Resolute — programme activity

Every workstream on one page: what is built, what is running, what it is producing right now, and what is still to be executed. For Chelsea, Caroline and John DeReamer.

Measured 2026-08-15 23:00:05 UTC (2026-08-15 19:00:05 EDT). This page is static: it shows the state at that instant and does not update itself. It is regenerated at 07:00 and 19:00 America/New_York.

Warehouse scope for every Snowflake figure on this page: account MLA96789, user RESOLUTE_ADMIN, role RESOLUTE_ADMIN, warehouse RESOLUTE_AI_WH, database RESOLUTE_MIND.RACING_DATA.

1 · Sale sites

The two sale-day field-sheet applications. Reachability was probed by HTTP GET from the Mac Studio at build time; everything else in this section was measured on 2026-08-12 and is carried forward.

Saratoga — Fasig-Tipton, 2026 — ABBI.023

Hips

226

across 2 sessions · recorded 2026-08-12

Reachable now

HTTP 200 165 bytes

HTTP GET, 20s timeout, from the Mac Studio at the measurement time

URL: https://abbi-m1-bigscreen.tailfb6577.ts.net:8443/ — Tailscale funnel :8443 -> 127.0.0.1:8787 (team build, served at path /FxEINNCkHa/)
Hosting: Mac Studio, supervised by a bare `while true` shell loop, not a LaunchAgent

Built and working

10 team tabs, a 22-column board, 8 filters and 5 sorts
Full-width horse card: a why-tile derivation drawer, Dr. Baker's vet reports, the consignor repository reports shown separately and labelled, and 65 vetdoc images
Purchased tab seeded from the Fasig-Tipton results feed: ownership %, partners, an 11-value shipper list, ship/farm dates, and a partner roll-up across the whole sale
Conformation module — 45 faults in 4 view groups, 13 walk observations, the Architect's 1-5 severity scale — BUILT AND WORKING BUT NOT SHIPPED to the live site

Still to be executed

BUG-20260812-0017 (P1): the entire generator source tree lives in two Claude Code session scratchpads under /private/tmp with no .git anywhere. /private/tmp is subject to OS temp reaping. Get it into version control.
Merge the conformation module into the production build: mk_full.py never imports conf/, the live page contains zero confTpl occurrences, and server.py has no conf branch — so a conformation POST to production is silently dropped and marks.json holds 0 conformation records.
mk_dev.py and mk_full.py have diverged in BOTH directions since 2026-08-11 14:50 — reunifying them is a merge, not a copy.
Port 8789 (the conformation build) is not funnelled, so it was never reachable from the sale grounds.
No auth of any kind: security is an unguessable path and a roster dropdown for self-asserted identity.

Keeneland — September 2026 — ABBI.027

Hips

4,635

across 12 sessions · recorded 2026-08-12

Reachable now

HTTP 200 4,970 bytes

HTTP GET, 20s timeout, from the Mac Studio at the measurement time

Parity gap outstanding

features Saratoga has that Keeneland does not

URL: https://v200-webapp-kee27-fieldsheet.pages.dev/ — Cloudflare Pages, so reading survives the Mac; live ticking still posts to the Mac over Tailscale
Hosting: Cloudflare Pages `v200-webapp-kee27-fieldsheet` + two LaunchAgents on the Mac

Built and working

19 selectable tabs: 12 session tabs (181/184/377/379/416/423/417/423/427/431/496/481 = 4,635 hips) and 7 cross-cutting views
9-column sortable board; hip drawer with 9 why-tiles and comparables over 27 Keeneland September sales 1999-2025, 121,933 lots after dedup
The guide is the session median. The fitted price model was built, back-tested and REJECTED, and is carried honestly as a tile labelled 'Model (rejected)'
Dam-side identity gate — every dam-derived field is gated on the catalogue's own broodmare sire agreeing: corroborated 1,935 / withheld_wrong_mare 1,485 / withheld_unverified 1,215
Append-only fsync'd event log with a sequence cursor, adaptive polling, a service worker and an offline mark queue

Still to be executed

11 parity gaps against Saratoga — enumerated in the table below.
The deployed build p202608120157 is NOT reproducible from disk: its source tree is gone and price_model.py's input parquet is missing.
Reading survives the Mac; TICKING DOES NOT — marks post to a Tailscale hostname on the Mac Studio.
GET /events and /healthz are unauthenticated: anyone who reaches the origin reads every mark.
The catalogue feed is an undocumented internal Keeneland Drupal endpoint and must be treated as unlicensed.

