Method and limitations
This site turns sub discussion into counts. Every count involves a judgment call. Here are the ones that change the numbers most, including the parts that don't work well.
739 scoresWhat the score is
These are the scores their authors gave. We did not interpret them. A lot of people who post a review on the sub write their own mark out of 10 in it — 8.5/10, or a line of sub-scores that they average themselves. That number is what this site collects, and the figure shown for a brand is the median of those numbers. There is no model in that chain, nothing is inferred from the wording, and no sentiment analysis happens anywhere on this site any more.
Of the 1,599 reviews counted, 739 (46%) carry a score. 206 brands have at least one; 46 have at least 5, which is the floor for showing a median at all. Every brand page prints every score it has, in order, so you can check the median yourself in about five seconds.
What this measures, exactly: among the products people chose to write a review of, how well they said it went. People review what they bought, and mostly what they were willing to buy again — 10% of all scores on the site are 6 or lower, and the site-wide median is 8.5. It is not a survey of everything on the shelf, and a brand with no scores here is not a bad brand, it is an unreviewed one.
Where the data comes from
Public posts and comments from r/NYSCannabis. Anything deleted or removed by mods is excluded from every figure on this site. It can't be verified and it can't be linked to, so counting it would mean asking you to take numbers on faith. The next section counts exactly how many review posts that costs.
Excluding it does bias the history a little: the deletion rate itself moved over the window, peaking early rather than late. Brand mentions are 89% comments, and comment deletion is far rarer and flatter than post deletion, so the drift this puts into the counts is small — but it is not zero, and it is not evenly spread across brands.
Why mentions and scores are separate
Mentions come from posts and comments. About 89% come from comments, so leaving them out would throw away most of the signal. A mention is not an opinion: a recall gets discussed as hard as a hit does.
Scores come only from the body of posts flaired Pic with Review or Craft/Micro Review, with the subject taken from the title. Comments are excluded here. In a thread, a line like "this one's not as good as X" mentions two products and rates neither cleanly, and a score written in a comment usually has no way to be attached to a specific purchase.
1,599 of 2,678 countedWhere the other review posts went
Every median on this site rests on the reviews it could actually read. That is a minority of the review posts in the window, and the reason is not mostly deletion:
| Counted | 1,599 | 60% | flaired as a review, still readable, and the title names a brand in the dictionary |
| Title names no known brand | 330 | 12% | alive and readable — the dictionary just doesn't have that brand yet. The largest gap, and a fixable one |
| Deleted from Reddit | 722 | 27% | of these, 274 named no known brand either, so only 448 could be attributed to a specific brand at all |
| Picture only | 27 | 1% | a review flair and a title, but no body text to read |
The middle row is the honest headline: the ceiling on this site is not what got moderated away, it is how much of the community's vocabulary the dictionary has caught up with. Deletion is the smaller problem and the one that cannot be fixed; the dictionary is the bigger one and can.
reporting limit n=5Why every median comes with its raw scores
A brand at 10.00 from 2 scores is not better than one at 9.00 from 50. Brands with fewer than 5 scores get no median at all — the page says how many they have instead. Above that floor, the median never appears on its own: the count, the range and, on the brand's own page, every individual score are printed next to it.
The range is an order-statistic interval: its two ends are real scores somebody posted, not a model's output, and the coverage it actually achieves is printed with it rather than rounded up to a comfortable 95%.
Do not compare the width of two brands' ranges. At 5 to 8 scores that range is literally the lowest and highest score posted; from 9 upwards it starts trimming the ends. So a narrow band on a brand with six scores and a narrow band on a brand with fifty are not the same statement, and the pages word them differently for that reason — “all 6 scores 8–9.5” against “n=70 · median 8.2–8.88 (96% coverage)”. What is always comparable is the scores themselves, which is why they are all printed.
The percentage in that band is not the share of scores inside it. It is the band's coverage: how often a fresh sample of the same size would put its median inside it. Rythm has 70 scores and a band of 8.2–8.88 at 96% coverage, while 18 of those 70 scores — 26% — actually fall in it. The two numbers answer different questions, and across the site the second one runs anywhere from 26% to 90%.
