Confluence startup seed: pins the15-symbol research universe + SPY benchmark
into the permanent demand set, and creates three system rack presets (Full
Confluence, Technical Momentum, Macro+Flows+Sentiment) if absent. Called once
from index.ts; idempotent on every restart.
Parse the annual FinFutYY.txt zip (leveraged-funds long/short + OI) into a
CotSeries. Registers the 'cot' source_kind, 'cot_weekly' TTL (7d), the 'cftc'
vendor family (1 req/1.5s pacing), and the adapter in the queue registry.
Uses the file-header column names so parser is robust to layout changes.
- Add await ctx.cache.subscribe() to addSymbol mutation so symbols
added via the sidebar get registered in symbol_demand and yfinance
jobs are queued immediately
- Backfill PEP, WYNN, STZ, CELH into symbol_demand + adapter_queue
- Upgrade yahoo-finance2 3.15.3 -> 3.15.4 and pass validateResult:false
to quoteSummary() to handle Yahoo schema drift
- Add error detail logging for analyst ratings schema failures
- Update .gitignore with common ignores
- SecLintAdapter implements SourceFetch, runs via shared queue/drain loop
- Two new SourceKinds: sec-lint-holders, sec-lint-insiders (weekly schedules)
- No-op cache handlers so drain->cache.set doesn't throw on lint keys
- tRPC admin.queueLint(symbol, kind) — run lint for one symbol, returns LintResult
- tRPC admin.queueLintAll(kind) — backfill ALL watched symbols at once
- tRPC admin.dataQualityList() — query data_quality rows (filterable by symbol/kind)
- InstitutionalDashboard: 'Lint holders' button + status badge in detail panel header
- Admin queue page: Data Quality section with per-row status badges, 'Lint all' buttons
- DEFAULT_RATE_MS includes 167ms (~6 req/s) for lint kinds matching EDGAR limiter
- Pause/resume persisted in queue_state; drain loop checks isPaused() and skips
- Error capture: every failed/backoff attempt logged to queue_errors with the
full stack trace; admin queue page expands a failed job to stream it
- Retry controls: retryJob(key), retrySource(kind), clearDone(olderThanMs)
- Per-source scheduling: queue_schedules table + 30s enqueueDueSchedules loop
(seed defaults sec-fetch 24h, yfinance 5min); admin UI lists/adds/deletes
- Startup recovery: interrupted in_flight jobs reset to pending on boot
- fix(edgar): archive URLs use the filer CIK (accession-number prefix), not the
company CIK — resolves Cloudflare 429 that left SEC backfills sparse/empty
- Migration: add queue_errors, queue_schedules, queue_state tables plus error/
scheduled_for columns on adapter_queue (idempotent ALTER on startup)
Verified: full sec-fetch backfill now succeeds for all watched symbols
(NVDA 14,675 institution filings, CIFR 274 insider txns, TSLA 6,011, etc.).
503 backend tests pass.
Co-Authored-By: Claude <noreply@anthropic.com>
auth.signup/login/logout/me with signed HMAC session cookies + scrypt hashing;
market.snapshot mega-endpoint (quote+candles+sector). node:http server mounts
tRPC at /api/trpc + /health + background drain loop. Live-verified: real NVDA
194.97/65 candles/Technology served after stale-while-revalidate drain. 36 tests green.