GET /api/v1/markets/search now leaves out Polymarket’s polyv2 copy of a game when the game itself is stored, so each game is one result. Before this change q=lakers&status=active returned both nba-lal-phx-2026-11-15 and polyv2-nba-lal-phx-2026-11-15, two Polymarket events for the same fixture with the same title and start time.- A copy is an event whose slug starts with
polyv2and whose remainder is a stored game with a moneyline market. Get game already answers404for such a slug; search now agrees with it. - A
polyv2event with no stored original is still returned. - A copy’s own Protocol V2 markets no longer come back from search, so their
condition_ids are not in its results; the stored game’s markets are. - Rows are removed after the page is cut, so a page can hold fewer than
limitresults whilehas_moreistrue. Keep paging untilhas_moreisfalse. - The
search_marketsMCP tool returns the same results.
polyv2- slugs itself can keep doing so. A client that stops paging on a short page should page on has_more instead.New:
GET /api/v1/pick-of-the-day/picks/{pick_id}/wallets/{address}/markets lists the decided markets behind a pick wallet’s printed count. Before this change a pick printed a wallet’s record as a bare count, such as “won 57.2% of 119”, with no way to see the markets in it.sample=edgelists the settled markets of $20 or more behind the edge;sample=recordlists every closed position behind the any-size record.withinpicks the row’s league or title (scope) or its sport (category).- Each market carries
side,price_paid,stake_usd,result,realized_pnl_usd,settled_atand a Polymarket link. Pages run newest first; passnext_cursorascursor. basis: "at_release"lists the markets settled byas_of, when the pick’s figures were counted.printedis the figure the pick shows andlistedthe list’s own count, withmatches_printed.listed.status: "counting"means read again in a few seconds.- The pick’s access rule applies: a pick your key cannot open returns
404. Theget_pick_of_the_day_wallet_marketsMCP tool serves the same list, and@0xinsider/sdk0.66.0 addsgetPickOfTheDayWalletMarkets.
Each wallet a full pick lists in
display_holders on GET /api/v1/pick-of-the-day now carries after_release: whether it held, added, cut or left its position before the game, and whether it redeemed after settlement. Before this change a listed wallet showed only its share count at release, so a client could not tell a wallet still in the position from one that sold an hour later.stateisheld,added,reducedorexitedfor the time from release to the game’s start, andredeemedonce the wallet redeemed backed shares after settlement.unavailablemeans the facts could not be read;unavailable_reasonsays why. Never read it asheld.shares_changeandshares_afterare the net backed shares gained after release and the position that leaves.other_shares_changeis the same for the other outcome.last_change_atis the wallet’s latest fill, merge or split before the game.redeemed_sharesandredeemed_atgive the redemption. On pick 2178, Harrydubois bought 1,100 more shares 14 minutes after release and two listed wallets redeemed after the market settled.- The share-change window ends at the game’s start (
changes_through), so trading during the game does not count. Redemptions count until settlement plus 72 hours (redemptions_through), when the facts are stored andfrozenturnstrue. - The release
sharesand the rest of the listed wallet are unchanged. Theget_pick_of_the_dayMCP tool returnsafter_releasein its full view, and@0xinsider/sdk0.65.0 adds it to the holder type.
after_release.state instead of the release shares.Each full pick on
GET /api/v1/pick-of-the-day now says how many graded wallets its backed_side_share counts. That share counts every S, A and B wallet holding the backed side when the pick was selected, listed or not, while display_holders lists only some of them, so a client reading the roster beside the share undercounted: pick 2178 lists 6 wallets and its 67.5% share counts 11.side_graded_wallet_countis the number of graded wallets on the backed side when the pick was selected. It can exceedholder_countand the length ofdisplay_holders.side_graded_usdis those wallets’ gross dollars on the backed side at that time, sent only beside the count and only when above zero. It can differ fromsharp_positioned_usd, which values the listed wallets’ positions.- Both are absent on picks recorded before the S, A and B wallet policy, on picks that stored no count, and on teasers. The
get_pick_of_the_dayMCP tool returns them in its full view, and@0xinsider/sdk0.64.0 adds them to the pick type.
side_graded_wallet_count instead of counting display_holders.Each full pick on
GET /api/v1/pick-of-the-day now carries the proof that it was published before its game, and the start it was published against. Before this change a pick named only the provider’s current kickoff, which moves when a game is rescheduled, and a client had to call the ledger entry to find the pick’s hash.commitment_hashis the pick’s sealed hash, the same value its ledger entry serves.sealed_atis when it was sealed, andledger_urlis the day’s file in the public 0xinsider/picks repository.committed_kickoffis the start the sealed commitment holds.kickoffstill follows the provider, so the two differ when a game moved after release: pick 2178 was committed against 09:30 UTC on November 12, 2026 and played at 10:50 UTC.game_start_timerepeatskickoffunder the name the 0xinsider website uses.holders_observed_atis when the listed wallets were read at release, andrecords_as_ofis when the daily pass that computed those wallets’ records began. Both are absent when the stored release roster is not available.- All four proof fields are absent on a pick published before sealing began. The
get_pick_of_the_dayMCP tool returns them and namescommitted_kickoffin its text, and@0xinsider/sdk0.63.0 adds them to the pick type.
