Skip to main content
These are the API changelog entries from November 15, 2026, newest first. The newest changes are on the API changelog.
Pick of the Day: hit_rate carries the implied win rate and the edge over it
GET /api/v1/pick-of-the-day/archive and GET /api/v1/pick-of-the-day/stats now send five fields on the record’s hit_rate, the ones each stats bucket already carries:
  • priced_decided: the settled picks with a recorded publication price.
  • priced_in_pct and priced_in_display: the average midpoint price of those picks when published, read as a win rate.
  • edge_pp and edge_display: the win rate over those picks minus priced_in_pct, in percentage points.
Before this change only the buckets carried the implied win rate, so the record’s 57.1% win rate had nothing to be read against. On November 15 the record’s 1,080 settled picks were all priced, at an implied 56.8%: the picks won about as often as their prices said, which is why the return on stake is near zero.
  • The fields leave out void and pending picks, as pct does. priced_in_pct, priced_in_display, edge_pp and edge_display are absent when no settled pick is priced.
  • The get_pick_of_the_day_archive and get_pick_of_the_day_stats tools on the remote MCP endpoint return the same, and @0xinsider/sdk 0.102.0 adds the fields to its types.
Backward compatible. The fields are new and every existing field is unchanged.What to change: Nothing. To show whether the picks beat their prices, read edge_display beside win_rate_display.
Game sharp money: since_start counts from the first live score when the scheduled start is late
GET /api/v1/games/{event_slug}/sharp-money and the sharp_money object on GET /api/v1/games?expand=sharp_money now count since_start from the earlier of the game’s scheduled_at and its first live score. Before, since_start.since was always scheduled_at.A scheduled start can be hours late. On November 15, two women’s college basketball games read scheduled_at 21:00 UTC and 00:00 UTC the next day, while both started at 19:00 UTC. While they were in play, since_start.since was later than since_start.as_of, so the reading covered no time and read bought_usd 0 with verdict no_sharp_money.
  • since_start.since is never later than since_start.as_of. A game in play before its scheduled_at with no live score recorded has since_start: null, which claims nothing.
  • scheduled_at is unchanged: it is still the provider’s scheduled start.
  • @0xinsider/sdk 0.101.0 carries the new description; its types are unchanged.
Backward compatible. The object’s shape and every value it can carry are unchanged.What to change: Nothing. If you compute the reading’s window from scheduled_at, read since_start.since instead.
Games: women's college basketball periods read as quarters, never OT in regulation
GET /api/v1/games and GET /api/v1/games/{event_slug} now return a women’s college basketball (cwbb) game’s status.period as its quarter, and a new status.provider_period carries the feed’s own period string on every game.Before this change status.period was the feed’s label, and the feed names a women’s game’s four quarters with the men’s halves vocabulary. A game in its 3rd quarter read OT, and a game in its 4th read End OT. On November 15 at 20:59 UTC, 11 of 24 live games read OT or End OT, and none of them was in overtime.
  • The feed sends no frame between the end of the 3rd quarter and the final, so End Q3 stays through the 4th quarter and any overtime, with scores_freshness unknown.
  • A frame whose status is final or stopped keeps the feed’s label in status.period.
  • Every other league’s status.period is unchanged and equals provider_period.
  • The get_game and list_games tools on the remote MCP endpoint return the same, and @0xinsider/sdk 0.100.0 types provider_period.
Backward compatible for every league except cwbb. Breaking for a client that read OT or End OT in a cwbb status.period as overtime: that reading was wrong.What to change: Nothing to show the right quarter. To keep the feed’s raw label, read status.provider_period.
Game sharp money: since_start states no buys with bought_usd 0, so null claims nothing
GET /api/v1/games/{event_slug}/sharp-money and the sharp_money object on GET /api/v1/games?expand=sharp_money now send since_start on a binary game in play when no S, A or B grade wallet bought since the scheduled start: bought_usd 0, trader_count 0, verdict.kind no_sharp_money, last_trade_at null, and every outcome with bought_usd 0, pct 0, avg_buy_price null and is_leader false.Before this change since_start was null both when none of those wallets bought and when the game had no reading at all, so a caller could not tell “nothing bought” from “not read”. A soccer 1X2 game in play read null, the same as a binary game with no buys, while graded wallets had bought on it.
  • null now claims nothing about what was bought: before the start, after the game, on a soccer 1X2 game, which the reading does not cover, and when no reading was available.
  • @0xinsider/sdk 0.99.0 carries the new description; its types are unchanged.
Backward compatible. Every value is one the field already documents, and the object’s shape is unchanged.What to change: Treat since_start: null as “no reading”, not as “no buys”. To show that nothing was bought, read since_start.bought_usd == 0.
Traders: stale after a held market settles or a trade moves the open positions, and expand=trust agrees with data_quality
GET /api/v1/trader/{address} and POST /api/v1/traders/batch now report three cases as stale that read fresh before:
  • sports_record reads stale when a market the wallet held settled after its last completed sync, even with no trade since. Settling removes the wallet’s open row for that market, and the settled row arrives with the next read of the wallet’s Polymarket positions, so until then pnl.in_scope, stats.markets_traded and stats.win_rate may leave that market out. The group’s owner is traders.last_synced, its as_of is synced_at, and its reason starts a market the wallet held settled after its last completed sync. On November 15, 198 wallets with a record and no trade since their sync held a market that settled in one six-hour window.
  • positions reads stale when the wallet traded after the positions clock (last_active later than the group’s as_of): pnl.unrealized comes from the open positions read then, and a new fill changes none of them. On November 15, 17,347 of 174,272 wallets with stored positions were in this case.
  • With expand=trust, a field served with a clock now takes its data_quality group’s verdict: freshness.status is the group’s fresh or stale, with the group’s as_of and max_age_s. Before, every such field read fresh whenever a clock existed, so trust.markets_traded read fresh beside a stale sports_record, and trust.synced_at read fresh beside a stale sync group.
  • @0xinsider/sdk 0.98.0 updates the descriptions.
Backward compatible. stale and every field are values these routes already document. A max_age_s request ceiling works as before. trust.*.freshness may now carry max_age_s, a key its schema already allows.What to change: Nothing, if you already handle stale. If you show the record or pnl.unrealized as current whenever their group reads fresh, these wallets now read stale, with the figures as of as_of. If you read trust.*.freshness.status, it now matches data_quality.
