Where we go after Trust-Final. For each task, tap YES to approve, NO to skip, or MODIFY and write what you actually want. Decisions auto-save. When you're ready, hit Submit — Claude reads everything and starts on what you approved.
This is what "running through the validated system" actually means. Each strategy goes through all seven, in order, halt-on-failure.
| Step | Function | Output |
|---|---|---|
| A Locate | Find canonical source code | Resolved path or fail-loud |
| B Port | Wrap in v3 bundle layout | bundles/<id>/manifest.yaml |
| C Harness | Run canonical_gated harness, full history | Trade ledger + equity curve in PG |
| D Pin | Lock metrics | upstream_validations/<id>.meta.yaml |
| E Binary test | One historical trade locked as CI | tests/v3/strategies/test_<id>_reference.py |
| F Register | Frontend picks up automatically | Dashboard row + trust_state badge |
| G Guard | Pre-commit blake3 hash check | Drift refused at commit |
Bundled as scripts/canonicalise_strategy.sh <bundle_id>. One command. Smoke-tested on a known-good (sma4_btc_gated) before any cohort run.
From Playwright inspection at 127.0.0.1:8098 on 2026-05-07. Backend numbers are correct; these are render-layer issues.
-41.66 on SMA-4 BTC — annualisation × tiny n = noise+8740% / 22 trades from legacy paper_lifetime_backfill— while API returns -3.932 PAPER badge in YTD hero (actual count: 8)0.00 across the board (should be —)## Wave 1A — Clock injection [RESOLVED · 2026-05-04 · ae2a705]
### What was broken
ema22_gate cooldown used time.time(); fast backtest replay locked the gate.
### What we did
Bar-timestamp clock injected. Test added.
### What that resolved
Jan 19 SMA-4 BTC ema22_gate_exit @ $86 625.60 fires correctly.
### What it surfaced
Wave 1B — pinned metrics computed against stale Mar 20 checkpoint.Append-only. Every closed stage gets an entry. Forward stages get a hypothesis entry on open and a resolution entry on close. Goes in the v3 git repo so it travels with the code.
Once submitted, Claude reads your responses and starts executing. You can still come back and edit, but rerunning the build manually after edits is on you.