Sydnee nightly — execution audit — 2026-04-24 P0 findings: 1 P1 findings: 1 Risks: 2 - Area: Execution Quality (Friday theme — day 4) - Bugs found (P0 / P1): 0 / 1 - Risks noted: 2 - API note: dev.sydnee.ai unreachable from audit host (host_not_allowed). All findings are static code analysis. --- Full report (dev branch): https://github.com/kanex1/sydnee.signals/blob/dev/docs/audit_2026-04-24_execution.md Reply FROM [email protected] to [email protected] to request fixes, e.g.: "code_task on sydnee-signals-dev: apply fix for the P0 about RVOL threshold in bot.py" Sydnee Agent will propose + you APPROVE (or plain 'approve') + auto-push to dev. --- Full audit below (first 12 KB) --- # Nightly Audit 2026-04-24 — EXECUTION QUALITY ## Summary - Area: Execution Quality (Friday theme — day 4) - Bugs found (P0 / P1): 0 / 1 - Risks noted: 2 - API note: dev.sydnee.ai unreachable from audit host (host_not_allowed). All findings are static code analysis. --- ## Findings ### BUG [P1]: FLAT_EXIT trim1 closes 100% of shares for ALL setups — trend runners killed at trim1 price **File:** `bot.py:3759-3760` (trim quantity) vs `bot.py:4927-4930` (_compute_exits guard) **Evidence:** `_compute_exits` correctly guards FLAT_EXIT to mean_revert and extension_fade only: ```python # bot.py:4927-4930 if os.environ.get("FLAT_EXIT", "").lower() in ("1", "true", "yes") \ and setup in ("mean_revert", "extension_fade"): trim_dist = 1.0 * daily_atr_dollar # exit distance override tgt_dist = 1.0 * daily_atr_dollar ``` But the trim1 execution block has no setup-type guard: ```python # bot.py:3759-3760 _flat_exit = os.environ.get("FLAT_EXIT", "").lower() in ("1", "true", "yes") trim_shares = t.shares if _flat_exit else t.shares // 2 # ← no setup check ``` When FLAT_EXIT=true (currently active on dev), `trim_shares = t.shares` for **all** setups. A trend trade — whose trim_dist in `_compute_exits` is kept at 0.5×daily_atr (line 4899, unaffected by FLAT_EXIT) — will have trim1 fire at the correct distance but close 100% of the position instead of the intended 50%, destroying the runner leg. Secondary: the TRIM1 activity log at lines 3782 and 3785 hardcodes "50%": ```python self._log_activity(sym, f"TRIM1 50% ({trim_shares}sh) at ${live_price:.2f} — ...") ``` When FLAT_EXIT causes a 100% close, the log still says "50%", masking the incorrect execution in the activity trail and making post-trade audit unreliable. **Impact:** - With FLAT_EXIT=true on dev (enabled 2026-04-18), any trend trade that hits trim1 is fully closed at +0.5×daily_atr instead of 50%-trimmed with a runner. Strategy_decisions.md line 143 explicitly states "Trend setup unchanged (ladder kept)." - Artificially inflates the trimmed_1 hit rate metric and skews the FLAT_EXIT validation comparison (P&L appears improved but a fraction is from prematurely cut trend runners). - If FLAT_EXIT is cherry-picked to main before 2026-05-01 go-live, trend runners on real money will be fully closed at +0.5×ATR. **Fix:** ```python # bot.py:3759-3760 — add setup-type guard matching the _compute_exits condition _flat_exit = os.environ.get("FLAT_EXIT", "").lower() in ("1", "true", "yes") _is_flat_setup = "TREND" not in (t.strategy_reason or "").upper() trim_shares = t.shares if (_flat_exit and _is_flat_setup) else t.shares // 2 # bot.py:3782 and 3785 — fix log to show actual percentage _trim_pct = "100%" if (_flat_exit and _is_flat_setup) else "50%" self._log_activity(sym, f"TRIM1 {_trim_pct} ({trim_shares}sh) at ...") ``` Uses `strategy_reason` (persisted, restart-safe), consistent with the HOLD_CAP_MIN approach at line 3398. --- ### RISK: Friday 3:50 PM ET flat depends on bar events — no independent timer fallback (carried from 2026-04-17) **File:** `bot.py:3088` (flat check inside `_evaluate_symbol`) **Evidence:** The `t_min >= 15 * 60 + 50` flat sweep lives inside `_evaluate_symbol(sym)`, which is only triggered by IBKR real-time bar events (`rtb.updateEvent`). There is no standalone periodic sweep in the main loop that guarantees every symbol is evaluated at 3:50 PM ET. If a symbol's RTB feed goes quiet between 3:40 and 3:50 PM ET (thin tape, IBKR routing hiccup, or brief connection drop), that position is not flattened by the 3:50 PM ET sweep. The AH fallback (7:50 PM ET) would catch it, but that is 4 hours of unintended weekend exposure. **Today is Friday 2026-04-24. This risk is actively open.