Methodology
Every brand score on this site is computed by one function, from inputs that are all published on that brand's own page. Nothing about the score is hidden or hand-adjusted per brand.
Scoring weights
| Input | What it measures | Weight |
|---|---|---|
| Depth | How many batches are indexed for the brand in this state. Log-scaled so the first batches count for more than the hundredth. | 20 |
| Span | How many months of unbroken testing history the indexed batches cover, from the earliest completed date to the most recent. | 15 |
| Pass rate | The scored component is the share of ALL indexed batches carrying a clean PASS verdict, so a record made mostly of verdict-less informational certificates earns fewer points. The displayed pass-rate percentage counts only verdict-carrying batches: a certificate whose panels print TESTED makes no pass/fail claim, and this index never converts "no claim" into a failure. | 25 |
| Variance | How consistent total THC is within a product line, batch to batch, measured as the mean coefficient of variation across every line with two or more batches reporting a THC value. Excluded from the total (not defaulted to zero) for a brand with no such line. | 20 |
| Panel breadth | How many pesticide analytes the brand’s widest-screened batch tests for, relative to the widest panel indexed anywhere in that state. | 10 |
| Public COA | The share of indexed batches whose source link resolves to an independently verifiable, batch-specific certificate rather than a laboratory’s bare homepage. | 5 |
| Recall record | A penalty from state recalls on file for the brand, weighted by mandatory vs. voluntary and decayed to zero over 24 months. Recalls follow the brand, not the state being scored: a recall in any covered state counts against every score the brand holds, because a recall speaks to brand practices rather than one market. | 5 |
| Total | 100 |
Data sources
Batch records are extracted directly from certificates of analysis published by licensed testing laboratories, one certificate at a time, preserving each certificate's own pass/fail verdicts, detection qualifiers and stated limits rather than re-deriving them. Recall records are extracted from a state cannabis regulator's public recall feed and reproduced without alteration; every recall page links to that regulator's own record as the authority.
Pesticide limits: two laboratories, two formats
The two California laboratories express the state's zero-tolerance pesticide limits differently. Infinite Chemical Analysis prints “Any amt”. Encore Labs prints a numeric limit equal to that analyte's own detection limit. Both describe the identical state rule. This index reproduces what each certificate printed rather than harmonising them, so the same limit appearing as “Any amount (zero tolerance)” on one batch and a number on another is not a difference in strictness between the laboratories.
Scores are never blended across states
A score is computed separately for every (brand, state) pair, filtered to that state's batches before any statistic runs. This index currently carries data for 3 states, and a brand sold in more than one of them gets one score per state, never a single blended number.
The same rule governs how rankings are presented (change noted 2026-09-28): brands are ranked within their own market, never in one cross-market list. Two scores from different markets are computed against different required panels, different regulators, and archives of very different public depth - some markets' brands publish multi-year certificate archives while others' public records begin much later - so ordering them in a single list would rank brands against tests they were never asked to run and history they were never able to publish. Nothing about the scores themselves changed: every (brand, market) score is computed exactly as before, and every input remains published.
Method changes
2026-10-10, product-line grouping corrected. The variance component compares batches within one product line. The original grouping keyed on the first segment of the printed product title, which mis-measured two real certificate conventions: titles that lead with the brand name ("Heavy Hitters: 1g Cart: Blue Dream") fused every product of a brand into one pseudo-line, so different strains were compared against each other and called inconsistency; and titles that embed per-batch tokens (lot codes, expiry dates) split one real product into unmeasurable singletons. Lines are now keyed on the full product name after stripping a leading own-brand prefix and trailing per-batch tokens, uniformly for every brand. Scores moved in both directions when this landed (brands with prefix-style certificates rose, brands whose products were previously unmeasured dropped); no records changed, and the old and new groupings can be compared from the public data export. Known limitation that stands: a beverage's THC percentage is a small number, so its batch-to-batch percentage variation reads noisier than a concentrate's at the same absolute spread.
Update cadence
There is no fixed publishing schedule. The dataset is rebuilt by re-running the ingest pipeline against each laboratory's published certificates and the state recall feed, and the site is rebuilt from that output; both happen whenever new certificates are indexed, not on a calendar. Every page states the date its own figures were current as of, so currency can be checked directly rather than assumed.
A consistency score publishes only once a brand has 10 or more finished-good batches indexed in a state. A score computed from one or two certificates would be statistically meaningless, so below that line the brand's batches are fully indexed and browsable but no score is shown. The threshold applies uniformly to every brand, including the brand that funds this site.