Pick of the Day can now list fewer graded wallets than hold the backed side, and
side_summary, sharp_wallet_count, traders, holder_count, holders, and display_holders describe that list.Before this change, a pick listed every S, A, and B grade wallet holding the backed side, the roster market holders returns.- A pick published after this change can count fewer wallets than one published before it for a similar game.
side_summarystill names the grades it counts, as in11 S and A grade wallets on Bills. - A pick published before this change keeps the list and counts it was published with.
- Market holders is unchanged and still returns every graded wallet on both sides.
@0xinsider/sdk0.47.1 updates the descriptions ofside_summary,sharp_wallet_count, andgetMarketHolders. The types are unchanged.
display_holders as every graded wallet on the side. Read market holders for that.@0xinsider/mcp 3.17.0 adds two CLI commands for the Pick of the Day ledger and changes the exit code for an id the API does not have.Before this release, only api call getPickOfTheDayLedgerEntry and api call getPickOfTheDayLedger reached the ledger, and a 404 not_found exited with code 6, the code for an upstream failure. A script could not tell a mistyped pick_id from a broken service.0xinsider pick proof <pick_id>reads one pick’s ledger entry. An opened entry has the payload and nonce: hash the payload bytes as received, followed by the hex-decoded nonce, with SHA-256, and compare the result withcommitment_hash.0xinsider pick ledgerreads every published commitment in one document.- Both commands need no sign-in and no API key.
- A
404not_foundfor an id you sent, such as apick_id, a pastdate, a wallet, or a market, exits with code2, as a400already did.pick_not_releasedandunknown_endpointstill exit with code6. - The MCP tools are unchanged.
get_pick_of_the_day_ledger_entryalready reads one pick’s ledger entry on the remote and local servers.
6 as “not found”. Every other change adds a command.What to change: Treat exit code 2 as “check the id or the parameters; do not retry”. Install with npm install --global @0xinsider/mcp@3.17.0 once it is on npm. 0xinsider --version prints 3.17.0 once the update is in place.Pick of the Day now returns
published_at on every pick, documents the public_sports_identity object it already returned, and counts S and A grade wallets in side_summary. display_holders[].wallet_age_days is now an integer.Before this change, the response and the OpenAPI spec disagreed in several places. Now:published_atis the RFC3339 instant the pick was published, beside the scheduledrelease_at: the same instant the archive returns for the pick. The daily read did not return it before.public_sports_identityis in the spec, withevent_matchup,league_name,league_logo, and ayes_teamandno_teamwith names, logos, records, and tennis headshots. The response returned it before this change, but the spec did not list it. A value that starts with/is a path onhttps://0xinsider.com.side_summaryon a pick dated August 12, 2026 or later reads11 S and A grade wallets on Bills(1 S or A grade wallet on ...for one wallet). It read11 profitable wallets on Bills, the wording the 0xinsider site uses for a count that also includes B grade wallets, so the API’s number looked wrong next to the site’s. The count itself is unchanged and equalssharp_wallet_count.wallet_age_daysis returned as271, not271.0, and the spec types it asinteger. The value is unchanged.- The spec now says exactly when
sharp_usd,sharp_pct, andbacked_sharp_usdare absent, and thatGET /api/v1/pick-of-the-day/ledger/{pick_id}needs no API key. The429description names the four rate-limit budgets, not three. @0xinsider/sdk0.47.0 typespublished_atandpublic_sports_identityonPickOfTheDay, and theget_pick_of_the_dayMCP tool lists them in its output schema.
side_summary text against profitable wallets, and for a strict JSON decoder that expects wallet_age_days to be a float. Every other change is additive.What to change: Read sharp_wallet_count instead of parsing side_summary. Decode wallet_age_days as an integer, or as any number.Pick of the Day now measures a pick’s closing line value (CLV) up to the real start of play. A pick whose game the live feed reported over before it was ever in play, as in a walkover, now has
clv_status ended_before_start and no CLV.Before this change, two cases gave a wrong close:- A
DelayedorPostponedreport from the live feed ended the pre-game window, as if the game had started. A rain-delayed baseball game postponed to the next day took its close from the night before. - A walkover reported over before kickoff kept measuring until kickoff, so its close was the price of the settlement, not of the game.
