Operations · nightly writer 18:30, retry 22:30
Data health
Freshness, fetch outcomes, the options backfill and the run ledger, read straight from the database. This page still renders, degraded, while a writer holds the lock.
Status
The facts that say whether last night's run did its job.
Checks this page can make from what it reads — a missing symptom here is not proof the run was sound.
How to read this
A lamp is steady when a check reads as expected and pulses amber when it does not; grey means nothing to judge yet. “On schedule” is judged in the browser, by its own clock: tonight's bar is due once the 18:30 run has had until 19:20 to finish (the 22:30 retry until 23:20), and the lag is counted in weekdays — exchange holidays are not known to this page, so a holiday reads as one weekday behind.
These are symptoms the page can see, not a certificate: a run that wrote the wrong numbers on time passes every check here.
Freshness
How far each table's newest date reaches, measured against the prices it is computed from. A bar that stops short of the prices line is a table that did not keep up.
- Prices 2026-11-04
- Indicators 2026-11-04
- Targets 2026-11-03
- Signals 2026-11-04
- Predictions 2026-11-04
- Last fetch 2026-11-04
Anchored on prices; indicators and targets legitimately lag (targets by the horizon, since a target needs the next bar).
How to read this
Each bar runs along the last few weekdays and stops at the table's newest date. The solid line is the newest price bar; the dashed line is the date tonight's schedule says should be in. A bar is amber only when it trails prices by more than its table legitimately may — targets by one weekday, everything else by none. The prices row itself is measured against the schedule instead.
Indicators are held to none although the note above says they may lag: the nightly run computes them through the bar it has just fetched, so the table's newest date should match prices, and a day's gap there is a step worth checking rather than the lag the note allows.
Lags are counted in weekdays by the browser; exchange holidays are not known to this page.
Fetch outcomes trailing 7 days of log
Every vendor call the ingest made, by what came back. Hover or focus a row, or hover a segment, to find its partner.
oka network error or throttlingany other logged outcome
| Outcome | Count | Share |
|---|---|---|
| ok | 19,547 | 89.9% |
| adjustment_suspect | 1,645 | 7.6% |
| network_error | 393 | 1.8% |
| not_served | 113 | 0.5% |
| empty_deferred | 33 | 0.2% |
Every fetch outcome is logged, so "nothing was silently dropped" is a query rather than a hope. Options-chain fetches share this log with prices.
Universes
The newest membership date of every universe the database holds, beside the prices it should reach.
| Universe | Symbols | Last date |
|---|---|---|
| liquid500 | 1,893 | 2026-11-04 |
| smallcap1000 | 4,232 | 2026-07-24 stale vs prices |
Only liquid500 is maintained by the nightly job; other universes go stale by design. Symbol counts are distinct members over the whole history.
Recent runs
fits, pre-registrations and research runs — the nightly scoring job logs to logs/daily.log, not here
- 2026-10-10
- 11:14 prereg registered < 1 s 31c04f53
- 10:55 prereg registered < 1 s a337f113
- 2026-10-09
- 15:54 prereg registered < 1 s 5b688882
- 11:43 prereg registered < 1 s 6a278167
- 2026-10-08
- 23:42 prereg registered < 1 s eecc0df1
- 22:50 prereg registered < 1 s e0d3be45
- 22:50 prereg registered < 1 s e0d3be45
- 19:34 prereg registered < 1 s 58ae8e55
- 12:53 prereg registered < 1 s 64ff6d3e
- 2026-10-02
- 16:39 prereg registered < 1 s ddc932a7
- 16:03 research done < 1 s 4e847631
- 2026-09-29
- 16:29 prereg registered < 1 s 7b8eba20
- 2026-09-20
- 15:38 prereg registered < 1 s 28311355
- 15:26 prereg registered < 1 s 8db71cf9
- 2026-09-19
- 15:42 research done < 1 s 37bac3a9
- 14:45 prereg registered < 1 s 9f9b60fc
- 14:23 research done < 1 s 9f9b60fc
- 14:21 prereg registered < 1 s 9f9b60fc
- 2026-09-10
- 09:23 research done < 1 s f66894fd
- 09:23 prereg registered < 1 s f66894fd