Traders: last_active counts trades seen on Polymarket's live trade stream
GET /api/v1/trader/{address} and POST /api/v1/traders/batch now serve last_active as the later of the wallet’s most recent recorded trade and its most recent trade seen on Polymarket’s live trade stream. A trade later than synced_at marks the sports_record group of data_quality.field_groups stale, as before, so the group now reads stale within about two minutes of such a trade.Before this change last_active was the most recent recorded trade only, and we record a small fill of many wallets only at their next sync. On November 15 a grade D wallet made 16 fills of $2.59 to $5.68 after its sync; none was recorded, so last_active stayed before synced_at and sports_record read fresh at -$686.38 while Polymarket showed -$180.12.
  • The stream carries the taking side of each match. A wallet whose only trades since its sync filled resting orders is not seen there, and its last_active waits for the next sync, as before.
  • The leaderboard entry’s last_active still reads the recorded trade alone, re-read hourly, so a trader body’s last_active can be later.
  • last_active can now be sent for a wallet with no recorded trade, when the stream saw one.
  • The get_trader tool on the remote MCP endpoint returns the same, and @0xinsider/sdk 0.97.0 updates the field’s description.
Backward compatible. The field keeps its type and format, and stale is a value the group already documents.What to change: Nothing. If you compare a trader body’s last_active with the leaderboard entry’s, expect the trader body’s to be later at times.
Game markets: sharp_money fills the lines on your page first, and not_computed holds between reads
GET /api/v1/games/{event_slug}/markets with expand=sharp_money changes in two ways. The list_game_markets tool on the remote MCP endpoint reads the same route.
  • A read now asks for the split lines on its own page, not every line of the game. On November 15, page 1 of nfl-min-gb-2026-11-15 had 0 of its 27 split lines computed six minutes after the first read, while 27 lines on pages 2 to 5, which nobody had read, were.
  • A line the game page does not split (a player prop, a line beyond the seven around its family’s main line, a closed or settled line) now reads not_computed on every read. Before, a read that came more than about 2.5 minutes after the game’s last read could report those lines unavailable: the same page read 100 of 100 unavailable, then 73 not_computed 29 seconds later.
  • The status description gives each status one meaning, and says to read the same page again after the 30 seconds its Cache-Control max-age states.
Backward compatible. The field and its three values are unchanged. Fewer lines read unavailable when no read will compute them.What to change: Nothing. A client that pages through a game gets each page’s lines computed after it reads that page, not before.
Market close: every moneyline market gets its closing line, not only markets a tracked wallet bought
GET /api/v1/market/{condition_id}/close and POST /api/v1/markets/close/batch now measure the close of every moneyline market, each leg of a soccer 1X2 included, whose game starts after this release. Before this change a market was measured only when a wallet we track had bought it before it settled. On November 15, 44,283 of the 46,336 moneylines that settled in the previous 30 days answered unavailable with unavailable_reason not_captured.
  • A moneyline whose game started after this release answers pending until its close is measured, then measured, within the same 30-day window after kickoff as before.
  • A market that is not a moneyline is measured as before: when a tracked wallet bought it before it settled.
  • Moneylines whose game started before this release keep not_captured.
  • The get_market_close tool on the remote MCP endpoint returns the same. The not_captured description in the spec and in @0xinsider/sdk 0.96.0 now says which markets the capture reads.
Backward compatible. No field or value was added or removed; more markets reach pending and measured.What to change: Nothing. A client that treated not_captured on a moneyline as final can read it again after kickoff.
Games: a finished game carries its traded volume
GET /api/v1/games/{event_slug} and GET /api/v1/games with status=ended, cancelled or postponed now return volume on a game read after it leaves the live board (freshness.source stored). The figure is the game volume summed from the stored volumes of the game’s markets, by the same game-lines rule the game page uses: basis game_lines, source gamma, unit taker_shares, and no observed_at.Before this change volume was omitted on every such game. On November 15 all 100 games of a status=ended page had none, while the game page for nba-det-mia-2026-11-14 showed 1,774,869.
  • The stored figure can trail Polymarket’s final figure by a few shares until the next read of the closed markets.
  • A match nobody traded has no volume on Polymarket and still omits volume.
  • @0xinsider/sdk updates the field’s description.
Backward compatible. The field keeps its type, and an omitted volume was already a documented state.What to change: Nothing. A client that skipped finished games when ranking by volume.value can now include them.
Game sharp money adds since_start: what graded wallets bought during a game
GET /api/v1/games/{event_slug}/sharp-money and the sharp_money object on GET /api/v1/games?expand=sharp_money now carry since_start on a binary game in play: what S, A and B grade wallets bought on its moneyline since the scheduled start, by side, in dollars put in.
  • since_start holds since (the scheduled start), as_of (when the trades were read), verdict, bought_usd, trader_count, last_trade_at, and outcomes[] with side, label, bought_usd, pct (of bought_usd, summing to 100), avg_buy_price (0 to 1, null with no buy on that side) and is_leader.
  • Its verdict is the share of those dollars alone, in the same vocabulary as the game’s verdict: all_on, leans, split, no_sharp_money or withheld, and withheld when the side it would name holds too little money. It makes no claim against the price: in play, buys follow the score, so read each side’s avg_buy_price beside it.
  • It moves only when one of those wallets trades. at_kickoff, the split before the start, still never moves after the start.
  • null before the start, after the game, on a soccer 1X2 game, and when none of those wallets bought since the start.
  • The three verdicts share one schema, GameSharpMoneyVerdict, in the OpenAPI document. The get_game_sharp_money, get_game and list_games tools on the remote MCP endpoint return the field, and @0xinsider/sdk 0.94.0 types it.
Backward compatible. The new key sits beside the old ones; a client that ignores it reads every game as before, verdict.kind withheld in play beside at_kickoff.What to change: Nothing. To show what graded wallets did during a game, read since_start.outcomes[] and its avg_buy_price.