kickoff keeps its meaning.What to change: Nothing. To check how long before its game a pick went out, compare sealed_at with committed_kickoff, not with kickoff.Each flag on
GET /api/v1/suspicious-trades and GET /api/v1/suspicious-trades/{id} now carries hedge: what else the wallet holds that pays if the flagged outcome loses, the hedge the Suspicious trades page shows. Before this change the API gave a flag but not whether the wallet had covered it, so an agent could not tell a hedged position from an outright one.stateishedged,none,sellorunknown. Ahedgedflag lists each opposing position inlegs, largest cost first, with itsplace,sharesandusd, andopposing_usdis their total.unknownnames its cause inunknown_reason; never read it asnone.basissays which positions were measured:live, the wallet’s current positions, orat_flag, the positions frozen when the trade was flagged, once the current ones can no longer show the hedge.positions_atis when Polymarket listed them.hedgeisnullfor a trade older than 90 days, the window the page measures.- A hedge is a fact about positions, not a verdict on the trade.
- The deprecated
/api/v1/insider-radaraliases carry the same field. The remote MCP toolsget_suspicious_tradesandget_suspicious_tradereturn it and name it in their text, as does@0xinsider/mcp3.23.0, and@0xinsider/sdk0.62.0 adds it to the suspicious trade type.
hedge.state and hedge.opposing_usd.Each settled entry on the keyless ledger and ledger entry now carries
withdrawn_at, the time the site withdrew the pick. Before this change an entry said why a pick was withdrawn but not when, so a reader could not tell a correction from a removal after the pick lost.withdrawn_atis an ISO 8601 UTC time beside a non-nullwithdrawal_reason, andnullwhen the pick counts in the record or the time is not stored. Each of the 4 withdrawn picks on November 12, 2026 carries it. A sealed entry and a pending entry omit it, as they omitwithdrawal_reason.- Compare it with
resolved_at. Pick 450 settled on October 2, 2026 at 04:57 UTC and was withdrawn on October 17, 2026 at 12:59 UTC, after its market settled. - It is not part of the hashed
payload, so everycommitment_hashverifies as before. - The
get_pick_of_the_day_ledger_entryMCP tool returns it, and@0xinsider/sdk0.61.0 adds it to the ledger entry types.
withdrawn_at.On
GET /api/v1/pick-of-the-day, public_sports_identity now carries side_order, the same value sports_context.side_order carries: yes_first or no_first, which of yes_team and no_team the game lists first, in Polymarket’s order for the event. It is absent when the order is unknown.Before this change, a client that used only public_sports_identity had no order to follow, so a spread, whose Yes side is the favorite, showed its teams in the opposite order to a total on the same game.@0xinsider/sdk 0.60.0 adds side_order to public_sports_identity on PickOfTheDay, and the remote MCP server’s Pick of the Day tool lists it in its output schema.Backward compatible. One optional field is added, and no field is removed or retyped.What to change: Nothing. To lay out a pick from public_sports_identity the way the game lists it, put no_team first when side_order is no_first, otherwise yes_team first.Each pick on
GET /api/v1/pick-of-the-day, in view=full and view=summary, now carries backed_side_share and sharp_positioned_usd: the “Side share” and “Sharp positioned” the 0xinsider site shows for the same pick. Before this change no API field held them for a pick dated August 12, 2026 or later, because sharp_pct and backed_sharp_usd count S and A grade wallets only and are absent from that date.backed_side_shareis the backed side’s share of graded-wallet money, a fraction from 0 to 1.sharp_positioned_usdis the dollars the counted graded wallets hold on that side.- Both count the grades of the pick’s own wallet policy: S, A and B grade wallets from August 12, 2026, S and A before. On an earlier pick they equal
sharp_pctandbacked_sharp_usd. - Each is absent when the pick recorded no value. Neither is a winning probability or a live balance.
sharp_pctandbacked_sharp_usdare unchanged.- The
get_pick_of_the_dayMCP tool lists both in its summary fields,@0xinsider/mcp3.23.0 says so, and@0xinsider/sdk0.59.0 adds them toPickOfTheDay.
backed_side_share and sharp_positioned_usd.On
GET /api/v1/pick-of-the-day with pick_id and no date, an id the key cannot read still returns 404 with error.code not_found and error.param pick_id, and its error.message now says why that can be: no published pick has the id, the pick was withdrawn, or the key’s plan does not open it. Before this change the message said the id was not among today’s picks and told the caller to send pick_id without date, which it had already done.- The message points to the keyless ledger entry, which says whether a pick with the id was published, and to the archive, which lists the published picks that were not withdrawn.