** FLAT_EXIT=true is also enabled today, meaning trend positions that hit trim1 (see P1 above) may already be 0-share zombies at 3:50 PM ET — so the flat sweep may see `t.shares < 1` and skip, leaving a DB-open trade that holds over the weekend until reconcile. **Mitigation already in place:** AH trim + flat (7:50 PM ET) catches anything missed at 3:50 PM ET. Friday AH is a separate code path (lines 3113–3166) with its own per-symbol iteration. Risk remains but is backstopped. --- ### RISK: FLAT_EXIT zombie parent persists in DB until reconcile — generates spurious 0-share SELL orders (carried from 2026-04-23) **File:** `bot.py:3760, 3793-3794` **Evidence:** When FLAT_EXIT=true and trim1 fires 100%: ```python trim_shares = t.shares # 100% remaining = t.shares - trim_shares # = 0 self._update_trade_field(t.trade_id, shares=0) ``` The parent trade (`exit_price IS NULL`) stays in `open_trades()` with `shares=0`. On every subsequent 5m bar, the exit loop re-evaluates the parent. If a stop fires while shares=0: - `_close_trade(t, ...)` → `MarketOrder(SELL, 0)` → IBKR rejects - Repeated until reconcile detects IBKR qty=0 and marks the trade closed (~5 min) No financial harm (IBKR rejects 0-share orders); nuisance log pollution and spurious `signals.audit` duplicate-close events. Carries from 2026-04-23 persistence audit where the fix was documented. --- ## OK (checked, working or recently fixed) **Fixed since 2026-04-17 execution audit:** - `fills[0]` P0: avgFillPrice now used at both entry path (bot.py:5376) and exec-qa patrol (bot.py:5809-5810), falling back to fills[0] only when avgFillPrice is unpopulated. ✓ Comments at both sites explicitly reference audit_2026-04-17_execution.md. **Working correctly:** - `total_scale` applied to shares before entry (bot.py:5098-5102): confirmed multiplication, not just logging. Zero-share guard at bot.py:5114-5116 fires after all sizing caps. ✓ - `HOLD_CAP_MIN` trend exemption (bot.py:3398): `_is_capped_setup` checks for MEAN_REVERT / EXTENSION_FADE in strategy_reason; trend (which has "TREND" in reason) is implicitly excluded. Hold time measured from `entry_time` via fromisoformat with timezone handling. ✓ - `OBV_LOSS_EXIT` loss-only guard (bot.py:3349): fires only when `profit_per_share < 0` in combination with OBV adverse. Profitable trades take the BE-stop-tighten path instead. ✓ - `FLAT_EXIT` distance in `_compute_exits` (bot.py:4927-4930): trim_dist = tgt_dist = 1.0×daily_atr correctly guarded to mean_revert and extension_fade only. Trend keeps original ladder distances. ✓ (But see P1 above for the trim quantity issue.) - Friday 3:50 PM ET flat (bot.py:3088-3105): no rsi_extreme exemption present. Comment at lines 3081-3087 explicitly notes the 2026-04-17 decision to drop the carve-out. ✓ - Entry duplicate protection: three-layer guard at bot.py:4985-5024 (in-store, IBKR cache, ib_insync wrapper). Deferred entry dedup at lines 4573-4584 collapses to latest-per-symbol within a drain cycle. ✓ - Stop-loss check (bot.py:3417-3421): uses current bar's close_price, correct direction comparison. Stop calculated from daily ATR at entry, stored in Trade and persisted to DB. ✓ - `RSI_TRIM_PROFIT_GUARD`: correctly skips BE-stop-tighten when at a loss, at all three fire sites (TRIM1 at 3780-3782, TRIM2, OBV rule F at 3347-3349). ✓ **Carried from prior audits — unfixed, no change:** - `signal_price` / `signal_time` not restored on restart (P1 filed 2026-04-17; slippage analytics only, no exit logic depends on in-memory value). - `entry_time` records fill-completion timestamp, not signal time (P1 filed 2026-04-17; analytics only). - `_setup_mode` not persisted to DB (RISK filed 2026-04-16; no current exit logic uses it directly — HOLD_CAP correctly uses strategy_reason string instead). - `_fri_ah_close_queued` not persisted (RISK filed 2026-04-23; protected by `_close_trade` is_open guard).