Phases numbered by execution priority. Each phase: goal, status, task summary. Cross-references ARCHITECTURE.md for technical detail.
Goal: Single compiled Rust binary with zero Python runtime. All hardcoded thresholds parameterised in config.yaml.
| Status | Count |
|---|---|
| ✅ Complete | 13/13 |
All core modules ported: WebSocket, order book mirror, arb math, position manager, constraint detection, market scanner, resolution validator, postponement detector, dashboard, notifications. Three rationalised scripts (start.sh, restart.sh, kill.sh). CHANGELOG.md and INCIDENT_LOG.md created.
Goal: Code-complete execution path — all modules built, unit-tested, and shadow-testable. No funded account required.
| Status | Count |
|---|---|
| ✅ Complete | 29/29 |
Depth gating, replacement chain analytics, closed position retention.
EIP-712 order signing, dynamic tick size handling. B2.3: FAK vs GTC precision verified with 21 comprehensive tests across all tick sizes and order types.
Market BUY quantity guard, LiveExecutor, rate limiting, timestamp normalisation, error handling, partial fills, batch orders. B3.2: Suspense accounting (MATCHED → suspense, CONFIRMED → real, FAILED → reverse + sell opposing legs). FillAction enum + process_fill_event().
Cross-asset fill matching, FOK/FAK overfill handling, live P&L tracking. B4.0/B4.1: Enhanced reconciliation with Data API freshness polling (query_clob_positions_fresh) + auto apply_reconciliation. B4.5: Parallel WS + Data API trade confirmation (fill_confirmation.rs). B4.6: USDC.e balance monitor (usdc_monitor.rs) with dashboard integration.
ARCHITECTURE.md, ROADMAP.md (this file).
Verification: 101/101 unit tests pass. Milestone D 8/8 PASS validates execution path end-to-end.
Goal: System can protect capital automatically and alert the operator. No funded account required.
| Status | Count |
|---|---|
| ✅ Complete | 10/10 |
Done: Circuit breaker (C1), POL gas balance monitoring (C1.1), kill switch (C2), Telegram notifications (C3), daily P&L report (C4), seamless close transition (C4.1), proactive near-resolution exit (C4.2), OPS_RUNBOOK.md (C5), USER_GUIDE.md (C6), retire PROGRESS_ROADMAP.md (C7).
Goal: Prove full execution path works against real Polymarket CLOB. Place, fill, cancel, and reconcile real micro-orders. Target: 1 week.
| Status | Count |
|---|---|
| ✅ Complete | 9/9 |
VPS: [vps-provider] Madrid ([madrid-vps-ip]). Interim location — Dublin (0.83ms latency) planned when capacity available. Funds: ~$50 USDC.e deposited, wallet 0x21f1...fb1. First real CLOB order: 2026-03-20 from Madrid VPS.
| Task | Status | Notes |
|---|---|---|
| D1: Deposit test funds | ✅ | REST API balance check |
| D2: Submit + cancel order | ✅ | GTC order, verify on book, cancel |
| D3: Real micro-fill | ✅ | FAK BUY at market price |
| D4: negRisk market fill | ✅ | negRisk-specific signing |
| D5: Multi-leg arb execution | ✅ | WS User Channel fill tracking confirmed. Auth fix: WS uses raw creds (not HMAC). |
| D6: Cold-start reconciliation | ✅ | Data API freshness polling, venue state change detection, apply_reconciliation (venue source of truth). FAK/GTC amount precision fix. |
| D7a: Circuit breaker | ✅ | Engine state validation |
| D7b: Kill switch | ✅ | Executor cancel-all |
| D8: Closeout positions | ✅ | Real SELL orders + accounting verification |
| D9: Partial fill test | ✅ | Validated decision logic against real CLOB via clob_test/src/tests/d9_partial_fill.rs (commit fae8786). SELL at 0.99 triggers reliable one-sided fill; evaluate_partial_fills() correctly returns Unwind for unprofitable partials. Unwind execution is F-pre-1 (G1). |
- Trade confirmation reliability: WS User Channel drops ~20% of CONFIRMED events. Current mitigation: accept MATCHED, fall back to Data API. Next: run WS + Data API poll in async parallel, take first confirmation. Needs integration in: fill_tracker, D6 helper, D8 closeout, production executor.
