Skip to content

PHASE_HANDOFF.md

Latest commit

 

History

History
180 lines (168 loc) · 10.5 KB

PHASE_HANDOFF.md

File metadata and controls

180 lines (168 loc) · 10.5 KB
Raw

Phase Handoff - Trader Risk Audit

Date: 2026-05-19

Use this file only for restart/context recovery. Detailed history lives in docs/IMPLEMENTATION_JOURNAL.md, docs/EVIDENCE_INDEX.md, docs/archive/, and docs/tasks.md.

Current State

  • Phase: 26 Private Pilot Evidence Collection
  • Last completed: T140 Dune Public Wallet Rehearsal Review
  • Active task: T116 Operator-Approved Private Run And Reviewed Report Evidence (blocked pending operator private input)
  • Baseline: 263 pass / 0 skip
  • Ruff: clean
  • Open findings: P2 carry-forwards only; no P0/P1. Phase 23, Phase 25, Phase 27, Phase 28, and Phase 32 caveats remain accepted/blocked limitations.
  • Constraint state: T93 deferred real local read-only exchange fetching; Phase 22 implementation tasks are blocked. Core is paused.
  • Branch: product-local working branch

Next Action

Read:

  1. docs/CODEX_PROMPT.md
  2. docs/tasks.md Phase 26, T116
  3. docs/HYPOTHESIS_VALIDATION_WITHOUT_PRIVATE_EXPORT_PLAN.md
  4. docs/OPEN_SOURCE_AUDIT_VALIDATION_ROADMAP.md
  5. docs/PRE_PRIVATE_HYPOTHESIS_EVIDENCE_MATRIX.md
  6. docs/PRE_PRIVATE_DISCOVERY_SCRIPT_RU.md
  7. docs/PRE_PRIVATE_REPORT_CONVERSATION_PACK.md
  8. docs/PRE_PRIVATE_EVIDENCE_CAPTURE_RUNBOOK.md
  9. docs/PRE_PRIVATE_OUTREACH_EVIDENCE_REVIEW.md
  10. docs/PAID_PILOT_READY_GATE.md

