Prediction Contracts: September 6, 2026 snapshot Methods and data notes for the settlement-source map and last-price audit SCOPE We collected 103,998 unique open, non-combination Kalshi contracts across 12,645 event tickers. Collection ran from 2026-09-06T03:32:50.853765Z to 2026-09-06T03:34:05.823531Z (late September 5 in New York). This was one sequential API sweep, not an atomic snapshot or time series. COLLECTION Public endpoint: https://api.elections.kalshi.com/trade-api/v2/markets Parameters: status=open, mve_filter=exclude, limit=1000; follow the cursor. 104 responses were collected, ending with an empty cursor. Each original response was saved locally with its URL and retrieval timestamp. Public /series metadata was collected separately: 13,839 unique series. Two identical API-record tickers are deduplicated by keeping the last record; there were no duplicate market tickers in this collection. SOURCE MAP Infer series from the portion of each event ticker before its first hyphen. Match that identifier to the retrieved series metadata. 103,836 contracts matched; 162 remained unknown. Unknown contracts stay in the denominator but contribute no source references. Read settlement_sources from series metadata. Parse URL hostnames, lowercase them, remove a leading www., and deduplicate within each contract. Count linked contracts, unique event tickers, and unique inferred series. The recent count includes linked contracts with positive volume_24h_fp. All-contract percentage denominator: 103,998. Recently-traded percentage denominator: 18,721. Source lists overlap: rows and percentages must not be added together. Hostnames are not corporate parents. Lists do not establish source priority, fallback status, source actually used, reliability, or exclusive dependence. We inspected examples of full contract terms, not every contract PDF. An independent check found 2,829 ESPN-linked series in the complete series response, of which exactly 1,000 had contracts in the open-market sample. The 1,000 count is not a truncation limit in the join. QUOTE AUDIT Use YES prices in dollars and fixed-point quantities from market responses. Valid two-sided: 0 < YES bid <= YES ask < 1, with positive quantities at both best prices. Endpoint prices and zero-quantity quotes are excluded. Last-price comparisons require positive reported lifetime trading volume. Recent cohort additionally requires positive volume_24h_fp. Outside: last < bid - 1e-8 OR last > ask + 1e-8. Distance: max(bid - last, last - ask, 0). More than five cents outside: distance > 0.05 + 1e-8 dollars. Midpoint: (bid + ask) / 2. Spread: ask - bid. The 10-90 cent midpoint restriction is inclusive and uses decimal arithmetic for boundary comparisons. Equal-event weighting averages each event's eligible-contract outside fraction, then averages those fractions across 4,954 represented events. The top-100 cohort is selected by descending volume_24h_fp across the entire collected sample; only its 83 valid two-sided contracts enter quote metrics. These cohorts overlap. Counts describe contracts, not independent events, people, dollar trading volume, or market accuracy. COPPER CASE Event: KXCOPPERMON-26SEP3017. Tickers: KXCOPPERMON-26SEP3017-T6.03 and KXCOPPERMON-26SEP3017-T7.47. Both use the same Pyth copper source and one-minute candle close for September 30, 2026 at 5 PM EDT, with matching secondary rules. Separate market, order-book, and trade endpoints were checked after the main sweep. The case-study JSON provides their times, prices, and rules. Kalshi order books contain YES and NO bids; YES asks can be read as 1 minus the corresponding NO bid price. Quantities are contracts. The historical trade prices were recorded about two minutes apart and roughly two hours before collection. They are not simultaneous forecasts. We did not inspect their consumer-website presentation or measure executable arbitrage, manipulation, or potential profits. LIMITATIONS Quotes and the set of open contracts can change while pages are collected. API cache timing was not independently measured. Completion of pagination is not proof of capturing every contract simultaneously open on the venue. The weekend timing and new, distant, and far-tail listings affect the sample. Open-only volume omits already-settled contracts and is not exchange volume. updated_time describes metadata changes, not the latest trade. The deprecated liquidity_dollars field was not used. Outside-range last prices do not establish why a discrepancy arose or how long it lasted. Repeated observations are needed for claims across time. No original Polymarket data is included in these calculations. FILES settlement-source-map.csv: all observed source hostnames and derived counts. quote-audit.json: quote metrics, categories, and sensitivity checks. copper-case-study.json: two corroborated cases with archive checksums. These downloads contain derived results and selected evidence, not the full contract-by-contract dataset. Original API responses and analysis scripts are preserved in the project's local dated research archive for reanalysis. The 110 gzip archives were checksum-verified. Initial offline reproduction matched the original outputs. A subsequent independent audit used exact decimal prices and found a boundary error in the midpoint sensitivity cohort. The corrected analysis and downloads now agree with that independent audit. PRIMARY DOCUMENTATION https://docs.kalshi.com/api-reference/market/get-markets https://docs.kalshi.com/api-reference/market/get-series https://docs.kalshi.com/api-reference/market/get-market-orderbook https://assets.kalshi.com/contract_terms/FOOTBALLSEASONSTAT.pdf RELATED PRIOR WORK https://il9.org/methodology The last-price/outside-quote issue is previously recognized. These articles contribute a dated measurement and examples, not a claim of first discovery. FACT-CHECK UPDATE (September 7 UTC / September 6 New York, 2026) The inclusive 10-90 cent midpoint cohort is 10,997 eligible contracts, of which 2,377 have last prices outside their quotes. Previously the download reported 10,994 and 2,376 because binary floating-point rounding excluded three contracts with midpoints exactly at 10 cents. The rounded 21.6% is unchanged. No headline counts changed. Original archives and pre-correction results are preserved locally. The football terms PDF was also checked: its document title is FOOTBALLENTITYSTAT despite its FOOTBALLSEASONSTAT.pdf filename, and it ranks the governing league before ESPN. Series metadata ordering is not a contractual source hierarchy. Copper timestamps describe the reference price observation, not a promised payout time.