- A request with
datekeeps its message, which names that date. - The
get_pick_of_the_dayMCP tool returns the samenot_found.
error.code and error.param are unchanged; only error.message changed.What to change: Nothing. Branch on error.code and error.param, not on the message.The description of
cursor on GET /api/v1/pick-of-the-day/archive said a cursor this route did not return was a 400. The route never did that: a cursor is a position in the record, pick_date.pick_rank (YYYY-MM-DD.N), and any value in that form returns the rows after that position. Only a value not in that form returns 400 with error.code bad_request and error.param cursor, and its error.message now names the form.- Following
page.next_cursornever skips or repeats a row, also when a pick is withdrawn between two pages. @0xinsider/sdk0.59.0 carries the corrected description.
400 message text.What to change: Nothing. Do not rely on a 400 to detect a cursor from another source.On
GET /api/v1/pick-of-the-day and the get_pick_of_the_day MCP tool, a backed pick’s thesis now reads Profitable wallets hold <side>, and their top trader grade is <grade>., for example Profitable wallets hold G2 (+1.5), and their top trader grade is S. The grade is the value top_grade carries, the best grade among the pick’s wallets, on every pick, past ones included.Before this change the clause read , led by a grade-<grade> trader, which says the largest position, while the grade was the best one among the wallets, often a small position. On pick 2153 the one grade-S wallet held 4% of the listed dollars and a grade-A wallet held 46%.- A pick with no grade to name still reads
Profitable wallets hold <side>., and a pick without holder proof still reads0xInsider's Pick of the Day is <side>.
thesis changed.What to change: Nothing if you display thesis as returned. If you parse it for the grade, read top_grade instead.GET /api/v1/suspicious-trades, one suspicious trade, suspicious_only=true on large trades, and the suspicious_trade_flagged webhook now judge a wallet by its history as it stood before the trade: its age and the other markets it had traded. Before this change the scorer read the wallet’s stored profile when it scored, and a wallet whose history had not loaded yet looked brand new. Most flags therefore went to wallets with long records, and a sell that closed a position counted as trading against the price.- A wallet whose history has not loaded never counts as new. When it loads, the wallet’s trades of the last 72 hours are scored again, so a flag and its webhook delivery can arrive up to 72 hours after the trade.
created_atis still the trade’s time, so withsort=recenta late flag takes its place by trade time, not at the top.- A sell that closes a position the wallet held no longer counts as trading against the price, on the
whaleandfresh_convictiontracks. fresh_convictionflags a wallet with almost no history buying at a long-shot price or opening a sell at a high one, on a trade of $10,000 or more. Before, it took smaller trades and few of them were flagged.sliced_positionflags only a wallet with little history before the trade. Itsevidenceaddsprior_markets, that count.whalekeeps its 4 signals and its size test; its new-wallet and price signals read the history as above.- On November 11, 2026 at 22:18 UTC, the fixed scorer was applied to the flags stored before this change. The 288 flags it rejects were removed: they no longer appear in either list, and one suspicious trade returns
404witherror.codenot_foundfor their ids. - 26 older trades now carry the fixed scorer’s score and
evidence. 17 of them are listed for the first time, and 9sliced_positionflags gainevidence.prior_markets. - The list went from 342 flags to 71, and the newest flag is now from 2026-11-07T18:37:06Z. No webhook delivery or stream event was sent for these removals and additions. A
mode=stablewalk in progress at 22:18 UTC gotcursor_expiredonce, as it does after any scoring run.
prior_markets is a new key in a sliced_position flag’s evidence.What to change: Nothing. If you poll for new flags by created_at, look back 72 hours and deduplicate by id, or subscribe to suspicious_trade_flagged. If you stored flag ids before November 11, 2026 22:18 UTC, expect some of them to return 404, and list the flags again with sort=recent to resync.GET /api/v1/suspicious-trades, GET /api/v1/suspicious-trades/{id} and their deprecated insider-radar aliases now say what followed each flag. Before this change a flag said a trade looked unusual when it filled and nothing about how it turned out, and scores.timing and scores.edge were always null.result.statusiswon(a buy whose outcome won, or a sell whose outcome lost),lost,void(resolved to neither side),pendingorunknown(resolved, but the trade’s outcome is not known).result.pnl_usdcis what the fill made or lost at settlement, gross of fees, set forwonandlost.result.resolved_atis set forwon,lostandvoid.clvis the closing line object large trades already carry, against the market’s closing line. It is always present; itsstatussays whether it is measured.scores.timingandscores.edgeare deprecated and staynull. How long before kickoff a trade filled isclv.close_anchor_atminuscreated_at; its edge over the close isclv.clv_cents.- The
get_suspicious_tradesandget_insider_radarMCP tools describe the new fields, and@0xinsider/sdk0.59.0 adds them to its types and marks the two scores deprecated.