The parity gap — 11 features Keeneland lacks that Saratoga has

#	Feature	State
1	Conformation module	Absent entirely — not a stub. The event schema has no slot for it: KINDS = ('vet','buy','out').
2	Photos	Absent; no <img> anywhere. Keeneland's feed media fields are 0% populated today.
3	Video	Absent; zero hits across the whole tree.
4	Notes / comments with autosave	Protocol-ready but UI-absent: the server accepts note[:280], the client's mark() never sends one, and there is no textarea and no autosave.
5	Prospect tick	Absent; only Vet list and Short list.
6	Purchased tab	Absent — no ownership %, no partners, no shipper, no dates.
7	Vet / repository panel	Absent; 'Vet list' is a tick, not a repository. No radiographs, no scope, no vet findings fields.
8	Consignor filter	Dead code: honoured in applyView() but there is no control in the DOM.
9	Broodmare-sire filter	Absent — a column and a search term only.
10	Per-hip URL / page	Absent; drawer only, and the hash does not change.
11	Per-person chips, removal-with-reason, the 40-entry audit trail, the twin public/team build, the sub-tabbed vet list	All absent.

Scope: enumerated by reading both source trees on 2026-08-12; each entry was confirmed by a grep of the Keeneland tree that returned no implementing code. Not re-measured on this build.

2 · Data warehouse

Total rows

54,152,461

122 base tables in RESOLUTE_MIND.RACING_DATA

Empty tables

base tables holding exactly 0 rows

Table	Rows	What it is
`PEDIGREE_LINKS`	40,701,853	Sire/dam edges. The single largest object in the estate.
`STARTERS`	1,609,270	One row per runner per race. LAST_RACE_DATE is 0.0000% populated.
`HORSES`	1,190,432	The horse dimension. FOAL_DATE is populated on 3 rows.
`DIM_HORSE`	1,153,270	The reconciled horse dimension built by the identity lane.
`SPEED_FIGURES`	958,063	Multi-source figures. 14,027 rows are dated in the future and 1,393 have a NULL race date.
`PRODUCE_RECORDS`	812,606	Mare produce records. FOAL_GRADE is 100% NULL and was rejected as a source.
`PAST_PERFORMANCES`	672,686	Past-performance lines — a two-year window only.
`SALE_ENTRIES`	488,349	Auction lots. The comparables backbone of both sale sites.
`COVERINGS`	437,998	Coverings. COVERING_DATE is empty; only COVERING_YEAR is usable.
`WORKOUTS`	367,860	Published workouts.
`RACE_RESULTS`	205,211	Result rows.
`RACES`	152,692	Race header rows.
`NARC_RATINGS`	85,430	North American Rating Committee ratings.
`SCRAPER_LOG`	26,285	The extraction fleet's own audit log — the source for section 3 of this page.

Scope: RESOLUTE_MIND.RACING_DATA. Every figure is a real SELECT COUNT(*) executed at the measurement time above — not the SHOW TABLES / INFORMATION_SCHEMA estimate. The total covers all 122 base tables in the schema, not only the rows listed here.

The gaps — base tables holding exactly zero rows

Empty table	What its emptiness costs
`EQUIPMENT_CODES`	Equipment codes cannot be decoded.
`HORSE_REGISTRY`	Superseded by HORSES / DIM_HORSE.
`NICK_RATINGS_INTL`	The international nick lane never landed.
`OWNERS`	Ownership joins return nothing, although STARTERS.OWNER_ID exists.
`P360_MATING_RESULTS`	The Pedigrees360 mating lane never landed, although its sibling P360 tables did — and matings are what Resolute's own breeding decisions turn on.
`POINTS_OF_CALL`	No running lines exist at all. Pace analysis is impossible.
`RACE_FRACTIONALS`	No sectional times exist at all. Pace analysis is impossible.
`RESOLUTE_SALES`	Resolute's OWN sales records are absent.

Scope: the same query as above. A table appears here only if its real count is exactly 0. An empty table is a measured fact; it is not the same thing as a table that failed to be measured, which would read ERROR.

3 · Extraction fleet

How status is decided on this page, and why it differs from the old dashboard. A lane's status here is rows written over a window. It is never the age of the last run that reported OK. The dashboard at :3005 measures last-OK age, which is why it shows dead lanes as green — allbreed-pedigree has been logging STATUS='OK' | fetched 38 | written 0 | errors 5 around the clock since it last wrote a row. That is BUG-20260812-0014. On this page a lane that ran and wrote nothing is DEAD, and its run count is printed next to its zero so the shape of the failure is visible.

