exclusively in Kyoto, Japan.

In early November 2015, the exchange rate for Bitcoin suddenly increased steadily to the point where it had gained 50% and would ultimately almost double; I hold more bitcoins than I want to at this point due to laziness about cashing out (I haven’t sold too much in part because I’m worried about going past some threshold and then having a headache of taxes to deal with), and I couldn’t see any good reason for the exchange rate to increase so much, so I decided to cash out some in the form of consumption. I prepaid some hosting costs, donated $540 to GiveDirectly, another$50 to the Brain Preservation Foundation, experimented with buying some household goods I needed off Amazon via Purse.io (which worked more smoothly than I had expected aside from their website not working in my Iceweasel web browser), and finally thought - I wonder if anyone sells good loose tea for bitcoins? Turned out at least 4 did: Tealet, New Mexico Tea Company, Beautiful Taiwan Tea, & Teanobi. Tealet struck me as extremely expensive (1g/$!) and I didn’t like the focus on blacker teas since that end of the spectrum has not worked out well for me in the past even when darker teas have been advertised as oolongs; New Mexico Tea Company looked OK, and I was thinking of trying the plum & orange blossom oolongs but I didn’t see anything else that grabbed me and I try to order at least 3 teas at a time to amortize S&H; Beautiful Taiwan Tea struck me as sketchy, somehow, so I moved on; finally, at Teanobi, which has an interesting emphasis on green teas as cooking ingredients & coffee blends, I found some of my old favorites and some new ones to try, so though fairly expensive (5g/$) I ordered 5 from them at ฿0.208 (a hojicha, 2 genmai-chas, a Uji Shibano, and a Tamaryokucha Koga), then $95, and a week later (after the teas had arrived), worth$70 (so by quickly buying, I got in effect a 25% discount). They all turned out to be good, although tasting so similar to the previous Upton’s that I find myself wondering if Teanobi & Upton’s are sourcing to the same tea farms (presumably there aren’t that many hojicha tea makers in Japan). I also wish all 5 had come in resealable foil packets; 2 did, but the other 3 are not resealable which is always a nuisance.

The Hojicha is, well, similar to the Upton’s Japanese Ku-Ki Ho-Ji Cha. The only difference is that it tastes somewhat sharper and more coffee-like.
• Heirloom Tea Flowers Organic (★★★☆☆ / ★★★★☆)

An unusual tisane made solely from organic Camellia sinensis flowers, which are carefully plucked from heirloom tea bushes in full bloom, and then sun-dried. Tests show the flowers to contain similar amounts of catechins and polyphenols as regular leaf tea, but with a significantly lower caffeine content. The flavor is surprisingly full, with notes of honey, caramel and citrus. This is a new lot of this very unique offering.

