Each flag on
GET /api/v1/suspicious-trades and GET /api/v1/suspicious-trades/{id} now carries the fill: side, outcome, price, size_usd, traded_at, and market.slug. Before this change, a flag named the wallet and the market and left out the fill, so you had to find the same trade in the large-trade history.sideisBUYorSELL, orUNKNOWNwhen the stored side is neither.outcomeis the traded outcome’s name, ornullwhen the stored outcome is not trusted or the market has no label for it.priceis the fill price of that outcome, from 0 to 1.size_usdis the fill’s USD notional, ornullwhen the trade stored none.traded_atis when the trade filled.created_atstays and is that same instant.market.slugis the market’s slug. It is omitted when the market has none.- The deprecated
/api/v1/insider-radaraliases carry the same fields. Theget_suspicious_tradesandget_suspicious_tradetools on the remote MCP endpoint and in@0xinsider/mcpdescribe them, and the next@0xinsider/sdkrelease types them.
side, outcome, price, size_usd, and traded_at on the flag.@0xinsider/sdk 0.71.0 types the game changes of November 14: GameMarket.yes_quote, of the new type GameMarketQuote, holds Polymarket’s best bid and ask for a market’s YES outcome, and GameSharpMoney.lifecycle adds "ended" and "in_play". The description of TrendingWallet.grade now says what a grade reads, without the grader’s weights.Before this release, npm served 0.70.0, which has none of these. Earlier entries on this page named 0.70.1 for some of them; 0.70.1 was never published, and those changes ship in 0.71.0.Breaking for TypeScript code that checks lifecycle exhaustively. yes_quote is optional, so code that does not read it compiles unchanged.What to change: Run npm install @0xinsider/sdk@0.71.0, and handle ended and in_play wherever you switch on lifecycle. Both come with a withheld verdict.GET /api/v1/games/{event_slug}/sharp-money now answers lifecycle: "in_play" from kickoff until the game is over, with verdict.kind: "withheld", every pct at 0 and no is_leader side. live now means a game that has not started. The in_play value also appears on List games with expand=sharp_money, in the get_game_sharp_money and list_games tools on the remote MCP endpoint and in @0xinsider/mcp, and in the CLI and @0xinsider/sdk types.Before this change a game in play kept lifecycle: "live" and a verdict computed from holdings valued at in-game prices, which follow the score, so the lean tracked whoever was ahead. On November 14 at 22:27 UTC, cfb-mich-ore-2026-11-14 at halftime, Michigan 7 and Oregon 21, answered leans Oregon Ducks at 95%.outcomes[].usd,outcomes[].wallets,total_usdandtrader_countare still set.usdvalues holdings at in-game prices.- A game with no graded money still answers
no_sharp_money. in_playmeans the scheduled start has passed, or the provider reports the game live, and the game is not over. It is not proof that play is under way: a suspended, postponed or cancelled game past its scheduled start answersin_playtoo, and so does a game with no live score feed.- When the game is over,
lifecyclebecomesended, as before.
lifecycle value other than live, ended, settled or indeterminate, and for clients that read pct or is_leader during a game.What to change: Accept in_play in lifecycle, and treat it as you treat a withheld verdict: show who holds each side, not a side. Read pct and the verdict before kickoff.The file from
GET /api/v1/trader/{address}/export/download now sends weight as null in every trader.skill_components entry. Each entry keeps its three keys in the same order (key, percentile, weight), and skill_components keeps its place in trader.Before this change, weight held a number: the signal’s share of the wallet’s grade. That share is internal to how grades are computed, and no other response sends it. key and percentile, the wallet’s standing on each signal, are unchanged.- A file built before the release keeps its values until it expires a day later.