PRODUCING = wrote at least one row in the last 24 h. STALLED = wrote rows within 7 days but nothing in the last 24 h. DEAD = wrote nothing at all in 7 days, however many times it ran.

External lanes producing

4 / 20

wrote ≥1 row in the last 24 h

External lanes dead

0 rows written in 7 days

External rows collected

4,366

last 24 h · 181.9 rows/hour

Derived rows

736,542

last 24 h · computed from data already held, not collected

Collected and derived are reported separately on purpose: the derived lanes manufacture hundreds of thousands of rows a day out of data the warehouse already holds, and adding the two together hides a collection failure completely.

External collection lanes

Lane	Status	Runs 24h	Rows written 24h	Runs 7d	Rows written 7d	Fetched 7d	Errors 7d	Days since last productive run
`allbreed-pedigree`	DEAD	95	0	671	0	25,498	3,408	24
`brisnet`	DEAD	0	0	0	0	0	0	never
`equibase_charts`	DEAD	0	0	0	0	0	0	27
`equibase_results`	DEAD	0	0	0	0	0	0	29
`equibase_workout_scraper`	DEAD	0	0	0	0	0	0	29
`hkjc`	DEAD	1	0	7	0	63	0	11
`pedigreequery`	DEAD	0	0	148	0	8,838	8,838	10
`pedigreequery-pod@3ba95bcad588`	DEAD	0	0	0	0	0	0	29
`pedigreequery-pod@dff74ceec122`	DEAD	0	0	0	0	0	0	29
`pedigrees360`	DEAD	1	0	7	0	7	7	34
`racelens`	DEAD	0	0	0	0	0	0	28
`ragozin`	DEAD	0	0	0	0	0	0	never
`sl-race-scraper`	DEAD	0	0	0	0	0	0	48
`thorograph`	DEAD	0	0	0	0	0	0	never
`jbis`	STALLED	1	0	7	108	84	0	1
`narc`	STALLED	96	0	672	60	4,012	130	3
`equinedge`	PRODUCING	7	3,664	38	20,311	867	0	0
`racingpost`	PRODUCING	1	490	7	3,245	333	0	0
`hrn-results`	PRODUCING	1	209	7	1,301	70	0	0
`horse_news`	PRODUCING	4	3	28	9	2,420	0	0

Derived / internal lanes

Lane	Status	Runs 24h	Rows written 24h	Runs 7d	Rows written 7d	Fetched 7d	Errors 7d	Days since last productive run
`daily-freshness-audit`	DEAD	1	0	7	0	49	11	never
`dedup`	DEAD	0	0	1	0	0	0	never
`derived-produce`	DEAD	1	0	7	0	0	0	10
`derived-inbreeding`	PRODUCING	1	379,614	7	2,657,298	0	0	0
`identity.reconcile`	PRODUCING	48	239,746	336	1,673,478	1,675,432	0	0
`derived-nicks-structure`	PRODUCING	1	117,162	7	820,134	0	0	0
`completeness-audit-refresh`	PRODUCING	1	20	7	140	0	0	0

Scope: RESOLUTE_MIND.RACING_DATA.SCRAPER_LOG, one row per run. The 24 h and 7 d columns are SUM(RECORDS_WRITTEN) and COUNT(*) over RUN_STARTED_AT within that window. “Days since last productive run” is DATEDIFF from MAX(RUN_COMPLETED_AT) WHERE RECORDS_WRITTEN > 0 over ALL TIME, so it is not capped by the window; a lane with no productive run anywhere in the log reads never. Nothing in this table reads the STATUS column.

4 · Freshness

How old the newest row is, per source, for the two data classes the sale sites and the analytics depend on.

Races — days stale by source

Source	Rows	Newest row	Days stale
`horseracingnation.com`	1,345	2026-08-14	1
`racingpost.com`	37,566	2026-08-14	1
`jbis.or.jp`	24	2026-08-13	2
`racing.hkjc.com`	14,667	2026-07-15	31
`netkeiba.com`	11,052	2026-07-10	36
`drf.com`	52,052	2026-07-09	37
`sportinglife.com`	24,877	2026-07-08	38
`equibase.com`	4,924	2026-07-08	38
`open-pmu-api`	3,760	2026-07-05	41
`sportinglife`	1,229	2026-06-28	48
`HRN`	275	2026-06-28	48
`timeform.com`	616	2026-05-14	93
`eprochasson/horserace_data`	305	1995-02-22	11,497

Scope: RESOLUTE_MIND.RACING_DATA.RACES, grouped by SOURCE_SITE. Days stale is DATEDIFF(day, MAX(RACE_DATE), CURRENT_DATE()) at the measurement time above. Note the two duplicated source spellings (sportinglife vs sportinglife.com, HRN vs horseracingnation.com): they fragment any GROUP BY, including this one, and are listed here exactly as they appear in the column rather than merged.