Visually resembles cut-up dry honeycomb. The use of the flowers is as strange as kukicha’s use of twigs & stems, and I had to try such an uncategorizable. The caffeine-free aspect is also welcome for expanding the minimal ranks of tea products without caffeine. Even more surprising is the flavor: it tastes like osmanthus oolong! I had expected it to taste like nothing at all, or perhaps a white tea. But no, it tastes good. The downside is that it is expensive: eg 200g for $48, compared with the osmanthus oolong’s 250g for$15 (so the tea flowers are 1/4th the g/$). • Organic China Green Sencha (★★★☆☆) In the style of Japanese Sencha, this China tea represents an affordable alternative to the more costly Japanese varieties. The pleasing cup is sweet and vegetal with a citrus-like brightness. This offering is an excellent choice for everyday consumption. Another mediocre Chinese clone. Stick with the Japanese senchas. • Food Lion: Nature’s Place Organic 100% Green Tea (★★☆☆☆) Commercial tea bag; apparently Food Lion’s generic in-store brand. Steep quickly and easily oversteep, but despite being careful, still not decent tea at all. And it’s disquieting when a tea makes a boast like 100%. • Uncle Lee’s Tea: green tea (★★☆☆☆/★★★☆☆) A better tea bag in comparison to the Food Lion. For a tea bag, maybe not that bad? ## White My general take on white tea is that they seem to be rather fragile and I generally prefer stronger flavors from green/oolongs. (Subtle flavors can be good, but for white teas, it seems that their subtlety usually comes across as weak or tasteless.) It’s possible I’ve either not had really good white tea or I’ve ruined the ones I had. • Special Grade Shou Mei (★★☆☆☆) Fairly twiggy (little in the way of leaves proper). Very white - tasted like a weak green with a certain floral overtone. In its favor, it handled re-steeping very well, not becoming bitter even slightly & tasting the same over multiple cups. • Organic Pai Mu Tan (★★★☆☆) The Pai Mu Tan tasted like the Shou Mei or Yin Zhen Bai Hao, but much more so, and so gets more approval from me; probably won’t buy it again, though. (I don’t actually dislike the general white tea flavor, it’s just usually far too weak to be worth drinking.) • As promised, the pekoe is indeed downy - the leaves & branches are downright fuzzy. However, it tastes almost identical to the Shou Mei. • Peach Momotaro (★★☆☆☆) A gift from the littler sister. I was amused at the clever title - an allusion to the Japanese folktale Momotarō (literally Peach Tarō or Peach Boy). I didn’t have much hope for this flowering tea, but it improved on my expectations: the bloomed tea ball was a lovely white stalk on a grassy green base, and the peach flavor was respectable and comparable to the other peach tea I have. Flavor-wise, the tea was pretty weak (I was under the impression it was either a green or oolong tea) and overpowered by the peach, but at least it had a flavor and so was better than the previous flowering teas. It improved a little bit by the 10 minute mark, having sweetened a little. The weak tea flavor was explained when I learned it was a white tea; such a flavor is pretty par for the course for whites. • Tranquil Tuesday: White Peony White Tea (★★★☆☆) • Pre-Chingming Ya Bao Special (★★★☆☆) Ya Bao is a special production Yunnan tea made from tender young buds that are hand plucked from ancient tea trees. The liquor has a unique flavor, with floral notes and hints of sweet corn and light honey. The aftertaste has a wonderful, light evergreen note, which lingers on the palate. A 2015 production. Not your usual white tea, this comes in the strikingly different-looking form of large slightly-moist greenish-white corn/pine-cone-shaped buds. There is a sweet aroma with a mysterious edge to it, which is even stronger when brewed. To my regret, the flavor did not entirely agree with me; it is interesting but not tasty to me. Perhaps it might work for white tea fans? • Clipper Ship Tea Company: White Peony/Pai Mu Tan (★★☆☆☆) Small NY retailer of tea. Resealable foil packets, 25g each, no prices or descriptions available online. Received a sampler of 5 teas covering the full range from Pu’erh to white. The teas themselves are average to good quality exemplars of their types, so my ratings are about the same as for other instances. • From Jinggu County in Yunnan province, these exquisite downy white tea buds yield a pale straw-colored infusion, notable for a nutty/toasty aroma with hints of honey. A sweet hay nuance complements the savory cup, which finishes with a gentle spice hint. I tried one last batch of 6 white teas from Upton’s. With the exception of the Jinggu Spring Buds, which was decent, they all tasted near identical to me and a waste of time. I am giving up on white teas - like the pu’erhs, the entire category just doesn’t work for me. • China White Tea Silver Needle (★★☆☆☆/★★★☆☆) In this elegant white tea selection, beautiful silver-sage buds yield a rose-gold infusion with a sweet, gentle character. The cup has a fresh herbaceous aroma with a suggestion of honey. Delicate notes of honey and melon as well as orange nuances are notable in the silky smooth liquor. • Pre-Chingming White Needle (★★☆☆☆/★★★☆☆) In this 2016 Pre-Chingming white tea selection from Fujian province, silvery sage-green buds yield a delicate cup with a light honeyed floral aroma. The flavor is sweet and refined with hints of melon and a gentle pine essence. A toasty nuance lingers in the smooth finish. • Shai Zhen Zhu Shou Mei China White Pearls (★★☆☆☆/★★★☆☆) The bold, olive-green leaves of this Shou Mei white tea are carefully crafted into large pearl shapes. A toasty/woody suggestion may be found in both the aroma and the sweet cup. The champagne-gold infusion is delicate and velvety smooth with notes of fruit, a honeyed sweetness and melon nuances. Each pearl weighs approximately 5 grams, which will yield about 2 cups of tea. • Downy silver tea buds are carefully crafted into large pearl shapes, which produce a pale straw-colored infusion with a delicate, honeyed fragrance. The sweet cup has a clean character, with hints of melon and flowers. Each pearl weighs approximately 5 grams, which will yield about 2 cups of tea. • Qing Zi You Yun (Lan Xiang)/China White Tea Cake (★★☆☆☆/★★★☆☆) Large leaves are molded into a segmented rectangular tea cake to create this unique white tea selection. The pale rose-gold liquor has a delicate, toasty aroma with honeyed floral hints. The cup is smooth and very sweet with melon notes, a suggestion of peach, and a crisp, clean finish. Each individual segment weighs 5-6 grams and will yield approximately 2 cups of tea. ## Black I am not a fan of black teas, but I still try them out occasionally: • Ginger Peach Tea, bag-tea by English Tea Shop (★★☆☆☆) It is a black tea mixed with ginger pieces and peach flavor. To my surprise, it was fairly good. The black tea is a pretty weak black and as far as I can tell, towards the oolong end of the spectrum. The peach flavor is entirely dominant over the ginger, which is as I would prefer, peach being an old favorite of mine. The first steep is good, but it falls off very quickly and needs replacing by the fourth steep or so. • Satori Tea Company’s Amali African Queen (★★★☆☆) Another gift; this one confused me because it was clearly labeled oolong, but when I tried it out, it tasted very much like a black tea and the leaves were pretty oxidized and produced a black-tea-looking liquor (extremely dark as opposed to amber), and quickly began thinking of Earl Grey. My confusion was resolved when I began to look up the teas and found that the African Queen was in fact a black tea (as opposed to a peculiarly black oolong). • Upton’s Traditional Masala Chai (★★☆☆☆) A traditional Indian spiced tea recipe with a warm and robust character. The full flavor notes of cinnamon, cardamom, ginger and clove hold up to milk and honey, the traditional way to take this tea. Christmas gift. Tastes like the description (which is really too much), and does indeed taste better with some milk & honey added. • Yunnan Fop Select (★★★☆☆) The dark brown leaf of this selection is accented with golden tip. Rich, earthy notes are present in the aroma and dark amber cup, and hints of white pepper and cassia add a pleasing accent. The smooth finish has a light tangy feel. A free sampler of a Chinese black tea that Upton’s threw into a 2015 order. Less bitter than most blacks, with the mentioned earthiness balanced by sweetness. I would not have paid for this, but I don’t mind drinking it. • Huang Shan Sunset Tea (★★★☆☆/★★★★☆) Rare China Keemun Mao Feng is scented with peach, and decorated with flower blossoms. Serve this for a truly decadent afternoon tea. A strong black tea like the Yunnan Fop Select, it lives up to its billing: the peach is strong, almost overwhelming, and it is sweeter than usual. There aren’t many peach-flavored selections, which makes this stand out for me. Better than any of the others like the Momotaro. • Peach Tea/Peach with Flowers (★★★★☆) A peach-flavored, bold whole-leaf black tea, decorated with flowers. Bolder leaf version of TF72 Much like the Huang Shan Sunset Tea, but better. • Apricot Tea (★★☆☆☆) An apricot flavored, whole-leaf black tea decorated with flowers and apricot pieces. • Black/Green Midsummer Dream (★★☆☆☆) China Black tea harmonizes with green Sencha. Blended with sunflower petals, cactus flowers and flavored with rhubarb. A refreshing tea for any time of day. I was skeptical that a black tea could ever harmonize with something like sencha - the black flavor is bitter and powerful and will pummel any green into indistinguishability. I was right, and it tastes black. The rhubarb and other additives also lend it a chemically taste, so it fails as a black tea in the same way as the apricot tea did. • Peach With Pieces (★★☆☆☆) A peach flavored whole-leaf black tea base, decorated with peach fruit pieces. Ingredients: black tea, peach bits (peach, sugar), osmanthus blossoms, natural flavor • Capital Teas, Himalayan Golden Monkey (★★★☆☆/★★★★☆) The confluence of crystal-cool zephyrs and fervid monsoons in Nepal bears a tea whose caramel liqueur courts the palate with honey, cocoa, and a muscatel kiss. Visiting a random mall, I stopped in at a tea store I didn’t recognize on my way to the Teavana shop. The salesman was knowledgeable and very persuasive, and when I was disappointed at the oolong selection and noted I didn’t really want any of the blacks, told me that he had a black tea which he had discovered, prepared correctly, was a lot like an oolong. The secret was to steep it for a short time, perhaps 30 seconds, and no more. He gave me a mug of it and yes, it did taste much more like an oolong than an unpleasant black! I was impressed enough I let him sell me a bunch of it and I’ve enjoyed it ever since as long as I remember to keep the steep time short. • Eastern Shore Tea Company: Blood Orange Tea (★★★☆☆) The unusually deep, rich flavor of the blood orange is sublime. Fabulous as an iced tea. Flavored black tea. Contains caffeine. 3 oz. loose tea with re-usable tea bag. I remembered a truly delicious blood orange sorbet I had once in SF, so when I happened to spot, on a shelf of otherwise offputting black teas, a blood orange black tea, I thought I’d give it a try though I was unfamiliar with the seller. The black tea is speckled with the blood orange bits and pieces. Flavor-wise, it is a harmonious balance of orange flavor and black tea, although my usual prejudice against black tea means it’s not as enjoyable as it might be. (An oolong version would be great if balanced properly, but unfortunately Eastern Shore Tea Company appears to specialized in flavored black teas, with only one listed oolong and 4-5 greens.) Resteeps like a champ. • Zaloni Estate Assam STGFOP ((★★☆☆☆)/★★★☆☆) This wonderful selection has a dry leaf with subtle, sweet aroma and rich, dark color. The liquor is also dark, with an amber color and medium-bodied intensity. The aroma has a malty hint and biscuity note. The cup has a very forward sweetness, reminiscent of light molasses or dark honey. Reminds me of Capital Teas’s Himalayan Golden Monkey. Very mild, sweet - bland. • Clipper Ship Tea Company: Nilgiri Black (★★☆☆☆) • Hunan Bloolong Tea (★★★☆☆) On our recent trip to China, we tasted this unusual tea and loved it. Our friends at Hunan Tea have made a black tea from a plant that normally would make oolong tea. It is much more aromatic than most black teas, and the peach flavors are lovely. It also coats your mouth with those luscious peach flavors. When we were naming it, we could not resist calling it Bloolong, i.e. Black Oolong. Details: A black tea made from an oolong varietal tea plant. Dry Leaves: Dark, chunky leaves. Liquor: Brown. Aroma: Boiled peaches, vanilla beans. Caffeine Level: Caffeinated. Body: Medium. Flavors: Cooked peaches. ## Pu-erh I have tried pu’erh tea from time to time, but without exception of brand or preparation method, I have not liked them at all. • Clipper Ship Tea Company: Pu’er (★☆☆☆☆) • This sun-dried, compressed green tea is made from the fine buds of Yunnan’s heirloom tea trees. The flavor is vegetal, with notes of wild honey and dried fruit. The sweet and complex aftertaste lingers on the palate. Appropriate for multiple infusions. A green pu-erh would seem to be a contradiction in terms: the point of green tea is that it is processed shortly after harvesting to stop oxidation within days, while pu-erh is both heavily oxidized and further fermented by bacteria & fungus for years. Nevertheless, Upton’s classifies this as a pu-erh so I was intrigued and made an exception to my usual ban. The sample comes as one tiny greenish-yellow cake, which crumbles reasonably easily. It yields something on the boundary between greens and oolongs, which resteeps well but doesn’t leave much of an impression other than tasting rather un-pu-erh-like. # Tisane Tisanes are any tea which does not incorporate Camellia sinensis - so this category includes herbals like mint tea, barley tea or red tea (rooibos) or honeybush. (I once ordered rooibos & honeybush from Upton’s for my mother; I found them so unmemorable I can’t even review them here.) • Roasted barley tea (★★☆☆☆) Like the Benshan oolong, bought from Rainbow Grocery Cooperative. I was initially going to only buy some genmaicha but then I saw their oolongs, so I went with plain roasted barley instead and combined it. The barley was very… nutty and barley-ish on its own. Not entirely drinkable, I thought, although it added some strength and robustness to the Benshan oolong in small amounts. • Ginger herbal tea (★☆☆☆☆/★★☆☆☆) This Royal King product was, as it promised, gingery. I’d have to say I don’t actually like the flavor of ginger that much, and couldn’t drink it very often. • Rote Grütze (★☆☆☆☆): disgustingly sweet and fruity (accented with dried blackcurrants, blueberries, strawberries and wild cherries is an understatement). The best I can liken it to is drinking one of those potpourri or stuffed pomegranates old women buy. It initially seemed to re-steep well but I realized it was somehow ineffably becoming more and more offputting with each steep. I can’t see it really motivates me to try any more kinds of rooibos. • Superior Organic (★★☆☆☆): much better than the Rote Grütze, with just the right amount of sweetness. • Honeybush: honeybush vanilla (★★☆☆☆) reminded me a little of rooibos (though different species entirely), but much toned down, sweet like its name suggests, and the vanilla combined nicely. I actually liked it a little. Good for occasional breaks or when I want something hot to drink but caffeine would be a bad idea (eg. past 7 PM). • Maracuja/Orange Fruit Tea (★★★☆☆) Contains fruit pieces, rose hips, hibiscus flowers, citrus peel and flavoring. Ingredients: apple bits, hibiscus, rose hip peels, beetroot bits, orange bits, citrus peels, artificial flavor A strongly-flavored, tangerine-like herbal tea. • Cape Cod Cranberry Fruit Tea (★★★☆☆) A special blend of dried cranberries, hibiscus and apple bits. Caffeine free and delicious. Ingredients: apple bits, hibiscus, cranberries (cranberries, sugar, sunflower oil), artificial flavor Despite the differing ingredient list, tastes very similar to the Maracuja. The cranberries add their own kick to the orange-like flavor. I noticed I could eat it straight out of the bag like it were trail-mix. • Lemon Myrtle (★★☆☆☆) Grown in the sub-tropical rainforests of Queensland, Australia, Lemon Myrtle (Backhousia citriodora) is a relatively new caffeine-free tisane. It is a natural source of citral essential oils, antioxidants which imparts a stout lemony aroma and flavor. Overwhelmingly sweet and lemon-tasting; lemon, lemon, and more lemon. A little goes a long way. I ultimately found it too much of a muchness, and couldn’t finish it. • Mulberry tea / Kuwa-cha: apparently mulberry tree leaves make a decent sencha-like powdered tea; an acquaintance described it as being like a good green tea. As it happens, I have a longstanding fondness for mulberries. Unfortunately, though mulberry trees are not rare trees (there were several growing wild within blocks of where I grew up), the prices are not as cheap as one would hope; apparently there’s a fad diet cluttering listings & driving up prices. Anyway, out of curiosity I ordered 45g of$12 mulberry tea (★★☆☆☆/★★★☆☆) from Kesennuma (grown in Miyagi Prefecture).