- A delayed, postponed, or canceled report no longer ends the window. The window ends at the first report of play, or at kickoff.
clv_statuscan beended_before_start.clv_explanationsays the game was reported over before kickoff, and the row counts as CLV unavailable in the summary.- Four stored picks from September to November 2026 had the wrong close. They were corrected on November 10, 2026: three now take their close at the real start of play, and one has
clv_statusended_before_start.
clv_status gains one value; no field is added, removed, or renamed.What to change: Nothing, if your client treats an unknown clv_status as CLV unavailable. If it matches clv_status against a fixed list, add ended_before_start.Pick of the Day now returns
more_picks_until, an RFC3339 instant, on the top level of today’s read while another pick can still be added to the day. It is absent once the day is final.Before this change, nothing in the response said whether today’s list was final. scheduled_picks lists only selected slots, so a day with every selected pick released and no scheduled_picks could still gain a pick later. Now:more_picks_untilis on theview=fullandview=summarybodies and on thestate=nonebody that lists onlylocked_picks.- It is absent for a past
date, inside eachpicks[]item, and on apick_idread. - The
get_pick_of_the_dayMCP tool returns it in its output, and@0xinsider/sdk0.47.0 types it as an optionalstringonPickOfTheDay,PickOfTheDayNoEntitledPicksand the summary body. The0xinsider pick todaycommand prints it with the rest of the body. - A later pick is possible, not promised.
more_picks_until instead of polling, and treat its absence as a final day.@0xinsider/sdk 0.46.0 adds protocol_version, "v1" | "v2", to MarketSettledData: the data of a market_settled webhook event and of a MarketSettled frame on Stream, which carry the field since today.Before this release, SDK 0.45.0 typed protocol_version only on SettledMarketData, the List settled markets row, so event.data.protocol_version did not compile on a MarketSettledEvent. Now SettledMarketData is the same type as MarketSettledData under its own name.Breaking for TypeScript code that builds a MarketSettledData value itself, such as a test fixture: the field is required. Code that reads events compiles unchanged, and SettledMarketData still exists.What to change: Run npm install @0xinsider/sdk@0.46.0 to read event.data.protocol_version with types. Add protocol_version to any MarketSettledData you build by hand.The
market_settled webhook’s data and the MarketSettled frame on Stream now carry protocol_version, v1 or v2: the Polymarket protocol the market settles and redeems on, the same value List settled markets returns for the market.Before this change, only the list rows carried protocol_version, so a webhook event and the list row for the same settlement had different fields. Now:v1is a ConditionalTokens market.v2is a Protocol V2 market: the new markets from November 9, 2026 and the Polymarketpolyv2copies of a game.MarketSettledDatain the OpenAPI spec and the SDK requiresprotocol_version.SettledMarketData, the list row type, is now the same type under its own name.- A redelivery resends an event exactly as it was first delivered, so an event first delivered before this change does not carry the field.
protocol_version from the event instead of looking the market up in the list.List settled markets now takes
condition_id (at most 20) and event_slug (at most 10), so you can check whether one market or one game settled without reading the whole list. Every row now carries protocol_version, v1 or v2, the Polymarket protocol the market settles and redeems on.Before this change, the endpoint took only since, sports_market_type, league, limit, and cursor, and a Polymarket polyv2 copy of a game was never listed. Now:condition_idandevent_slugwork like themarket_settledwebhook’scondition_idsandevent_slugs. Apolyv2copy is listed when you name its own condition id or its own slug, withcopy_of_event_slugset to the original game’s slug.- With
condition_id, an omittedsports_market_typekeeps every type, not onlymoneyline. - A cursor keeps the two new filters like the others; sending a different value with it is a
400witherror.reasoncursor_expired. - The
list_settled_marketsMCP tool takes the same two arguments.