Traders: the sports record reads stale when the wallet traded after its last sync
GET /api/v1/trader/{address} and POST /api/v1/traders/batch now mark the sports_record group of data_quality.field_groups as stale when the wallet traded after its last completed sync, that is, when last_active is later than synced_at. The group’s owner is then traders.last_synced, its as_of is that sync, and its reason says what the figures leave out. The group covers pnl.in_scope, stats.markets_traded and stats.win_rate. The top-level data_quality.status follows, as for any stale group. The get_trader tool on the remote MCP endpoint returns the same.Before this change the group always read fresh. Its clock, trader_product_scope_records.computed_at, moves when a fill or a resolution makes us total the record again. The record’s realized P&L, cost basis and market counts come from the per-market values the sync reads from Polymarket, and a new fill changes none of them. On November 15 one wallet’s record was timed 11:16 UTC and labeled fresh while it still showed the realized P&L of its 04:06 sync: +$361.38, where Polymarket showed -$686.38. When we checked, about one wallet in nine with a record had traded after its last sync.The definition of stale grows by one case. A group without max_age_s can now be stale when the source changed after as_of, and reason names the change. Until now such a group was never stale.Backward compatible. stale and every field are values these routes already document. A max_age_s request ceiling works as before: it accepts a stale body whose oldest clock is inside the ceiling, and data_quality.as_of is unchanged, because the sync group already carries the same clock.What to change: Nothing, if you already handle stale. If you show pnl.in_scope or the win rate as current whenever sports_record reads fresh, a wallet that traded since its sync now reads stale. Its figures are as of synced_at until the next sync, and opening the wallet’s profile on 0xinsider requests one.
Edge Hunter: the one-way share leaves out wallets whose positions cannot be read
GET /api/v1/edge-hunter, GET /api/v1/sports/pre-game-sides and its alias GET /api/v1/sports-edge-signals now compute directional_confidence over the backed-side graded dollars whose wallets could be classified. Hedged and two-sided dollars still count against it, and dollars held by wallets no source could classify are left out of both sides. It is null when those wallets hold more than 10% of the backed-side graded dollars.Before this change, unclassified dollars counted against the share, so it moved with whether a wallet’s positions could be read on that compute. On November 15 one Jaguars vs. Titans side read 0.8682 and then 0.1114 two and a half minutes apart, with the same backed dollars and the same hedged holders.
  • one_way_holder_count, hedged_holder_count and one_way_graded_usd are unchanged.
  • directional_rank_score follows the new share. A null share ranks as it did before, as if it were 0.
  • GET /api/v1/sports/pre-game-side-observations records the new share on rows observed after this change; earlier rows keep the share they recorded.
  • The get_edge_hunter tool on the remote MCP endpoint returns the same value, and @0xinsider/sdk 0.92.0 updates the field’s description.
Backward compatible. The field keeps its type and its range of 0 to 1, and null was already a documented value.What to change: Nothing. A client that treats null as zero should treat it as unknown.
Holder edges count every settled market of every size, and say so in every_size
category_price_edge and category_scope_price_edge on GET /api/v1/pick-of-the-day holders and on GET /api/v1/market/{condition_id}/top-holders now count every settled market of every size where the wallet’s side and price are known. markets, hit_rate and avg_price describe that one set, and edge is still hit_rate minus avg_price.Before this change the three figures counted only markets where the wallet put in $20 or more, and the site printed the rule as `$20+`.
  • A new boolean, every_size, is true on figures that count every size. It is false, or absent, on figures that still count markets of $20 or more: holders frozen on a pick published before this change, and a wallet whose record the daily rebuild has not yet recounted.
  • sample=edge on GET /api/v1/pick-of-the-day/picks/{pick_id}/wallets/{address}/markets lists every settled market of every size behind the row’s edge. On a pick frozen before this change, listed can differ from printed, and matches_printed is then false.
  • The get_pick_of_the_day_wallet_markets tool on the remote MCP endpoint and @0xinsider/sdk 0.91.0 describe the same rule and type every_size.
  • A wallet’s markets count usually grows, and its hit_rate, avg_price and edge can move either way. category_avg_price, the leaderboard’s price_edge, and a profile’s Edge still count markets of $20 or more.
Backward compatible. No field was renamed or removed, and every_size is additive.What to change: Nothing. If you print a holder’s market count with a size rule, read every_size and print the rule only when it is not true.
Game sharp money: settled games withhold the verdict and carry the split before the start
GET /api/v1/games/{event_slug}/sharp-money now answers a settled game with verdict.kind: "withheld", every pct at 0 and no is_leader side, and returns at_kickoff, the split as the site last recorded it before the scheduled start. The get_game_sharp_money tool on the remote MCP endpoint returns the same. List games with expand=sharp_money is unchanged: it answers a settled game not_computed, as before.Before this change every settled game answered all_on its winner with pct 100. A settled game’s holders are the winning side’s wallets that have redeemed in profit, so the losing side always read 0: on November 15 all five settled NBA games of November 14 read that way, though graded wallets had held both sides before the start.
  • outcomes[].usd, outcomes[].wallets and trader_count are unchanged: the winning side’s realized profit and wallets.
  • at_kickoff counts both sides’ holders and has the same fields as on an in_play game. It is null when no split with graded money was recorded before the start.
Backward compatible. No field or value is added or removed. A settled game now answers values the route already documents. @0xinsider/sdk 0.90.0 carries the new descriptions in its types.What to change: If you read a settled game’s verdict or pct as where graded money stood, read at_kickoff instead.
Trading styles: two_sided counts markets bought on both sides
The two_sided trading style now counts the markets a wallet bought on both sides, as the page has always described it. Before this change it counted only the markets where the wallet still held shares of both outcomes. A settled market rarely holds both, so on November 15 no wallet carried two_sided, and a wallet that bought both sides in most of its markets still read 0%.Styles are recomputed over the next hours. Some wallets move to two_sided, or gain it in their other qualifying patterns, so GET /api/v1/leaderboard with strategy=two_sided, and every strategy_type field, can now return wallets where it returned none.Backward compatible. The field names and the six style values do not change.What to change: Nothing. A client that filtered on two_sided and read an empty list now gets wallets.
Max accounts get four times Pro's API rate, webhooks, integration keys and exports
A Max account’s allowances on the API are now four times Pro’s: 400 requests and 10,000 batch items a minute, 40 webhook endpoints, 40 named integrations, and 80 exports a day with 20 per wallet an hour. Before this change Max had 200 requests and 5,000 batch items a minute, and the same 10 endpoints, 10 integrations, 20 exports a day and 5 an hour as Pro. Pro’s allowances and both monthly quotas are unchanged.