result and clv; stop reading scores.timing and scores.edge.GET /api/v1/suspicious-trades and its deprecated alias GET /api/v1/insider-radar take a new sort query parameter. sort=recent lists the newest flagged trade first, by created_at. Before this change the list was ordered by suspicion_score only, so the newest flags sat behind every older flag with a higher score, and reading the latest ones meant walking every page.sort=suspicion_scoreis the default and keeps the order every earlier release served, byte for byte.- Each sort returns its own
next_cursorin eachmode. A cursor from the other sort returns400witherror.paramcursor; restart from page 1 when you changesort. mode=stablebinds asort=recentwalk to one published scoring run, as it does for the score order, and expires it witherror.reasoncursor_expiredthe same way.- Any other
sortvalue returns400 bad_requestwitherror.paramsort. - The
get_suspicious_tradesandget_insider_radarMCP tools, the CLI’slistSuspiciousTradesoperation and@0xinsider/sdk0.58.0 acceptsort.
sort gets the same order and cursors as before.What to change: Nothing. To read the latest flags, send sort=recent.GET /api/v1/trader/{address}?expand=breakdowns now names a best and a worst hour, and a best and a worst entry price band, among only those whose win rate rests on 5 or more decided markets:hourly_performanceaddsbest_measured_hour,best_measured_win_rate,worst_measured_hour,worst_measured_win_rateandmeasured_spread_pp.entry_price_analysisaddsbest_measured_bucketandworst_measured_bucket.- Each
entry_price_analysis.buckets[]band addsmarkets_decidedandwin_rate_status.
win_rate_status is measured from 5 decided markets and not_enough_data below, the floor stats.win_rate_status already uses. A measured pick is null when no hour or band reaches the floor. Before this change the only picks were best_hour and worst_hour, taken over every hour, so an hour with 1 decided market, a win, came back as best_hour with best_win_rate: 100.best_hour,worst_hour,spread_pp,best_bucketandworst_bucketkeep their meaning and are still sent.- The output schemas of the
get_trader,get_trader_contextandbatch_get_tradersMCP tools describe the new fields, and@0xinsider/sdk0.57.0 adds them to its types.
null say there is not enough data yet.GET /api/v1/trader/{address}?expand=breakdowns now says when an hour’s or a band’s win_rate rests on too few markets to read as a rate:- Each
hourly_performance.entries[]addsmarkets_decidedandwin_rate_status. - Each
time_breakdowns.by_duration[]andtime_breakdowns.by_hour_bucket[]band addsmarkets_won,markets_decidedandwin_rate_status.
win_rate_status is measured from 5 decided markets and not_enough_data below, the floor stats.win_rate_status already uses. Before this change an hour with 1 decided market, a win, returned win_rate: 100 with nothing beside it to say so, and an hour whose settled markets all broke even returned win_rate: 0.win_ratekeeps its meaning and is still sent below the floor.- The output schemas of the
get_trader,get_trader_contextandbatch_get_tradersMCP tools describe the new fields, and@0xinsider/sdk0.56.0 adds them to its types.
win_rate as a rate, check win_rate_status first, and below the floor show the won and decided counts instead.GET /api/v1/leaderboard and GET /api/v1/positions now say when a row’s win_rate rests on too few markets to read as a rate:pnl.in_scopeaddsmarkets_decidedandwin_rate_status. It is the same object onGET /api/v1/trader/{address},POST /api/v1/traders/batchand each leaderboard row’sin_scope.- Each leaderboard row adds
win_rate_statusbesidewin_rate. It is the row’sin_scope.win_rate_status; the count isin_scope.markets_decided. - Each position’s
traderaddsmarkets_decidedandwin_rate_status.
win_rate_status is measured from 5 decided markets and not_enough_data below, the floor stats.win_rate_status on the trader endpoint already uses. Before this change these rows returned win_rate: 1.0 for a wallet with 1 decided market, a win, with nothing beside it to say so.win_ratekeeps its meaning and is still sent below the floor.- The
get_leaderboardandget_positionsMCP tools describe the new fields, and@0xinsider/sdk0.55.0 adds them to its types.
win_rate as a rate, check win_rate_status first, and below the floor show the decided count instead.GET /api/v1/trader/{address} and POST /api/v1/traders/batch now return three more fields in stats:markets_won: sports and esports markets the wallet closed with a profit.markets_decided: sports and esports markets it closed with a profit or a loss. A market that broke even is in neither count.win_rate_status:measuredfrom 5 decided markets,not_enough_databelow.
win_rate is markets_won / markets_decided. Before this change a wallet with 1 decided market, a win, returned win_rate: 1.0 with nothing to say it rested on one market. The 0xinsider leaderboards, profile and trade rows show no win rate below 5 decided markets; the profile now shows “1 of 1 won” in its place. On November 11, 2026, 40,306 wallets had 1 to 4 decided markets, and 11,013 of them read 100%.- The three fields are sent exactly when
pnl.in_scopeis, and omitted until the wallet’s sports and esports record is computed. win_ratekeeps its meaning and is still sent below the floor./context.mdstates the record, such as “1 won of 1 decided”, in place of a percent below the floor.- The
get_traderMCP tool describes the same fields, and@0xinsider/sdk0.54.0 adds them to the traderstatstype.