data does not gain protocol_version.What to change: Nothing. To find one market, send condition_id; to read one game, send event_slug.When stdin closes while a request is still running, the local MCP server in
@0xinsider/mcp 3.16.2 now writes one line to stderr before it exits: 0xinsider: stdin closed with 1 request running; cancelled without an answer. Keep stdin open until every answer arrives.Before this change, the server cancelled the request and exited silently, which looked like a tool call that never answered. Closing stdin still ends an MCP stdio session: the server cancels every request still running and answers none of them. Only the line is new, and a session with nothing running when stdin closes prints nothing.Backward compatible. stdout and every answer are unchanged; the line goes to stderr.What to change: Nothing. A script that pipes requests to the server must keep stdin open until it has read every answer. In zsh, a { ...; } group whose last command redirects its output, such as ping -c 20 127.0.0.1 >/dev/null, closes stdin as soon as that command starts; end the group with :.@0xinsider/sdk 0.44.1 exports PropsListParams and SettledMarketsListParams from the package root. They are the parameter types of listProps(), which calls List props, and listSettledMarkets(), which calls List settled markets.Before this release, the SDK defined both types but did not export them, so import type { PropsListParams } from "@0xinsider/sdk" did not compile. PropsListParams was missing from 0.43.0 and 0.44.0, and SettledMarketsListParams from 0.44.0.Backward compatible. The release adds two type exports and changes no method, type, or field.What to change: Run npm install @0xinsider/sdk@0.44.1. Code that uses OperationQuery["listProps"] or OperationQuery["listSettledMarkets"] keeps working, because those are the same types.Ctrl-C during
0xinsider login in @0xinsider/mcp 3.16.1 now ends it after its cleanup, also under npx: the CLI prints its Interrupted. error, exits 130 and removes session.lock. Every other command that calls the API ends the same way, and so do SIGTERM and closing the terminal (SIGHUP).Before this change, Ctrl-C under npx reached the CLI twice, once from the terminal and once forwarded by npm, and the second copy ended the process before it removed its lock. The next login then exited 6 with Another CLI operation holds <dir>/session.lock. SIGTERM sent to the whole npx process group, and closing the terminal with any install, left the lock the same way.Backward compatible. An interrupted command exits 130, as documented. The one exit code that changes: SIGTERM sent to a whole npx process group now exits 130, not 143.What to change: Nothing. A lock an earlier version left behind stays until you remove it: confirm the process whose pid is in its owner.json has stopped, then delete the session.lock directory.@0xinsider/sdk 0.44.0 is on npm with two new reads: listMarketTrades() calls List market trades, and listSettledMarkets() calls List settled markets. It also types the fields other responses gained since 0.43.0.Before this release, SDK 0.43.0 had no method or type for either endpoint, so a TypeScript client had to call them with its own fetch. Its types had no oracle_updated_at, no indexed_graded_holders, and none of the four PickHolder price edge fields.listMarketTrades(conditionId, { from, to, limit, cursor })returns one market’s taker fills of $100 or more in a time window, newest first, each fill once.paginate(client, "listMarketTrades", ...)walks a window, andlistMarketTradesConditional()takes the lastETagasIf-None-Matchand returns anot_modifiedresult when the page has not changed. The types areMarketTradesParamsandMarketTrade.listSettledMarkets({ since, sports_market_type, league, limit, cursor })returns the game market settlements recorded at or aftersince, oldest first. Each row is aMarketSettledData, thedataof amarket_settledwebhook event, so a receiver can catch up after downtime.PickHolderPriceEdgeandPickHolderPriceEdgeHoldtype the optionalPickHolderfieldscategory_price_edge,category_scope_price_edge,category_price_edge_hold, andcategory_scope_price_edge_hold.EdgeHunterObservationgainsindexed_graded_holders, and thesettlementblock on game markets, market snapshots, andMarketSettledDatagainsoracle_updated_at.
npm install @0xinsider/sdk@0.44.0. If you call either endpoint with fetch, switch to the new method and drop your own response types.@0xinsider/mcp 3.16.0 is on npm and in the MCP Registry. It brings the November 7 and November 8 props, settled markets, and market trades reads to the CLI and the local MCP server. It also carries the login change in the entry below.Before this release, 0xinsider market trades in 3.15.0 printed Error: Unknown product command, api list named 67 operations, and the local MCP server listed 71 tools.- CLI:
market trades <condition_id>reads a market’s taker fills of $100 or more, with--query from=andto=.api call listProps,api call listSettledMarkets, andapi call listMarketTradesreach the props board, settled markets, and market trades, soapi listnames 70 operations.api call getStreamtakes the stream’sMarketPriceMovedfiltersleague,min_move_cents, andwindow_seconds. - Local MCP: the tools
list_props,list_settled_markets, andget_market_trades. The local server lists the same 74 tools as the remote server. Its tool descriptions also match the API changes since November 6, such as settlements, Edge Hunterindexed_graded_holders, trader categoryscopes, and every stored fill in a position timeline.