It is a nice slightly-dark green, shredded finely like confetti, and reminds me a bit of how sencha green tea looks; the smell is faint and the best my impoverished scent vocabulary can come up with is a bit musty. Steeped, the water is also a nice green; my first impression of the taste is that it’s slightly sweet. Beyond that… it tastes perhaps like a white or green would if one removed all hint of bitterness and grass ie. there’s not much of a flavor beyond the slight sweetness. The directions suggest preparing with hot water, but the mulberry tea tastes much the same prepared with cold or cool water and the flavor is easier to taste without heat in the way. (You can make iced tea with it, but I don’t advise brewing for more than a week - it seems to gain an offputting aftertaste after that.)

Do I like it? Well, I don’t dislike it but a very inoffensive green tea isn’t something I have a pressing need for. I think it would make a decent summer tea since you could prepare & serve it cold, but I like barley tea and genmai-cha better, so I don’t need a mulberry tea. Still, interesting to try out - who knew mulberry leaves could be used to make an OK tea?
• Daesang organic barley tea ($4.13, 100g in 15 packets; available on Amazon as$9.99, 300g) (★★★☆☆); advertising copy from the 30-bag Amazon offering inasmuch as I can’t translate the Korean on the bag itself:

100% organic barley tea comes in 30 unbleached teabags. Healthy, naturally caffeine-free, sugar-free, and deliciously refreshing. A caffeine-free coffee substitute drink. Zero calories. Serve year-round, hot in winter and cold in summer. Each tea bag makes 2 liters of barley tea.

I picked this up while at a Korean grocery store hunting for kimchi; I couldn’t buy too much since I was leaving for a long trip shortly thereafter, and I didn’t want to leave much food behind, so I picked up some other things while browsing the tea section. I was intrigued by the offering of a corn tisane but I chickened out because it was too big and I didn’t want to be stuck with a lot of undrinkable tisanes, so I went instead with a barley, to see if perhaps it improved over the other barley and was something I’d like.

Each teabag is quite considerable since it’s intended for making large batches, and are overkill for a single mug of hot tea (but you can just resteep it several times). As a hot tea, my reaction on the first sip was - sweet! It claims no additives or other ingredients of any kind, and 0g of sugar, but I have to wonder, since it’s really sweet compared to the previous one, offputtingly so. I did not like that.
• This fine-cut leaf grade of organic Holy Basil, (aka Tulsi), produces a rich cup with a complex and spicy character. The dominant flavor notes are anise and pepper, with nuances of citrus and cinnamon.

Holy Basil is an Indian herb similar to the more familiar sweet basil (which I grow to put on my tomatoes and in my BLTs). The taste is immediately familiar from sweet basil, but with a more peppermint and peppery, almost licorice-sort of flavor. It’s meh.
• Holy Basil Green Leaf Organic (★☆☆☆☆/★★☆☆☆)

Our green whole-leaf grade of Holy Basil (aka Tulsi) is a mix of leaf and tiny dried flowers. The smooth, full-bodied infusion is sweet and spicy. Hints of anise and clove are present in the aroma as well as the flavor.

A coarser and fuzzier blend then the Purple Leaf, the Green Leaf tastes much the same but somewhat sweeter and much less flavorful.
• A pleasant beverage that we recommend for its sweet flavor. Made from the inner bark of the lapacho tree, this herbal contains no caffeine.

A coarse cinnamon/tan-yellow bark from Brazil, my sample was lighter in color than the Upton photo & like that of the WP article, which mentions some fun rat research into lapacho’s toxicity. Something about the musty (but admittedly sweet) flavor deeply put me off and I threw it away.
• Organic Spearmint (★★★★☆)

A new lot of our organic, coarse cut spearmint (formerly BH43). Great for blending with green tea or steeping alone as a refreshing, caffeine-free beverage.

An instant favorite of mine. One of my friends swore off all mint products like mint ice-cream or Girl Scouts thin mints or York peppermint patties, arguing that they tasted like toothpaste; I thought this was too bad, and didn’t quite get his argument (wouldn’t it be more accurate to say that toothpaste tastes like mint?). Still, with flavors like mint and cinnamon, there is always a danger of going too far. The spearmint tea does not go too far and is just right, with enough subtlety to the whole flavor to be drinkable on its own. The caffeine-free aspect is also a bonus since it lets me drink it past 7PM when I avoid anything caffeinated. I am more used to peppermint than spearmint, though, so next I tried Upton’s peppermint offering to see how it compares; the difference wasn’t notable.
• Ethereal Cocoa Tea Canister (★★☆☆☆)

Unique steeped-cacao beverage. Contains organic ingredients, not USDA certified… Rare heirloom cacao nibs & shells are used to produce this delicious and delicate chocolate tea. Makes 16 servings.

Caffeine & sugar-free. The cardboard cylinder contains cocoa nibs/shells, and smells like a box of cocoa powder. Steeped at both green and black temperatures, it tastes like it smells: like a watered down weak hot chocolate. Pointless.
• Golden Chrysanthemum (★★☆☆☆)

Beautiful chrysanthemum flowers have long been part of Chinese culture. When dried flowers are brewed as a tisane, they create a bold, sweet and tangy cup. Dry Leaves: Full, dried chrysanthemums. Liquor: Bright green. Aroma: Menthol, liquorice. Caffeine Level: Caffeine Free. Body: Medium. Flavors: Sweet liquorice.

• Guayusa (★★☆☆☆)

Guayusa is a traditional Ecuadorian beverage from the Amazon. Like Mate, it contains elevated levels of caffeine and many antioxidants. It has a mild vegetable flavor. Only recently has it been introduced to the States. Details: At an industry event we met the earnest young people that import this traditional herbal from Ecuador. We liked the taste, so we bought some. The is one the few herbals that contains caffeine. Dry Leaves: Crushed guayusa leaves, small and varying lighter shades of green. Liquor: The liquor of our Guayusa is a greenish-brown. Aroma: Sharp vegetal notes reminiscent of peppers. Caffeine Level: Caffeinated. Body: Medium bodied. Flavors: Assertive vegetal flavors.

# Tea kettles

Besides the teas themselves, kettles are key equipment. An occasional drinker may use a stove-top kettle, which have some advantages:

• cheap / free
• nearly-indestructible
• even simpler to operate

(In the sense that everyone knows how to turn on & turn off a stove burner already, not that electric kettles require a PhD to operate - typically, it’s a single button to push to boil some water.)
• somewhat more compact
• often picturesque
• always ceramic or metal, so no possibility of the water tasting like plastic

But for regular tea-drinkers, stove-top kettles come with serious disadvantages compared to electric kettles, and some of the advantages are negated:

• they are much slower to heat compared to electric kettles

Even in the USA, an electric kettle will be faster than stove-top. I haven’t tested it with a timer, but I think the difference between the T-fal and the metal kettle I was was ~2-3x. The delay is irritating
• they are energy-inefficient: much of the heat of the stove-top is not transferred into the water but the air, which is a waste of electricity or gas, and during summer will unpleasantly warm the house
• picturesque means that they are not always designed with safety or ergonomics in mind; I’ve seen more than a few which bade fair to burn users somewhile
• temperature control is difficult without a thermometer, so one must either compromise the simplicity & convenience of a stove-top kettle or risk destroying white/green/oolong teas by overheating the water
• it’s often easy to not notice when the water has reached a boil, or to not be present when the kettle does begin to whistle

This results in a waste of time, of water (increasing humidity, incidentally), overheating & de-oxygenating the water, and poses a safety hazard if the kettle boils dry

On the downside, the electric kettles lose most of the stove-top kettle advantages (they cost real money, can break, take up counter space, and the pretty ones are more expensive). But on net, I prefer an expensive electric tea kettle which will heat fast, not dump excess heat into the room, has boil-dry protection, and different heat settings.

I bought my first electric tea kettle on 7 January 2008 from Upton’s. It was their (since-discontinued) AK16 model (Upton Tea Imports® Variable Temp. Electric Kettle), and cost $43.80. Besides my daily tea, I used it for heating water for ramen, speeding up cooking of soups & stews, humidifying rooms, and unclogging drains. It worked well for years until the handle snapped off somewhere in winter 2013 or so. (It was my fault: I had been using it to humidify the room and had placed a book on the handle to keep the kettle boiling past the temperature shut-off.) This wasn’t a fatal problem because it was easy to take a small screwdriver and wiggle the switch inside the base. The kettle finally broke fully on 19 January 2014, having given me ~2203 days of loyal service at 2¢ a day (ignoring the electricity consumption). I would have bought it again except Upton’s no longer sold it or a replacement electric tea kettle; their website noes New kettle sources are being evaluated. I made do with a stove-top kettle laying around, but eventually the hassle of waiting twice or thrice as long, occasionally burning a green/oolong, and the upcoming hot summer spurred me to buy another. I had an unused Target gift card, so on 15 April 2014, I spent$36.04 to buy an Oster Digital Electric Tea Kettle model BVST-EK5967 from Target; it was the only electric tea kettle they had in stock with temperature control. I liked the digital temperature control (the Upton’s was an analogue knob with tea ranges), and it worked well. My main complaint was that the digital control would forget the previous temperature setting after use, and would reset to 212° so one had to set the temperature every use. By 20 May 2014, after 35 days, it had broken: it would turn on, but the water would never get hot. I thought perhaps it was a loose connection but a great deal of wiggling & experimenting failed to help matters, and I noticed the inside of the base seemed partially melted - so perhaps it couldn’t withstand its own heat? Target’s return policy didn’t seem to allow a return, so I had to give up. Much too late, I checked the Amazon page for the Oster BVST-EK5967 and saw that I was far from alone in having a bad experience with BVST-EK5967s dying unreasonably early. Oh well!

