Sports and exchanges: status

Where every sport stands on every exchange: what data we hold, what prices it, what has been backtested and how far the result can be trusted, and what runs live or on paper. Reviewed in full on 2026-10-14 against the code on staging at ba839f3 (what production runs once PR #180, the staging-to-main promotion, merges; "promotion pending" marks the few cells where that matters) and against production's database, read-only, through the MCP server (overview() and sql, 04:50 UTC). Every count below carries its query date; a count with no date is from an earlier review and says so.

Nothing here trades real money: POLYMARKET_TRADING_ENABLED and KALSHI_TRADING_ENABLED are unset, and OG.com has no order code at all. The fantasy league trades F$ on recorded books (Alpha layout). Venue prices and tapes are shown to signed-in users with their source and fetch time, under the terms-pending (#aintcare) warning (owner, 2026-10-14).

The short version. Only F1 on Polymarket has a real model, a backtest that held up in a held-out season, and live paper trading wired up; Polymarket lists F1 race markets again, from Italy (6 Sep) through São Paulo (8 Nov). Every other cell is one or more steps short: F1 on Kalshi has a tuned maker and a taker result that rests on two weekends a season; NASCAR and MotoGP price from the one engine's results core with no calibration check yet (C48), so their prices are unvalidated; OG.com is recorded and replayable with no strategy tested on it; downhill has a real model and no exchange; IndyCar, Le Mans and SailGP are tapes with no model; road cycling has a first model, admin-only. The three recorders run every 5 minutes, every day, on all three venues (books in Postgres to 04:49 UTC today on each).

The 3 × 3 the soft launch was planned around (F1, NASCAR, MotoGP × Polymarket, Kalshi, OG.com), counts from production on 2026-10-14:

	Polymarket	Kalshi	OG.com
F1	Real model, robust taker A; 8,602 links, 898 open; race markets for rounds 13 to 20	Real model; maker K tuned; 4,272 links, 74 open (futures; the US GP lists 2 to 4 days out); books every 5 min since 30 Sep	Season futures plus per-GP race markets since Malaysia (22 links each for rounds 16 and 17); 808 trades since 9 Oct
NASCAR	1,721 links, 927 with a race, 229 open; books and trades recorded; no backtest of its own	Results core, in-sample demo paper; 14,490 links, 7,994 with a race, 529 open; next race (18 Oct) linked	17 champion contracts, 69 trades since 9 Oct; uncalibrated season sim, offline
MotoGP	209 links, 131 with a race, 29 open; 49 trades since 7 Oct; no strategy tested	Results core, race win only, in-sample demo paper; 344 links, 289 with a race, 33 open; no round after Austria (18 Sep) in production until #134's calendar step runs (promotion pending)	Empty: not listed

ProphetX (2026-10-21, not in the table): F1 race winners only (the Austin event first), book snapshots and quotes from the recorder once the owner switches it on (Exchanges: ProphetX). Admin-only, read-only; no trades, history or settlement; priced by the same race-win model, with no backtest or strategy of its own.

How to read the status

Word	Means
Real model	A model with its own features, walk-forward backtested on results and compared with a baseline and the market (F1: F1 evaluation; downhill: Evaluation)
Results core	The one engine (GlobalModel, Backtest core) at the results level: a probit strength model on recent finishing positions and the schema's declared inputs, one price per race before the weekend. It runs in the replays and prices the board's columns. Its calibration by probability bucket has not been checked (C48), so read its prices as unvalidated, not as edge
Quote only	The market is synced and recorded; no model prices it
Tape only	The sport's schema has model_family = "none": links are unmodeled, prices, trades and books are recorded, nothing is priced
Robust	Positive in the target season (2026) and the held-out season (2025), beyond the noise floor (promotion rule)
In-sample	The settings were chosen on the same races they are scored on
Unvalidated	Priced, but never scored by probability bucket against results or closing prices

Counts of links, trades and books are what production's Postgres held on 2026-10-14; the recorder's hourly archive pass moves cold rows to Parquet (data/archive/markets/), so a trade count is the hot window, not the whole tape. racinglines backtest coverage (CLI) counts both stores per sport × exchange × kind × season; run it on the VM for the full picture.

The status matrix

One row per sport and exchange that lists it (a venue that lists nothing for a sport is left out; see Not listed anywhere). Links: total / with a race / open, production 2026-10-14.