@0xinsider/mcp@3.16.0.Backward compatible. Each change adds a command, a flag, an operation, a filter, or a tool, or updates a description.What to change: For the CLI and the local MCP server, run npm install --global @0xinsider/mcp@3.16.0, or change a client’s @0xinsider/mcp@3.15.0 argument to @0xinsider/mcp@3.16.0. 0xinsider --version prints 3.16.0 once the update is in place.0xinsider login in @0xinsider/mcp 3.16.0 now opens your default browser on the approval page with the code already in the field, and prints the same code in the terminal. Check that the page shows the code your terminal shows, then approve. The OAuth device authorization response also carries verification_uri_complete, which is verification_uri with ?user_code= added (RFC 8628 section 3.2).Before this change, login opened nothing: you copied the printed URL into a browser and typed the code.The CLI opens nothing, and prints the URL and the code, when:- you pass
--no-browser, a newloginflag; - stdin or stderr is not a terminal, as in an agent’s shell or a pipe;
CIis set;- Linux has no
DISPLAYorWAYLAND_DISPLAY; - you are connected to macOS over SSH.
Open <url>, Enter code: <code>, and the review line, and stdout carries only the JSON result. The link only fills in the code: you still select Review request and Approve agent yourself.Backward compatible. verification_uri_complete is a new field, and the output an agent reads is unchanged.What to change: Nothing. Pass --no-browser to keep a terminal from opening a browser.A custom
from/to range on GET /api/v1/reports/weekly now leaves out markets that were reclassified out of sports and esports after their trades were stored. A daily, weekly or monthly report built for the first time from now on does the same. Each report is checked against a market’s current classification when it is built, as market_scope: "sports_and_esports" already said.Before this change, a report of a period that ended before October 21, 2026 still counted those markets. The custom range from February 1 to February 28, 2026 counted 85,748 trades, 166 of them in 20 markets moved from Boxing to Finance on October 29, 2026. It now counts 85,582.- Totals, volume, active traders, categories and the top trades all leave the moved markets out.
- A stored daily, weekly or monthly report keeps the numbers it was published with.
- A range that holds no such market returns the same numbers as before.
The description of
GET /api/v1/reports/resolutions and of the get_resolution_report MCP tool now says that a day from October 21, 2026 counts sports and esports markets only, and an earlier day counts every resolved Polymarket market. Before, both said every day counted sports and esports markets only. market_scope on each report already said which, and the data does not change.Backward compatible. No field, value or response changes.What to change: If you read a day before October 21, 2026 as sports and esports only, check its market_scope: all_markets means it also counts other markets.report.market_scope on GET /api/v1/reports/weekly, GET /api/v1/reports/daily, GET /api/v1/reports/monthly, and GET /api/v1/reports now names the markets the numbers actually count. A custom from/to range says sports_and_esports whatever its dates, because 0xinsider no longer stores large trades outside sports and esports markets.Before this change, a range that ended before October 21, 2026 said all_markets while it counted only sports and esports trades. For March 2026 the custom range counted 148,967 trades and the stored monthly report counted 210,394, and both said all_markets.sports_and_esports: every report whose period ends on or after October 21, 2026, every custom range, and an earlier daily, weekly, or monthly report first built after October 24, 2026.all_markets: an earlier stored report built before then. Its published numbers do not change.
market_scope on both. When they differ, the all_markets report also counts markets outside sports and esports, so its totals are higher.GET /api/v1/market/{condition_id}/trades now keeps each fill for coverage.retention_days (365) and then deletes it. coverage.recorded_from is now where the record starts: the later of when recording began and 00:00 UTC on the day 365 days before today.Before this change, recorded_from was when recording began and never moved, and no stored fill was ever deleted.- Recording began on November 1, 2026, so
recorded_fromstays at that time until November 2027. From then on it moves forward one day at 00:00 UTC each day. - A page serves no fill before
recorded_from, even whenfromis earlier. Before, a window that started earlier could return a fill stored seconds beforerecorded_from. - The
get_market_tradesMCP tool reads the samecoverage.
GET /api/v1/market/{condition_id}/trades is new. It lists one market’s taker fills of $100 or more between from and to, newest first, 100 a page by default and up to 500, with a cursor for the next page. Each fill appears once: its id is the transaction hash and log index of its fill event.Before this change, the API had no way to read every trade on a market in a window. Large trades history lists only the large trades, and Game trades only a game’s newest fills.- Each fill carries
traded_at, the taker’sside,outcome,outcome_index,token_id,price,shares,size_usd,taker,transaction_hash, andlog_index. - Every page carries
coverage:min_size_usd(100),recorded_from(when recording began, about 7 days before launch),complete_through(a few minutes behind the current time), andretention_days(365). A window beforerecorded_fromreads empty because it was not recorded. - The remote MCP server adds the
get_market_tradestool. The CLI commandmarket tradesand the TypeScript SDK’slistMarketTrades()arrive in their next releases.
from set to the last complete_through and drop fills whose id you already have.GET /api/v1/markets/settled and the list_settled_markets MCP tool now serve each settlement’s event_slug, game_id, league, sports_market_type and outcomes as they stood when we recorded it. They stay on the row however long ago the market settled.Before this change, these fields were read from the market when the page was served. We stop holding a market that never traded about 15 days after it resolves, so its settlements then read event_slug, game_id, league and sports_market_type as null and outcomes as []. On November 8, 52 of the 100 rows of one sports_market_type=* page read that way.- A settlement recorded from now on keeps these fields.