win_rate as a rate, check win_rate_status first, and below the floor show markets_won of markets_decided.GET /api/v1/edge-hunter now returns data.unranked on the default board view: the markets the board looked at for your window and selection that got no row, counted by reason.Before this change an empty board returned total: 0 and nothing else, so you could not tell a quiet slate from a board that had stopped working. On November 11, 2026 the board was empty for hours because its one game had S, A and B money split between both sides.split_holder_pile: S, A and B money on the two sides is too even to back either side.no_graded_holders: no S, A or B wallet holds either side now.below_grade_floor: one side is backed, but no wallet on it is at or abovemin_grade.data.totalplus the three counts is every market the board looked at. Three zeros withtotal: 0means no game in the window drew a large trade from an S, A or B wallet in the last 24 hours.data.unrankedisnullonly on a snapshot built before this release.- The
get_edge_hunterMCP tool returns the field with the board, and@0xinsider/sdk0.53.0 adds it toEdgeHunterBoard.
view=signals and view=observations are unchanged.What to change: Nothing. To explain an empty or short board, read data.unranked instead of treating total: 0 as an outage.GET /api/v1/pick-of-the-day (with or without date, pick_id or view), /archive, /stats, GET /api/v1/me and GET /api/v1/usage now send Cache-Control: private, no-store on a 200 or 304, and Cache-Control: no-store on an error. Every other endpoint that needs an API key now adds Vary: authorization unless it sends no-store.Before this change these reads sent private, max-age=30, stale-while-revalidate=60 and varied only on Accept-Encoding. An HTTP cache that sends requests for several API keys could reuse one key’s response for another key for up to 90 seconds. A Pick of the Day read at a release could also return the pick set from before the release.- The
ETagand304 Not Modifiedon the Pick of the Day routes are unchanged. - The keyless ledger and ledger entry keep their headers.
If-None-Match, keep the body and its ETag in your own storage, because an HTTP cache will no longer store it. If your HTTP cache used to answer these reads, your client now sends a request every time, and each Pick of the Day request counts toward your plan’s usage, a 304 included.GET /api/v1/pick-of-the-day now returns kickoff on every pick: when the backed game starts, in RFC3339 UTC. It is the same time the 0xinsider site shows for the pick.Before this change, a pick stated no start time, although game_started was described from one. To learn when a pick’s game started you had to call Get game for it, or read entry_authorization.expires_at, which is when an entry authorization lapses, not a start time.kickoffcomes from Polymarket’s current schedule, so a rescheduled game shows its new start. Without one, it is the start recorded with the pick. It is absent only on an old pick with neither, and never estimated.- Every pick in
picks[], the top-level copy of the first pick, apick_idread and a pastdatecarry it, inview=fullandview=summary. kickoffis the namescheduled_picks[]andproof_pending_picks[]already use for the same pick’s start.game_startedis unchanged. It is checked against the start recorded with the pick, which can trail a reschedule thatkickoffalready shows by a short while.locked_picks[]and the public ledger’s entries for unresolved picks carry no start, as before.- The
get_pick_of_the_dayMCP tool and the CLI’spick todayreturn the field.@0xinsider/sdk0.52.0 addskickofftoPickOfTheDay, and@0xinsider/mcp3.22.0 names it in the tool’s text.
kickoff instead of calling Get game or reading entry_authorization.expires_at.Large positions
avg_entry_price, Market flow top_positions[].size_usd, and price_edge.avg_price on Trader and Leaderboard now price a share with the taker fee the wallet paid on its buys. That holds on a position still held as on a settled one.Before this change, a held position’s entry price was Polymarket’s average before the fee, and a settled one’s included it. So a position’s entry price rose by its fee when its market settled: 24.44 shares bought at 0.45 with 0.74 of fees read 0.4499 while held, and would read about 0.4803 settled.- The fee a share is the position’s buy fees over the shares it still holds. A position that paid no fee keeps Polymarket’s average unchanged.
- On fee-paying positions held on November 11, 2026, the fee was 0.0113 a share at the median and 0.0175 at most.
avg_entry_priceon an open position can read above the average Polymarket shows for it, by that fee.share_count,value_usdand the P&L fields are unchanged.size_usdis the larger side’s shares times this price, so it now includes the fee paid on them.price_edge.avg_priceaverages the same prices over settled markets, so it includes the fee, andedgeis lower by it.- Some settled positions take their price from the wallet’s trades, because Polymarket’s record gives no usable one. From the wallet’s next refresh, those include the fee as well, read from the buys in its Polymarket activity. Where that activity cannot show the fee (a combo on the market, or buys that do not match the trades), the price stays before the fee, a lower bound.
