High conviction: validation, Python wiring & audit filter tweaks

What we validated

• Closed-book slice — Running classify_hf_conviction_tier on alpha_engine/data/closed_picks.json showed crypto picks that received HF tiers (mostly tier B in that sample) with materially higher win rate than the crypto baseline; non-crypto rows in the file were too sparse to recalibrate thresholds from outcomes alone.
• HF quality validator — audit_trail/hf_pick_validator.py remains a strict global gate (confidence, R:R, empirical strategy WR, freshness, concentration); it is complementary to tier labels, not a substitute.
• Repeatable script — tools/validate_hf_by_asset_class.py prints tier and baseline win rates by normalized asset class for ongoing checks.

What we shipped in code

• Non-crypto HF tiers in Python — alpha_engine/conviction_stack.py now assigns hf_conviction_tier A/B for PEAD/quality-style strategies using config/hf_conviction_tiers.json (non_crypto_tier_a_strategies, non_crypto_tier_b_strategies, confidence floors). Previously only a patch file described this; it is now wired for real payloads.
• Audit UI parity — passesHighConvictionPick in audit_dashboard/template.html and audit_dashboard/index.html: equity/forex/ETF score-only paths tightened (require trust or forward-history support where appropriate); commodity symbol list typo fixed (GOLD).
• Tests — tests/test_hf_conviction_tier.py covers non-crypto tier A/B examples.

Deploy note

The live audit page loads /audit/index.html; both template and index are uploaded by tools/deploy_to_ftp.py so the High Conviction button logic stays in sync.