Sales — days stale by source

Source	Rows	Newest row	Days stale
`keeneland.com`	298,667	2026-04-24	113
`inglis.com.au`	4,377	2026-03-01	167
`arqana.com`	1,569	2026-01-01	226
`goffs.com`	2,699	2026-01-01	226
`tattersalls.com`	92,899	2025-10-01	318
`fasigtipton.com`	73,020	2024-11-04	649
`obssales.com`	11,533	2024-03-01	897
`capethoroughbredsales.com`	3,585	2022-02-01	1,656

Scope: RESOLUTE_MIND.RACING_DATA.SALE_ENTRIES, grouped by SOURCE_SITE. Days stale is DATEDIFF(day, MAX(SALE_DATE), CURRENT_DATE()) at the measurement time above. This is the most degraded lane in the warehouse, and it is the table the Keeneland September comparables are built on.

5 · What remains

The work still to be executed, counted from the three companion documents that sit alongside this dashboard, plus the defects filed and not yet fixed.

Companion document	Open items	What is counted
Data Extraction Completion List Data-Extraction-Completion-List.html	21	numbered work items (X1-Xn), in four blocks: Restoration, the licensing decision, genuine new coverage, and quality fixes
Keeneland Sale Site Completion List Keeneland-Completion-List.html	23 + 11	numbered work items (K1-Kn), in four tiers: Preservation, the conformation module, availability, and parity features; plus 11 recorded risks (R1-Rn), carried rather than resolved
Continuous Data Refresh — Research Decision Continuous-Refresh-Research-Decision.html	n/a	a decision document — it carries no numbered work items. The open decision it holds: obtain the negotiated commercial licence price, between the two measured anchors of ~$3,000/yr (TrackMaster retail) and ~$60,000/yr (the Equibase per-file equivalent)

Scope: these counts are parsed out of the companion documents themselves at build time, from /private/tmp/claude-501/-Users-abbiadmin/276e123c-6e99-46e3-a2c1-01644f8e81a5/scratchpad, by counting distinct item identifiers. They are not transcribed by hand and so cannot drift from the lists. If a document is unreadable the last successful count is shown with its date and marked stale; if there has never been one the cell reads ERROR.

Open defects — filed, none fixed

Bug	Severity	What it is
`BUG-20260812-0012`	P2 / HIGH	The collection controller has been inert for 24 days. Every guard in monitor/racing_monitor.py — proxy recovery, bandwidth, duplicate recovery, silent-stall detection, the staleness sweep, alerting and paging — runs only in the else-branch that requires all six docker containers up. The 2026-07-19 Azure migration put worker and flaresolverr behind a compose profile, so they are permanently absent and the loop takes the missing: branch every 300s forever. Last healthy pass 2026-07-18T23:31:17.
`BUG-20260812-0013`	P2 / HIGH	RESOLUTE_SCRAPEDO_KEY is absent from the live Azure worker. The four Equibase-family lanes fetch exclusively via scrapedo_get, and scrape.do is the only method proven to clear PerimeterX. With the key absent the code logs 'RESOLUTE_SCRAPEDO_KEY not set' and writes nothing. RESOLUTE_RESIDENTIAL_PROXY is also absent, disabling the IPRoyal metered-escalation path.
`BUG-20260812-0014`	P3 / MEDIUM	A dead lane reads green. Both health consumers — the old dashboard's status column and check_all_scrapers — measure the AGE OF THE LAST OK RUN, never rows produced. THIS DASHBOARD IS BUILT AGAINST THAT DEFECT: every status in section 3 is rows-written-over-window.
`BUG-20260812-0017`	P1 / CRITICAL	The Saratoga generator source root is a Claude Code session scratchpad under /private/tmp, 502 files, no .git anywhere, supervised by a bare `while true` loop rather than a LaunchAgent. /private/tmp is subject to OS temp reaping; if either scratchpad is reaped the sheet cannot be rebuilt and the conformation module source is lost outright.

Scope: the four defects filed against the Resolute programme and open as of 2026-08-12. Recorded, not re-measured on this build.