- A settled position whose realized profit is more than its shares could pay at Polymarket’s average price, and whose sales do not account for it, now takes its price from the wallet’s buys of that outcome, with their fee as above. After a sale, Polymarket’s average is that of the last shares held: one position bought 4,899.84 shares for $2,731.86, a price of 0.5575, and read 0.9902, the price of its last 454.53 shares. Its cost stays unknown. With no buys of the outcome on record (shares from a split or a conversion), the price stays Polymarket’s average. This is the
avg_priceof a settled row in themarketsof a trader export that took one outcome. - A settled position whose cost at Polymarket’s average price is less than the wallet paid for the buys that make up its shares now takes its cost and price from those buys, with their fee. After the wallet merges shares and buys again, Polymarket’s average is that of the shares then held: one position held to settlement bought 63,946.64 shares for $17,880.55 with fees and read a cost of $11,109.08, a price of 0.1737. It now reads $17,880.55, a price of 0.2796. Where the wallet’s activity does not show those buys, the position keeps Polymarket’s figures. In a trader export’s
markets, this moves a settled row’sinitial_value,initial_value_yesorinitial_value_no,entry_price_yesorentry_price_no, andavg_price, andprice_edge.avg_priceandedgefollow it. - Positions recorded before this change take the new price as they are refreshed: an open position at its next update, a settled one when it is redeemed or within 90 days.
- The
get_large_positions,get_market_flow,get_traderandget_leaderboardMCP tools return the same values.
avg_entry_price with Polymarket’s average on an open position, expect it to be higher by the fee a share.GET /api/v1/pick-of-the-day with pick_id and no date now returns that pick whatever day published it. The response is the one date=<its pick_date>&pick_id=<id> returns, under the same plan rules.Before this change, pick_id without date looked only among the current product day’s picks. An id from an earlier day returned 404 not_found with error.param: "pick_id", so you had to look up its pick_date in the archive first. A pick URL carries the id, not the date.datewithpick_idstill works. Adatethat is not the pick’s own day still returns404 not_foundwitherror.param: "pick_id".- A pick your plan does not open on its day is withheld as it is with
date: an unresolved pick outside your plan returns the same404. - An id that names no served pick (unknown, skipped or withdrawn) returns the same answers as before. A call without
datenever returnserror.param: "date". - The
get_pick_of_the_dayMCP tool and the CLI’spick today --query pick_id=<id>return the same answer.@0xinsider/mcp3.21.0 describes it in the tool’s text. @0xinsider/sdk0.51.0 updates thepick_iddescription.
200 returns the same body. The one change is that an earlier day’s pick_id sent without date now returns 200 instead of 404.What to change: Nothing. If you look up a pick’s pick_date only to send it with pick_id, you can drop that lookup and send pick_id alone.Pick of the Day stats now return
ask_comparison, the record at the ask. It answers what a buyer who paid the ask got, which the stats page already prints.Since October 17, 2026, each pick also records the best ask on its side when it is published. ask_comparison takes the settled picks that carry one and values them twice, at the midpoint the record uses and at that ask, so the two figures differ only by the price paid.valued_decidedcounts those picks, andfrom_dateis the first day among them.midpointandaskeach carryunits,roi_pctand their printed forms, at full precision. A figure is null when no such pick exists or its sum is not finite.- Fees are not deducted from either figure.
- The record itself,
hit_rateandoutcome_stats, stays at the midpoint. ask_comparisonis inview=summary, the MCP tool’s default, andsectionsaccepts it.@0xinsider/sdk0.50.0 addsask_comparisontoPickOfTheDayStatsandPickOfTheDayStatsSections.@0xinsider/mcp3.20.0 returns the field frompick statsandget_pick_of_the_day_stats.
ask_comparison.ask.roi_pct beside ask_comparison.midpoint.roi_pct to see what paying the ask cost.Pick of the Day now returns
resolved_at on each pick beside published_at: the instant outcome was last set to win, loss or void, in RFC3339 UTC. It is the same instant the archive returns for the pick.Before this change, the endpoint returned when a pick was published but not when it settled, so you had to read the archive to date a result.- A corrected market that re-settles a pick moves
resolved_at. - The field is absent while the pick is pending, and on picks that settled before the instant was recorded. Absent means unknown, not unsettled;
outcomesays which. - The MCP Pick of the Day tool returns the same field.