weight as a number.What to change: Stop reading trader.skill_components[].weight, or accept null. Read percentile for the wallet’s standing on a signal.GET /api/v1/games/{event_slug}/sharp-money now answers lifecycle: "ended" for a game that is over while its markets have not settled, with verdict.kind: "withheld", every pct at 0 and no is_leader side. The ended value also appears on List games with expand=sharp_money, in the get_game_sharp_money and list_games tools on the remote MCP endpoint and in @0xinsider/mcp, and in the CLI and @0xinsider/sdk types.Before this change such a game kept lifecycle: "live" and a verdict computed from holdings valued at prices that already reflected the result. On November 14, cs2-lun-tsa-2026-11-14 ended at 18:38 UTC and two hours later still answered leans Spirit Academy at 99%, which showed where the shares ended up, not where the wallets stood before the result.outcomes[].usd,outcomes[].wallets,total_usdandtrader_countare still set.usdvalues holdings at current prices.- A game with no graded money still answers
no_sharp_money. - When the markets settle,
lifecyclebecomessettled, orindeterminatefor void, partial or contradictory outcome facts, as before.
lifecycle value other than live, settled or indeterminate.What to change: Accept ended in lifecycle, and treat it as you treat a withheld verdict: show who holds each side, not a side.GET /api/v1/leaderboard now orders every B wallet by its score from the latest hourly grading pass, as the November 13, 2026 change did for S and A. Before this change, a pass republished a B wallet’s score only when it moved far enough, so the B section compared scores from different passes, and a wallet could sit above one whose record ranked higher in the same pass.GET /api/v1/trader/{address}, batch traders, trader context and the whale trade endpoints return the samescore, so a B wallet’sscorethere is the latest pass’s too.- For a B wallet, the
rankingdata-quality group’sas_ofis now the latest pass or later, somax_age_smeasures it against one pass. - The first pass after the change can move B wallets a few places within their grade. No wallet changes grade because of it.
- C, D and F wallets keep the October 14 rule.
GET /api/v1/games and GET /api/v1/games/{event_slug} now return markets[].yes_quote: Polymarket’s best bid and best ask for the market’s YES outcome, with source and observed_at. The list_games and get_game tools on the remote MCP endpoint advertise it in their output schema, and the @0xinsider/sdk 0.71.0 types name it GameMarketQuote.Before this change, a game nobody had traded carried no price at all: prices.provider.state read incomplete with display_price_pair_unavailable, while Polymarket held resting orders on both sides. On November 14, ufc-jak-tabdul-2026-11-21 had a 60 cent bid and a 78 cent ask on Polymarket and no price on the game.best_bidandbest_askare Polymarket’s figures, from 0 to 1. Each isnullwhen Polymarket reports no order on that side. We compute no midpoint.sourceis alwaysgamma.observed_atis when we read the board data the quote came from, as forliquidity.- The object is omitted when Polymarket reports neither side, and on a finished game read after it leaves the live board.
- In a two-outcome market the NO book mirrors the YES one, so it is not repeated. List game markets with
expand=quotesstill returns each outcome’s book with depth.
yes_quote when prices.provider.state is incomplete.A live game whose holders have not been read no longer answers
no_sharp_money. GET /api/v1/games/{event_slug}/sharp-money now reads that game’s holders and returns the verdict from that read, and List games with expand=sharp_money answers status: "not_computed" for it. The get_game_sharp_money and list_games tools on the remote MCP endpoint and in @0xinsider/mcp, the CLI and @0xinsider/sdk return the same answers.Before this change both endpoints served status: "computed" with verdict.kind: "no_sharp_money", total_usd: 0 and money_as_of: null for such a game. On November 14, three live games answered that way while A and B wallets held them, and answered split and all_on five minutes later.- The first per-game read of such a game can answer
503withread_model_warmingandRetry-After: 3while the holders are read. It is the503this endpoint already returns for a game that is not on the current board. - When that read cannot reach the holders, the per-game answer is
not_computedas well. - On a live game,
no_sharp_moneynow always hasmoney_as_ofset: the holders were read and no graded wallet holds the game.