Having learned my lessons about ignoring user reviews, this third time I’m going with one of the top-reviewed electric tea kettles with temperature control on Amazon: the T-fal BF6138US Balanced Living 1-Liter 1750-Watt Electric Mini Kettle ($23.62). Unfortunately for me, my first order arrived broken. Reasoning that since it’s one of the top-reviewed models, it’s more likely that I got a bad item than it’s a rubbish model like the Oster, I decided to return it for a replacement (which Amazon makes reasonably easy: you print out an address and a bar code, slap it on the shipping box, and mail it at your local post office). The second order arrived working, and aside from the garish green-black coloring, seems to do its job well (although I miss the built-in thermometer of the Oster, which made it easier to find optimal temperatures for particular teas). # TODO • Upton’s shopping list: • ZG03: Season’s Pick Green Fannings Organic https://secure.uptontea.com/store/item.asp?itemID=ZG03 • Jasmine White Monkey https://secure.uptontea.com/store/item.asp?itemID=ZJ88 • Jasmine Yin Zhen https://secure.uptontea.com/store/item.asp?itemID=ZJ95 • Pu-Erh Sheng Cha https://secure.uptontea.com/store/item.asp?itemID=ZH44 • China Green Silver Spiral https://secure.uptontea.com/store/item.asp?itemID=ZG74 • http://www.teabox.com/samples ? • try gyokuro & genmai from http://www.hibiki-an.com/ ? # Appendix ## Electric vs stove kettle: fight! Electric kettles are faster, but I was curious how much faster my electric kettle heated water to high or boiling temperatures than does my stove-top kettle. So I collected some data and compared them directly, trying out a number of statistical methods (principally: nonparametric & parametric tests of difference, linear & beta regression models, and a Bayesian measurement error model). My electric kettle is faster than the stove-top kettle (the difference is both statistically-significant p≪0.01 & the posterior probability of difference is P≈1), and the modeling suggests time to boil is largely predictable from a combination of volume, end-temperature, and kettle type. ### Experiment My electric kettle is a T-fal BF6138US Balanced Living 1-Liter 1750-Watt Electric Mini Kettle, plugged into a normal electrical socket. The stove-top is a generic old metal kettle with a copper-clad bottom (it may have been intended to be a coffee percolator, given the shape, but it works well as a kettle) on a small resistance-heating coil stove burner (why not one of the 2 large coils? because the kettle bottom doesn’t cover the full surface area of the large burners); the stove is some very old small Gaffers-Sattler 4-burner stove/oven (no model name or number I could find in an accessible spot, but I’d guess it’s >30 years old). I began comparing them on the afternoon of 16 February 2015; the sea-level kitchen was at a warm 77.1° Fahrenheit & 49% relative humidity (as measured by my Kongin temperature/humidity data logger). For measuring water volume, I used an ordinary 1-cup kitchen measuring cup (~235ml). And for measuring water temperature, I used a Taylor 9847N Antimicrobial Instant Read Digital Thermometer (Amazon), which claims to measure in units of .1°F up to 450° and so can handle boiling water; I can’t seem to find accuracy numbers for this particular model, but I did find a listing saying that a similarly-priced model (the Taylor 9877FDA Waterproof Pocket Digital Thermometer) is accurate ±2°, which should be enough. The relevant quantity of water for me is at least one of my fox tea mugs, which turns out to be almost exactly 2 cups of water ( or ~0.5l marking on electric kettle). My testing procedure was as follows: 1. rinse out each kettle with fresh cold water from the tap (43°), fill with some, let sit for a few minutes 2. dump out all water from kettles 3. pour in with measuring cup 2-4 cups of cold water, put onto respective spots 4. adjust temperature setting on electric kettle if necessary I divide the T-Fal temperature control into min/medium-low/medium/max. 5. start timer software, then turn on kettles as quickly as possible (I’d guess this was a delay of ~3s; 3s has been subtracted from the times, but there’s still imprecision or measurement error in how fast I looked at the stopwatch or how long it took me to react or whether I jumped the gun.) 6. wait until electric kettle clicks, record time in seconds; turn electric kettle off, insert thermometer, and read final temperature of electrical kettle; then insert thermometer into stove-top kettle to measure stove-top kettle’s intermediate temperature 7. record time and 2 temperatures 8. place thermometer back into stove-top kettle, and watch the temperature reading until the stove-top’s temperature has reached the electric kettle’s final temperature; record the time 9. turn off stove heat, dump out hot water, return to step #1 This ensures that both kettles start equal, and the stove-top kettle is run only as long as it takes to reach the same temperature that the electric kettle reached; the intermediate temperature could also be useful for estimating temperature vs time curves. I ran 12 tests at various combinations of water-volume and temperature setting. I wound up not testing temperature settings thoroughly because once I began measuring final temperatures, I was dismayed to see that the T-fal temperature control was almost non-existent: 3/4s of the dial equated essentially to boil, and even the minimum heat setting still resulted in temperatures as high as 180°! - which makes the temperature control almost useless, since I think one needs colder water than that to prepare white teas and the more delicate greens… I am not happy with T-fal, but at least now I know what temperatures the dial settings correspond to. To deal with the poor temperature control, I bought a cheap mechanical cooking thermometer, calibrated it, drilled a narrow hole through the plastic of the T-fal tea kettle, and inserted the probe down into the water area. Now I can see what the current water temperature is as it heats up, and shut it off early or dilute it with fresh water to get it to the target temperature. Perhaps not as convenient as a fully digital electric tea kettle, but it’s much cheaper and probably more reliable/durable; I may do this with future tea kettles as well. ### Data The data from each kettle (time in seconds, temperatures in Fahrenheit): boiling <- read.csv(stdin(),header=TRUE) Test,Setting,Type,Volume.cups,Time,Temp.final,Temp.intermediate.stove 1,max,electric,2,144,210,120 1,max,stove,2,344,210,120 2,max,electric,3,199,211,108 2,max,stove,3,665,211,108 3,max,electric,3,210,211,120 3,max,stove,3,701,211,120 4,max,electric,2,141,211,107 4,max,stove,2,524,211,107 5,medium,electric,2,142,209,130 5,medium,stove,2,399,209,130 6,medium-low,electric,2,145,207,131 6,medium-low,stove,2,408,207,131 7,min,electric,2,110,190,101 7,min,stove,2,423,190,101 8,min,electric,3,146,180,108 8,min,stove,3,435,180,108 9,min,electric,4,178,172,99 9,min,stove,4,567,172,99 10,min,electric,2,105,186,106 10,min,stove,2,362,186,106 11,min,electric,2,109,187,108 11,min,stove,2,331,187,108 12,min,electric,2,113,187,114 12,min,stove,2,341,187,114 ### Analysis There many ways to analyze this data: are we interested in the mean difference in seconds over all combinations of volume/final-temperature, as a two-sample or paired-sample? In modeling the time it takes? In the ratio or relative speed of electric and stove-top? In correcting for the measurement error (±2° for each temperature measurement, and perhaps also how much water was in each)? We could look at all of them. #### Hypothesis testing Means, ratios, and tests of difference: abs(mean(boiling[boiling$Type=="electric",]$Time) - mean(boiling[boiling$Type=="stove",]$Time)) # [1] 313.1666667 boiling[boiling$Type=="electric",]$Time / boiling[boiling$Type=="stove",]$Time # [1] 0.4186046512 0.2992481203 0.2995720399 0.2690839695 0.3558897243 0.3553921569 0.2600472813 # [8] 0.3356321839 0.3139329806 0.2900552486 0.3293051360 0.3313782991 summary(boiling[boiling$Type=="electric",]$Time / boiling[boiling$Type=="stove",]$Time) # Min. 1st Qu. Median Mean 3rd Qu. Max. # 0.2600473 0.2969499 0.3216191 0.3215118 0.3405722 0.4186047 wilcox.test(Time ~ Type, paired=TRUE, data=boiling) # # Wilcoxon signed rank test with continuity correction # # data: Time by Type # V = 0, p-value = 0.002516 # alternative hypothesis: true location shift is not equal to 0 wilcox.test(Time ~ Type, paired=FALSE, data=boiling) # # Wilcoxon rank sum test # # data: Time by Type # W = 0, p-value = 7.396e-07 # alternative hypothesis: true location shift is not equal to 0 t.test(Time ~ Type, data=boiling) # # Welch Two Sample t-test # # data: Time by Type # t = -8.2263, df = 12.648, p-value = 1.983e-06 # alternative hypothesis: true difference in means is not equal to 0 # 95% confidence interval: # -395.6429254 -230.6904079 # sample estimates: # mean in group electric mean in group stove # 145.1666667 458.3333333 t.test(Time ~ Type, paired=TRUE, data=boiling) # # Paired t-test # # data: Time by Type # t = -11.1897, df = 11, p-value = 2.378e-07 # alternative hypothesis: true difference in means is not equal to 0 # 95% confidence interval: # -374.7655610 -251.5677723 # sample estimates: # mean of the differences # -313.1666667 So the electric kettle is, as expected, faster - by 5 minutes on average, ranging from 4x faster to 2x faster, and the advantage is statistically-significant. (Nothing surprising so far.) #### Linear regression How much variance do the listed variables capture? summary(lm(Time ~ Test + Temp.final + as.ordered(Setting) + Type + Volume.cups, data=boiling)) # ... # Residuals: # Min 1Q Median 3Q Max # -89.37786 -28.86586 12.18942 29.12054 93.64656 # # Coefficients: # Estimate Std. Error t value Pr(>|t|) # (Intercept) -2619.607015 1179.160067 -2.22159 0.04108708 # Test 5.808909 9.295290 0.62493 0.54082745 # Temp.final 11.696298 5.422822 2.15687 0.04657229 # as.ordered(Setting).L 124.978494 109.450857 1.14187 0.27031014 # as.ordered(Setting).Q 88.043187 54.791923 1.60686 0.12763657 # as.ordered(Setting).C 24.081255 48.488960 0.49663 0.62620135 # Typestove 313.166667 24.560050 12.75106 8.4966e-10 # Volume.cups 157.883148 38.444409 4.10679 0.00082473 # # Residual standard error: 60.15959 on 16 degrees of freedom # Multiple R-squared: 0.9257358, Adjusted R-squared: 0.8932452 # F-statistic: 28.49242 on 7 and 16 DF, p-value: 6.927389e-08 summary(step(lm(Time ~ Test + Temp.final + as.ordered(Setting) + Type + Volume.cups, data=boiling))) # ...Residuals: # Min 1Q Median 3Q Max # -115.480920 -41.874831 -3.183459 38.182981 125.963323 # # Coefficients: # Estimate Std. Error t value Pr(>|t|) # (Intercept) -835.055578 206.880137 -4.03642 0.00064609 # Temp.final 3.607011 0.934039 3.86174 0.00097187 # Typestove 313.166667 24.083037 13.00362 3.247e-11 # Volume.cups 111.948746 20.132673 5.56055 1.922e-05 # # Residual standard error: 58.99115 on 20 degrees of freedom # Multiple R-squared: 0.9107407, Adjusted R-squared: 0.8973518 # F-statistic: 68.02206 on 3 and 20 DF, p-value: 1.138609e-10 Because I controlled water volume and volume and final-temperature, the mean difference should be identical, and it is, 313s. The signs are also appropriate and coefficients sensible: each additional degree is +3.6s, a cup is +111s, and the setting variable drops out as useless (as it should since it should be redundant with the final-temperature measurement) as does the test number (suggesting no major change over time as a result of testing). We can plot the electric and stove-top data separately as 3D plots with residuals to see if any big issues jump out: library(scatterplot3d) plot3D <- function(k) { with(boiling[boiling$Type==k,], {
