Corrections
Last updated: 2026-09-03
A dated ledger of public claims SquawkFlow published and later withdrew or corrected, each tied to repository evidence. Each entry cites repository evidence. The machine-readable data manifest lists every public dataset, whether it is measured or modeled, and which endpoint serves it.
2026-10-08
The /market-tide page published signed net call and net put premium curves, splitting premium into bought and sold by where each contract’s last trade sat against the quote.
What was true
A delayed chain snapshot records no aggressor side, and on 2026-10-08 the last-trade position was measured against next-day open interest (3,049 contract-sessions): contracts labelled bought and sold opened new positions at the same rate, so the label carried no measurable signal. The curve also read only the 20 newest scanner rows every 15 seconds, so it skipped much of what the scanner saw.
What changed
The curve is now two unsigned totals, estimated call premium and put premium, with the number of symbols actually contributing shown on every reading. Its source also changed: it previously summed the scanner’s flagged rows (contracts trading at least 1.5 times their open interest); it now sums every published full-chain capture cell, each with at least 3 traded contracts and published at least 15 minutes after capture.
Published: 2026-09-08 · Corrected: 2026-10-08 · Evidence: backend/app/services/options_snapshots/flow_view.py
2026-09-03
SquawkFlow comparison pages listed congressional-trading data as an included capability.
What was true
On 2026-09-03 both upstream congressional trade feeds returned 403 and the endpoint behind the CongressTracker panel returned an empty list, so the panel rendered no trades.
What changed
On 2026-09-03 the vs comparison row for congressional-trading data was set to no, with an inline comment in frontend/app/vs/data.ts documenting the upstream failure.
Published: 2026-08-14 · Corrected: 2026-09-03 · Evidence: b0e5f58
2026-09-02
Public gamma and heatmap pages presented exchange open-interest matrices built on whatever spot quote the adapter returned.
What was true
Before provenance labelling shipped, the delayed-chain adapter could substitute synthetic quotes on CDN failure without SQUAWKFLOW_DEMO_MODE, and the heatmap checked chain provenance but not quote provenance, so a real OI book could be archived on a fabricated spot.
What changed
Every adapter return now carries provenance, demo generators run only when SQUAWKFLOW_DEMO_MODE is set, and the heatmap refuses to compute or archive unless both chain and quote are real.
Published: unknown · Corrected: 2026-09-02 · Evidence: df3d003
2026-09-01
squawkflow.com and api.squawkflow.com were unreachable during US market hours on 2026-08-28.
What was true
The production host was down from 07:52 to 19:04 PDT; on-box health checks died with the machine and nothing alerted for eleven hours.
What changed
docs/ops/runbook.md now documents the incident, the on-box heartbeat and two-sink notifier, and the known gap that only an off-box checker can detect a dead host.
Published: unknown · Corrected: 2026-09-01 · Evidence: docs/ops/runbook.md
2026-08-31
The /gex page, landing funnel, and public GEX APIs published how often each call wall and put wall held over recent sessions.
What was true
The published rate was measured on a third party's levels from a source we cannot redistribute, so it is withdrawn and not replaced.
What changed
Hold-rate fields and prose were removed from /gex, the landing page, session-log output, and every public GEX API response; overnight open-interest change replaced the slot on the page.
Published: 2026-08-17 · Corrected: 2026-08-31 · Evidence: 84205dc
2026-08-27
The landing page presented April 2026 rows as current behind a health envelope that reported the sources as fresh, and terminal panels showed a hardcoded LIVE label beside a badge that could simultaneously read STALE.
What was true
The landing page short-circuited to a frozen shared table instead of running the live-compute path, and source health was asserted rather than computed, so a stale or absent source still reported as fresh.
What changed
The short-circuit was removed so the live-compute path always runs, sourceHealth is computed instead of asserted, channel and third-party data vendor artifacts gained freshness gates, and panels no longer hardcode LIVE.
Published: unknown · Corrected: 2026-08-27 · Evidence: d5b1682
2026-08-25
/api/v1/dark-pool/prints described its payload as recent public dark-pool prints.
What was true
The endpoint served modeled block data from darkpool_service, not licensed FINRA ADF prints; only the daily DIX series is a real measured input.
What changed
The route now returns available:false with an empty prints array and a docstring dated 2026-08-25 explaining the retirement; callers receive a clear reason instead of modeled rows.
Published: 2026-08-25 · Corrected: 2026-08-25 · Evidence: ba7bdcf
2026-08-25
/api/insider returned Form 4-style transactions attributed to real, named executives and stamped source: sec_edgar.
What was true
The transactions were generated with random share counts, prices, and dates; SquawkFlow holds no licensed Form 4 feed.
What changed
The fabricating generators and real-person CIK table were deleted from sec_service.py; the endpoint now returns an empty list with available:false.
Published: unknown · Corrected: 2026-08-25 · Evidence: 168bd30
2026-08-17
The public /gex page displayed Net GEX $0, flip N/A, and identical call and put wall levels on every visit.
What was true
The broker returned an SPX chain with zero open interest on every contract, which passed the usability gate and computed to confident zeros instead of failing over to the delayed exchange chain, which carries real SPX open interest.
What changed
get_options_chain gained require_open_interest for positioning callers, the reader prefers the D2DT daily SPX snapshot when present, and /api/public/gex began serving real headline metrics.
Published: 2026-08-14 · Corrected: 2026-08-17 · Evidence: 116ada8