  • RateLimit-Limit and X-RateLimit-Limit read 400 on a Max key.
  • The daily and hourly export counts are shared by trader exports and dataset jobs, as before.
  • large_export_policy.rate_limit on the export snapshot now states the caller’s plan: 80 and 20 on Max. It read 20 and 5 for every key.
  • On Pro, the 400 at the endpoint limit now reads maximum 10 webhook endpoints allowed on Pro; Max allows 40, and the 409 at the integration limit reads 10 integrations are active, the Pro limit; Max holds 40. Revoke one before creating another.
  • @0xinsider/sdk 0.89.0 carries the new rates in its endpoint descriptions; its types are unchanged.
Backward compatible. Limits only rose, and every field and status code is the same.What to change: Nothing. A client that hard-codes 200 a minute for Max can read RateLimit-Limit instead.
Games: an ITF singles match in play reads live with its score
GET /api/v1/games and GET /api/v1/games/{event_slug} now report an ITF singles match in play as status.state live, with each player’s set score in competitors[].score, coverage.scores available, and freshness.source live. It is listed under status=live. The get_game and list_games tools on the remote MCP endpoint return the same.Polymarket’s live data never marks an ITF singles match live, and we read its state from API-Tennis. Before this change that state did not reach these routes: on November 15 all four ITF matches in play read scheduled with no score, and status=live&sport=tennis listed only the one ATP match. For these matches status.provider_status is API-Tennis’s own status, such as Set 1.Backward compatible. Every value is one the routes already document.What to change: Nothing. A client that skipped ITF games because they never read live can now follow them.
Market candles: 5m and 1h fill the newest buckets from Polymarket's 60-second prices
GET /api/v1/market/{condition_id}/candles at resolution=5m and resolution=1h now fills the buckets Polymarket has not published yet from its 60-second prices, which trail the clock by 4 to 11 minutes. Before this change the newest 5m candle started 16 to 31 minutes before the read, and the newest 1h candle started 1 to 2 hours before it, because Polymarket publishes those buckets in batches.
  • Each 5m and 1h candle now carries basis. provider_bucket is Polymarket’s own bucket price, so o, h, l and c are equal, as before.
  • provisional_minutes marks a bucket after Polymarket’s newest published one, in a window that ended under 3 hours ago. It is built from Polymarket’s 60-second prices inside it: o is the first minute’s price, h the highest, l the lowest, and c the last.
  • A provisional candle changes. The bucket in progress gains minutes until it closes, and a later read replaces the candle with Polymarket’s own bucket once it is published.
  • A window that holds a provisional candle is re-read after at most 60 seconds. A read that is not cached of a window that ended under 3 hours ago spends 2 of your account’s cold-read share, 6 a minute in bursts of up to 30.
  • 1d and 1w candles have no basis, and their answers are unchanged. The get_market_candles tool on the remote MCP endpoint returns the same field, and @0xinsider/sdk 0.88.0 types it as "provider_bucket" | "provisional_minutes".
Backward compatible. basis is a new field, and every candle keeps t, o, h, l and c.What to change: Nothing. To draw the newest candles differently, check basis == "provisional_minutes", and replace candles by t on each read rather than appending.
Market candles add 5m and 1h resolutions from Polymarket's own price buckets
GET /api/v1/market/{condition_id}/candles now accepts resolution=5m and resolution=1h, read from Polymarket’s price history at that bucket width. Before this change both values returned 400, and the only resolutions were 1d and 1w, built from stored daily prices.
  • Each 5m or 1h candle is one Polymarket bucket: one price, so o, h, l and c are equal, unrounded. A bucket Polymarket did not price has no candle.
  • Without from, 5m covers the last day and 1h the last 7 days, ending now or when the market resolved. A window spans at most 15 days, 5m reaches back 60 days and 1h 90; a wider or older window returns 400.
  • data.window_start, data.window_end and data.observed_at appear only at 5m and 1h. A window that reaches the present is re-read after at most 60 seconds.
  • Polymarket publishes these buckets in batches, so the newest candle trails the clock: on November 15 the newest 5m candle started 16 to 31 minutes before the read, and the newest 1h candle was the last completed hour. For the current price, read GET /api/v1/market/{condition_id}/snapshot.
  • A 5m or 1h read that is not cached spends your account’s share of cold reads, 6 a minute in bursts of up to 30. Over it the answer is 429 with error.reason cold_read_limited and Retry-After; cached windows still answer. If Polymarket does not answer, the route returns 503 with reason read_model_warming.
  • The get_market_candles tool on the remote MCP endpoint accepts the same values, and @0xinsider/sdk 0.87.0 types them and the three new fields on getMarketCandles.
Backward compatible. 1d stays the default, and 1d and 1w answers are unchanged. The new fields appear only at the new resolutions.What to change: Nothing. To chart a game in play, request resolution=5m, pair it with the snapshot for the latest price, and handle 429 cold_read_limited by waiting for Retry-After.
Pick of the Day: each pick carries its market type, as the archive row does
GET /api/v1/pick-of-the-day now returns sports_market_type on each pick your key can open, in view=full and view=summary, with or without pick_id. It is the backed market’s type, such as moneyline, spreads or totals, as recorded with the pick: the same value GET /api/v1/pick-of-the-day/archive returns on that pick’s row.Before this change only the archive row carried the market type, so a client reading one pick could not tell a moneyline from a spread or a total.
  • The field is omitted when the market states no type, and on a pick your key cannot open, which the read leaves out as before.
  • The get_pick_of_the_day tool on the remote MCP endpoint returns it, and @0xinsider/sdk 0.86.0 adds it to PickOfTheDay and to the view=summary pick type. The tool’s description drops the sentence that repeated what its view argument says; no argument changed.
Backward compatible. One optional field is added; no field was removed or changed.What to change: Nothing. To label a pick’s market, read sports_market_type from the pick instead of its archive row.
Market snapshot: the reference now says what a first-read cache_miss means and when to read again
GET /api/v1/market/{condition_id}/snapshot reads top of book from our book cache. An open market that no reader holds has no cached book yet, so its first read answers top_of_book.status unavailable with reason cache_miss on each outcome. That same read starts the market’s live book. In our reads on November 15 the book was served within 620 ms of the first read on 7 of 7 markets.The OpenAPI document now describes each top_of_book.reason value, and the get_market_snapshot MCP tool description carries the same instruction: on cache_miss for an open market, read again after 1 second. A closed market starts no book, so its cache_miss stays.Backward compatible. No field, value or status changes. Only the descriptions do. @0xinsider/sdk 0.85.0 carries the new description in its types.What to change: If your client reads cache_miss as “this market has no book”, read the snapshot once more after 1 second before you decide.