Sport × exchange	Data we hold	Model	Backtest (caveats)	Live / paper	Main gap
F1 × Polymarket	8,602 links / 5,070 / 898 (4,323 unmodeled: props and markets no kind covers); 2025–26 race weekends and championships; 2026 race markets for every round from Italy (138 links) to São Paulo (266), Singapore 272; books every minute since 3 Oct (919,164 snapshots in Postgres, last 04:49Z); 6,199 trades in the hot window since 4 Oct	Real model (the one engine's laps core, stored as f1_sector_sim) for race kinds; season forecast for champion, season-wins and standings head-to-head	Robust. Taker profile A (+$1,389 in 2026 / +$1,432 in 2025 at 16k sims, PR #85) and maker C; the overnight 106-job sweep kept A. Championship checkpoints lost in 2026 (−$566 to −$797, every variant). Five alpha styles (T1, T2, T5, T6, T8) carry their 2025 and 2026 backtests on the picker; T8 is in-sample (Alpha layout)	Signal engine and paper positions for A and C, stage by stage; last stored signal Singapore after qualifying (2026-10-11 14:37Z). Paper positions 2,264 (2,182 settled, 41 weekends). Fantasy: every signed-in account trades F$ by hand on these recorded books (Trade tab) or follows a style	G1 evidence: CLV over 100 live bets (Paper trading); #181's golden comparison against the Probalytics tape
F1 × Kalshi	4,272 links / 4,165 / 74 (925 unmodeled); 2025–26 history, trades and hourly candles; books every 5 minutes since 2026-09-30 (923,282 snapshots, last 04:48Z; record_venues.sh); 47,034 trades in the hot window since 9 Oct	Real model (same as above)	Maker K tuned (2026 +$490, 2025 +$659; tunable, each change logged). Taker fills at the touch since 2026-10-05 (book, else candle bid/ask): profile A at 16k +$2,332 / +$1,257, negative without its best two weekends in both seasons, so no edge shown (Kalshi history). Demo maker's Kalshi record 2025 +$193.39, 2026 −$118.84 (backfill)	Paper signals wired (venue = kalshi, U1; sprints on by default since 2026-10-06). Paper positions 2,632, all settled, last written 2026-09-30 (the backfill): at Singapore profile A took no Kalshi trade after FP1	Books before 30 Sep: the Probalytics backfill (Kalshi books from 9 May 2026, PR #181); a taker result that survives dropping its best weekends
F1 × OG.com	108 links / 44 / 20: 20 open season futures and, new since the 3 Oct review, per-Grand-Prix race markets (22 links each for Malaysia and Singapore, both closed); 74,213 book snapshots since 8 Oct; 808 trades since 9 Oct; settlement feed read hourly (markets --exchange og settle)	Real model for the champion markets (season forecast) and race kinds	None run. OG.com is a replay venue (#88); no champion or race replay has been reported on it (Championship markets)	Fair-price indicator only (RACINGLINES_OG_VENUE; on on the VM per the 48-hour report, not re-checked today). No trading: needs FCM onboarding	Thin books; fee $0.02 unverified; per-GP contract rules in exchanges/og.toml still to write from the two race weekends seen
NASCAR × Kalshi	14,490 links / 7,994 / 529 (14,071 unmodeled); 2026 season 40 events, 36 completed (results through Charlotte, 11 Oct); books in the 5-minute recorder; 33,814 trades in the hot window since 9 Oct; Phoenix (18 Oct) already 200 links	Results core ([model] inputs = ["results", "qualifying", "fastest_lap"], NascarCupRace a challenger) for race win, top 3/5/10/20 and h2h; champion: #93 season sim, not calibrated, offline only	Not robust. Taker grid 2026 +$4,226 to +$5,479, 2025 −$242 to −$390 in every setting (Kalshi last-trade prices, so indicative). Nothing tuned. Calibration unchecked (C48)	Demo paper rows, in-sample (#110): Pro 2025 −$278.31 / 2026 +$6,525.45; Basic −$357.23 / +$2,657.23. Next-race fair price from the latest stored forecast, run 9937 (2026-10-11, data through 4 Oct); the forecast-refresh timer runs every 30 minutes and fetches results on race weekends	A held-out-positive result; the calibration check; the pre-race field from NASCAR's entry list (C46)
NASCAR × Polymarket	1,721 links / 927 / 229 (1,611 unmodeled); each Sep–Oct race 42 to 44 links; books and trades recorded by pm_sync.sh (243 trades since 7 Oct)	Race kinds: same results core; champion: #93, not calibrated	None of its own. The demo backfill reused Kalshi's selected settings (matched 1 race, 30 Sep)	None since the demo backfill	A replay on its own tape, now that it is recorded
NASCAR × OG.com	17 Cup Champion contracts, all open; books since 8 Oct; 69 trades since 9 Oct	Champion: #93, not calibrated, offline (nascar season --quotes)	None	None	Calibration of the season sim; OG.com lists no NASCAR race markets
MotoGP × Kalshi	344 links / 289 / 33; 408 trades since 17 Sep; books in the recorder; production holds 2026 rounds through Austria (18 Sep) only: 15 events, no scheduled round, so no next-race price. The forecast-refresh MotoGP step in #134 files the calendar and results on its first pass after #180 deploys (promotion pending)	Results core (inputs = ["results", "qualifying"]; MotoGPRaceChallenger a challenger) for race win. Sunday race only: sprints not ingested. Champion and teams: quote only	No verdict. No held-out season (2025 lists no MotoGP race market on Kalshi); race-win log loss 0.0796 with no baseline comparison; calibration unchecked (C48: on the Indonesia GP the model's top three summed to 42.5% against Kalshi mids of 83.5%)	Demo paper rows, in-sample, 2026 only (11 races): Pro +$31.25, Basic +$7.70. Visible to every signed-in user after #180 (audience = "all", owner's data-terms decision 2026-10-14)	The calendar on production; the calibration check; sprint results; champion classifiers
MotoGP × Polymarket	209 links / 131 / 29 (all unmodeled); 49 trades since 7 Oct; books by pm_sync.sh (PM_SPORTS)	Champion: quote only. Race winners: the results core prices them on the board; the links stay unmodeled	None	None	A first replay; the champion needs sprints, schedule, standings (Championship markets)
Downhill × private book	Whistler 2026 live event (7,703 fills); results 2021–26 from ChronoRace; 2026 season complete (8 rounds)	Real model (the one engine's run-times core, stored as timed_runs), walk-forward on 43 rounds	Walk-forward on results (Final-qualification error 38% below a uniform guess). No exchange tape, so no market backtest	Private book only (318 paper positions, 4 events). Off the Markets board since #163 (markets_board = false)	No exchange lists downhill; next World Cup spring 2027
IndyCar × Kalshi	1,374 links, all closed, none with a race	Tape only	None	None	Results source's terms forbid automated use (#75); season over until March 2027
Road cycling × Kalshi	0 links and 0 events on production (2026-10-14); schema road_cycling (KXCYCLING*) built, no VM sync run	Results core, admin-only on the board since #163 (audience = "admin"); a stage-race GC model for the Eneco Tour is a draft (PR #177); PCS results live on the owner's Mac only	None on production	None	A results source with usable terms; the GC model's decision-log entries; a first VM data step (backup first)
Le Mans × Kalshi	Schema built (le_mans), no links on production	Tape only	None	None	No results source or model
SailGP × Kalshi, OG.com	OG.com 13 Championship Winner links, all open; Kalshi schema built, no links	Tape only	None	None	No results source or model