@0xinsider/sdk0.49.0 addsresolved_at?: stringtoPickOfTheDay.
resolved_at here instead of looking it up in the archive.GET /api/v1/pick-of-the-day with pick_id, before the current product day’s first pick is published, now returns 404 not_found with error.param: "pick_id". It is the same answer the endpoint gives that id after the day’s first release, with no error.reason, no retry_at and no Retry-After.Before this change, that call returned 404 with error.reason: "pick_not_released", a retry_at and a Retry-After of up to several hours. Retrying never helped: an id you hold names a pick of an earlier day, which is not among the current day’s picks. The get_pick_of_the_day MCP tool reads the same answer, and the CLI’s pick today --query pick_id=<id> now exits 2, a parameter error, instead of 9, “not available yet”.- A call without
pick_idstill returnspick_not_releasedwith its retry guidance. datewithpick_idis unchanged: send a pick’s ownpick_dateasdateto read it.- The OpenAPI description of the
404response says this.
param: "pick_id" 404 already. If you scheduled a retry on pick_not_released after a pick_id read, read that pick with its pick_date as date instead: the archive lists both.On
GET /api/v1/pick-of-the-day, sports_context now describes every pick on one game the same way, whatever the market. Before this change a total and a spread on the same game could name its teams in opposite orders and in different letter case.side_orderis new:yes_firstorno_first, which ofyes_teamandno_teamthe game lists first, in Polymarket’s order for the event. It is absent when the order is unknown.matchup_titlefollows that order on every market. Before, a spread put its Yes side, the favorite, first: on November 10 the total on the Clippers at Suns game readLA Clippers – Phoenix Sunsand the spread readPhoenix Suns – LA Clippers.short_labelis upper case on every market, for examplePHX. Before, a moneyline or spread could carry Polymarket’s lower-casephxwhile a total on the same game carriedPHX.game_idis present on a total, or any pick whose market does not name a team, from the moment it is published. Before, it was absent until the game started.
yes_team and no_team keep their meaning: the team on the market’s Yes and No outcomes. @0xinsider/sdk 0.48.0 adds side_order to PickSportsContext, and the remote MCP server’s Pick of the Day tool describes the same field. A settled pick recorded before this change can lack side_order.Backward compatible. One optional field is added, and no field is removed or retyped.What to change: Nothing. To show the picks on one game the same way, lay out no_team first when side_order is no_first, otherwise yes_team first. If you compared short_label values with lower-case abbreviations, compare without case.Pick of the Day issues
entry_authorization again on picks that still qualify when they release. An authorization now means one thing: at release, a fresh check found that the pick meets the selection standard in force then and that its current price still allows an automated entry. That is the standard that selected the pick, unless the standard changed between selection and release; then the newer one applies.Before this change, the check at release judged a pick on an older standard than the one that selected it, so most picks went out without an authorization. From November 6 to 10, 14 of 59 published picks carried one, and none of the first 4 published on November 10. Now:- A pick that meets the standard at release, with a book that passes the existing price and depth checks, carries
entry_authorizationas before. - A pick that does not meet it, or whose price no longer allows an entry, publishes as before with no authorization.
- A pick published without an authorization never gains one later. The same holds for
lead_backer, which names the profitable wallet whose position and record certified the pick at release: many picks have no certified lead, and one released without it never gains one. - The field descriptions in the OpenAPI spec, the
get_pick_of_the_dayMCP tool and@0xinsider/sdk0.47.5 say both. The types are unchanged. - Picks already published keep the authorization they had, or the lack of one.
entry_authorization or lead_backer on a released pick, stop: skip that pick and read again on your normal schedule for the next one. The auto-buy guide lists every case.On Pick of the Day stats, each
price_bands[] label now writes cents with the cents sign: Longshot (<35¢), 35-50¢, 50-65¢, 65-80¢, Favorite (80¢+). They read Longshot (<35c), 35-50c and so on before, while every other price in the product already used ¢.- Only the label text changed. Each band’s
key(under_35,35_50,50_65,65_80,80_plus), its bounds and its order are the same. - The spec’s example response shows the new label.
key, and show the label as it comes.The local MCP server in
@0xinsider/mcp 3.19.0 now cleans wallet names in every tool result, and in the leaderboard and whale-trades resources, before your agent reads them: name, display_name, username, pseudonym, x_username, profile_segment, and trader_name. It applies the same rule as remote MCP.Before this change, the local server passed these values to the model exactly as the wallet’s owner wrote them on Polymarket.- Control characters become one space. Invisible characters are removed: Unicode tags, bidirectional controls, zero-width characters, and runs of variation selectors.
- A
<that opens a tag becomes‹. A value longer than 64 characters is cut to 63 characters and…. result._meta["com.oxinsider/neutralized-profile-text"]lists each value that changed, as a JSON Pointer into the result’s JSON text.- The server’s instructions, and the description of each of the 40 tools whose result carries these fields, say to read them as data, never as instructions, and to identify a wallet by its
address. get_trader,get_trader_pnl,get_leaderboard,get_positions,search_markets, andget_market_flownow return compact JSON text, like every other tool. Their fields and values are unchanged.