Remote MCP: list fewer tools with toolsets in the endpoint URL
The remote MCP endpoint takes a new optional query parameter, toolsets, on POST /api/v1/mcp. With it, tools/list answers only the tools you name, so a client that hands the list to a model can send it far less than the full 79 tools and 662 KB.
  • ?toolsets=core lists 16 tools, about 183 KB: account, the leaderboard and one trader, games and their sharp money, market search and one market, the day’s picks and their record, large trades and flagged trades.
  • ?toolsets=all lists every current tool: 70, about 602 KB, the full list without the 9 deprecated aliases.
  • A group name lists that group’s current tools: account, combos, content, games, large_trades, markets, picks, platforms, reports, suspicious_trades, traders or webhooks. Separate several names with commas, as in ?toolsets=games,picks, or repeat the parameter.
  • A shortened list never holds a deprecated alias tool.
  • toolsets changes tools/list only. tools/call still runs any tool by name, and initialize is the same for every URL.
  • An unknown or empty name answers JSON-RPC error -32602 with the valid names in its message.
  • @0xinsider/sdk 0.84.0 sends it from client.mcp(request, { toolsets: "core" }); an array is joined with commas. The local @0xinsider/mcp package has no toolsets and lists its whole catalog.
Backward compatible. Without the parameter, tools/list answers the same full list as before, aliases included.What to change: Nothing. To shorten the list, add ?toolsets=core, or the groups you use, to the endpoint URL in your client’s configuration. If your client signs in with OAuth and sends its own endpoint URL as the token resource, keep the URL without the query for that value.
Positions: a wallet's age and is_new_wallet now count from its first trade
GET /api/v1/positions now counts trader.wallet_age_days and trader.is_new_wallet from the wallet’s first trade, the rule large trades and the site’s New wallets view use, and each row carries that first trade as trader.first_trade_at. Before this change both fields counted from when 0xinsider first recorded the wallet, so a wallet that had traded for months could read as new: 98 of 4,822 wallets on the feed on November 15 did.
  • trader.first_trade_at: the wallet’s first trade 0xinsider recorded, the same value large trades return. null until the wallet has one.
  • trader.wallet_age_days: days since first_trade_at. Absent when first_trade_at is null, as before when the age was unknown.
  • trader.is_new_wallet: true when first_trade_at is under 30 days old. It stays a boolean and is false when first_trade_at is null, where large trades return null.
  • The get_positions tool on the remote MCP endpoint returns the same fields, and @0xinsider/sdk 0.83.0 types trader.first_trade_at.
Backward compatible. One field is added and no field, type or required field changed. The values change meaning: wallets 0xinsider recorded after their first trade now show a larger wallet_age_days and, if that trade is 30 or more days old, is_new_wallet false.What to change: Nothing, unless you stored either field and compare it with older reads. To filter new wallets, keep reading trader.is_new_wallet; it now selects the same wallets as large trades.
Pick of the Day stats: the OpenAPI document and the MCP output schema name the row shapes they repeated
The schemas for GET /api/v1/pick-of-the-day/stats, PickOfTheDayStats and PickOfTheDayStatsSections, repeated the same row shapes inline 28 times. The OpenAPI document now defines each shape once and both schemas refer to it by name. No response changes.
  • Eight new components, listed under Objects: PickOfTheDayStatsBucket (the rows of categories, sports, tiers, months, price_bands, market_types and total_sides), PickOfTheDayStatsDay, PickOfTheDayStatsVoid, PickOfTheDayMarketExpectations, PickOfTheDayOutcomeStats, PickOfTheDayStatsRecords, PickOfTheDayStatsStreaks and PickOfTheDayAskComparison.
  • The get_pick_of_the_day_stats tool on the remote MCP endpoint advertises an 18,599-byte outputSchema, down from 40,637, and tools/list is 662,236 bytes, down from 684,275. The schema’s references stay inside the tool’s own $defs.
  • @0xinsider/sdk 0.82.0 exports the eight shapes as types.
Backward compatible. No field, type or required field changed, and a response that passed the old schema passes the new one.What to change: Nothing. If you generate code from the OpenAPI document, your generator now emits these component names where it made up a name for each inline shape, so regenerate and update any reference to the old generated names.
Large trades: each row carries the wallet's age, and history takes new_wallets
Every large trade on GET /api/v1/large-trades, GET /api/v1/large-trades/history and GET /api/v1/large-trades/{id} now carries trader.first_trade_at and trader.is_new_wallet. History also takes new_wallets=true, which returns only trades by wallets whose first recorded trade is under 30 days old, the wallets the site’s New wallets view shows.
  • trader.first_trade_at: the wallet’s first trade 0xinsider recorded. null until the wallet has one.
  • trader.is_new_wallet: true when first_trade_at is under 30 days old, false when older, null when first_trade_at is null.
  • With new_wallets=true the window starts 30 days back, because no older trade can belong to such a wallet. A from earlier than that is moved up to it.
  • GET /api/v1/large-positions now fills trader.username the way the large-trade rows do: the wallet’s username first, else its Polymarket pseudonym, and no name when the only one is the address. Before this change a wallet without a profile name could read as Polymarket’s placeholder, its address followed by a number.
  • The MCP tools follow the routes: the large-trade tools return both fields, get_large_trades_history and get_whale_trades_history take new_wallets, and get_large_positions names wallets the same way. The four large-trade list tools describe their arguments more briefly; no argument, type, default or bound changed. @0xinsider/sdk 0.82.0 types the fields and the parameter.
  • A trade GET /api/v1/events/feed/since expands carries the same two fields.
  • The deprecated /api/v1/whale-trades routes return the same fields and take the same parameter.
Backward compatible. Fields and a parameter were added. A position’s trader.username can now differ from the value it had before for the same wallet.What to change: Nothing. To follow new wallets, send new_wallets=true to large trades history, or read trader.is_new_wallet on any large-trade row.