not_computed.What to change: Nothing if you handle not_computed and retry a 503 after its Retry-After. For a not_computed game on the list, read the game with Game sharp money for a measured answer.GET /api/v1/trader/{address} and the batch trader read now return forecast_evidence truncated to two decimal places: 0.87, where they returned 0.871244635193133. The remote MCP get_trader tool, @0xinsider/mcp and @0xinsider/sdk pass the same value through.Backward compatible. The field is still a number from 0 to 1 with the same meaning, and it is still omitted when the wallet has no forecasting signal. A value is at most 0.01 lower than before.What to change: Nothing, unless you compare forecast_evidence past its second decimal.The
suspicious_only parameter of GET /api/v1/large-trades and GET /api/v1/large-trades/history now reads “return only the trades the unusual-activity check flagged” in the OpenAPI spec, the @0xinsider/sdk 0.71.0 types, the remote MCP endpoint’s tools/list and @0xinsider/mcp 3.24.1. Before this change, the description printed the score a flag starts at.min_suspiciononGET /api/v1/suspicious-tradesand theget_suspicious_tradestool says that every listed trade already meets the flag floor, so a lower value changes nothing.- The deprecated
whale-tradesandinsider-radarpaths and their MCP aliases carry the same text. - 0xinsider.com/llms.txt and 0xinsider.com/llms-full.txt say the same.
The
smart_label field on GET /api/v1/markets/explore, in the OpenAPI spec and the @0xinsider/sdk 0.71.0 types, now says in words when it names an outcome: the graded money’s share of that side sits well past the side’s price-implied share, enough graded money backs the leaning side that a few small positions cannot name it, and the side is not a long shot holding nearly all of the graded money. Before this change, the description printed the lean’s point gap and dollar floor. Those are internal settings, and 0xinsider no longer publishes them.The agent API reference at 0xinsider.com/llms-full.txt also describes forecast_evidence on GET /api/v1/trader/{address} in words: the 0-1 share of forecast_score that the trader’s own resolved markets support, which rises toward 1 as more of them resolve and never reaches it. It no longer prints the formula.Backward compatible. Only descriptions changed. Both fields, their values and every response are unchanged.What to change: Nothing.GET /api/v1/games/{event_slug} now returns 503 with error.reason read_model_warming and Retry-After: 5 for a slug it cannot find while a live or upcoming board source is unavailable. Before this change, it returned 404 not_found, “No covered game has that event_slug”, for a game that existed on the source that did not answer, and the CLI exited 2, which tells a script not to retry.- A
404now means every board source answered.GET /api/v1/gamesnames an unavailable source incoverage.sources_unavailable. - A finished game is still served from its stored record during the outage.
- The sub-resources (
/markets,/sharp-money,/trades,/holders) find a game from stored markets and are unchanged.
503 shape is the one this route already returned when the whole board did not answer; @0xinsider/sdk retries it and the CLI exits 6.What to change: Nothing, if you already retry a 503 after Retry-After. A client that treats a 404 from this route as final can keep doing so.The
grade field of TrendingWallet, in the OpenAPI spec and the @0xinsider/sdk 0.71.0 types, now says what a grade reads: realized profit on settled sports and esports markets, ranked against other graded wallets, with win rate, forecasting edge, record length, drawdown and consistency counting too. S, A and B need banked profit on these markets, and the top grades also check record depth, money at stake, edge and the worst drawdown. Before this change, the description printed profit’s share of the grade as a percentage and called the other signals a tie-breaker. Those are the grader’s internal settings, and 0xinsider no longer publishes them.Backward compatible. Only the field’s description changed. The field, its values and every response are unchanged.What to change: Nothing.@0xinsider/mcp 3.24.0, the CLI and local MCP server, now reads five routes it could not reach before, so its stdio server lists the same 79 tools as the remote one. Before this change, 3.23.0 served 74 tools over stdio, and its CLI knew none of these routes, not even through api call.get_fair_pricereadsGET /api/v1/fair-price: one game outcome’s Polymarket fair price, to check a sportsbook bet against. In the CLI:api call getFairPrice.get_clvreadsGET /api/v1/tools/clv: a Polymarket price or sportsbook odds graded against the market’s stored close. In the CLI:market clv <condition_id>.list_price_movesreadsGET /api/v1/price-moves. In the CLI:price-moves list, andmarket price-moves <condition_id>for one market.get_trader_marketsreadsGET /api/v1/trader/{address}/markets: one wallet’s sports and esports markets, settled ones included. In the CLI:traders markets <wallet>.get_pick_of_the_day_wallet_marketsreadsGET /api/v1/pick-of-the-day/picks/{pick_id}/wallets/{address}/markets: the decided markets behind a pick wallet’s record. In the CLI:api call getPickOfTheDayWalletMarkets.- Tool text now matches the remote server:
get_pick_of_the_daynames a sealed pick’s proof fields and each listed wallet’safter_release, and saysholder_countandtop_gradecount S and A grade wallets only;get_pick_of_the_day_archivesays the same oftop_grade;search_marketssays a game is one result;get_positionspoints toget_trader_marketsfor settled markets.
