The Flower Index

theflowerindex.com

A directory of state-licensed cannabis dispensaries in New York City and Westchester County.

New York has a licensed cannabis market and a much larger unlicensed one. The value of this project is not the size of the list — it is that every entry carries its OCM licence number and the sources behind it, so a reader can check our work. One fabricated record would send someone to an unlicensed shop believing we had verified it, and that would make the whole register worthless.

That is why one rule outranks everything else in this repo:

An empty field beats a plausible guess. Unknown is null, never false.

Layout

data/
  schema/dispensary.schema.json    the data contract — read it before touching data
  schema/municipality.schema.json  opt-out status of cities, towns and villages
  schema/flower-listing.schema.json    flower on a shelf — no prices, flower only
  schema/strain-reference.schema.json  fallback terpene profiles by strain
  dispensaries.json                the delivered dataset (empty until research lands)
  dispensaries.demo.json           7-record sample so the site has something to show
  municipalities.json              Westchester opt-outs + NYC boroughs (hand-checked)
  territories.json                 the three territories, kept apart — and what each
                                   one does and does not know about municipal opt-outs
  upstate/, long-island/           the two newer New York territories, collected separately
  massachusetts/                   a second state, as an experiment — docs/MASSACHUSETTS.md
  raw/                             raw registry snapshots, kept as provenance
docs/
  AGENT_RESEARCH_BRIEF.md          phase 1 — collecting the register from the state
  AGENT_STORE_ATTRIBUTION_BRIEF.md open — twenty licences whose shelf was read from
                                   another store's menu, and the shops that declared
                                   more than we read
  AGENT_UPSTATE_WEBSITES_BRIEF.md  open — 262 upstate and Long Island shops the state
                                   records as open and we have no address for
  AGENT_ENRICHMENT_BRIEF.md        phase 2 — menus, geocodes, phones, services
  AGENT_MENU_BRIEF.md              phase 3 — reading flower off the shelves
  RESEARCH_REPORT.md               what phase 1 found, and what it could not
  STATEWIDE-PLAN.md                covering the rest of New York, and what a 2nd state costs
  MASSACHUSETTS.md                 the 2nd state, measured: what broke, what it cost, what
                                   stands between it and its menus
  FROZEN-municipal-opt-out.md      opt-out outside NYC: why it is NOT collected, and
                                   the exact procedure for when it is needed
  SOURCE_COLUMNS.md                the registry's real columns and the filter used
scripts/
  validate-data.ts                 schema + semantic validation (runs in CI)
  ingest/                          pulls the state registry into the schema
  build-preview.ts                 renders everything into one shareable HTML file
src/
  app/                             the Next.js site — / and /westchester/ are the
                                   NYC register; /upstate/, /long-island/ and
                                   /massachusetts/ are registry imports only and say
                                   so on the page
  lib/data.ts                      NYC + Westchester, borough-shaped
  lib/territory-data.ts            the newer territories, county-shaped, kept apart
preview/index.html                 generated single-file snapshot

Commands

npm install

npm run dev        # site at http://localhost:3000
npm run build      # static export to out/
npm run validate   # check the datasets against schema + semantic rules
npm run ingest     # pull the NY OCM registry into data/dispensaries.json
npm run ingest:dry # same, but write nothing
npm run preview    # regenerate preview/index.html
npm test           # typecheck + validate

How data gets in

1. npm run ingest pulls the state registry (Current OCM Licenses, resource jskf-tt3q), filters it to retail licences in the six in-scope counties, and writes the skeleton records.
2. Enrichment adds what the registry does not hold — geocodes, phone numbers, opening hours, services, menu platform. This is the research agent's job; see docs/AGENT_RESEARCH_BRIEF.md.
3. npm run validate must pass with zero errors before anything ships.

Re-running ingest refreshes the registry facts and carries enrichment forward for licences already in the file, so hand-collected detail is not lost.

The ingest adapter does not hard-code Socrata column names. It resolves each logical field against the columns actually present and, when a required one is missing, prints the real column list rather than silently emitting nulls.

Repository size

Raw registry snapshots in data/raw/ are the provenance for every record, and they are large in line count — one snapshot is about 4.5 MB across ~117k lines of pretty-printed JSON. That reads alarming and is not: JSON of this shape compresses to roughly 9%, so git stores a snapshot in about 0.6 MB, and the whole repository including history is under 10 MB.

The thing to watch is not any single snapshot but the rate. Git history is permanent — a large file committed weekly cannot be shrunk later by deleting it. At roughly 0.6 MB per snapshot, weekly ingest adds about 30 MB of history a year, which is comfortable; keep that cadence in mind rather than committing a snapshot on every experimental run.

Nothing here is slow: validation of 456 records takes under a second, the static build of 462 pages takes about 25 seconds.

Scope

Phase 1 covers six counties: New York (Manhattan), Kings (Brooklyn), Queens, Bronx, Richmond (Staten Island) and Westchester. Adding a region means adding a source adapter under scripts/ingest/sources/ and widening the county enums in the schemas.

Flower listings

Phase 3 reads what is actually on the shelves, and two decisions shape it.

Flower only, and no prices. Pre-rolls, vapes, concentrates and edibles are out of scope. Price is not collected in any form — the register does not compare cost, and not collecting it also keeps the project clear of the part of a menu that platforms defend hardest.

A terpene profile must say where it came from. A profile measured from the batch's own certificate of analysis and a profile taken from the strain's typical reference are different claims about the world, so terpenes.source is required and the validator refuses a LAB_COA with nothing to back it or a STRAIN_REFERENCE that cannot name its reference. Reference entries built from someone else's dataset cannot ship without the licence that permits the use.

Listings are perishable and belong in Postgres, not in git — weekly snapshots of tens of thousands of rows would bloat a history that cannot be reclaimed. The schema is the exchange and validation contract for one snapshot.

Relationship to SŌMA

SŌMA matches a person's sensory taste profile against strains on a dispensary menu. It needs to know which shops are real and which of them publish a machine-readable menu — which is what the menu.provider field in this schema records. This register is the location layer under that product.

What this is not

Not a shop, not a ranking, and not affiliated with or endorsed by the Office of Cannabis Management. A shop missing from the list is not proof it is illegal, and a listing here is not a substitute for checking the QR verification decal at the door or the state tool at cannabis.ny.gov/dispensary-location-verification.