Pick of the Day archive: each row carries the kickoff, market and backing figures the single pick returns
GET /api/v1/pick-of-the-day/archive with view=full now returns, on each row whose backed side your key can open, the kickoff, the market and the backing figures GET /api/v1/pick-of-the-day?pick_id= returns for the same pick. Before this change those fields were only on the single pick, so reading them for the whole record took one request per pick, more than 1,000.
  • kickoff: when the backed game starts, as recorded with the pick. The single pick states the new start when a game moved after publication.
  • condition_id: the backed market, the value the single pick returns as entry_authorization.condition_id.
  • sports_market_type: the market type, such as moneyline, spreads or totals, when the market states one.
  • backed_side_share, sharp_positioned_usd, sharp_wallet_count, side_graded_wallet_count and side_graded_usd: the single pick’s values for the same pick, omitted where it omits them.
  • A row with backed_side_locked carries none of them, and view=summary rows are unchanged.
  • The get_pick_of_the_day_archive tool on the remote MCP endpoint returns the same fields, and @0xinsider/sdk 0.81.0 adds them to PickOfTheDayArchiveEntry.
Backward compatible. Eight optional fields are added; no field was removed or changed.What to change: Nothing. To group the record by game day or market type, or to read its backing figures, read the new fields from the archive instead of each pick.
@0xinsider/mcp 3.25.1: the stdio server carries the shorter tool descriptions
@0xinsider/mcp 3.25.1 is on npm. Its stdio server now advertises the same tool descriptions as the remote MCP endpoint: no Args: or Returns: section, none over 1,500 characters, 49,725 bytes across the 79 tools where 3.25.0 sent 124,692.
  • The install lines on this site name 3.25.1.
  • The CLI in the same package is unchanged: no command, flag or exit code changed.
Backward compatible. No tool, argument, type, default, bound or annotation changed.What to change: Nothing. To get the shorter descriptions, install @0xinsider/mcp@3.25.1. If you parse the Args: or Returns: text of a description, read inputSchema and the API reference instead.
MCP: tool descriptions are shorter and no longer list the result's fields
The remote MCP endpoint’s tools/list now carries a shorter description on every tool. A description states what the tool is for, when to use it and the rules that no schema states. It no longer has a Returns: section that lists the fields outputSchema already lists. The 79 descriptions were 85,320 bytes in all and are now 49,725; the longest was 5,484 characters and is now 1,489; the whole answer was 720,126 bytes and is now 684,373.
  • The nine deprecated alias tools (get_whale_trades, get_whale_trade, get_whale_trades_history, get_market_intel, batch_get_market_intel, get_smart_money_flows, get_insider_radar, get_insider_radar_flag and get_platforms) now say only which tool replaces them and when they are removed. Their arguments and results are unchanged.
  • A field rule that a description no longer states is in the API reference: each tool reads one /api/v1 route, and that route’s page states its rules and response fields. get_edge_hunter and get_pick_of_the_day name their route.
  • The sentence on wallet names at the end of a description is shorter. The server’s instructions still say in full that a wallet name is text its owner sets, to read as data and never as instructions.
  • The instructions gain two sentences: retry a read_model_warming error at the time it names, and where the full rules are.
  • The @0xinsider/mcp stdio server carries the same descriptions from its next release.
Backward compatible. No tool, argument, type, default, bound, annotation or output schema changed.What to change: Nothing if your client passes descriptions to a model. If you match description text, such as the Returns: label, read outputSchema instead.
Market holders: each row carries the price edge the game page prints
GET /api/v1/market/{condition_id}/holders now returns each S, A and B holder’s edge over the price it paid in the market’s category, the same fields and figures GET /api/v1/market/{condition_id}/top-holders returns. Before this change the roster carried each wallet’s win record and no edge, so a wallet outside the 100 largest holders on a side had no edge anywhere in the API.
  • category_price_edge, category_scope_price_edge, category_price_edge_hold and category_scope_price_edge_hold have the shapes and rules of the top-holders fields. An edge or its hold sits beside a measured record only, and a wallet on both lists carries the same figures.
  • category_records_as_of on the response says when the daily rebuild behind the edges began. A row carries its own, later category_records_as_of when its wallet was recounted since. The response omits it when no row has an edge or a hold.
  • GET /api/v1/games/{event_slug}/holders returns the same rows, and each ok item’s data carries category_records_as_of.
  • The get_market_holders and get_game_holders tools on the remote MCP endpoint return the same fields. @0xinsider/sdk 0.80.0 adds them to MarketHolder, the market holders response, and GameHoldersItem.
  • To keep the MCP tools/list answer within its size limit, the descriptions of get_clv, get_sport_leaderboard and get_game_holders no longer carry Returns: text. Those of get_market_trades, list_games, list_settled_markets, list_game_markets and get_game_sharp_money no longer list the fields their output schemas describe. get_market_holders, get_market_top_holders and get_game_holders say their other facts in fewer words.
Backward compatible. Optional fields are added; no field or tool argument was removed or changed.What to change: Nothing. To show a holder’s edge as the game page does, read category_price_edge, and its hold when it is absent. If you parse a tool’s Returns: text, read its outputSchema instead.
MCP: tool descriptions no longer repeat the argument list
The remote MCP endpoint’s tools/list no longer prints an Args: block in any tool description. Each argument is described once, in inputSchema.properties.<name>.description. Before this change, 73 of the 79 descriptions listed every argument a second time, and the answer was 752,842 bytes; it is now 720,327.
  • 37 argument descriptions in 24 tools gained the facts that only the Args: block had stated, such as the wth_ cursor prefix of the large-trade history tools.
  • The rest of each description, its Returns: text included, is unchanged.
  • The @0xinsider/mcp stdio server carries the same descriptions from its next release.
Backward compatible. No tool, argument, type, default, bound, annotation or output schema changed.What to change: Nothing if your client reads arguments from inputSchema. If you parse the Args: text of a description, read inputSchema.properties instead.
Price moves: in-play and phase-unknown moves are kept 14 days
GET /api/v1/price-moves and GET /api/v1/market/{condition_id}/price-moves now keep a move by its phase: a pre_game move for 90 days, as before, and an in_play or unknown move for 14 days. Before this change every move was kept 90 days. In-play and unknown moves were 93% of the 427,438 moves stored on November 14.
  • With phase=in_play, the walk ends 14 days back: has_more is false on the page that reaches it, whatever since says.
  • Without phase, pages older than 14 days list pre_game moves only.