Buy-all is never a result. buy-all (and the replays' --buy-all) buys one YES and one NO of every market to prove it is found, priced and settled; a pair loses exactly its fees. Per the owner it is never shown in reporting or the web app, and none of the numbers above come from it.

Not listed anywhere

• MotoGP on OG.com, IndyCar, road cycling and Le Mans on OG.com: OG.com lists none of them (probed 2026-09-29).
• Downhill, XCO/XCC, track, cyclocross, BMX: no exchange lists mountain biking or track cycling. XCO results are fetchable from ChronoRace on the same weekend slugs, nothing is ingested, and there is no mass-start model.
• Alpine skiing: Kalshi and Polymarket list the February 2026 Olympic events only, no World Cup (owner's probe, 2026-10-14); re-probe in November. Horse racing (blocked by law on both US venues), Formula E and rally: out of scope.

Model status by sport

One engine, by owner rule (2026-10-11, One engine): every modelled sport prices with GlobalModel, and the schema's [model] inputs picks the core (results, laps or run times). The older classes stay as challengers. What is left to retire is tracked as Core C56: the first package release ships a single engine, which means the laps core's qualifying and grid knowledge must reach the results core first (F1 race-win Brier 0.1134 for the laps core against 0.1476 for the results core in the 2 Oct validation, Novig and the bake-off); the owner chose "compare first", and the F1 walk-forward comparison on production is Lab jobs 1 to 4 (2026-10-11; jobs 1 and 3 done, 2 and 4 failed, not re-run).