Start T116 only after operator private input exists outside git. If it does not exist, run the Phase 30 concierge outreach loop outside git and validate aggregate logs with evidence aggregate-validate before promoting summaries. The Dune report can support report-review conversations, but it is not private/customer/paid evidence. T98 created docs/OPEN_SOURCE_CASE_BANK.md with the source-selection protocol, allowed/excluded source classes, license/terms notes, anti-cherry-pick batch composition, and artifact-quality-only evidence boundary. T99 added trader_risk_audit.validation.open_source_case, case-bank validate, focused validator tests, and demo/open_source_sec_form4_001/output/reproducibility_status.json so the SEC Form 4 pack passes the directory contract. T100 listed five candidate packs in docs/OPEN_SOURCE_CASE_BANK.md: SEC reference, public_sample_001, risk_audit_case_001, synthetic_limit_leverage_001, and synthetic_schema_reject_missing_price_001. T101 generated complete runnable artifacts/status for SEC, public sample, risk audit, and synthetic leverage limitation packs; rejected the missing-price schema pack with only output/run_status.json; and wrote docs/OPEN_SOURCE_AUDIT_BATCH_INDEX.md. T102 added per-pack review notes under docs/audit/open_source_case_reviews/, docs/audit/PHASE23_ERROR_REGISTER.md, and reviewed-report limitation/provenance headers. Current register is P0:0, P1:0, P2:3. T103 archived docs/archive/PHASE23_REVIEW.md, updated audit index/state docs, and advanced to Phase 24 with no stop-ship findings. T104 added docs/REPORT_QUALITY_SCORECARD.md; SEC Form 4 reviewed report scored 17/18 with no fail condition as internal reference. T105 added docs/OPEN_SOURCE_RULE_COVERAGE_MATRIX.md, mapping packs to rules, fields, unsupported fields, limitations, output sections, and explicit follow-up coverage cases. T106 added docs/OPEN_SOURCE_AUDIT_QUALITY_DASHBOARD.md, classifying three controlled internal demo-quality packs, one internal reference pack, and one blocked/rejection-only pack. T107 added evidence_overclaim claim-guard phrases/tests and docs/PHASE24_REGRESSION_DECISIONS.md; baseline is now 258 passing tests. T108 added docs/INTERNAL_DEMO_PACK_OPEN_SOURCE_AUDITS.md with positive, limitation, and schema-reject examples plus safe excerpts, talk track, buyer promise, and paid-pilot ask. T109 archived Phase 24 with Stop-Ship: No, P0:0, P1:0, P2:3 accepted carry-forward caveats, and advanced to Phase 25. T110 added docs/PRIVATE_PILOT_INTAKE_CHECKLIST.md and linked it from docs/PILOT_INTAKE_CONTRACT_RU.md; it maps private intake to existing local CLI commands and forbids committing private rows, identifiers, credentials, private paths, and unapproved screenshots. Build the private report review checklist next. T111 added docs/PRIVATE_PILOT_REPORT_REVIEW_CHECKLIST.md and scorecard linkage; unresolved P0/P1 truth, privacy, policy, advice, live-control, or performance-claim issues block delivery. T112 added docs/private_pilot_runs/ with a safe run-note template and a blocker note: no operator-approved private/anonymized input was supplied, no private report was run or reviewed, and delivery remains blocked_do_not_deliver. T113 added docs/PAID_PILOT_PACKAGE.md and docs/PRIVATE_PILOT_FEEDBACK_LOG_TEMPLATE.md for one manual reviewed audit, safe feedback capture, no checkout, no SaaS, no live-control, no order blocking, and no advice/performance claims. T114 added docs/PAID_PILOT_READY_GATE.md with decision needs_fixes, not ready, because no operator-approved private/anonymized report has been run and manually reviewed. T115 archived Phase 25 with Stop-Ship: No, P0:0, P1:0, P2:4 and WARN health. T116 remains the specific blocking fix for private delivery readiness: one operator-approved private/anonymized local run and manually reviewed report, with only safe metadata committed. T117-T121 completed Phase 27 real-open-data rehearsal using real Ethereum mainnet Uniswap V2 WETH/USDC pair-level Swap logs. The pack demo/real_open_dex_swaps_001/ is reproducible and manually reviewed with P0:0, P1:0, P2:2 accepted caveats; it is development rehearsal only and does not close T116. T122 added demo/real_open_dex_contract_sequence_001/, using the same real Uniswap V2 WETH/USDC source filtered to one repeated public contract recipient. It is reproducible and manually reviewed with P0:0, P1:0, P2:2 accepted caveats. It is development rehearsal only and does not close T116. T123-T125 started Phase 29 with an evidence ladder, evidence matrix, RU discovery script, report conversation pack, and evidence capture runbook. T126 found no supplied aggregate outreach/report-review evidence and chose continue_concierge_validation. T127 archived Phase 29 with WARN health, Stop-Ship: No, P0:0, P1:0, P2:1. T128-T132 added the Phase 30 concierge validation execution kit and archived it with WARN health, Stop-Ship: No, P0:0, P1:0, P2:1. T133-T136 added Phase 31 aggregate evidence safety tooling and archived it with OK health, Stop-Ship: No, P0:0, P1:0, P2:0. T137-T140 added demo/dune_public_wallet_dex_001/, a Dune public Ethereum dex.trades case pack with 80 real public rows, 76 max-position findings, one unsupported leverage limitation, passed reproducibility, and P0:0, P1:0, P2:2 accepted caveats. It is report-review/development evidence only and does not close T116. T94-T97 are blocked until future privacy-safe market evidence reopens T93. Phase 17 added deterministic local intake_session.json, sanitized schema_profile.json, safe intake_report.md, and Cycle 22 deep review/archive with no stop-ship items. T74 added trader_risk_audit/policy/rule_catalog.py for supported deterministic rule types, threshold units, safe descriptions, starter profile applicability, and availability requirements. T75 added trader_risk_audit/policy/builder.py and policy build for deterministic generated policy.yaml. T76 added policy flow for structured local rule-builder input and unavailable-rule explanations. T77 added policy unsupported append and trader_risk_audit/policy/unsupported_register.py for sanitized manual-review limitations. T78 archived Cycle 23 with no stop-ship findings. T79 added trader_risk_audit/audit_session/runner.py and audit-session run, which gates on runnable intake/policy status, runs the deterministic audit internals, and writes safe complete/blocked run_status.json without raw rows in status/stdout. T80 added trader_risk_audit/audit_session/artifact_bundle.py and audit-session bundle, which writes/validates bundle_index.json with safe refs, hashes, status, preview state, and limitation register refs. T81 added trader_risk_audit/audit_session/reproducibility.py, which reruns a session in a separate output directory and writes safe reproducibility_status.json before preview/delivery. T82 archived Cycle 24 with P0:0, P1:0, P2:0 and no stop-ship items. T83 added trader_risk_audit/preview/model.py and preview build, which renders claim-safe redacted previews from bundle refs without source-row ids, symbols, raw rows, or private output paths in CLI stdout. T84 added eligible manual paid-pilot CTA packaging without checkout/SaaS/payment processing. T85 added privacy-safe preview conversion events and summary. T86 added local paid-unlock state transitions with manual evidence references only. T87 archived Cycle 25 with P0:0, P1:0, P2:0 and no stop-ship items. T88 added HypothesisFunnelEvent, combined legacy/new evidence loading, and docs that separate gate evidence from vanity/demo events. T89 added evidence hypothesis-dashboard, which summarizes legacy and funnel evidence locally with market/demo separation, ratios, gate status, blocker tags, and privacy-safe next action output. T90 added evaluate_hypothesis_gate, which returns proceed, needs_more_evidence, or pivot with concrete reasons and next action. T91 added evidence export, which writes aggregate-only CSV/Markdown evidence summaries with gate decision, safe tags, source log names, and source log hashes while rejecting identifiers, phone numbers, credentials, payment ids, raw-row-like fields, and private paths. T92 archived Cycle 26 with P0:0, P1:0, P2:0 and no stop-ship items. Do not implement real exchange network fetching unless T93 explicitly returns proceed. T93 returned defer because no market customer evidence log shows CSV/export friction as the binding blocker.

Guardrails

  • Raw private/customer trade data must not be committed.
  • Phase 14/15 exchange-import work is complete but should be used only when it directly supports the selected real audit artifact or the T93/T94 friction gate.
  • ADR-002 allows only local read-only historical ingestion.
  • No live exchange control, order placement, hosted uploads, hosted storage, checkout, order blocking, SaaS expansion, or trading advice.
  • Open-source validation must preserve positive, limitation/reject, and edge-case examples to avoid cherry-picking.

Historical Pointers

  • Completed through T140. T94-T97 blocked. Active route is T116 if approved private input appears; otherwise run Phase 30 concierge outreach outside git.
  • New active roadmap: docs/OPEN_SOURCE_AUDIT_VALIDATION_ROADMAP.md.
  • Detailed validation command history is in docs/IMPLEMENTATION_JOURNAL.md.
  • Prior review artifacts are in docs/archive/.