There is deliberately no spread number — no standard deviation, no interquartile range. We tested them: at these sample sizes their own uncertainty is as large as the numbers themselves, and their apparent ability to separate brands falls as the sample grows, which is the signature of a statistic that is measuring sample size rather than the brand. Counting the low scores is a count, and counts survive small samples.
Two different things are shown about low scores. Every brand gets a plain count: how many of its scores came in at 6 or less. That is information at any sample size and it is never hidden. Separately, a brand is flagged ⚑ low scores beyond chance only if that count is higher than the site-wide rate of 10% by more than luck explains — one-sided binomial test, corrected for testing all 46 rankable brands at once (α = 0.0011). Having a low score and having more of them than chance explains are not the same claim. Exactly one brand on the site currently carries the flag.
Last 3 mo · Last moWhy the time windows only slice volume
The brand list can be narrowed to the last few complete months, and the front page has a list of what is being discussed right now. Both count mentions only. Scores are always all-time, whichever window is selected — and the reason is the rule in the section above, applied honestly:
| Window | Reviews site-wide | Brands that could reach 5 | Most-reviewed brand |
|---|---|---|---|
| Last 3 mo Apr–Jun 2026 | 343 | 18 | 26 MFNY |
| Last mo Jun 2026 | 98 | 4 | 6 MFNY |
The third column is generous on purpose: it counts brands whose reviews in that window reach the floor, as though every one of those reviews carried a score. Site-wide only 46% of them do, so the real number is lower still — and the shortest window's best-covered brand has 6 reviews to begin with, against a floor of 5 scores. Slicing the median by window would mean either dropping the floor for one filter, or taking a median of two numbers. Volume is a count and survives being cut thin; a median needs bodies.
These figures are computed on every build rather than written down. If the sub starts posting enough scores for a 1-month median to be defensible, this paragraph will say so on its own.
299 brands in dictionaryHow a trend is called
Comparing the first six months of the window to the last six. The comparison is made on mentions per 1,000 surviving posts and comments, not on raw counts and not as a share of total brand mentions. That second denominator is tempting and wrong: it changes every time a brand is added to the dictionary — dropping just the single largest brand shifts everyone else's share by 16%, dropping the top five shifts it by 71%. The dictionary is at 299 brands and still growing, so tying published history to its current version would mean quietly rewriting past numbers. The denominator used here doesn't depend on the dictionary at all.
Charts show raw monthly mentions, because that's the unit everything else on this site uses. The direction is decided in the per-1,000 space, and when the two disagree the brand is left unlabelled rather than called either way.
299 brands tested at onceWhy so few brands get labelled
We test 299 brands for a trend at the same time. If each one got its own 5% error rate, we'd expect around six false calls per run — and the front page shows the most extreme results, which is exactly where that problem bites hardest. So the bar is set on the whole set instead: the entire labelled list has a 95% chance of containing no false call at all.
The bar isn't a formula. On every build the underlying discussion threads are reshuffled across months 4,000 times — whole threads at a time, since several mentions in one thread aren't independent events — and the threshold comes out of how far a brand can drift by chance alone. A brand needs at least 10 mentions a month to get a line at all, and clearing the bar on top of that. Most brands clear neither, and their pages say so instead of showing a number: on this build 47 of the 299 get past the mention floor and are actually judged. The correction is still sized for all 299, because the choice of which ones to look at was made after seeing the data.
Three things the charts deliberately don't show
An exact first mention. The Sustained from column and the mark on every brand chart are the first month whose mentions carried into the next two — one mention followed by silence does not start it. That is on purpose: a brand named once in January and then not again until August did not enter the conversation in January. The cost is that this is not the date a brand was first ever named here — on this build 74 of the 299 brands have at least one mention before the month shown. Nothing is hidden by it: every monthly count, including the earlier ones, is printed under the chart on each brand's page.
A first month, for brands that were already here. The data starts Jan 2025, but the sub didn't. A brand already at volume in that first month is marked ≤ Jan 2025 — 'as far back as we can see', not 'new'.
The current month. Charts stop at Jun 2026. Data runs into Jul 2026, but a partial month reads as a collapse even when nothing has changed — per day it's running level with the month before. Leaving it off and saying so here beats drawing a cliff that isn't there.