Sport	[model] inputs (core)	Results source	How it was checked	Honest status (2026-10-14)
F1	results, qualifying, grid, practice, laps, sectors (laps core, runs stored as f1_sector_sim)	FastF1 (cloud-blocked); the schedule-sync timer keeps upcoming sessions current (#169)	129 races, 2021 to Azerbaijan 2026, three stages each; after qualifying top-10 Brier 0.149 vs 0.163 grid-only. Polymarket is as sharp on race winners (sharper after qualifying, and in 2025 at every stage)	Real model; its edge is where the gap to the market is large. Live forecast run 9929 (2026-10-11, data through Singapore); live prices for rounds 18 to 20 re-run on the VM 2026-10-13 (runs 9952 to 9954, gridq+pretrain+reset, 4,000 sims)
Downhill	run_times, practice (run-times core, timed_runs)	ChronoRace web feed (primary), PDFs opt-in	43 World Cup rounds walk-forward (Evaluation)	Real model, nothing to trade it against; forecast run 1287 (2026-09-27)
NASCAR Cup	results, qualifying, fastest_lap (results core; NascarCupRace a challenger)	cf.nascar.com feeds, fetched by the forecast-refresh timer on race weekends	Taker replay on Kalshi only (above); no calibration table beside a naive baseline	Unvalidated. Race-day pricing ignores practice and track type; the replay's race spread (2.0 places) is tighter than NASCAR's measured ~7, an open owner decision; the pre-race field is the last race's (C46)
MotoGP	results, qualifying (results core; MotoGPRaceChallenger a challenger)	api.motogp.pulselive.com (unofficial, public); calendar and results step in #134, promotion pending	Race-win log loss 0.0796, no baseline comparison; Kalshi replay has no ranked result	Unvalidated, race win only. No sprints, so no season or champion pricing; no next race on production yet
Road cycling	results (results core)	PCS, on the owner's Mac; terms not cleared	Eneco Tour 2026 GC paper picks (not in the ledger), settle 18 Oct; the GC model is draft #177	Admin-only; first model, no backtest on production
IndyCar, Le Mans, SailGP	none	None cleared (IndyCar's forbids scraping)	–	Tape only

A market_implied or Plackett-Luce baseline is planned (engine roadmap E7); Plackett-Luce was tested on 7 Oct and lost to the probit engine in all four sports, so it is not the baseline. The generic leak guard (E2) is F1-only today: the NASCAR and MotoGP replays avoid leakage by pricing each race from results strictly before its start, a convention in pipelines/position_replay.py, not a checked guard.

The venues

Venue	What it is	Can we read it?	Can we trade it?	Take
Polymarket	Crypto exchange, CLOB, public Gamma + CLOB APIs, full trade and price history	Yes: markets sync/record/archive (F1 books every 60 s) and pm_sync.sh every 5 minutes for NASCAR and MotoGP links, books and trades	Paper and F$; V2 order path built, POLYMARKET_TRADING_ENABLED unset	Our first venue. The gap in F1 race listings (28 Aug to Italy) is over: production holds race links for rounds 13 to 20. Championships deep. The Bahrain GP (held in Malaysia) is a permanent hole in Probalytics' archive, not ours
Kalshi	CFTC exchange, public REST API with candlesticks and trades, per-market maker fee	Yes: markets --exchange kalshi, F1 2025–26 (Kalshi history); NASCAR, MotoGP, IndyCar with --sport; books every 5 minutes since 30 Sep; trades every 15 minutes	Paper; KALSHI_TRADING_ENABLED unset	Listed every 2025–26 F1 weekend and every NASCAR Cup race, 2 to 4 days out. Hourly candles are last trades, which spike; the recorded books fix that from 30 Sep on
OG.com	Crypto.com's US prediction market (CFTC contracts via Crypto.com Derivatives North America). Sports-first, $0.02 flat fee per contract (from reviews)	Yes, public market data with no key (api.crypto.com/fcm/v1/public/*), read by a schema (exchanges/og.toml, #44); books every 5 minutes since 8 Oct, settlements hourly; a replay venue since #88. History is short: about a month	Not without FCM onboarding and HMAC-SHA256 signing (owner waiting on an API key)	Readable and replayable, not tradable. F1 season futures and, since Malaysia, per-GP race markets; NASCAR and SailGP season futures. See OG.com
Probalytics	A third-party recorder of Polymarket and Kalshi fills and books (fills from 2022; books from 2025-11-20 on Polymarket and 2026-05-09 on Kalshi, owner's probe 2026-10-13)	Download by exact key from our links (scripts/probes/probalytics_pull.py, PR #181, open into staging); a backtest venue behind RACINGLINES_TAPE_SOURCE=probalytics	No (not a venue)	A reference tape, not a replacement: our recorder stays primary (owner, 2026-10-12). Pull order: F1 complete, then NASCAR; any other sport only on the owner's explicit go. Their terms allow in-app use, no redistribution
Private book	Our own simulated pool (racinglines live): the demo maker quoting to a simulated crowd and the demo taker	Yes, replayable to the fill	Yes (the T3 tier)	The fallback where no venue lists the event. Proves pricing, not an edge. Admin-only under the alpha layout