npx -y @0xinsider/mcp@3.24.0 or npm install --global @0xinsider/mcp@3.24.0.GET /api/v1/edge-hunter with view=observations, cohort=wider_holder, and admission=most_held_48h or admission=most_held_48h_by_type now tells you how far each row was from an admission=most_held_48h place. Before this change, a most_held_48h_by_type row could not show whether it missed those 8 places by one market or by fifty.most_held_48h_position, on each row: the market’s place in its sport’sadmission=most_held_48hqueue, every market type ranked together, from the same snapshot. Places 1 to 8 are the selection.nullmeans the market has no place, because its event already held 4 markets ranked above it.most_held_48h_cutoffs, at the top level: for each sport the response covers, the last marketadmission=most_held_48htook, with itsposition,condition_id,indexed_graded_holders,volume, andgame_start_time.- The page now states the full order: the most
indexed_graded_holders, then the mostvolume, then the soonestgame_start_time, then the lowestcondition_id, at most 4 per event. No market type is weighted or dropped before ranking. - The
get_edge_hunterMCP tool returns the same body, and the@0xinsider/sdk0.70.0 types carry both fields asEdgeHunterObservation.most_held_48h_positionandEdgeHunterMostHeld48hCutoff.
most_held_48h_by_type card was from a place, compare its most_held_48h_position and indexed_graded_holders with its sport’s entry in most_held_48h_cutoffs.GET /api/v1/leaderboard now orders every S and A wallet by its score from the latest hourly grading pass. Before this change, a pass republished an S or A wallet’s score only when it moved far enough (the October 14, 2026 change), so an idle wallet kept a score from a pass hours or days old. The board then compared scores from different passes, and a wallet could sit above one whose record ranked higher in the same pass.GET /api/v1/trader/{address}, batch traders, trader context and the whale trade endpoints return the samescore, so an S or A wallet’sscorethere is the latest pass’s too.- For an S or A wallet, the
rankingdata-quality group’sas_ofis now the latest pass or later, somax_age_smeasures it against one pass, not against the wallet’s last large score move. - The first pass after the change can move S and A wallets a few places within their grade. No wallet changes grade because of it.
- B, C, D and F wallets keep the October 14 rule.
score now reflect one pass.GET /api/v1/fair-price now fails stale_quote only when the book has gone more than 60 seconds without being confirmed current, and quote carries confirmed_at and confirmed_age_seconds. Before this change, the check counted from observed_at, when the bid or ask last changed, so a deep pre-game book that nobody moved read stale_quote while it still matched Polymarket’s.quote.confirmed_atisobserved_ator later: the last time the live Polymarket stream confirmed the unchanged book. The stream confirms it about every 10 seconds while it stays connected and its copy of the book still matches.quote.confirmed_age_secondsis the seconds fromconfirmed_attoas_of, andstale_quotereads it.- A dropped or stalled stream, a book whose copy stopped matching, and a book read from Polymarket’s REST API (
source: clob_rest) confirm nothing. For themconfirmed_atisobserved_at, so the check is the one it was before. observed_atandage_secondskeep their values. Their descriptions now say that they date the last change of the bid or ask.- The
ETagleaves out both new fields, as it leaves outage_seconds. - The
get_fair_priceMCP tool returns the same body, and the@0xinsider/sdk0.69.0 types carry both fields. - The
Server-Timingheader now names the read’s stages besideapi;dur:game_slug(a market asked bycondition_id),scope_gate,member_read,child_scope, andquotes.