b3d <- scatterplot3d(x=Temp.final, y=Volume.cups, Time, main=k);
b3d$plane3d(my.lm <- lm(Time ~ Temp.final + Volume.cups), lty = "dotted"); orig <- b3d$xyz.convert(Temp.final, Volume.cups, Time);
plane <- b3d$xyz.convert(Temp.final, Volume.cups, fitted(my.lm)); i.negpos <- 1 + (resid(my.lm) > 0); segments(orig$x, orig$y, plane$x, plane$y, col = c("blue", "red")[i.negpos], lty = (2:1)[i.negpos]); }) } png(file="~/wiki/images/tea-kettle-electricvstove.png", width = 680, height = 800) par(mfrow = c(2, 1)) plot3D("electric") plot3D("stove") invisible(dev.off()) It looks pretty good. But in general towards the edges the points seem systematically high or low, suggesting there might be a bit of nonlinearity, and the fit seems to be worse for the stove-top results, suggesting that’s noisier than electric (this could be due either to slight differences in setting the analogue temperature dial on the stove or perhaps differences in positioning on the burner coil). #### Beta regression Regressing on the relative times / ratios, using the unusual beta regression, might be interesting; if electric was always 1:3, say, then one would expect the ratio to be constant and independent of the covariates, whereas if the ratio increases or decreases based on the covariates then that suggests some bending or flexing of the plane: boilingW <- read.csv(stdin(),header=TRUE) Test,Setting,Volume.cups,Time.electric,Time.stove,Time.ratio,Temp.final,Temp.intermediate.stove 1,max,2,144,344,0.4186046512,210,120 2,max,3,199,665,0.2992481203,211,108 3,max,3,210,701,0.2995720399,211,120 4,max,2,141,524,0.2690839695,211,107 5,medium,2,142,399,0.3558897243,209,130 6,medium-low,2,145,408,0.3553921569,207,131 7,min,2,110,423,0.2600472813,190,101 8,min,3,146,435,0.3356321839,180,108 9,min,4,178,567,0.3139329806,172,99 10,min,2,105,362,0.2900552486,186,106 11,min,2,109,331,0.329305136,187,108 12,min,2,113,341,0.3313782991,187,114 library(betareg) summary(betareg(Time.ratio ~ Temp.final + as.ordered(Setting) + Volume.cups, data=boilingW)) # Standardized weighted residuals 2: # Min 1Q Median 3Q Max # -3.0750244 -0.9096389 0.1394863 0.7292433 2.8146091 # # Coefficients (mean model with logit link): # Estimate Std. Error z value Pr(>|z|) # (Intercept) 7.49624231 4.15031174 1.80619 0.070889 # Temp.final -0.03782627 0.01921650 -1.96843 0.049019 # as.ordered(Setting).L -0.73487186 0.36540077 -2.01114 0.044311 # as.ordered(Setting).Q -0.47817927 0.19405990 -2.46408 0.013737 # as.ordered(Setting).C -0.18696030 0.16717555 -1.11835 0.263419 # Volume.cups -0.22934320 0.13345514 -1.71850 0.085705 # # Phi coefficients (precision model with identity link): # Estimate Std. Error z value Pr(>|z|) # (phi) 201.71768 82.16848 2.45493 0.014091 # # Type of estimator: ML (maximum likelihood) # Log-likelihood: 24.01264 on 7 Df # Pseudo R-squared: 0.3653261 # Number of iterations: 751 (BFGS) + 3 (Fisher scoring) Excluding the Setting variable, it looks like the temperature and volume may affect the timing, but not much. #### SEM Moving on to measurement error, one favored way of handling measurement error is through latent variables and a structural equation model, which in this case we might model in lavaan this way: library(lavaan) Kettle.model <- ' Temp.final.true =~ Temp.final Time.true =~ Time Volume.cups.true =~ Volume.cups Time.true ~ Test + Temp.final.true + as.ordered(Setting) + Type + Volume.cups.true ' Kettle.fit <- sem(model = Kettle.model, data = boiling) summary(Kettle.fit) # lavaan (0.5-16) converged normally after 120 iterations # # Number of observations 24 # # Estimator ML # Minimum Function Test Statistic 65.575 # Degrees of freedom 6 # P-value (Chi-square) 0.000 # # Parameter estimates: # # Information Expected # Standard Errors Standard # # Estimate Std.err Z-value P(>|z|) # Latent variables: # Temp.final.true =~ # Temp.final 1.000 # Time.true =~ # Time 1.000 # Volume.cups.true =~ # Volume.cups 1.000 # # Regressions: # Time.true ~ # Test 6.572 6.394 1.028 0.304 # Temp.final.tr 5.237 0.547 9.568 0.000 # Setting 0.385 16.210 0.024 0.981 # Type 313.163 18.009 17.390 0.000 # Volume.cps.tr 126.450 15.055 8.399 0.000 But the latent variable step turns out to be a waste of time (eg Temp.final.true =~ Temp.final 1.000), presumably because I don’t have multiple measurements of the same data which might allow an estimate of an underlying factor/latent variable, and so it’s the same as the linear model, more or less. #### Bayesian models What I need is some way of expressing my prior information, like my guess that the temperature numbers are ±2° or the times ±3s… in a Bayesian measurement error model. JAGS comes to mind. (Stan is currently too new and hard to install.) library(rjags) model1<-" model { for (i in 1:n) { Time[i] ~ dnorm(Time.hat[i], tau) Time.hat[i] <- a + b1*Test[i] + b2*Temp.final[i] + b3*Setting[i] + b4*Type[i] + b5*Volume.cups[i] } # intercept a ~ dnorm(0, .00001) # coefficients b1 ~ dnorm(0, .00001) b2 ~ dnorm(0, .00001) b3 ~ dnorm(0, .00001) b4 ~ dnorm(0, .00001) b5 ~ dnorm(0, .00001) # convert SD to 'precision' unit that JAGS's distributions use instead sigma ~ dunif(0, 100) tau <- pow(sigma, -2) } " j1 <- with(boiling, jags(data=list(n=nrow(boiling), Time=Time, Temp.final=Temp.final, Volume.cups=Volume.cups, Type=Type, Setting=Setting, Test=Test), parameters.to.save=c("b1", "b2", "b3", "b4", "b5"), model.file=textConnection(model1), n.chains=getOption("mc.cores"), n.iter=100000)) j1 # Inference for Bugs model at "5", fit using jags, # 4 chains, each with 1e+05 iterations (first 50000 discarded), n.thin = 50 # n.sims = 4000 iterations saved # mu.vect sd.vect 2.5% 25% 50% 75% 97.5% Rhat n.eff # b1 2.156 10.191 -18.143 -4.569 2.085 9.022 22.003 1.001 4000 # b2 -0.406 1.266 -2.906 -1.265 -0.373 0.419 2.061 1.001 4000 # b3 -40.428 26.908 -96.024 -58.060 -40.351 -22.395 11.220 1.001 2700 # b4 308.501 28.348 253.075 289.747 308.563 327.221 363.598 1.001 4000 # b5 80.985 23.082 33.622 66.062 81.784 96.441 125.030 1.001 4000 # deviance 269.384 4.058 263.056 266.473 268.887 271.673 279.300 1.001 4000 # # For each parameter, n.eff is a crude measure of effective sample size, # and Rhat is the potential scale reduction factor (at convergence, Rhat=1). # # DIC info (using the rule, pD = var(deviance)/2) # pD = 8.2 and DIC = 277.6 # DIC is an estimate of expected predictive error (lower deviance is better). ## Stepwise-reduced variables: model2<-" model { for (i in 1:n) { Time[i] ~ dnorm(Time.hat[i], tau) Time.hat[i] <- a + b2 * Temp.final[i] + b4 * Type[i] + b5 * Volume.cups[i] } a ~ dnorm(0, .00001) b2 ~ dnorm(0, .00001) b4 ~ dnorm(0, .00001) b5 ~ dnorm(0, .00001) tau <- pow(sigma, -2) sigma ~ dunif(0, 100) } " j2 <- with(boiling, jags(data=list(n=nrow(boiling),Time=Time, Temp.final=Temp.final, Type=Type, Volume.cups=Volume.cups), parameters.to.save=c("b2", "b4", "b5"), model.file=textConnection(model2), n.chains=getOption("mc.cores"), n.iter=100000)) j2 # mu.vect sd.vect 2.5% 25% 50% 75% 97.5% Rhat n.eff # b2 1.830 0.941 -0.140 1.235 1.875 2.474 3.594 1.001 4000 # b4 302.778 27.696 246.532 285.082 303.193 321.338 355.959 1.001 4000 # b5 90.526 22.306 46.052 75.869 90.848 105.883 133.368 1.001 4000 # deviance 268.859 4.767 261.564 265.238 268.276 271.854 279.749 1.001 4000 # ... # DIC info (using the rule, pD = var(deviance)/2) # pD = 11.4 and DIC = 280.2 # DIC is an estimate of expected predictive error (lower deviance is better). The point-estimates are similar but pulled towards zero, as expected of noninformative priors. With a Bayesian analysis, we can ask directly, what is the probability that the difference stove-top vs electric (b4) is >0? A plot of the posterior samples shows that no sample is ≤0, so the probability that electric and stove-top differs is ~100%, which is comforting to know. Measurement-error for Temp.final; we need to define a latent variable (true.Temp.final) which has our usual noninformative prior, but then we define how precise our measurement is (tau.Temp.final) by taking our two-degree estimate, converting it into units of standard deviations of the Temp.final data (2/14.093631), and then converting to the precision unit (exponentiation followed by division): model3<-" model { for (i in 1:n) { true.Temp.final[i] ~ dnorm(0, .00001) Temp.final[i] ~ dnorm(true.Temp.final[i], tau.Temp.final) Time[i] ~ dnorm(Time.hat[i], tau) Time.hat[i] <- a + b2 * Temp.final[i] + b4 * Type[i] + b5 * Volume.cups[i] } a ~ dnorm(0, .00001) b2 ~ dnorm(0, .00001) b4 ~ dnorm(0, .00001) b5 ~ dnorm(0, .00001) sigma ~ dunif(0, 100) tau <- pow(sigma, -2) tau.Temp.final <- 1 / pow((2/14.093631), 2) } " j3 <- with(boiling, jags(data=list(n=nrow(boiling), Time=Time, Temp.final=Temp.final, Type=Type, Volume.cups=Volume.cups), parameters.to.save=c("b2", "b4", "b5"), model.file=textConnection(model3), n.chains=getOption("mc.cores"), n.iter=100000)) j3 # mu.vect sd.vect 2.5% 25% 50% 75% 97.5% Rhat n.eff # b2 1.843 0.936 -0.106 1.238 1.889 2.497 3.570 1.001 4000 # b4 303.132 27.966 248.699 285.056 302.651 321.256 359.027 1.001 2900 # b5 89.503 21.985 43.631 75.713 90.076 104.306 130.206 1.001 4000 # deviance 242.944 8.204 228.833 237.013 242.222 248.095 260.670 1.001 2500 # ... # DIC info (using the rule, pD = var(deviance)/2) # pD = 33.6 and DIC = 276.6 # DIC is an estimate of expected predictive error (lower deviance is better). In this case, ±2° degrees is precise enough, and the Temp.final variable just one of 3 variables used, that it seems to not make a big difference. Another variable is how much water was in kettle. While I tried to measure cups as evenly as possible and shake out each kettle after rinsing, I couldn’t say it was hugely exact. There could easily have been a 5% difference between the kettles (and the standard deviation of the cups is not that small, it’s 0.653). So we’ll add that as a measurement error too: model4<-" model { for (i in 1:n) { true.Temp.final[i] ~ dnorm(0, .00001) Temp.final[i] ~ dnorm(true.Temp.final[i], tau.Temp.final) true.Volume.cups[i] ~ dnorm(0, .00001) Volume.cups[i] ~ dnorm(true.Volume.cups[i], tau.Volume.cups) Time[i] ~ dnorm(Time.hat[i], tau) Time.hat[i] <- a + b2 * Temp.final[i] + b4 * Type[i] + b5 * Volume.cups[i] } a ~ dnorm(0, .00001) b2 ~ dnorm(0, .00001) b4 ~ dnorm(0, .00001) b5 ~ dnorm(0, .00001) sigma ~ dunif(0, 100) tau <- pow(sigma, -2) tau.Temp.final <- 1 / pow((2/14.093631), 2) tau.Volume.cups <- 1 / pow((0.05/0.6538625482), 2) } " j4 <- with(boiling, jags(data=list(n=nrow(boiling), Time=Time, Temp.final=Temp.final, Type=Type, Volume.cups=Volume.cups), parameters.to.save=c("b2", "b4", "b5"), model.file=textConnection(model4), n.chains=getOption("mc.cores"), n.iter=600000)) j4 # mu.vect sd.vect 2.5% 25% 50% 75% 97.5% Rhat n.eff # b2 1.832 0.962 -0.213 1.210 1.885 2.488 3.603 1.001 4000 # b4 302.310 27.773 245.668 284.605 302.509 320.646 355.748 1.001 4000 # b5 89.655 21.768 44.154 75.561 90.428 104.743 129.892 1.002 2100 # deviance 188.133 10.943 169.013 180.519 187.413 195.102 211.908 1.001 4000 # ... # DIC info (using the