Other venues seen and not tracked: Robinhood's prediction-market tab (Kalshi's contracts resold), FanDuel Predicts (CME and OG contracts), PrizePicks (picks, not a book), Octagon, and Coinbase's F1 weekend markets (a temporary fallback in markets/exchange_driver.py and pipelines/live_f1.py, no exchange schema, no links on production). None adds a market Kalshi or Polymarket doesn't already list.

The gaps, ranked

Effort: S under a day, M one to three days, L a week or more. Ordered by what moves the G1 goal (CLV above +2% over 100 live bets, Oct to early Nov) and then a cell of the matrix soonest. None starts without the owner's go (Roadmap).

1. CLV on every live and replayed fill (C4) — M. racinglines backtest clv reports it for stored paper bets; the sweep stats, f1 reconcile, f1 scorecard and the weekend report do not. G1 is judged on it.
2. Calibration by probability bucket, every sport (C48) — M. Favourites, mid-field and long shots, walk-forward against results and against closing prices where recorded; a schema-set temperature or shrink if it fails, with a decision-log row. Until then NASCAR and MotoGP prices are unvalidated and no pick rests on them alone.
3. Golden sweep on the Probalytics tape (#181) — S once merged. Same jobs, two tapes; pass marks: mid within 1¢ on 95% of snapshots, weekend P&L within 10% or $5, trades within 5%. Runs on racinglines_staging_demo.
4. MotoGP's calendar on production — promotion pending (#134 in #180). The first forecast-refresh pass after the deploy files the scheduled rounds (it takes its own daily backup); then MotoGP gets a next-race price.
5. Kalshi taker on real touch prices before 30 Sep — M. Our books start 2026-09-30; Probalytics' Kalshi books start 2026-05-09, so Canada to Azerbaijan 2026 can be re-run at the touch once the pull lands.
6. NASCAR's pre-race field and start lists (C46, C5) — S to M, Mac probe first. The forecast copies the last race's field; part-time entries are missing until the weekend feed is ingested.
7. Champion replays (#95, #96) on the VM — S. Merged 30 Sep; no run is recorded in Data changes. They would be the first backtest of any OG.com cell.
8. MotoGP sprints and champion — M. Sprint classifications (a Mac probe first), then the champion classifiers and a season sim (Championship markets).
9. OG.com follow-ups — S each. Per-Grand-Prix contract rules from the Malaysia and Singapore listings, confirm the $0.02 fee, join OG.com to the U7 disagreement log. Trading only after FCM onboarding.
10. Road cycling — L, with the owner. A results source whose terms allow it, the GC model's decision-log rows (#177), then a first VM data step. Le Mans and SailGP stay tape only; XCO / XCC only if a venue ever lists it.

Done since the 3 Oct review: the three recorders on every day (owner rule 2026-10-07; trades every 15 minutes, Polymarket NASCAR and MotoGP books, backfill_tape.sh, PR #91); the Kalshi book recorder (30 Sep) and the OG.com settlement feed (hourly); the forecast-refresh timer with NASCAR results and the F1 championship forecast after each result; Polymarket MotoGP links (209); the one engine on every modelled sport (2026-10-11); road cycling on the board for admins (#163); the coverage counter backtest coverage (C0, #81); the fantasy alpha on F1's recorded books (#146, #160, #161); the data-terms decisions (2026-10-14).

