Orca — multi-book US earnings paper forward test

Paper-only side-by-side forward test of 14 strategies (12 since 2026-11-10, plus bigmove_unified_v1 and bigmove_eu_v1 from 2026-11-13; EUR 1,000 each). Live: orca-paper.vercel.app.

Safety: mode: paper only. No broker API, no real orders, no keys with trade/withdraw rights. ORCA_MODE must stay paper.

Books

Book	Trader	Rule
agent	Cursor agent (stocks.decide)	Discretionary baseline (earnings + news)
bigbeat_gap	Engine	surprise >15%, gap >5%, SPY>MA50 · 10% · max 8 · hold 3 · no stop
guidance_beat	Engine	raised guidance + surprise >8%, release out · 10% · max 8 · hold 5
beat5_raise	Engine	surprise >5% + raised outlook · 10% · max 8 · hold 5
miss_short	Engine	surprise <-10% AND gap <-5% · 5% · max 4 · hold 2 · 8% stop
news_event	Engine	config-driven (config/strategies/news_event.yaml) · defaults: bullish/bearish · 5% · max 6 · hold 2 · 8% stop
guidance_raise_tuned / dilution_shorts_tuned / quality_longs_tuned / guidance_raise_score / gap_filtered_guidance / wide_capacity	Engine	tuned news configs from box backtest · each €1,000 paper · next_open · mostly no stop
bigmove_unified_v1	Engine + bigmove loop	Fade offering / PR-pop / M&A gaps + selective gap-down fades across pre / regular / after · 5% · max 20 · hold 5 · 0.5%/fill + 0.5% ext slippage · mega-cap shocks alert only · no earnings/guidance chase · start 2026-11-13
bigmove_eu_v1	Engine + bigmove-eu loop	EU liquid large/mid caps: fade ≥10% gap-downs (earnings / other-PR / no-news) on primary hours 09:00–17:30 Berlin · 5% · max 20 · hold 5 · 0.5%/fill (+0.5% outside primary) · STOXX/DAX regime · alert-only mega-caps · no gap-up shorts (EU event study) · start 2026-11-13

Shared candidates + state/stocks/news.json; independent ledgers. Cost 0.5%/fill (bigmove adds configurable 0.5% outside primary/RTH); EUR sizing via EURUSD=X (EU quotes converted when needed). Paper shorts have no borrow model.

Commands

python -m stocks scan                  # earnings calendar candidates
python -m stocks news                  # 24/7 news → news.json + candidate merge
python -m stocks bigmove --once        # EDGAR 8-K/424B/SC TO + move triggers (bigmove book)
python -m stocks bigmove --loop        # ~22 min poll every ~2.5 min + engine ticks
python -m stocks bigmove-eu --once     # EQS/DGAP aggregators + EU movers (bigmove_eu book)
python -m stocks bigmove-eu --loop     # ~22 min poll every ~2.5 min + engine ticks (Berlin primary)
python -m stocks.decide --ticker TICK --side long --size 0.08 --hold 5 --reason "..."  # agent book only
python -m stocks engine --apply        # all books: auto-decide rules + fill/exit/mark
python -m stocks engine --apply --book bigmove_unified_v1
python -m stocks engine --apply --book bigmove_eu_v1
python -m stocks init                  # ensure all books exist (€1,000; does not wipe existing)
python -m stocks replay --live         # rule dry-run over sample + live candidates

See AGENT.md and stocks/DECISION_PROMPT.md for the deciding-agent loop (agent book only).

Listeners (GitHub Actions)

Workflow	When	Hours covered
stocks-news	~every 15 min, all days (5,20,35,50 * * * *)	24/7 news ingest
stocks-scanner	~every 30 min weekdays UTC (7,37 * * * 1-5)	earnings calendar
stocks-engine	every 15 min during US RTH window + after close	regular session fills/marks for all books
stocks-bigmove	every ~30 min 08:00–00:00 UTC weekdays, each run loops ~22 min polling every ~2.5 min	US pre 04:00–09:30 ET, regular 09:30–16:00 ET, after 16:00–20:00 ET (weekdays). Concurrency group stocks-bigmove; rebase/retry on push.
stocks-bigmove-eu	every ~30 min 08:00–16:59 UTC weekdays, each run loops ~22 min polling every ~2.5 min	EU primary ≈09:00–17:30 Berlin (no free Tradegate/L&S). Skips EU holidays. Concurrency group stocks-bigmove-eu; rebase/retry on push.
stocks-decide	workflow_dispatch	manual agent helper

Bigmove hours & cadence

• Intended coverage: weekdays 04:00–20:00 America/New_York (pre + regular + after).
• UTC schedules: crons at :05 and :35 for hours 8–23 UTC Mon–Fri, plus early 0 UTC Tue–Sat to catch the last after-hours slot under EST/EDT.
• Intra-run loop: each workflow invocation polls EDGAR + movers for ~22 minutes every ~150s and runs the bigmove_unified_v1 engine each cycle, so coverage does not depend on cron firing exactly on the minute.
• Overlap safety: concurrency.group: stocks-bigmove with cancel-in-progress: false; commits rebase/retry to avoid ledger write conflicts.
• Actions minutes: ~16 scheduled fires/weekday × ~25 min ≈ ~6–7 runner-hours/weekday for this workflow alone — dense but within a typical public-repo free budget when combined with the lighter news/scanner/engine jobs.

If schedules never run (Jonas)

Cloud agents cannot flip repo Actions settings or workflow_dispatch. Do this once:

1. Open Settings → Actions → General
2. Actions permissions → Allow all actions and reusable workflows → Save
3. Workflow permissions → Read and write permissions (required so the bot can commit state) → Save
4. Open Actions → stocks-scanner → Run workflow (branch main)
5. Same for stocks-engine / stocks-bigmove

After one successful run, weekday schedules should keep firing. Pushing the workflow YAML to main also triggers a one-shot run via the paths filter.

Fallback if schedule never fires: workflows also listen for repository_dispatch types orca-stocks-tick / orca-stocks-news / orca-stocks-engine / orca-stocks-bigmove. Point an external cron (e.g. cron-job.org) at:

curl -X POST -H "Authorization: Bearer $GH_PAT" -H "Accept: application/vnd.github+json" \
  https://api.github.com/repos/Prgrmmrjns/orca/dispatches \
  -d '{"event_type":"orca-stocks-tick"}'

(GH_PAT needs actions:write. Cloud agent tokens cannot dispatch.)

Layout

demo/                      # Vercel root — multi-book comparison dashboard
stocks/                    # scanner, engine, decide CLI, rule books, bigmove
config/strategies/         # frozen YAML for config-driven books
state/stocks/candidates.json
state/stocks/news.json
state/stocks/alerts.json   # mega-cap shock alerts (alert_only)
state/stocks/books/<id>/   # ledger.json + decisions.json per book

Disclaimer

Paper results are not predictive. Educational software only.