— PASSING A STALE KEY RETURNS 401. Passing NO key returns 200 with full data. We wrongly declared this source dead twice by testing with an expired key.

— Trailing slash is required on the metric path or it 307s.

— Without startDate/endDate it returns a single scalar, not a series.

— COMPANY-NOT-PRODUCT TRAP: for any venue whose main business is a crypto exchange, the symbol returns the EXCHANGE's volume. cryptocom ($112.9B) is the spot exchange, not its prediction market (nadex/CDNA). Same for gemini, hyperliquid, drift. Including those four would have reported ~$245B instead of $111B for Q2. Check metadata.about.description before trusting any symbol.

— Kalshi open interest reads ~$874M on a notional basis where our own priced figure is ~$110M — an 8x gap that is purely definitional, not an error in either.

— ~1.1 GB uncompressed per day. Cannot be JSON.parse'd — exceeds Node's max string. Must be streamed.

— Volume and open interest are CONTRACT COUNTS, not dollars. Multiply by price for USD; never by a flat $1 unless you mean notional.

— Carries only daily high/low, no per-trade price — so fees computed from it are per-market, not true row-level.

— A market that did not trade has no high/low, so its open interest cannot be priced. Report the coverage rather than assuming a price.

— min_ts/max_ts filtering returned empty in testing — ticker filtering works, time filtering did not.

— Per-ticker only in practice, so a whole-day sweep is impractical. This is the row-level source the fee methodology wants, but only for targeted markets.

— Use /events, NOT /markets — the flat /markets feed is dominated by MVE parlay shards (millions of rows, mostly zero open interest).

— Fields are suffixed _fp (open_interest_fp, volume_fp) and arrive as strings.

— Payload is large: ~2.6 MB per 100 events, which exceeds Next's 2 MB fetch-cache limit.

— 89 fields but NO open interest — liquidity is order-book depth, which is a different thing.

— Volume is native USD, so it is directly comparable to nothing on Kalshi without conversion.

— No history whatsoever. interval, startTs, fidelity, range and history params were all tried; every one returns the same single current value. There is no oi-history route.

— Accepts a comma-separated market list, so per-market OI is one round trip — but repeated ?market= params silently return only the last one.

— Reads ~24% below Artemis for the same venue and day ($359.5M vs $446.7M) — an unexplained basis gap worth resolving before either is published.

— Fee adapters embed the OLD Kalshi schedule (the one effective 4 Apr 2026) — so its fee figures lag schedule changes.

— Uses a hardcoded flat 25/50/25 split as an ASSUMPTION, not a measurement, in at least one adapter. Treat its fee output as modelled, not measured.

— Runs 2–4x below Artemis for the same venues because it reports matched rather than notional volume. Both are right; the basis differs.

— kalshi.trade_report has FIVE columns and none of them is a fee. No measured Kalshi fee figure exists anywhere; every number in circulation is a modelled schedule.

— EFFECTIVE-PRICE CORRECTION REQUIRED: (CASE WHEN price = 0 THEN 0.367 ELSE price + 0.45 END)/100.0 with price BETWEEN 0 AND 99. Omitting it drops sub-1c trades and HALVES combo notional ($46.6B → $26.9B).

— polymarket_polygon.combo_trades_raw writes one row per side; taker shares are exactly half of all rows and fees appear only on taker rows. Verified arithmetically, not documented — treat as inferred.

— Key lives in ~/Downloads/Pricedin/.env

— Not yet parsed. This is the dark-pool backstop — the only route to volumes for regulated venues that publish nothing themselves.

— Hardest to build and the most defensible thing on the list. Nobody else in the sector does it.

— PLAY MONEY. Must never enter a dollar league table. Its volume is not comparable to anything else here.

— Name collides with 'Manifold' the venture fund in our own investor data — this already caused a wrong-data bug on a venue profile.

— Self-reported free text, not measured. Preserve the raw claim verbatim and never promote it to measured.

— Strict USD-only parsing required; contract-count claims must be excluded from dollar rankings.

— ~50% success rate. The SAME url returns 200 and 404 minutes apart, and behaves identically under a browser User-Agent, so it is not bot filtering.

— Requires retry with widening backoff plus a last-good disk snapshot, or shows will blink in and out of the site.

— The YouTube Data API key (free quota) is the robust upgrade and is still unclaimed.