OG.com

Corrected 2026-09-29. An earlier draft of this page, written from 2026 reviews, said OG.com has no public API. That is wrong. The owner found the docs and a probe from the owner's Mac confirmed them.

What it is: Crypto.com's US prediction market (CFTC contracts via Crypto.com Derivatives North America). Its API is the "Crypto.com GEN4 FCM US B2C API" at exchange-developer.crypto.com/fcm-b2c/v1/docs/, which og.com's footer links to. It is a different product from Crypto.com's crypto derivatives exchange.

Public, no key (100 requests per second per method per IP; a UAT sandbox at uat-api.3ona.co also answers):

Need	Endpoint under https://api.crypto.com/fcm/v1/public/	Note
Market listing	get-events, get-instruments, get-instrument-lifecycle-states	12,359 events, 54,676 live instruments on 2026-09-29
Prices	get-tickers, get-book	50-level book; tick 0.01, contract size 1
Trades	get-trades	Retained about one month
History	get-ticker-histories	1-minute bid/ask/last; at most 31 days and 1,000 rows per call. No candlestick endpoint for event contracts
Settlements	get-expired-settlement-price	Probed from the Mac 2026-10-06; read hourly by record_venues.sh (Exchanges)
Stream	wss://stream.crypto.com/fcm/v1/market	No auth. Channels book, ticker, trade, instrument, instrumentall, settlement, tradingstatus

Private endpoints (orders, positions) need FCM onboarding and HMAC-SHA256 request signing. The owner is waiting on an API key. No API terms page was found on the docs site; the legal pages are og.com/legal and og.com/document/legal_us.pdf (the owner checked them on 2026-09-29: the terms allow this read-only use).

What it lists (live instruments, sports grouping, 2026-09-29): college football 23,938, soccer 9,706, NFL 4,400, NHL 3,015, MLB 555, tennis 297, WNBA 152, golf 144, NBA 110, esports 52, F1 20, NASCAR 17, sailing 13, chess 2; plus politics, culture, economics, crypto, companies and climate. Not listed: MotoGP, IndyCar, any cycling or mountain bike, Le Mans.

Where it matters to us. F1 first: 20 season futures (Drivers' champion 9, Constructors' 11, expiring 2027-01-31), and per-Grand-Prix race markets that appear around a weekend: production holds 22 links each for the Bahrain GP in Malaysia (round 16) and Singapore (round 17), both closed, and 808 F1 trades since 9 Oct (2026-10-14). Liquidity is thin: on 29 Sep the Norris and Verstappen futures books were asks only at 1–4¢ (1–5k contracts), no bids, about 4 trades a week. NASCAR is 17 Cup Champion contracts (69 trades since 9 Oct) and SailGP 13 Championship Winner contracts. Nothing else on our list. The F1 futures are model-priced (the season forecast); the NASCAR champion is priced only offline by the uncalibrated #93 season sim; SailGP is tape only.

Connector built as a schema, not code (PR #44, 2026-09-29). exchanges/og.toml describes the API and one generic driver reads it, so OG.com needs no OG-specific code. It syncs the futures and race markets into market links, records the tape, minute prices and books (every 5 minutes since 8 Oct, record_venues.sh), reads the settlement feed hourly, and prints a fair-price indicator (model fair vs quote, net of the fee). It is read-only and off by default (RACINGLINES_OG_VENUE=1 shows the venue). Since #88 it is also a replay venue (nascar replay --venue og, --venue all). What is left:

1. Per-Grand-Prix contract rules in exchanges/og.toml, now that two race weekends' instruments have been recorded (their names are unchecked). Effort S.
2. The fee: $0.02 per contract is from reviews; confirm it in OG.com's fee schedule.
3. Cross-venue comparison (U7): OG.com joins the championship comparison once markets/disagree.py, which knows Polymarket and Kalshi only, is generalised. Effort S to M.
4. A replay on its tape: the champion replays (#95, #96) and a race replay on the Malaysia and Singapore markets; nothing has been run.
5. Trading: only after FCM onboarding, signed private endpoints and the validation rules. Not started.

Spot-check, 2026-09-29 (VM database, read-only)

Historical: this is the state before the first VM syncs of the new sports. The matrix above supersedes it.