  • The first moves past 14 days are deleted from November 27, 2026, 14 days after storing began.
  • The list_price_moves tool on the remote MCP endpoint answers the same way. Its description no longer lists the arguments, which its input schema already describes. @0xinsider/sdk 0.79.0 updates the listPriceMoves() description only.
Backward compatible. No field, parameter or value changed.What to change: If you study in-play or unknown moves over more than 14 days, read them within 14 days and keep your own copy.
Market top holders: each row carries the price edge the game page prints
GET /api/v1/market/{condition_id}/top-holders now returns each holder’s edge over the price it paid in the market’s category, the figure the game page’s Top holders row prints beside the wallet, or the reason the page prints none. Before this change the row carried the wallet’s win record in the category and no edge, so an agent could not read the figure the page shows.
  • category_price_edge: { markets, hit_rate, avg_price, edge }, the shape display_holders[] on Pick of the Day already uses. hit_rate is how often the side the wallet bought won and avg_price the average price it paid, over its settled markets in the category of $20 or more at a known entry price; edge is hit_rate minus avg_price. All four are 0-1 probabilities at full precision: edge 0.022 is the page’s +2.2 pp.
  • category_scope_price_edge: the same edge over the game or league category_win_rate_game names, beside that record only.
  • category_price_edge_hold and category_scope_price_edge_hold: { hold, record? } when the page prints no edge beside a measured record, with hold one of category_loss, too_few_markets, sample_mismatch or no_markets.
  • The edges are as of the row’s category_records_as_of, or the response’s when the row has none. The page’s 95% range is 1.96 times the square root of avg_price * (1 - avg_price) / markets, either side of edge.
  • The get_market_top_holders tool on the remote MCP endpoint returns the same fields, and @0xinsider/sdk 0.78.0 adds them to MarketTopHolder. GET /api/v1/market/{condition_id}/holders and the game holders batch carry no edge.
Backward compatible. Four optional fields are added; no field was removed or changed.What to change: Nothing. To show a holder’s edge as the page does, read category_price_edge, and its hold when it is absent.
Games: an ended game no longer reports an in-play match status
GET /api/v1/games and GET /api/v1/games/{event_slug} now omit status.match_status and status.provider_status on a game whose status.state is ended when the last status the live-score feed sent is one it sends only before the end. Before this change, a game we recorded as ended after its feed went quiet kept that status: on November 15, itf-cross1-bulgar1-2026-11-14 read state: ended beside match_status: in_progress and provider_status: Set 2, and 53 ended games read in_progress or scheduled.
  • The statuses omitted beside ended are scheduled, in_progress, halftime, penalty_shootout and delayed. A final, awarded, forfeit, postponed, cancelled, suspended or unknown status is still served.
  • freshness.scores_observed_at is now when the score was last observed. Recording the end of such a game no longer moves it. A game ended that way before this change keeps the time of that write, because the earlier observation was not kept.
  • The list_games and get_game tools on the remote MCP endpoint answer the same way. @0xinsider/sdk 0.77.1 updates the three field descriptions only.
Backward compatible. Both status fields were already optional; no field was added or removed.What to change: Nothing if you read status.state for whether a game is over. If you read match_status alone, treat a missing value on an ended game as no status from the feed.
Games: a game held in play with no update for 3 hours reads suspended, in more sports
GET /api/v1/games and GET /api/v1/games/{event_slug} now report status.state suspended and status.live false for a soccer, hockey, basketball, American football or cricket game, or a series in one of eight esports titles, that Polymarket still lists in play when no live-score feed reports it and its period and score have not changed for 3 hours. Before this change only tennis had that rule, so a game whose feed had stopped read live for hours or days. On November 15 a Valorant series read live 16 hours after its kickoff, and two Sheffield Shield matches 4 days after theirs.
  • The 3 hours start no sooner than 2 hours after kickoff, or 6 for esports, so a game reads suspended only after the longest game measured in its sport has ended. The esports titles are CS2, Dota 2, League of Legends, Valorant, Mobile Legends: Bang Bang, Rainbow Six Siege, Overwatch and Call of Duty.
  • Cricket counts the 3 hours only after the match’s scheduled days: 5 for a match with a draw market, 1 for a limited-overs match, and one fewer when the score shows runs. A multi-day match stays live through its overnight breaks.
  • Every other sport or esports title, such as baseball, MMA, table tennis or StarCraft 2, keeps Polymarket’s in-play flag.
  • window=now leaves such a game out, as it does every suspended game whose kickoff is more than 3 hours old.
  • The game reads live again when its period or score changes, or when a live-score feed reports it.
  • The list_games and get_game tools on the remote MCP endpoint answer the same way. @0xinsider/sdk 0.77.0 updates the two field descriptions only.
Backward compatible. suspended was already a value of status.state; no field was added or removed.What to change: Nothing if you already handle suspended. If you count every game Polymarket lists in play as live, read status.state instead.
Game sharp money: in-play and ended games carry the split before the start
GET /api/v1/games/{event_slug}/sharp-money now returns at_kickoff on an in_play or ended game: the split as the site last recorded it before the scheduled start, valued at that moment’s prices. The same object 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/sdk 0.76.0 as GameSharpMoney.at_kickoff.Before this change a game in play or waiting to settle showed pct 0 and no leader on every side, and nothing said where graded wallets stood when it began.
  • at_kickoff has as_of, when the split was recorded, plus verdict, total_usd, trader_count and outcomes[] with side, label, usd, pct and is_leader.
  • outcomes[].pct and verdict at the top level keep describing the game now, so they stay 0 and withheld.
  • at_kickoff is null on a live, settled or indeterminate game, and when no split with graded money was recorded before the start. Soccer games have none more often.
  • A postponed or cancelled game past its scheduled start answers in_play, as before, and its at_kickoff is the split before that start. The lifecycle descriptions in the OpenAPI spec and the MCP tools now say so.
Backward compatible. One field is added; no field was removed or changed.What to change: Nothing. To show where graded money stood going into a game that has started, read at_kickoff and its as_of.
Suspicious trades take from and to, so one day's flags are one request
GET /api/v1/suspicious-trades now takes from and to, bounds on the fill time traded_at, the same bounds GET /api/v1/large-trades/history takes. Before this change the list took no time window, so reading one day’s flags meant paging sort=recent until traded_at passed the day.