- A settlement recorded before November 8, 2026 still reads them as before: from the market as it stands, and
nullonce we no longer hold it. - The
leagueandsports_market_typefilters read the same values, so they keep matching a settlement after its market closes. - The
market_settledwebhook and theMarketSettledframe on Stream send the same values as before: they are sent seconds after a settlement is recorded.
null on an older settlement carry their values on a new one.What to change: Nothing. If you looked up a settled market elsewhere because these fields were null, you can stop for settlements recorded from November 8, 2026.A game market’s
settlement now carries oracle_updated_at: when Polymarket last changed the market’s oracle state, which Polymarket calls its latest lifecycle change. It is on List game markets, List games with expand=settlement, Get game, List settled markets, Get market snapshot, the market_settled webhook and the MarketSettled frame on Stream.Before this change, a market settled through a resolved oracle question, such as college basketball and golf markets, carried no Polymarket time: its question row has no resolved_at, so that field read null.- On a resolved oracle question,
oracle_updated_atis when the answer settled the question. Polymarket’s ConditionalTokens contract pays at the same second or up to about 85 minutes later. - On a settled market, it is the value we recorded with the settlement, and it never changes. A settlement recorded before November 8, 2026 reads
null. - It is not a resolution time.
resolved_atkeeps its meaning and staysnullwhere Polymarket gives none.
null. No other field changes.What to change: Nothing. To time a settled oracle question, read oracle_updated_at where resolved_at is null.GET /api/v1/markets/settled, the market_settled webhook, and the MarketSettled frame on Stream now carry only sports and esports markets, the scope every other route of this API serves. A sports market with no market type, such as a racing or golf future, is still included, with sports_market_type: null.Before this change, a Finance or Politics market that Polymarket gives a game start time could be recorded as settled: 10 states of 8 such markets on November 7 and 8, among them will-qqq-hit-week-of-april-20-2026 and elon-musk-of-tweets-november-2026. A condition that no 0xinsider market names was recorded too, with event_slug, league and sports_market_type all null. Neither is recorded now.Backward compatible. No field, parameter or filter changes. These rows had no market type, so the default moneyline filter never selected them; the Stream frame, sports_market_type=* and a "*" filter could.What to change: Nothing. If you dropped rows outside sports yourself, you can stop.On
GET /api/v1/pick-of-the-day, game_ended now stays true once a pick’s game is over, on later reads and on past dates. The same holds for the summary view and the Pick of the Day MCP tool.Before this change, game_ended came only from the live scoreboard, which keeps a finished game for about a day. A day after the game, the field was dropped from the pick, so a settled pick read as “unknown”, and the response and its ETag changed for a date that had already settled.- A pick whose
outcomeiswinorlossreadsgame_ended: true, unless the live scoreboard shows the game still in play. - Any other pick past kickoff reads
truewhen its game was recorded as finished, also after the live scoreboard has dropped the game. - The field is still absent, never
false, when the game’s state is unknown.
true to absent.What to change: Nothing. If you treated an absent game_ended on a settled pick as “finished”, you can read the field directly.On
GET /api/v1/pick-of-the-day, GET /api/v1/market/{condition_id}/holders and GET /api/v1/market/{condition_id}/top-holders, a holder’s wallet_age_days now counts UTC calendar days from the date of the wallet’s first trade. Every holder’s age changes at the same moment, 00:00 UTC. A wallet whose first trade was at 23:00 UTC reads 1 an hour later.Before this change, wallet_age_days counted full 24-hour periods since the first trade, so each wallet’s age changed at its own time of day. On a Pick of the Day pick set, that changed the response and its ETag many times a day, even for a past date, so a request with If-None-Match often got 200 instead of 304 Not Modified.wallet_age_dayscan now read one more than before, from 00:00 UTC until the time of day of the wallet’s first trade.is_new_walleton these holders followswallet_age_days: it turns off at the 00:00 UTC on which the count reaches 30, up to a day sooner than before. Other endpoints that serveis_new_walletare unchanged.
