Codex Handoff: Pam — Next Tasks

Date: 2026-04-16 Branch: Work from main Test baseline: 181 pass, 7 fail, 22 error (210 total) Goal: All 210 tests passing, research loop producing valid gate metrics, live scanner surfacing at least 1 deployable_trade per scan cycle.

Priority 1 — Fix Failing Tests

1a. Book dataclass missing contract_side field (22 errors)

The Book dataclass in arbitrage/liquidity.py now requires a contract_side field, but three test files still construct Book without it.

Fix: Add contract_side="yes" to every Book(...) call in: - tests/test_liquidity.py - tests/test_pricing.py - tests/test_executor_paths.py

This is a mechanical find-and-replace. Once done, 22 errors should clear.

1b. equivalence_score() signature changed (3 failures)

equivalence_score() in matcher/align.py now requires 4 positional args: (pm, pm_parsed, ks, ks_parsed, expiry_tol_sec=90). The 3 call sites in tests/test_matcher.py still pass only 2 args.

Fix: Update each call to pass ks and ks_parsed dicts as the 3rd and 4th arguments. Check tests/test_matcher.py for the context around each call to determine appropriate test fixture values.

1c. Stale expiration fixture in tests/test_scanner_gates.py (1 failure)

Line 48 uses "2026-04-10T12:00:00Z" as an expiration date, which is now in the past.

Fix: Change "2026-04-10T12:00:00Z" → "2027-04-10T12:00:00Z". One-line change.

1d. Equivalence score regression in tests/test_parsers_matcher.py::test_equivalence_close (1 failure)

The BTC price pair test expects a score of >=0.9 but currently returns 0.76. This is a real logic regression, not a stale fixture.

Investigation path: - Look at recent changes to matcher/align.py, specifically the penalty weights used by equivalence_score() - The score drop from >=0.9 to 0.76 suggests a penalty was added or an existing weight was increased - Run the test in isolation with verbose output to see which sub-score is dragging it down - Restore the penalty weights for the BTC-style case, or adjust the scoring so a close price-pair match still scores >=0.9

Priority 2 — Research Loop State

research/latest_state.json currently shows confident: True with score: None, stable_cycles: None, hit_rate: None. The deployment gate is passing on empty metrics — this is a false positive.

Steps: 1. Run: python main.py research-loop --max-cycles 1 2. Verify research/latest_state.json is repopulated with non-null values for score, stable_cycles, and hit_rate 3. If the gate passes again after the run, confirm it’s passing on real numbers, not nulls 4. If the run fails or produces nulls, check research/continuous_loop.py for where these fields are written — the write path may have a silent exception or early exit

Context: research/best_strategy_params.json currently holds the best known baseline (score: 17.6, hit_rate: 0.8). The research loop should be able to evaluate against these params and produce a valid state snapshot.

Priority 3 — Live Scanner Effectiveness

The live scanner (strategy/kalshi_opportunity_scanner.py) produces mostly watch rows. The replay strategy is healthy (score: 17.6), but the live signal path doesn’t translate to deployable trades.

Current blockers (from CODEX_HANDOFF_STRATEGY_FIX.md): - Global: stable_cycles < 3 — gate requires 3 consecutive cycles of stable confidence before deploying - Local: live_edge_bps < 100.0 — individual candidates aren’t crossing the edge threshold

Task: Add a quantitative bypass path in the live scanner analogous to the replay-side quantitative_eventful_reversal sleeve.

The replay sleeve bypasses heuristic_direction entirely, gating only on eventful_score, price_move_bps, tradeability, and low live price. The live scanner should have a parallel code path that does the same: if a market meets the quantitative criteria, classify it as actionable regardless of heuristic_direction.

Key files: - strategy/kalshi_opportunity_scanner.py — add the bypass path here - research/candidate_review.py — check how live candidates are classified; may need mirroring - research/strategy.py — reference implementation of quantitative_eventful_reversal sleeve

Verification:

cd /Users/joshua/Projects/Unicorn.Land/pam
source .venv/bin/activate
python main.py opportunity-scan --limit 20
python main.py trade-candidates --limit 20 --json

Target: at least 1 row with deployable_trade status. If still zero, document the exact remaining blocker mix: - live heuristic direction - live edge estimation - stable-cycles gate - sleeve matching / semantic mismatch

Priority 4 — Cleanup (Low Priority)

Remove stale debug and config files from the project root once the above are complete.

Debug scripts to delete: - debug_dates.py - debug_final_issue.py - debug_polymarket_discovery.py - algorithm_diagnosis.py - fix_algorithm.py - fix_discovery_patterns.py

Stale config files to delete: - config_backup_20250913_143604.yaml - config_backup_20250913_143647.yaml - config_backup_20250913_143722.yaml - config_fixed.yaml - config_optimal.yaml

Use trash not rm for all deletions.

Success Criteria

Key Files Reference