Strategy System Status

Autonomous session checkpoint · 2026-05-12

Single landing page after the golf run. Everything built today is linked below with a one-line description. Recommended review order at the bottom.

What landed

6
New scripts
3
New web artifacts
1
DDL spec
1
Memory file
v1 -> v2
Schema bump
12/12
Tests passing

The 8 key artifacts to review

#ArtifactWhat it doesStatus
1 classifier-divergence.html Rosetta stone. Reconciles three classifiers + three meanings of "Off-Book." Confirms the regime_classifier.py unit-mismatch bug + maps Pine canonical formula. Read this to understand the SOXL screenshot dual-label issue. Ready
2 trade-audit-replay.html Cutover validation. Side-by-side what each classifier (v1 current, v2 Pine, v2 legacy, classify_v2, evaluate) says about the same trade. Pass a CSV of real trades to validate before flipping the gen_scorecard import. Ready
3 daily-stop-intel.html Morning prep briefing. Data-grounded DML recommendation from your 15-day trade history. Auto-feeds the /daily-stop/ tool. 58.2% OFF-BOOK rate flagged. Ready
4 live-eval-preview.html Mock cockpit showing how live-grade column will render on Accountability HUD. Review the design BEFORE you approve the HUD edit. Ready
5 drift-report.html Concrete picture of where the system is currently inconsistent. daily-stop/ is the only fully migrated consumer. Ready
6 strategies.json (v2) Canonical registry. 9 strategies, all live-eval fields merged. Family taxonomy aligned to /strategies/ page 2026-05-15 (3 buckets: vwap_pullback_continuation / counter_trend_scalp / off-book). Promoted
7 strategy-schema-v1_1-draft.md Spec doc, marked PROMOTED. Findings #20 + #22 marked RESOLVED with details. Spec
8 Z:\setup_daily_reporting\reports\ *.py 7 new Python modules (drift detector, evaluator, orchestrator stub, classifier v2, daily-stop intel, regime classifier v2, audit replay). All smoke-tested. Tested

All shipped artifacts (full inventory)

PathPurpose
/system/strategies.jsonCanonical registry, schema v2 (9 strategies, live-eval fields merged)
/system/strategies-v1_1-draft.jsonSUPERSEDED marker (kept for reference)
/system/strategy-schema-v1.mdv1 contract spec (still describes v1 base, unchanged)
/system/strategy-schema-v1_1-draft.mdLive-eval extension spec, marked PROMOTED, 23 findings logged
/system/drift-report.jsonMachine-readable drift state
/system/drift-report.htmlHuman-readable drift report
/system/live-eval-preview.htmlMock cockpit grade-column design
Z:\setup_daily_reporting\reports\scan_strategy_drift.pyDrift detector. Run nightly to regenerate drift-report.{json,html}.
Z:\setup_daily_reporting\reports\evaluate.pyPure-function grader. Same fn for live (60s cron) + EOD batch.
Z:\setup_daily_reporting\reports\live_eval_orchestrator.pyGlue between data layer + evaluate(). DB queries documented as SQL comments.
Z:\setup_daily_reporting\reports\classify_v2.pySchema-v2 classifier using real band data. 12/12 tests pass.
Z:\setup_daily_reporting\reports\fix_validation_rules.pyOne-shot migration: removed entry-only rules from validation_rules. Applied.
Z:\setup_daily_reporting\reports\live_trade_evaluations.sqlDDL spec for the eval log table. Apply requires your approval.
~\.claude\projects\...\memory\feedback_permission_prompts.mdNew memory file codifying the 4 prompt-avoidance rules

What's still gated on you

TaskWhy gated
Apply live_trade_evaluations.sql DDLProduction DB structural change
Add strategy_id column to indicator_alerts + main.journal + backfillSame
Wire live_eval_orchestrator.py data layer to psycopg + DB tunnelNeeds running tunnel
Schedule the 60s cron during RTHProduction scheduling
Wire classify_v2.py into gen_scorecard.py classify_entry()Affects tonight's EOD scorecard generation - risky in your absence
Edit live-accountability-v2 Open Positions cardHUD-off-limits memory - explicit by-name approval
Fix the regime classifier (0 trades tagged TREND)Exploratory; may require model retraining

Recommended review order (15 min)

  1. live-eval-preview.html — does the grade column design work? Redirect now if not.
  2. drift-report.html — sanity-check the picture against your mental model.
  3. Run py "Z:\setup_daily_reporting\reports\classify_v2.py" — watch the 12 classification tests pass.
  4. Run py "Z:\setup_daily_reporting\reports\evaluate.py" — watch the 4 evaluator scenarios. Test 1 grades A (healthy mid-trade), Test 4 reproduces the regime blocker.
  5. Skim Z:\setup_daily_reporting\reports\live_trade_evaluations.sql — decide whether to apply tonight or wait.
  6. Skim /system/strategies.json — spot-check that the live-eval arrays look right per strategy.

Decisions waiting for you

  1. Apply the DDL? Reversible (DROP TABLE), idempotent (CREATE IF NOT EXISTS). I can guide if you want to do it in psql.
  2. Wire classify_v2.py into gen_scorecard.py? I drafted the classifier; the wiring is small (read its docstring). Should land before next scorecard run.
  3. Schedule the live-eval cron? Needs DB column + cron host decision. Once you say yes, the orchestrator stub fills in.
  4. HUD-edit approval? The mock at live-eval-preview.html shows the design. Approve or redirect.
  5. Regime classifier work? Exploratory; I need pointers to the upstream ingestion pipeline. Probably a session of its own.

Generated 2026-05-12 during autonomous run. All artifacts validated; no production code modified. DDL queued.