- USDC balance monitor: Add periodic on-chain USDC.e balance check to production engine (like gas_monitor for POL).
- Suspense accounting: MATCHED trades should go to suspense account, CONFIRMED promotes to real position. Failed matches reverse + sell opposing arb legs.
Goal: Prove complete system works without real money. Determine optimal parameters via 6 parallel shadow instances.
| Status | Count |
|---|---|
| ✅ Complete | 6/12 |
| 🔄 In Progress | 2/12 |
| ⬚ Remaining | 4/12 |
- E1: Multi-instance support —
--instance <n>flag, per-instance config overlays. - E2.5: Stress test harness —
scripts/stress_test.pywith--profile worst-case, crash simulation, safe-zone analysis, config auto-update./metricsendpoint (35 fields). - E3: Comparison dashboard — Strategies tab with Sharpe/Sortino/Recovery Factor/Profit Factor/Max Drawdown/Capital Utilisation.
--test-periodauto-stop. Test period countdown banner. - Audit v5-v7: 22 findings fixed (API auth, accounting, robustness, fill validation). 101/101 tests pass.
- Risk mitigations (external risk analysis, 2026-03-23):
- P1: Geoblock health check (
scripts/geoblock_check.sh) — 15-min cron, CLOB + Gamma 403 detection. - P2: Risk register expanded 14 → 19 risks, corrected ratings (INC-001, INC-012).
- P3: Dashboard security documented (SSH tunnel, kill switch exposure).
- P4: Suspicious arb flagging — arbs > 20% profit flagged WARN + dashboard
[!]. - P5: negRisk correlated exposure cap — proportional scaling to 50% max.
- P6: UMA dispute active monitoring — detects
proposed/disputedstatus, excludes from replacement.
- P1: Geoblock health check (
- Sports WebSocket (
sports_ws.rs): Persistent connection to Polymarket Sports WS. Pre-screens postponement detection before AI calls. SQLite persistence. 6 unit tests. 107/107 tests pass.
- E2.6: Running on VPS (~35hrs, worst-case profile, target-cpu=native). CRITICAL: Review results and update config.yaml parameters BEFORE starting E4.
- E4: 14-day shadow run started 2026-03-24 14:16 UTC. v0.20.0 added
evaluated_opportunitiesSQLite table and strategy eval counters for post-run threshold calibration.
E5 (execution validation), E6 (reconciliation clean), E7 (performance baseline), E9 (parameter selection), E10 (CTO sign-off).
Shadow-F targets short-lived crypto price markets (5-15 min resolution) with fast rebuild intervals (60s) and low replacement cooldown (10s).
Goal: First real money trades with optimised parameters from shadow validation.
| Status | Count |
|---|---|
| ⬚ Remaining | 15/15 (9 pre-live gates + 6 go-live tasks) |
Funds required: $1,000 USDC additional (~$1,050 total including D1 deposit).
Pre-live gates (F-pre-1 through F-pre-9): Wire executor, S1/S2 security, fill quality logging, capital velocity metric, geoblock runbook, pre-trade Gamma freshness check, resolve deferred audit items, D9 partial fill test.
Go-live tasks (F1-F6): deposit funds, configure winning parameters, switch to live mode, supervised first trades, 48h monitoring period, steady state confirmation.
Goal: Post-launch improvements. Not blocking go-live. Prioritise based on operational experience.
| Status | Count |
|---|---|
| ⬚ Remaining | 4/4 |
- Increase capital allocation (after 30 days profitable)
- Historical performance dashboard
- Dashboard control panel (mode switch, config edit)
- Multi-exchange support (Kalshi)
| Milestone | Tasks | Status |
|---|---|---|
| A: Rust Port | 13/13 | ✅ Complete |
| B: Execution Infrastructure | 29/29 | ✅ Complete |
| C: Safety | 10/10 | ✅ Complete |
| D: CLOB Integration Test | 9/9 | ✅ Complete |
| E: Shadow Validation | 6/12 | 🔧 In Progress |
| F: Go Live | 0/15 | ⬚ Planned (9 pre-live gates + 6 tasks) |
| G: Scale | 0/4 | ⬚ Future |