Sydnee nightly — exit_flow audit — 2026-05-26 P0 findings: 0 P1 findings: 2 Risks: 4 - Area: Exit flow (how and when open positions are closed) - Bugs found (P0 / P1): 0 / 2 - Risks noted: 4 Branch: `dev` (dc2d9b4). Cross-checked against `git log --oneline -30` and `docs/strategy_decisions.md`. --- Full report (dev branch): https://github.com/kanex1/sydnee.signals/blob/dev/docs/audit_2026-05-26_exit_flow.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 2026-05-26 — EXIT FLOW ## Summary - Area: Exit flow (how and when open positions are closed) - Bugs found (P0 / P1): 0 / 2 - Risks noted: 4 Branch: `dev` (dc2d9b4). Cross-checked against `git log --oneline -30` and `docs/strategy_decisions.md`. --- ## Findings ### BUG [P1]: OBV Rule F permanently disabled after in-loss skip **File:** `bot.py:4295–4334` **Evidence:** ```python if _obv_against: ... if _new_stop != _cur_stop: if _profit_guard and profit_per_share < 0: # In-loss path — skip BE-tighten if _loss_exit: hit = True elif not getattr(t, '_obv_be_applied', False): self._log_activity(...) t._obv_be_applied = True # ← set regardless (line 4332) self._update_trade_field(..., obv_be_applied=True) else: # In-profit path — tighten stop ... t._obv_be_applied = True ``` The outer guard at line 4265–4266 is `not getattr(t, "_obv_be_applied", False)`. Once `_obv_be_applied = True` is set, Rule F is permanently skipped for that trade's lifetime. The in-loss path (profit guard) sets this flag WITHOUT tightening the stop — consuming the one-shot protection. **Scenario that breaks:** Trade enters, goes underwater, OBV turns adverse → profit guard fires, flag is set, stop unchanged. Trade recovers to profit later. OBV remains adverse or turns adverse again. Rule F never fires. The wide entry stop defends until it's hit or another exit fires. **Impact:** Missing OBV-based stop tighten for trades that temporarily go in-loss before recovering. The Rule F protection intended to cut `bxt_reversal` damage (−$774 → −$217 avg in the original backtest) is silently disabled for these trades. **Fix:** Move `t._obv_be_applied = True` inside the `else:` branch (in-profit, tighten fired). In the in-loss case, log and skip without consuming the flag: ```python if _profit_guard and profit_per_share < 0: if _loss_exit: hit = True ... else: self._log_activity(...) # no-op for this bar only # Do NOT set _obv_be_applied — let Rule F retry next bar when profitable else: ... t._obv_be_applied = True # only mark applied when stop actually moved ``` Note: OBV_LOSS_EXIT (spot-exit) path still sets the flag because the trade is being closed — that's correct. --- ### BUG [P1]: TRIM2 `_rsi_force_trim2` flag never cleared for 1-share parents **File:** `bot.py:4728–4762` **Evidence:** ```python if not hit and is_trimmed and not is_trimmed_2 and (_trim2_hit or _force_trim2): trim = True # ← set unconditionally trim2_shares = t.shares // 2 if trim2_shares > 0: # ← guard ... t._trimmed_2 = True # ← inside guard t._rsi_force_trim2 = False # ← inside guard (never clears if shares=1) self._update_trade_field(t.trade_id, trimmed_2=True, force_trim2_pending=False, ...) ``` When TRIM1 reduces a position to 1 share, `trim2_shares = 1 // 2 = 0`. The `if trim2_shares > 0:` guard prevents the trim from executing. However `t._trimmed_2` and `t._rsi_force_trim2` are inside that guard — they are NOT updated when `trim2_shares == 0`. **Impact:** On every subsequent bar: 1. `_force_trim2 = getattr(t, '_rsi_force_trim2', False)` returns `True` (never cleared) 2. TRIM2 condition fires: `trim = True` but no close is queued 3. Fruitless re-evaluation until the stop or EOD exits the position The `force_trim2_pending` column in DB also stays `True`, surviving restarts and re-triggering the same no-op indefinitely. No crash or incorrect close — but CPU wasted every bar and the activity log cannot distinguish "TRIM2 executing" from "TRIM2 stuck at zero-share no-op." **Fix:** Add an `else` branch to set `_trimmed_2 = True` and clear the force-trim flag when `trim2_shares == 0`: ```python if trim2_shares > 0: ... # existing logic else: # Can't trim (0 shares) — mark done so we don't retry every bar t._trimmed_2 = True t._rsi_force_trim2 = False self._update_trade_field(t.trade_id, trimmed_2=True, force_trim2_pending=False, force_trim_reason=None) ``` --- ### RISK: OBV trail stop unavailable for early pre-market 5m-naked positions **File:** `bot.py:3517–3556` **Evidence:** ```python def _tv_atr_obv(self, ss): df5 = ss.bars.get(5, pd.DataFrame()) n = len(df5) if n < 15: return 0.0, 0 # ← no ATR, no OBV slope def _obv_trail_stop_hit(self, my_pos, ss, price): ... atr5, obv_slope = self._tv_atr_obv(ss) if atr5 <= 0: return None # ← stop silently disabled ``` 15 bars = 75 minutes of 5m data. The 5m-naked BXT engine runs from pre-market open (04:15 ET). For the first 75 minutes, both the catastrophic stop (`catx × ATR`) and the OBV-trail stop are disabled. A position opened at 04:15 ET has no mechanical stop until ~05:30 ET. **Impact:** Pre-market 5m-naked positions opened in the 04:15–05:30 ET window carry unlimited intra-session loss risk if a gap or fast move occurs. The engine's only protection during this window is the session-boundary force-flat (which only fires at 09:15 ET or the next session gap). This risk increases with the current 3% NLV risk / 15% cap (ca4b173 + b3fedce). **Mitigation:** Could add a `min_bars_for_entry` guard in `_eval_bxt_1m_naked` that skips new entries (but not exits) when `len(ss.bars.get(5, [])) < 15`. Or gate the `_eval_bxt_1m_all` call to only the RTH + post-market windows where 5m history is sufficient. --- ### RISK: Friday EOD progressive-trim exemption comment is stale/misleading **File:** `bot.py:3935–3950` **Evidence:** ```python # Skip RSI extreme entries from after 12:00 PM PT — keep those intact. # ← stale comment _is_rsi_trig = _reason_chk.startswith('RSI extreme') if _is_rsi_trig: if _et_et2.hour * 60 + _et_et2.minute >= 15 * 60: continue # skip — late RSI extreme, keep overnight # ← "keep overnight" is removed behavior ``` The comment "keep those intact" and inline `# skip — late RSI extreme, keep overnight` reflect the pre-2026-04-17 behavior (RSI extreme held through the weekend). That exemption was removed in commit 52f1bb1: the 3:50 PM final sweep (bot.py:4003–4020) now closes ALL positions with no exemption. The current exemption only skips the *progressive 25% trimming* (3:00–3:50 PM window) for late RSI extreme entries. The position still closes at 3:50 PM. This may be intentional (fresh entries shouldn't be trimmed immediately), but the comments are wrong — they still say "keep overnight." **Impact:** No runtime bug — positions don't actually get held overnight. But the misleading comments create audit ambiguity and could confuse a future developer into thinking the overnight hold is still present. **Mitigation:** Update comments at bot.py:3935, 3940, 3948 to reflect current behavior: late RSI extreme entries are exempt from progressive trimming but are fully closed at 3:50 PM like all other positions. --- ### RISK: FLAT_EXIT parent closed with empty `exit_reason` in DB **File:** `bot.py:4845`, `bot.py:1264–1268` **Evidence:** ```python # FLAT_EXIT 100% close: parent closed DB-only after IBKR order via child self._close_trade_in_store(t.trade_id, live_price) ``` ```python def _close_trade_in_store(self, trade_id, exit_price): ... self.db.close_trade(trade_id=trade_id, ..., exit_reason="") # ← always empty ``` For FLAT_EXIT setups (FLAT_EXIT=true, non-trend), TRIM1 closes 100% of shares via a child trade record (which gets the proper exit reason), and then `_close_trade_in_store` is called on the parent with `exit_reason=""`. The parent's DB row shows a closed trade with no exit reason. **Impact:** Post-trade analytics (trade review, P&L attribution by exit type) cannot classify FLAT_EXIT parent rows. The SQL regex in `bot.py:12356–12368` uses `exit_reason` matching — these rows fall through to `'other'`. Only the `_trim` child row has the correct reason. **Mitigation:** Pass the reason into `_close_trade_in_store` as an optional parameter, or update the parent via `db.update_trade(parent_id, exit_reason="FLAT_EXIT parent — see _trim child")` after the child is confirmed. --- ### RISK: `trim3` parent-DB update comment says "no DB column" — now incorrect **File:** `bot.py:7622–7623` **Evidence:** ```python elif trade.trade_id.endswith("_trim3"): pass # no DB column for trim3 yet — state tracked in-memory ``` The `trimmed_3` column WAS added in the 2026-04-15 persistence fixes (commit `a6be962`, strategy_decisions.md line 344). The column is populated at EXIT 3 (bot.py:4692–4693) before the trim child is queued. The `pass` here is correct (parent already updated), but the comment "no DB column for trim3 yet" is false and could cause confusion. **Impact:** No runtime impact. But a developer reading this comment might conclude the DB schema is incomplete and try to "fix" it by adding the update here, causing a benign duplicate write. **Mitigation:** Update comment to: `# trimmed_3 already set via _update_trade_field in EXIT 3 (bot.py:4692) — no second update needed`. --- ## OK (checked, working or recently fixed): - **FLAT_EXIT trim1 DB-vs-memory discrepancy** — fixed in ff4ac0d (bot.py:6615–6625): monotonicity guard's collapsed trim1 level is restored to `trim_target_price` for FLAT_EXIT setups before persisting. Verified in code. - **Friday pullback cutoff in `_check_pullback`** — fixed in the 2026-05-19 audit chain (bot.py:6025–6035): ALL pending pullbacks including rsi_extreme are cancelled Friday ≥ 15:00 ET. - **RSI in-loss exit flow gate + MIN_FLOW_BLOCK floor** — shipping correctly (bot.py:4519–4552): gate blocks in-loss RSI exit only when flow magnitude > 10K (default). Quiet-bar noise band is filtered. - **TRIM2 stop tighten direction (SHORT)** — verified correct: `new_stop2 = entry + 0.25×ATR×direction` gives `entry − 0.25×ATR` for shorts; stop fires at `close_price >= stop_price`, locking $0.25×ATR profit for shorts as intended. - **Friday final flat sweep** — no exemptions (bot.py:4003–4020): closes ALL positions at 3:50 PM ET on Fridays regardless of trigger type. The progressive trim exemption (3:00–3:50 PM) for late RSI extreme is scoped only to trimming, not the final flat. - **`pending_reversal` persistence on restart** — confirmed via `_load_trades_from_db` (bot.py:1100–1101): `t._pending_reversal = True` restored from DB on bot restart. - **`force_trim_pending` / `force_trim2_pending` persistence on restart** — confirmed (bot.py:1104–1109): both flags and `_force_trim_reason` are restored from DB. - **OBV-trail ratchet persistence (Redis)** — confirmed (bot.py:3573–3580): ratchet is written to Redis with 7-day TTL on each move; restored in `_load_trades_from_db` (bot.py:1129–1140). - **`_obv_trail_stop_hit` shared between TV and 5m-naked** — correct: called in `_eval_tv_bxt_sym` (bot.py:3730) and `_eval_bxt_1m_naked` (bot.py:3774). - **BXt1m/TV-BXt exemption from `_evaluate_symbol`** — confirmed (bot.py:3912–3915): exempt positions bypass all normal stop/trim/EOD logic. Identification via `strategy_reason.startswith("BXT1m")` / `startswith("TV-BXt")` is reliable. - **Deferred-close dedup** — confirmed (bot.py:5932–5946): `seen_close_ids` prevents duplicate closes in the same drain; `if not trade.is_open: continue` catches already-closed trades. - **HOLD_CAP_MIN trend exemption** — correct: applied only when `"MEAN_REVERT"` or `"EXT_FADE"` or `"EXTENSION_FADE"` in `strategy_reason.upper()`. Trend setups (BXT, pullback, RSI) don't match. - **`consec_losses` rebuilt on restart** — confirmed from 2026-05-22 persistence audit fixes. - **OBV Rule F `entry_time` parse fix** — confirmed (bot.py:4279–4281): `datetime.fromisoformat` used to convert string `entry_time` before timezone operations. The silent no-op bug (commit 782e3b6) is fixed.