rule, pD = var(deviance)/2) # pD = 59.9 and DIC = 248.0 While the DIC seems to have improved, the estimates look mostly the same. In this case, it seems that the variables are precise enough (measurement-errors small enough) that adjusting for them doesn’t change the results too much ## Water experiment The kind of water used in tea is claimed to make a difference in the flavor: mineral water being better than tap water or distilled water. However, mineral water is vastly more expensive than tap water. To test the claim, I compared my tap water, 3 distilled water brands (Great Value, Nestle Pure Life, & Poland Spring), 1 osmosis-purified brand (Aquafina), and 3 non-carbonated mineral water brands (Evian, Voss, & Fiji) in a series of n=67 blinded randomized comparisons of water flavor. The comparisons are modeled using a Bradley-Terry competitive model implemented in Stan; comparisons were chosen using an adaptive Bayesian best-arm sequential trial method designed to locate the best-tasting water in the minimum number of samples by preferentially comparing the best-known arm to potentially superior arms. Blinding & randomization are achieved by using a Lazy Susan to physically randomize two identical (but marked in a hidden spot) cups of water. The final posterior distribution indicates that some differences between waters are likely to exist but are small & imprecisely estimated and of little practical concern. Water for tea experiment: tap water reportedly differs a ton in taste between cities/states. This is plausible since tea tastes so much worse when microwaved, which is speculated to be due to the oxygenation, so why not the mineral/chlorine content as well? This question hearkens back to R.A Fisher’s famous lady tasting tea experiment and turns out to provide a bit of a challenge to my usual blinding & methods, motivating a look into Bayesian best-arm identification algorithms. ### Waters Well water & distilled water already on hand, so I need good commercial spring water (half a gallon or less each). I obtained 8 kinds of water: • tap water (I ran the tap for several minutes and then stored 3.78l in an empty distilled water container, and stored at room temperature with the others) • Walmart: • Great Value distilled water • Nestle Pure Life • Aquafina • Vos • Evian • Fiji Natural Water • Amazon: Price, type, pH, and contents: Water brand Water kind Country Price ($/l) Price ($) Volume (L) pH Total mg/l Calcium Sodium Potassium Fluoride Magnesium Nitrate Chloride Copper Sulfate Arsenic Lead Bicarbonates Silica tap water well USA 0 0 3.78 Great Value distilled USA 0.23 0.88 3.78 Nestle Pure Life distilled USA 0.26 0.98 3.79 6.95 36 7.6 6.8 1 0.1 3.6 0.4 13.45 0.05 13.5 0.002 0.005 Voss Still Water mineral Norway 3.43 2.74 0.8 5.5 44 5 6 1 0.1 1 0.4 12 0.05 5 0.002 0.005 Evian mineral France 1.78 1.78 1 7.2 309 80 1 26 6.8 12.6 36 15 Fiji Natural Water mineral Fiji 1.88 1.88 1 7.7 222 18 18 4.9 0.28 15 0.27 9 0.05 1.3 0.002 0.005 152 93 Aquafina osmosis USA 1 1 1 5 5 4 10 250 1 250 0.010 0.005 Poland Spring distilled USA 3.15 9.45 3 7.2 61 7.5 5.9 0.6 0.115 1.145 0.6 10.05 0.05 3 0.0014 0.005 Notes: • the pH & mineral content of my tap water is unknown; it is well water untreated with chlorine or fluoride, described as very soft • the Great Value/Walmart distilled water doesn’t report any data on the label and there don’t seem to be any datasheets online (the pH of distilled water can apparently vary widely from the nominal value of 7 and cannot be assumed to be 7, but should the mineral contents should all at least be close to 0) • the Nestle Pure Life numbers are not reported on the packaging but in the current online datasheet (pg4, 2015 Water Analysis Report); I have taken the mean when a range is reported, and the upper bound when that is reported (specifically, ND3) • Voss Still reports some numbers on the bottle, but more details are reported in an undated (metadata indicates 2011) report on the Voss website; for ND I reuse the upper bound from Nestle Pure Life • the Evian label reports a total of dissolved solids at 180C: 309ppm (mg/l). • Fiji Water provides a 2014 datasheet which is more detailed than the label; ND as before • Aquafina labels provide no information beyond using reverse osmosis; they provide a 2015 datasheet, which omits pH and several other minerals; various online tests suggest Aquafina samples have pHs of 4-6, so I class it as 5 • Poland Spring 2016 datasheet; note that the price may be inflated considerably because I had to order it online instead of buying in person from a normal retailer; like Nestle Pure Life, ranges are reported and ND taken as ceiling Reporting of the mineral contents of waters is inconsistent & patchy enough that they’re unlikely to be helpful in predicting flavor ratings. ### Design In rating very subtle difference in flavor, the usual method is binary forced-choice comparisons, as they cannot be rated on their own (they just taste like water). So the measured data would be the result of a comparison, better/worse or win/loss. Fisher’s original lady tasting tea experiment used permutation tests, but he was only considering two cases & was testing the null hypothesis, while I have 8 waters where I am reasonably sure the null hypothesis of no difference in taste is indeed false and I am more interested in how large the differences are & which is best, so the various kinds of permutation or chi-squared tests in general do not work. The analogy to sporting competitions suggests that the paradigm here should be the Bradley-Terry model which is much like chess’s ELO system in that it models each competitor (water) as having a performance variable (flavor) on a latent scale, where the difference between one competitor’s rating and another translates into the probability it will win a comparison. (For more detailed discussion of the Bradley-Terry model, see references in Resorter.) To account for ties, the logistic distribution is expanded into an ordered logistic distribution with cutpoints to determine whether the outcome falls into 1 of 3 ranges (win/tie/loss). With 8 waters to be ranked hierarchically using uninformative binary comparisons (which are possibly quite noisy) and sampling being costly (to my patience), it would be nice to have an adaptive experiment design which will be more sample-efficient than the simplest experiment design of simply doing a full factorial with 2 of each possible comparison (there are $\binom{8}{2} = 28$ possible pairs, since order doesn’t matter, so 2 samples each would give n=56). In particular, I am interested less in estimating as accurately as possible all the waters (for which the optimal design minimizing total variance probably would be some sort of full factorial experiment) than I am in finding out which, if any, of the waters tastes best - which is not the multi-armed bandit setting (to which the answer would be Thompson sampling) but the closely connected best-arm identification problem (as well as, confusingly, the dueling multi-armed bandit and preference learning/preference ranking areas; I didn’t find a good overview of them all comparing & contrasting, so I’m unsure what would be the state-of-the-art for my exact problem or whose wheel I am reinventing). Best-arm identification algorithms are often called racing algorithms because they sample by racing the two best comparisons against each other, focusing their sampling on only the arms likely to be best, and periodically killing the worst ranked arms (in successive reject). So they differ from Thompson sampling in that Thompson sampling, in order to receive as many rewards as possible, will tend to over-focus on the best arm while not sampling the second-best enough. Mellor 2014 introduces a Bayesian best-arm identification algorithm, Ordered-Statistic Thompson Sampling, which selects the arm to sample each round by: 1. fitting the Bayesian model and returning the posterior distribution of estimates for each arm 2. taking the mean of each distribution, ranking them, and finding the best arm’s mean4 3. for the other arms, sampling 1 sample from their posteriors (similar to Thompson sampling); add a bonus constant to tune the aggressiveness and sample more or less heavily from lower-ranked arms 4. select the action: if any of the arm samples are greater than the best arm mean, sample from that arm, otherwise, sample again from the best arm 5. repeat indefinitely until the experiment halts (indefinite horizon) This works because it frequently samples from any arm which threatens to surpass the current best arm in proportion to their chance of success, otherwise it concentrates on making more precise the best arm. The usual best-arm identification algorithms are for the binomial or normal distribution problems, but here we don’t have 21 arms of pairs of waters, because it’s not the pairs we care about but the waters themselves. My suggestion is that to adapt Mellor 2014’s algorithm to the Bradley-Terry competitive setting, one instead samples from each water, set the first water to be the highest mean water and then sample from the posteriors of the other waters and compare the best arm to the highest posterior sample. This is simple to implement, and like the regular best-arm identification algorithm, focuses on alternatives in proportion to their probability of being superior to the current estimated best-arm. (Who knows if this is optimal.) One alternative to my variant on ordered-statistic Thompson sampling would be to set a fixed number of samples I am willing to take (eg n=30) and define a reward of 1 for picking the true best water and 0 for picking any other water - thereby turning the experiment into a finite-horizon Markov decision process whose decision tree can be solved exactly by dynamic programming/backwards induction, yielding a policy for which waters to compare to maximize the probability of selecting the right one at the end of the experiment. This runs into the curse of dimensionality: with 28 possible comparisons with 2 possible outcomes, each round has 56 possible results, so over 67 samples, there are 5667 possible sequences. With such subtle differences, subjective expectations become a serious issue and blinding would be good to have, which also requires randomization. My usual methods of blinding & randomization, using containers of pills, do not work with water. It would be possible to do the equivalent by using water bottles shaken in a large container but inconvenient and perhaps messy. A cleaner (literally) way would be to use identical cups, one marked on the bottom to keep track of which is which after rating the waters - but how to randomize them? You can’t juggle or shake or mix up cups of water. It occurred to me that one could use a spinning table - a Lazy Susan - to randomize pairs of cups. And then blinding is trivial. ### Implementation Modified version of Ken Butler’s btstan Stan code for Bradley-Terry models and my own implementation of best-arm sampling for the Bradley-Terry model: ## list of unique competitors for conversion into numeric IDs: competitors <- function(df) { unique(sort(c(df$Type1, df$Type2))) } fitBT <- function(df) { types <- competitors(df) team1 = match(df$Type1, types)
team2 = match(df$Type2, types) y = df$Win
N = nrow(df)
J = length(types)
data = list(y = y, N = N, J = J, x = cbind(team1, team2))