Run from the owner's Mac against the live APIs and the VM's database, before the deploy of the overnight batch. Kalshi F1: 3,793 links, 37,416 trades, 16,083 price rows, archived. OG.com, NASCAR, MotoGP and IndyCar on Kalshi: 0 links, no archive folder, as expected until the VM runs the syncs. Polymarket NASCAR, MotoGP, IndyCar, Kalshi cycling and Le Mans: no schema. Kalshi's cycling and Le Mans series are still listed. Trap: total Kalshi trades by joining market_trades to market_links on token_id only. condition_id is the whole event ticker there, and joining on it overcounts about 20 times (Kalshi history).

What to verify on a device

The counts in the matrix come from production's database (the MCP server's sql, admin-only, or racinglines backtest coverage on the VM). The venues' own listings need a device that reaches them (the VM or the owner's Mac; the cloud sandbox does not). Each is a read-only GET.

Kalshi: every sports series that touches our sports, then the open and settled counts for cycling.

curl -s 'https://api.elections.kalshi.com/trade-api/v2/series?category=Sports&limit=200' \
  | python3 -c 'import sys,json; [print(s["ticker"], "|", s["title"]) for s in json.load(sys.stdin)["series"]]' \
  | grep -iE 'cycl|tour|giro|vuelta|mtb|mountain|downhill|f1|nascar|motogp|indy|sail|mans|formula|rally|ski'

for s in KXCYCLING KXCYCLINGSTAGE KXCYCLINGTEAM KXNASCAR KXMOTOGP KXINDYCAR; do
  for st in open settled; do
    n=$(curl -s "https://api.elections.kalshi.com/trade-api/v2/markets?series_ticker=$s&status=$st&limit=200" | python3 -c 'import sys,json; print(len(json.load(sys.stdin)["markets"]))')
    echo "$s $st $n"
  done
done

Polymarket: what it lists for each sport.

for q in cycling "tour de france" giro vuelta "mountain bike" nascar motogp indycar "formula 1" sailgp "le mans"; do
  echo "== $q"
  curl -s "https://gamma-api.polymarket.com/public-search?q=$(python3 -c "import urllib.parse,sys; print(urllib.parse.quote(sys.argv[1]))" "$q")" \
    | python3 -c 'import sys,json; d=json.load(sys.stdin); [print(" ", e.get("slug"), "active" if e.get("active") else "closed", e.get("volume")) for e in d.get("events", [])[:15]]'
done

OG.com: this was probed on 2026-09-29. To refresh its F1 and NASCAR counts:

curl -s 'https://api.crypto.com/fcm/v1/public/get-events' | python3 -c 'import sys,json; d=json.load(sys.stdin); print(json.dumps(d)[:1500])'

Write the answers into the matrix above, with the date.

Calendar: what could be listed before year end

Production's scheduled events on 2026-10-14, with the links already stored for each.

When	Event	Listed today	Our side
18 Oct	NASCAR Phoenix (Freeway Insurance 500)	Kalshi 200 links, Polymarket 43	Results core price on the board (unvalidated); recorded by the timers
23–25 Oct	F1 round 18, United States GP	Polymarket 266 links; Kalshi none yet (lists 2 to 4 days out)	A, C and K paper; fantasy trading on the recorded books; forecast run 9929
25 Oct	NASCAR Talladega (YellaWood 500)	None yet	As Phoenix
30 Oct – 1 Nov	F1 round 19, Mexico City; NASCAR Martinsville (Xfinity 500)	Polymarket 266 links for Mexico; nothing for Martinsville yet	As above
6–8 Nov	F1 round 20, São Paulo; NASCAR Championship Race (Homestead)	Polymarket 266 links for São Paulo; nothing for Homestead yet	As above; NASCAR's champion priced only offline
~12 Nov	Sizing review		After the live weekends on real markets (Roadmap)
20–22 Nov	F1 round 21, Las Vegas (Saturday race)	Nothing yet	The CLV-gated milestone (Roadmap); check the night-time stage cut-offs
27–29 Nov, 4–6 Dec	F1 rounds 22 and 23, Qatar and Abu Dhabi	Abu Dhabi: 1 Kalshi link	Championships settle; U8 rules if a race is cancelled
Nov	MotoGP's last rounds (Qatar, Portugal, Valencia as listed on 28 Sep; not in production's calendar until #134 runs)	Kalshi 33 open MotoGP links, Polymarket 29	Replay and board price only, unvalidated
from 23 Nov	UCI Cyclocross World Cup	Probably none	Nothing
Spring 2027	Classics, Giro, UCI MTB World Cup opener	Kalshi cycling series, Polymarket	Decide over the winter from the 2026 tapes and #177