Known limitations
| This is talk, not sales Everything here is share of discussion on one subreddit. Reddit users are not a sample of New York consumers, and a brand getting talked about more is not a brand selling more — a recall gets discussed as much as a hit does. Wide distribution inflates mentions, and so does controversy. |
| Nothing here can spot coordinated posting A brand that suddenly appears everywhere looks identical to one that genuinely caught on, and a self-reported score is as easy to write as any other sentence. The site shows the first month, the volume, every individual score and a link to every post behind them, so you can judge; it does not make that call itself. |
| Who writes reviews is not who buys People post reviews of things they chose, and mostly of things that worked out. That is why almost every brand here sits above 8 and why the interesting signal is at the bottom of the list rather than the top. Nothing on this site corrects for that, because there is nothing to correct it against. |
| The top of the list is a near-tie 40 of the 46 rankable brands have a median of 8 or better, and the largest single group of them share the exact same median. Ranks are shown as ties where they are ties. Do not read a gap of a few places near the top as a difference in quality — read the scores. |
| Four brand names are ordinary English words Dank, Matter, Trout and Olio all match text that isn't about the brand. Before any of them gets a direction label, their unambiguous brand uses and unambiguous non-brand uses are tracked as separate series — if both move together, the trend is language drift and no label is shown. |
| Scores are found by pattern, not judged A script reads the review body for a mark out of 10 — a stated overall, a single item score, or the mean of the sub-scores the author listed. Which of the three it was is printed next to every quote and in every evidence panel. A review whose author wrote no number contributes nothing to the median, and shows up as no self-rating when you open the samples. This site no longer runs sentiment analysis of any kind. The previous version did, and measuring it honestly showed it added nothing: it missed 41% of the reviews whose own authors scored them 6 or lower. |
| Brand and strain names are a hand-maintained list A new brand won't show up until it's added, and nothing said about it before that gets counted. This is the single biggest thing standing between the numbers on this site and the sub as it actually is: of the 2,678 review posts in the window, 330 are alive and readable but name no brand this list recognises. That is 12% of them, and it is a bigger hole than deletion. |
| A few real brands aren't counted at all One is named after a common English verb. Four out of five matches were the verb, so counting it would have been worse than skipping it. |
| Product-level samples are thin For a specific strain from a specific brand, most combinations have single-digit score counts. Those are shown as the scores themselves rather than a median, and never ranked. |
| Co-mention is not inventory A brand and a shop appearing in the same discussion doesn't confirm the shop carries it. The shop names come from a matcher that is right about three times in four, which is why they no longer appear on individual brand pages. |
How this is paid for
It isn't. No ads, no paid placement, nothing sold. I'm not in the industry and don't work with any brand listed here. Getting a brand's reputation wrong is worse than leaving it out entirely, so if something here is wrong — a number, a brand attribution, a judgment call I haven't listed — write to nyccannaboard@gmail.com and it gets fixed.
Legal and privacy
21 and over. New York limits adult-use cannabis to adults 21 and older, and so does this site.
Not advice. Nothing here is medical, legal, or purchasing advice. These numbers describe what people wrote on a public forum. They are not a measure of product quality, potency, or safety, and a high rank is not a recommendation to buy.
Whose words these are. Every quote was written by a Reddit user, not by this site, and it stays theirs. Quotes are short, attributed to the post they came from, and each one links back so you can read it in context. Nothing is reproduced in full. Deleted and removed content is excluded everywhere, including from the quotes. Brand names belong to their owners; this site is not affiliated with, endorsed by, or speaking for any of them, nor with Reddit. If you wrote something quoted here and want it gone, that is reason enough — write to nyccannaboard@gmail.com and it comes out.
Privacy. No accounts, no cookies, no analytics, no third-party requests of any kind — the page's Content-Security-Policy blocks them outright rather than promising not to make them. Nothing you do here is logged or sent anywhere. Two values are stored in your own browser and never leave it: which theme you picked, and that you confirmed your age. Clearing your browser data removes both. No Reddit usernames appear anywhere on this site.
Contact. nyccannaboard@gmail.com — for corrections, attribution disputes, and takedown requests.