m <- "data {
int<lower=0> N; // number of games
int<lower=1> J; // number of teams
int<lower=1,upper=3> y[N]; // results
int x[N,2]; // indices of teams playing
}
parameters {
vector[J] beta;
real<lower=0> cc;
}
model {
real nu;
int y1;
int y2;
vector[2] d;

beta ~ normal(0,5);
cc ~ normal(0,1);

for (i in 1:N) {
y1 = x[i,1];
y2 = x[i,2];
nu = beta[y1] - beta[y2];
d[1] = -cc;
d[2] =  cc;
y[i] ~ ordered_logistic(nu,d);
} }"
library(rstan)
model <- stan(model_code=m, chains=1, data=data, verbose=FALSE, iter=30000)
return(model)
}

sampleBestArm <- function(model, df) {
types <-  competitors(df)
posteriorSampleMeans <- get_posterior_mean(model, pars="beta")
bestEstimatedArm      <- max(posteriorSampleMeans[,1])
bestEstimatedArmIndex <- which.max(posteriorSampleMeans[,1])
## pick one row/set of posterior samples at random:
posteriorSamples <- extract(model)$beta[sample.int(nrow(extract(model)$beta), size=1),]
## ensure that the best estimated arm is not drawn, as this is pairwise:
posteriorSamples[bestEstimatedArmIndex] <- -Inf
bestSampledArm <- max(posteriorSamples)
bestSampledArmIndex <- which.max(posteriorSamples)
return(c(types[bestEstimatedArmIndex], types[bestSampledArmIndex]))
}