  • from is inclusive and to is exclusive, both RFC 3339. They apply in either sort and either mode, before the page is cut, so a bounded page is never short.
  • A value that is not RFC 3339 returns 400 with error.param from or to. A to that is not after from returns 400 with error.param to.
  • In mode=stable the cursor carries the window: change it while paging and the next page returns 400 with error.reason cursor_expired. Cursors issued before this change keep working.
  • The deprecated alias GET /api/v1/insider-radar takes both parameters too.
  • The get_suspicious_trades tool on the remote MCP endpoint and in @0xinsider/mcp 3.25.0 takes from and to, and @0xinsider/sdk 0.75.0 types them on listSuspiciousTrades.
  • In the CLI from 3.25.0, 0xinsider api call listSuspiciousTrades --query from=2026-11-12T00:00:00Z --query to=2026-11-13T00:00:00Z reads one day’s flags, and api list names both parameters.
Backward compatible. Two optional parameters are added; a request without them returns what it returned before.What to change: Nothing. To read one day’s flags, send from at 00:00 that day and to at 00:00 the next.
Get one game reads a started game Polymarket never finished, instead of 404
GET /api/v1/games/{event_slug} now returns 200 for a game whose kickoff has passed and that Polymarket never reported in play or finished. Before this change it answered 404 “No covered game has that event_slug”, although GET /api/v1/markets/search listed the game’s markets as active.
  • The game comes from its stored record, with freshness.source stored, and has the same shape as a finished game read after it leaves the live board: no prices, yes_quote, volume, liquidity, title or tournament.
  • status.state is suspended or delayed when that was the provider’s last verdict, and unknown otherwise. status.ended is false and status.ended_at is null.
  • GET /api/v1/games does not list these games. A game whose kickoff is still ahead and that the board does not carry still answers 404.
  • While a board source is unavailable, such a slug still answers 503 read_model_warming, because the missing source could carry the game in play.
  • The 404 message now reads “a game stays readable here by its slug once its kickoff has passed, finished or not”. The get_game tool on the remote MCP endpoint answers the same way.
Backward compatible. unknown, suspended and delayed were already values of status.state; no field was added or removed.What to change: If you treat a 404 from this route as “the game does not exist”, nothing. If you treat every 200 as a game that is live, upcoming or finished, handle status.state unknown and read each market’s current book from Market snapshot.
Combo leg markets say whether the game has ended
Each leg market on GET /api/v1/combos/fills and GET /api/v1/combos/{condition_id} now carries game_ended. Before this change a client could only read kickoff_date_passed, and a game that kicks off late in the UTC day is still in play after midnight UTC, so a past kickoff date did not say the game was over.
  • game_ended is true when the game the market belongs to has ended, from the same stored game state GET /api/v1/games/{event_slug} reads. A market in a child event, such as a soccer game’s more-markets event, reads its parent game.
  • It is false while the game is scheduled, in play or postponed.
  • It is null when no state is stored for the game, when the leg has no local market, or when the market has no event. Some games have no live feed, so null means unknown, not in play.
  • kickoff_date_passed keeps its meaning. Its description now says that true does not mean the game is over.
  • The get_combo_fills and get_combo tools on the remote MCP endpoint return the field, and @0xinsider/sdk 0.74.0 types it on the leg market.
Backward compatible. One field is added to each leg market; no field was removed or changed.What to change: If you read kickoff_date_passed as “the game is over”, read game_ended instead. A pending leg with game_ended or market_closed true is waiting on its result. A pending leg with game_ended null is unknown, not finished.
Game market sharp money: a line with no trade keeps its figures for up to 15 minutes
GET /api/v1/games/{event_slug}/markets with expand=sharp_money now keeps a line computed for up to 15 minutes after its positions were read, as long as the line has not traded since that read. Before this change, every line turned unavailable 5 minutes after its read, traded or not, and at a busy hour many lines turned unavailable before they could be read again.
  • money_as_of still dates the read, so on a computed line it can now be up to 15 minutes old.
  • A line that trades is read again, or turns unavailable, within about 5 minutes of the trade.
  • usd is valued at the price when the figures were computed, so on an older line it can differ from the line’s price now.
  • The list_game_markets tool on the remote MCP endpoint returns the same figures.
Backward compatible. No field was added, removed or renamed; more lines read computed where they read unavailable before.What to change: Nothing. If you need a recent read, check money_as_of before you use a line’s figures.
Combo leg markets carry the kickoff and whether that date has passed
Each leg market on GET /api/v1/combos/fills and GET /api/v1/combos/{condition_id} now carries game_start_time and kickoff_date_passed. Before this change a client could not tell a pick whose kickoff date has passed from one that kicks off today, and a postponed game’s question can still show the earlier date.
  • game_start_time is the kickoff stored on the market, in UTC. It is null when the leg has no local market or the market has no kickoff.
  • kickoff_date_passed is true only when that kickoff’s calendar date is before today (UTC). It is false on the kickoff’s own date, including while the game is in play, and false when there is no kickoff.
  • state stays pending until the market resolves. kickoff_date_passed does not change the result.
  • The get_combo_fills and get_combo tools on the remote MCP endpoint return both fields, and the next @0xinsider/sdk release types them on the leg market.
Backward compatible. Two fields are added to each leg market; no field was removed or changed.What to change: Nothing. To tell a past-date open pick from one that kicks off today, read kickoff_date_passed and game_start_time, not the date in the question.
Game sharp money: a side with too little graded money before kickoff answers withheld
GET /api/v1/games/{event_slug}/sharp-money now answers verdict.kind: "withheld" before kickoff when the side an all_on or leans verdict would name holds too little graded money to carry a claim. outcomes[].usd, pct and is_leader stay set, so the response still shows where the money is. The same verdict appears on List games with expand=sharp_money, and in the get_game_sharp_money and list_games tools on the remote MCP endpoint and in @0xinsider/mcp.Before this change, such a game answered all_on or leans on any amount of graded money. In games that kicked off from November 2 to 14, more than half the sides named before kickoff held that little, and those sides won about as often as their prices said.
  • lifecycle stays live, and no other field changes.
  • leans_over_price and no_clear_lean are unchanged.
Backward compatible. withheld is an existing value, and no field was added or removed.What to change: Nothing. To show the leading side on these games yourself, read is_leader and usd.