null. Only the moment it changes moves.What to change: Nothing, unless you compare wallet_age_days to your own count of elapsed hours. Count UTC dates instead.GET /api/v1/games and GET /api/v1/games/{event_slug} now return status.ended_at, with status.ended_at_source sports_ws, for a women’s college basketball (cwbb) game that ends after this change. The same holds for a game in any league whose live updates paused for up to 90 minutes before its final.Before this change, an end was recorded only when the game had sent a live update in the 15 minutes before its final. All 46 cwbb games that ended on November 7 had gone longer without one, so each has status.ended_at null, as do a few college football, soccer and esports games.- This change does not fill in a game that has already ended.
- A final that reaches us more than 90 minutes after the game’s last live update is still not used as its end.
- The
list_gamesandget_gameMCP tools return the same values.
null status.ended_at as unknown, not as still in play.The CLI now installs on macOS and Linux with
curl -fsSL https://0xinsider.com/install.sh | sh. The script installs the 0xinsider npm package when Node.js 22 or newer is present, into ~/.local when npm’s global folder is not writable, and never with sudo. Without Node.js it installs the 0xinsider/tap/oxinsider Homebrew formula, and with neither it names what to install and exits with an error.Before this change, you installed with npm install --global or Homebrew yourself. Both still work.- Each command that changes your machine is printed before it runs.
sh -s -- --dry-runprints them and runs none. - Running it again upgrades. A CLI installed as
@0xinsider/mcp, or by Homebrew, is upgraded as itself, so no second copy is installed. - It then runs
0xinsider --versionand prints0xinsider loginand0xinsider initas the next steps, with the command that uninstalls it.
On
GET /api/v1/pick-of-the-day, each holder’s last_traded_at in holders and display_holders is now the UTC date of the wallet’s most recent trade, sent as midnight UTC on that date, for example 2026-11-07T00:00:00Z. It changes at most once a day for each wallet. Today’s pick set and a past date read with date both change.Before this change, last_traded_at was the exact time of the wallet’s most recent trade on any market. On a pick set with active holders the response, and so its ETag, changed every few seconds, even for a past date whose picks no longer change. A request with If-None-Match almost always got the full body back with 200. Now it gets 304 Not Modified until a pick or a holder’s details change.Backward compatible. The field keeps its name, type, and format: an RFC 3339 timestamp, or null when no trade is recorded. Only the time of day is dropped.What to change: Nothing, if you show the date or only check how recently a holder traded. If you need a wallet’s exact last trade time, read last_active from GET /api/v1/trader/{address}.GET /api/v1/edge-hunter with admission=most_held_48h or most_held_48h_by_type now reads a market only that selection takes, when it has no holder read to serve, right after every default market with none. It no longer waits behind default markets whose last read still serves. The two selections take turns. The get_edge_hunter MCP tool returns the same data.Before this change, those markets waited until every default market had been read, including default markets that already had a read to serve. On a quiet evening of November 7, after a first request for Soccer totals, 32 markets that selection took had no holder read, and the last of them waited about 20 minutes for one. Meanwhile they counted as holder_deadline_unavailable.- A market that still has a read to serve is read again at most once every 10 minutes, as before.
holder_snapshot_atkeeps its bound: up to 15 minutes old, or up to 30 minutes for a market more than an hour before kickoff with no trade since that read.
Edge Hunter observation rows carry indexed_graded_holders, the count the most-held selections rank by
GET /api/v1/edge-hunter with view=observations adds indexed_graded_holders to every row: the S, A, and B graded wallets in the 0xinsider index that hold either outcome of the market. admission=most_held_48h and most_held_48h_by_type rank markets by this count, and so do the default selection’s most_held places. The get_edge_hunter MCP tool returns the same field.Before this change, rows carried only graded_holders, which counts the holders of the row’s side on Polymarket’s holder list. That count can be larger or smaller than the one the selection used, so a selection could look out of order when you checked it against graded_holders.rank is unchanged. It orders rows by directional_rank_score, highest first, and never by either holder count. The admission and rank descriptions now say so.Backward compatible. The field is added; no field is removed or changed. It is null only on a row from a snapshot computed before this change.What to change: Nothing. To check which markets a most-held selection took, compare indexed_graded_holders, not graded_holders. To order rows, keep using rank.