plotBT <- function(df, fit, labels) {
library(reshape2)
posteriors <- as.data.frame(extract(fit)$beta) colnames(posteriors) <- labels posteriors <- melt(posteriors) colnames(posteriors) <- c("Water", "Rating") return(ggplot(posteriors, aes(x=Rating, fill=Water)) + ggtitle(paste0("n=",as.character(nrow(df)),"; last comparison: ", tail(df$Type1,n=1), " vs ", tail(df$Type2,n=1))) + geom_density(alpha=0.3) + coord_cartesian(ylim = c(0,0.23), xlim=c(-12,12), expand=FALSE)) } It would be better to enhance the btstan code to fit a hierarchical model with shrinkage since the different waters will surely be similar, but I wasn’t familiar enough with Stan modeling to do so. ### Experiment To run the experiment, I stored all 8 kinds of water in the same place at room temperature for several weeks. Before running, I refrained from food or drink for 5 hours and brushed/flossed/water-picked my teeth. For blinding, I took my two identical white Corelle stoneware mugs, and put a tiny piece of red electrical tape on the bottom of one. For randomization, I borrowed a Lazy Susan table. The experimental procedure was: 1. empty out both mugs and the measuring cup into a tub sitting nearby 2. select two kinds of water according to the best-arm Bayesian algorithm (calling fitBT & sampleBestArm on an updated data-frame) 3. measure a quarter cup of the first kind of water into the marked mug and a quarter cup into the second 4. place them symmetrically on the Lazy Susan with handles inward and touching 5. closing my eyes, rotate the Lazy Susan at a moderate speed (to avoid tipping over the mugs) for a count of at least 30 6. eyes still closed for good measure, grab the mug on left and take 2 sips 7. grab the mug on the right, take 2 sips 8. alternate sips as necessary until I decide which one is slightly better tasting 9. after deciding, look at the bottom of the mug chosen 10. record the winner of the comparison, and run the Bayesian model and best-arm algorithm again. Following this procedure, I made n=67 pairwise comparisons of water: water <- read.csv(stdin(), header=TRUE, colClasses=c("character", "character", "integer")) "Type1","Type2","Win" "tap water","Voss",3 "Voss","Great Value distilled",3 "Great Value distilled","Poland Spring",1 "Poland Spring","Nestle Pure Life",3 "Nestle Pure Life","Fiji",1 "Fiji","Aquafina",3 "Aquafina","Evian",3 "Evian","tap water",1 "Great Value distilled","Poland Spring",1 "Great Value distilled","tap water",1 "Great Value distilled","Nestle Pure Life",1 "Fiji","Poland Spring",3 "Fiji","Poland Spring",1 "tap water","Poland Spring",1 "Poland Spring","Fiji",3 "Poland Spring","Fiji",3 "Poland Spring","Fiji",3 "Poland Spring","Fiji",1 "Poland Spring","Fiji",3 "Poland Spring","Fiji",1 "Poland Spring","Fiji",3 "Poland Spring","Fiji",1 "Poland Spring","Aquafina",1 "Poland Spring","Fiji",3 "Poland Spring","Aquafina",1 "Aquafina","Fiji",1 "Poland Spring","Aquafina",1 "Aquafina","Fiji",1 "Fiji","Aquafina",3 "Fiji","Poland Spring",3 "Fiji","Aquafina",3 "Fiji","Poland Spring",3 "Fiji","Poland Spring",3 "Fiji","Aquafina",3 "Fiji","tap water",1 "Fiji","Aquafina",3 "Fiji","Poland Spring",1 "tap water","Aquafina",1 "Fiji","Poland Spring",1 "Fiji","Poland Spring",3 "Fiji","Poland Spring",3 "Great Value distilled","Evian",3 "Evian","Voss",1 "Voss","Nestle Pure Life",3 "Nestle Pure Life","tap water",3 "tap water","Aquafina",1 "Aquafina","Poland Spring",3 "Evian","Great Value distilled",1 "Great Value distilled","Nestle Pure Life",3 "Nestle Pure Life","Voss",3 "Voss","tap water",3 "tap water","Aquafina",1 "Aquafina","Poland Spring",1 "Aquafina","Poland Spring",3 "Evian","Great Value distilled",1 "Great Value distilled","Nestle Pure Life",3 "Nestle Pure Life","tap water",3 "tap water","Voss",3 "Voss","Aquafina",1 "Evian","Great Value distilled",3 "Great Value distilled","Nestle Pure Life",3 "Nestle Pure Life","tap water",3 "tap water","Aquafina",3 "Evian","Great Value distilled",1 "Great Value distilled","Nestle Pure Life",3 "Nestle Pure Life","tap water",3 "tap water","Aquafina",1 ### Analysis types <- competitors(water); types # [1] "Aquafina" "Evian" "Fiji" "Great Value distilled" "Nestle Pure Life" # [6] "Poland Spring" "tap water" "Voss" fit <- fitBT(water); print(fit) # 8 chains, each with iter=30000; warmup=15000; thin=1; # post-warmup draws per chain=15000, total post-warmup draws=120000. # # mean se_mean sd 2.5% 25% 50% 75% 97.5% n_eff Rhat # beta[1] 1.18 0.01 1.89 -2.52 -0.09 1.18 2.46 4.89 23182 1 # beta[2] -2.69 0.01 2.11 -6.95 -4.08 -2.66 -1.25 1.33 27828 1 # beta[3] 1.83 0.01 1.89 -1.89 0.55 1.83 3.10 5.56 23147 1 # beta[4] -0.48 0.01 1.90 -4.21 -1.76 -0.48 0.81 3.21 23901 1 # beta[5] -0.40 0.01 1.88 -4.11 -1.66 -0.40 0.88 3.27 23487 1 # beta[6] 1.42 0.01 1.89 -2.29 0.15 1.42 2.70 5.14 23173 1 # beta[7] -0.49 0.01 1.87 -4.17 -1.74 -0.49 0.78 3.17 23084 1 # beta[8] -0.48 0.01 1.93 -4.29 -1.78 -0.48 0.82 3.29 24494 1 # cc 0.03 0.00 0.03 0.00 0.01 0.02 0.05 0.13 83065 1 # lp__ -49.58 0.01 2.27 -54.90 -50.87 -49.24 -47.91 -46.19 40257 1 sampleBestArm(fit, water) # [1] "Fiji" "Poland Spring" plotBT(water, fit, types) library(animation) saveGIF( for (n in 7:nrow(water)) { df <- water[1:n,] fit <- fitBT(df) p <- plotBT(df, fit, types) print(p) }, interval=0.6, ani.width = 1000, ani.height=800, movie.name = "tea-mineralwaters-bestarm-sequential.gif") Means in descending order: 1. Fiji 2. Poland Spring 3. Aquafina 4. Evian 5. Nestle Pure Life 6. Great Value distilled 7. Voss 8. tap water For the first 7 comparisons, since I didn’t want to insert any informative priors about my expectations, the best-arm choice would be effectively random, so to initialize it, I did a round-robin set of comparisons: put the waters into a quasi-random order ABCD, then compared A/B, B/C, C/D and so on. For the next 4 comparisons, I made a mistake in recording my data since I forgot that 1 coded for the left water winning and 3 for the right water winning, and so I had reversed the rankings and was actually doing a worst-arm algorithm, as it were. After fixing that, the comparisons began focusing on Fiji and Poland Spring, eventually expanding to Aquafina as it improved in the rankings. Comparing water turns out to be quite difficult. In some cases, a bad taste was quickly distinguishable - I quickly learned that Evian, Great Value distilled, and Nestle Pure Life had distinctly sour or metallic overtones. Despite repeated sampling, I had a hard time ever distinguishing Poland Spring/Fiji/Aquafina/Voss, but I thought they might have been ever so slightly better than my tap water in a way I can’t verbalize except that they felt cooler somehow. By n=41, Fiji continued to eek out a tiny lead over Poland Spring & Aquafina, but I ran out of it and could no longer run the best-arm algorithm (since it would keep sampling Fiji). I was also running low on the Poland Spring. So at that point I went back to round-robin, this time using the order of posterior means. With additional data, the wide posterior distributions began to contract around 0. At around n=67, I was bored stiff and not looking forward to sampling Evian/Great Value/Nestle many more times, and looking at the posterior distributions, it increasingly seemed like an exercise in futility. That is, further testing would probably show Evian/Great Value/Nestle as worse than the rest, but be unable to meaningfully distinguish between my tap water and the decent ones, which answered the original question - no, the decent mineral waters & waters are almost indistinguishable under even the most optimal taste-testing conditions, would be less distinguishable when used in hot tea, and there was zero chance they were worth their enormous cost compared to my free tap water & were a scam as I expected. I ended the experiment there, dumped the remaining water (except the remaining sealed Poland Spring bottles which are conveniently small, so I kept for use in my car), and recycled the containers. 1. Looking through my history 2006-2012, I order tea on a roughly annual or semi-annual basis: 1. 2006-10-16 ($19.30)
2. 2007-12-17 ($19.70) 3. 2008-08-01 ($43.80)
4. 2010-02-15 ($34.92) Thinking back, that 2 year gap between orders #3 and #4 was probably due to a Christmas where I received more than 2 pounds of tea, which took me a very long time to drink. 5. 2010-07-05 ($32)
6. 2011-05-14 ($51) 7. 2011-04-12 ($39)
8. 2012-07-15 ($75) 9. 2013, overall ($88)
10. 2015-06-06 ($90) 11. 2015-07-10 ($75)
12. 2015-09-01 ($111) 13. 2016-02-16 ($61)
14. 2016-05-09 ($86) 15. 2016-08-10 ($151)

I tried broadening my horizons and ordered 9 teas from a number of retailers through Amazon.com (principally Tao of Tea, Pure Herbal, & Summit Tea), and while some of Tao of Tea was good, overall I was not impressed.

2. Much of the research is of poor quality and from East Asia & China in particular, which is always a red flag for anything to do with traditional Asian treatments; reviews/meta-analyses, like the Cochrane reviews on Shengmai (a traditional Chinese herbal medicine) for heart failure & Ginseng for cognition typically find few high-quality studies & small inconsistent benefits.

3. MRL - Minimum Reporting Limit. Where available, MRLs reflect the Method Detection Limits (MDLs) set by the U.S. Environmental Protection Agency or the Detection Limits for Purposes of Reporting (DLRs) set by the California Department of Health Services. These values are set by the agencies to reflect the minimum concentration of each substance that can be reliably quantified by applicable testing methods, and are also the minimum reporting thresholds applicable to the Consumer Confidence…ND - Not detected at or above the MRL. –Nestle Pure Life 2015

4. This use of the posterior mean of the best arm distinguishes it from the simplest form of Thompson sampling for pairwise comparisons, which would be to simply Thompson sample each arm and compare the arms with the two highest samples, which is called double Thompson sampling by Wu & Liu 2016. Double Thompson sampling achieves good regret but like regular Thompson sampling for MABs, doesn’t come with any proofs about best-arm identification.