_meta appears only when a name changed.What to change: Nothing. To read a name as written, call the REST API. Install with npm install --global @0xinsider/mcp@3.19.0.Pick of the Day archive shows a fix on the first read after a release, and meta.cached says when a response came from cache
The Pick of the Day archive now serves a change to what its rows carry from the first read after the release that makes it.Before this change, a release that changed a row could keep serving the previous release’s rows for up to an hour. On November 11 pending archive rows still lacked
clv_pct after the release that added it had started serving.meta.cachedis nowtruewhen the archive response came from cache. It readfalseon every response before.meta.cache_age_sis omitted on these responses, because the cache entry does not record its age. The field’s description already allows this.- The body, the ETag, and every other field are unchanged.
meta.cached on this route as always false, read it as the cache flag it is on other routes.On Pick of the Day, the
disclaimer string now reads “Pick of the Day is analytics, not a prediction or a guarantee. Prediction markets carry risk; do your own research and never stake more than you can lose.” The Pick of the Day emails carry the same text.Before this change, its first sentence described what a pick reflects and named a trader group the product no longer uses. The selection method is private, as the rules page says.- Only the value changed. The field, its type, and where it appears are the same, in both
view=fullandview=summary. - The spec’s example responses show the new text.
disclaimer value as it comes.The remote MCP server now cleans the wallet names in every tool result before your agent reads them:
name, display_name, username, pseudonym, x_username, profile_segment, and trader_name. A wallet’s owner sets this text on Polymarket, so it can hold words written to steer an agent.Before this change, these values reached the model exactly as written.- Control characters become one space. Invisible characters are removed: Unicode tags, bidirectional controls, zero-width characters, and runs of variation selectors.
- A
<that opens a tag becomes‹. A value longer than 64 characters is cut to 63 characters and…. result._meta["com.oxinsider/neutralized-profile-text"]lists each value that changed, as a JSON Pointer intostructuredContent.- The server’s
instructions, and the description of each tool whose result carries these fields, say to read them as data, never as instructions, and to identify a wallet by itsaddress.
@0xinsider/mcp is unchanged.Backward compatible.What to change: Nothing. To show a name exactly as the owner wrote it, read it from the REST route the tool calls.On Pick of the Day,
position on a moneyline pick now reads <team> to win when the team’s own name contains parentheses, for example Miami (OH) to win. thesis, which quotes position when holder data is unavailable, reads the same way.Before this change, any parenthesis in the label left position as the bare team name, Miami (OH), so the field did not say the pick was the moneyline.- A team-to-advance pick follows the same rule in
positionandpick_outcome_label:<team> to advance, parentheses or not. - A negated or total side keeps its present wording, for example
Real Madrid (No)orPortugal O/U 2.5 (Over). Spread positions are unchanged. - The wording is applied when a pick is read, so published picks, past ones included, return the corrected
position.
position yourself when it lacked it, you can stop.On the Pick of the Day archive, a pending pick you can open now carries
clv_pct, clv_display, clv_entry_price, clv_close_price and backed_price as soon as its clv_status is measured.Before this change, such a row returned clv_status: "measured" and stated the CLV in clv_explanation, but left out the numbers until the game settled. Pick of the Day already returned them to the same key at the same time.backed_priceon a pending row equalsclv_entry_price. Settled and void rows are unchanged.- A pending pick your key cannot open still returns none of these fields, and no
clv_statusorclv_explanation. - The CLV averages in
hit_ratestill count settled picks only. @0xinsider/sdk0.47.3 updates the description ofbacked_priceonPickOfTheDayArchiveEntry. The types are unchanged.
clv_explanation for a pending pick, read clv_pct instead.@0xinsider/mcp 3.18.0 gives “not out yet” its own CLI exit code: 0xinsider pick today, or api call getPickOfTheDay, before the day’s first pick publishes now exits with code 9.Before this release, the API’s 404 pick_not_released exited with code 6, the code for an upstream failure. That state is expected every morning, so a script that alerts or retries on 6 did the wrong thing every day.- Stderr carries one JSON line with the API’s
retry_at, for example{"retry_at":"2026-11-11T08:00:00Z"}, before theError:line. It is the same line that already reports where a failed--pagestraversal can resume. unknown_endpointstill exits with code6. Exit code5still means rate limited.- The MCP tools are unchanged. They already return
pick_not_releasedwithretry_at.
6 from pick today as “not out yet”.What to change: On exit code 9, schedule the next read at retry_at instead of polling. Install with npm install --global @0xinsider/mcp@3.18.0 once it is on npm. 0xinsider --version prints 3.18.0 once the update is in place.On Pick of the Day, each
display_holders entry’s record fields now keep the values they had when the pick was published: category_win_rate, category_win_record, category_avg_price, category_scope_avg_price, category_win_rate_game, category_win_rate_status, category_price_edge, category_scope_price_edge, and the two _hold objects.Before this change these fields were recomputed from the wallet’s current record each time the pick was read, so a wallet’s rate and edge on a published pick could change after the game started or settled.- A pick published after this change returns the same record values on every read, before and after it settles.
- A pick published before this change keeps the old behavior: its record fields are recomputed when it is read.
@0xinsider/sdk0.47.2 updates the descriptions of these fields onPickHolder. The types are unchanged.