quote. An answer that read stale_quote on a current book now reads ok or its next failed check.What to change: Nothing. To show how current a quote is, read confirmed_age_seconds. age_seconds says how long the bid and ask have stood.Each holder row on
GET /api/v1/market/{condition_id}/top-holders can now carry category_records_as_of. A row has it when the wallet’s category record was recounted after the daily rebuild. A wallet’s record is now recounted about an hour after its markets settle, so one row’s record can be newer than data.category_records_as_of. Before this change, no field said which rows were newer.- A row’s
category_records_as_ofis when the recount of its wallet began. The row’scategory_win_record,category_win_rate, andcategory_avg_priceare as of that time. - A row without it is as of
data.category_records_as_of, the start of the daily rebuild, as before. That time is also the oldest any row’s record can be. - The MCP tool
get_market_top_holdersreturns the same rows, and the@0xinsider/sdk0.67.0 types carry the field. records_as_ofon Pick of the Day picks keeps its meaning, the start of the daily rebuild behind the listed wallets. Its description now says that it is the oldest their records can be.
category_records_as_of, and use data.category_records_as_of when the row has none.New:
GET /api/v1/tools/clv grades one Polymarket price, or sportsbook odds, against the stored close of a sports or esports market outcome. Before this change a client read the close from GET /api/v1/market/{condition_id}/close and did the arithmetic itself.- Send
condition_id,outcome, and eitherprice(0 to 1) oroddswithformat(americanordecimal).side=sellgrades a sold Polymarket price;placed_atmarks a bet placed at or after kickoff asin_play. clvcarriescents(price entries only),points,price_pctandev_at_close_pct, gross of fees. Each isnullunlessstatusismeasured, never0.statusismeasured,in_play,pendingorunavailable;unavailable_reasonaddsanchor_stalewhen the kickoff moved after the close was stored.- Pro and Max keys. The
get_clvMCP tool and the CLI’smarket clvserve the same body, and@0xinsider/sdk0.68.0 addsgetClv. A walkthrough on real NBA games is at 0xinsider.com/learn/how-to-use-clv-calculator.
GET /api/v1/trader/{address}/categories now counts a wallet’s settled markets in its category and scope records about an hour after they settle. The same is true of category_win_record and category_win_rate on GET /api/v1/market/{condition_id}/holders, GET /api/v1/market/{condition_id}/top-holders, and GET /api/v1/games/{event_slug}/holders. Before this change, the records were rebuilt once a day, so a record could leave out up to a day of settled markets.computed_aton/trader/{address}/categoriesis when the wallet’s records were last recounted. It can be any time of day, and it is often later than the daily rebuild.category_records_as_ofon top-holders is still the start of the daily rebuild. A wallet’s record in the same response is as of that time or later, never earlier.- Most wallets are recounted within about an hour of a settlement. A wallet with thousands of settled markets waits longer, up to 12 hours.
- No field changes meaning. Win records, win rates, and edges follow the same rules as before.
New:
GET /api/v1/price-moves and GET /api/v1/market/{condition_id}/price-moves list the stored market_price_moved detections, newest first, kept 90 days. Before this change a move existed only as a webhook delivery or a stream frame, so a client that missed one could not read it back.- Each item is the
market_price_movedpayload unchanged, plusmove_id,episode_started_atandoutcome_after. One item per market, window and move: the largest step of the move. outcome_afterholds outcome 0’s midpoint 15 and 60 minutes after the move and at a pre-game move’s kickoff, withheld_15m,held_60mandheld_kickoff:truewhen the midpoint kept at least half of the move. Anullflag means not measured;*_unavailable_reasonsays why.- Filters: one of
league,event_slugorcondition_id, plusmin_move_cents,window_seconds,phase,held,min_book_usd_within_2c,sinceanduntil. A page can hold fewer thanlimitmoves whilehas_moreistrue; follownext_cursoruntil it isfalse. - A
condition_idorevent_slugthat no market we store carries is a404witherror.codenot_found, never an empty page. A market or event outside sports and esports is a404witherror.reasonmarket_out_of_scope. - The record starts when storing began, on the release day. The
list_price_movesMCP tool, the CLI commandsprice-moves listandmarket price-moves, and@0xinsider/sdk0.67.0 (listPriceMoves,listMarketPriceMoves) read the same moves.
market_price_moved pushes you missed, read GET /api/v1/price-moves with since 15 minutes before your last delivery’s detected_at, since a move can grow for up to 15 minutes after it starts.New:
GET /api/v1/fair-price returns the Polymarket fair probability of one game outcome, the odds it implies, the taker fee and a status that says whether the book is fit to price against. Before this change, a client had to read List game markets with expand=quotes, normalise the midpoints and check the book itself.- Name the outcome by
event_slug,market_typeandoutcome_label(withlineon a spread or total), or bycondition_idandoutcome. fair.probis the outcome’s midpoint over the sum of its market’s midpoints, or over the three Yes midpoints on a soccer 1X2 leg (three_way).fair.decimal_oddsandfair.american_oddsare the break-even price.statusisok,closed,unavailable, or the first offailed_checks:wide_spread,thin_bookorstale_quote.max_spread_c(default 3) andmin_book_usd_within_2c(default 500) set the limits, andthresholdsrepeats them.feegives the taker fee of a buy at the ask andtaker_all_in_ask, andgame.startedsays whether the kickoff has passed.- The
get_fair_priceMCP tool serves the same answer, and@0xinsider/sdk0.67.0 addsgetFairPrice. Price a sportsbook bet against Polymarket turns it into an edge and a stake.
GET /api/v1/trader/{address}/markets lists one wallet’s sports and esports markets, the same rows as its profile positions table, one page at a time. Before this change the wallet position list was GET /api/v1/positions, which leaves resolved markets out.statusisclosed(the default),open, orall.closedis the profile’s Closed tab,openis the Open tab, andallis both, newest last trade first.- Each row has
condition_id,title,category,status,cost,realized_pnl,resolved_at, andlast_trade_at. costandrealized_pnlare exact decimals (value,unit,scale,basis).nullmeans the figure is not stored, never zero.resolved_atis null when the market has not settled.- Order is
last_trade_atdescending, thencondition_iddescending. A row with nolast_trade_atcomes after every dated row, as on the profile.next_cursorbelongs to the samestatus. - The MCP tool
get_trader_markets, the CLI commandtraders markets, andgetTraderMarketsin@0xinsider/sdk0.66.0 call the same endpoint.GET /api/v1/positionsis unchanged.
GET /api/v1/market/{condition_id}/holders, GET /api/v1/market/{condition_id}/top-holders, and GET /api/v1/games/{event_slug}/holders now return wallet_age_days as an integer, the number of UTC dates from the wallet’s first trade to today (822), and the spec types it as integer. The age changes for every wallet at 00:00 UTC, once a day, as the field’s description already said.Before this change, the two holders routes returned that whole number as 822.0. The top-holders route returned the days elapsed since the first trade as a fraction (822.3558917121296 for the same wallet), so one wallet could read different ages on the two routes, and its age changed every time the board was rebuilt. On top-holders, is_new_wallet now follows the whole-day count, as it already did on the holders routes: it can turn off up to a day before 30 full 24-hour periods have passed.The remote MCP tools get_market_holders, get_market_top_holders, and get_game_holders type the field as integer. The TypeScript SDK still types it as number | null, which accepts an integer.Backward compatible for a client that accepts any JSON number. Breaking for a strict decoder that requires wallet_age_days to be a float, and for a client that read a fraction of a day from top-holders.What to change: Decode wallet_age_days as an integer, or as any number. Do not read a fraction of a day from it on any route.GET /api/v1/webhooks, GET /api/v1/webhooks/{id}, GET /api/v1/webhooks/{id}/deliveries, GET /api/v1/webhooks/{id}/verification-attempts/{attempt_id} and GET /api/v1/webhooks/events now answer Cache-Control: private, no-store, on errors and 304 included.Before this change, those reads answered Cache-Control: private, max-age=30, stale-while-revalidate=60, so a client could keep showing an endpoint for up to 90 seconds after you created, changed, or deleted it, and a client polling a verification attempt could keep reading queued after it had finished.Backward compatible. The response body is unchanged. The header is stricter, so a client that stored the list has to drop that copy.What to change: Do not store these responses. Read the list again after you create, change, or delete a webhook.