Sydnee nightly — persistence audit — 2026-06-04 P0 findings: 0 P1 findings: 2 Risks: 2 - Area: Persistence - Bugs found (P0 / P1): 0 / 2 (1 new + 1 carried from 2026-05-28) - Risks noted: 2 (both carried from 2026-05-22 / 2026-05-28) --- Full report (dev branch): https://github.com/kanex1/sydnee.signals/blob/dev/docs/audit_2026-06-04_persistence.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 2026-06-04 — PERSISTENCE ## Summary - Area: Persistence - Bugs found (P0 / P1): 0 / 2 (1 new + 1 carried from 2026-05-28) - Risks noted: 2 (both carried from 2026-05-22 / 2026-05-28) --- ## Findings ### BUG [P1] NEW: `_tv_scale_state` not persisted — scale-in silently abandoned on restart **File:** `bot.py:4201-4202` (init), `bot.py:4230` (only in-memory write) **Evidence:** ```python # bot.py:4201-4202 — lazily initialised, never restored from Redis/DB if not hasattr(self, "_tv_scale_state"): self._tv_scale_state = {} # ← resets to {} on every restart # bot.py:4230 — only in-memory write self._tv_scale_state[sym] = st # no Redis setex, no DB update_trade_field # bot.py:4026-4033 — scale_active gate depends on in-memory state _scale_active = ( os.environ.get("TV_BXT_SCALE_IN", "").lower() in ("1", "true", "yes") and hasattr(self, "_tv_scale_state") and sym in self._tv_scale_state # ← False on restart and self._tv_scale_state[sym].get("dir") == sig and self._tv_scale_state[sym].get("slots_done", 10) < 10) if not _scale_active: return # already in direction, no more slots due ``` `_tv_scale_state[sym]` tracks the full scale-in window per symbol: `sig_ts`, `dir`, `target_shares`, `slot_shares`, `slots_done`, `parent_trade_id`, `next_slot_ms`, `start_ms`. No part of this dict is written to Redis or the DB. The slot-advance-on-skip change (ec7ed26) means `slots_done` increments on every call regardless of fill outcome, making the counter even more important to survive a restart. **Failure path on mid-scale restart (e.g. at slots_done=4/10):** 1. TV webhook fires → `tv_bxt_signal:{sym}` set in Redis; scale starts. 2. Slot 1 opens parent trade via `_open_trade`; slots 2-4 call `_tv_scale_add_shares`, mutating `parent.shares` + `parent.entry_price` in DB. Parent trade correctly persisted (4 × slot_shares shares, weighted-avg price). 3. Bot restarts at slot 5/10. 4. `_tv_scale_state` resets to `{}`. 5. `_load_trades_from_db` reloads the parent trade correctly from DB (shares, entry_price, strategy_reason preserved). 6. Next tick: `my_pos` found, `cur == sig`. Check `_scale_active`: `sym not in _tv_scale_state` → `_scale_active = False`. Returns early at line 4033. 7. `_tv_scale_step` is never reached. Remaining 6 slots (5-10) silently abandoned. 8. Parent trade survives with 4-slot shares, managed normally by OBV-trail stop and opposite-TV-flip exit. **Impact:** Scale-in does not complete after any mid-session restart. Position is undersized vs intent (40% of target in the above example). No duplicate orders, no incorrect closes — behavior is safe but incomplete. Active risk: `TV_BXT_SCALE_IN=true` on dev as of 2026-06-04 re-enable (strategy_decisions.md line 1009). **Fix:** ```python # 1. bot.py — persist at write site (bot.py:4230), after every st mutation: self._tv_scale_state[sym] = st if getattr(self, "_regime_redis", None): try: self._regime_redis.setex( f"tv_scale_state:{sym}", 660, # 10×60s + grace json.dumps(st)) except Exception: pass # 2. bot.py — restore on init (bot.py:4201-4202): if not hasattr(self, "_tv_scale_state"): self._tv_scale_state = {} _r = getattr(self, "_regime_redis", None) if _r: for _sym in self._tv_bxt_active_syms(): _raw = _r.get(f"tv_scale_state:{_sym}") if _raw: try: self._tv_scale_state[_sym] = json.loads( _raw.decode() if isinstance(_raw, bytes) else _raw) except Exception: pass ``` After restore, the next call to `_eval_tv_bxt_sym` sees `sym in _tv_scale_state` with `slots_done=4` and `parent_trade_id` set → `_scale_active=True` → falls through to `_tv_scale_step`. The step function checks `st["sig_ts"] == _sigts` (Redis signal timestamp still matches) and `st["slots_done"] >= 1` → jumps to the slot-2+ path, finds the existing parent by `parent_trade_id`, and resumes correctly. Eviction (TTL 660s = 11 min) ensures a restart 11+ minutes after a signal does not attempt to resume a dead window. --- ### BUG [P1] CARRIED (from 2026-05-28): `tv_bxt_signal` no-TTL + `_tv_confirm_state` not persisted **File:** `bot.py:12533-12534` (no-TTL set), `bot.py:4113-4114` (`_tv_confirm_state`) **Evidence:** ```python # bot.py:12533-12534 — plain r.set(), no TTL (unchanged since 2026-05-28 filing) r.set(f"tv_bxt_signal:{sym}", sig) r.set(f"tv_bxt_signal_ts:{sym}", str(int(time.time()))) # bot.py:4113-4114 — flow-confirm retry counter, in-memory only if not hasattr(self, "_tv_confirm_state"): self._tv_confirm_state = {} # ← resets to {} on every restart ``` Status note: with `TV_BXT_SCALE_IN=true` on dev, the `_tv_confirm_state` code path is bypassed — scale-in handles its own per-slot flow-confirm inside `_tv_scale_step`. The no-TTL `tv_bxt_signal:{sym}` concern remains active for both paths: a stale signal can survive a restart indefinitely, causing a fresh 10-attempt confirmation window (or a fresh scale window) to open on a signal that may be 15+ minutes old. The fail-open path (`_fresh = False` on stale WS at restart) bypasses flow confirmation and immediately attempts entry. **Impact / Fix:** Unchanged from 2026-05-28 filing. Apply `r.setex(f"tv_bxt_signal:{sym}", 900, sig)` and `r.setex(f"tv_bxt_signal_ts:{sym}", 900, ...)` at bot.py:12533-12534. Persist and restore `_tv_confirm_state` via Redis as described in the 2026-05-28 audit. --- ### RISK: `_fri_flat_queued` not persisted — RTH force-flat dedup missed (CARRIED from 2026-05-22 → 2026-05-28) **File:** `bot.py:4610-4615` **Evidence:** ```python # bot.py:4610-4615 — dedup guard in-memory only; no _update_trade_field if getattr(t, '_fri_flat_queued', False): continue t._fri_flat_queued = True # ← no persist self._deferred_closes.append((t, f"Friday FLAT: ...")) ``` Unchanged from 2026-05-28. On restart during the 3:50 PM ET Friday-close window, dedup resets and a second close order is enqueued. `is_open` guard in `_close_trade` provides secondary protection if the first fill processes first. Narrow window (Friday only, 3:50–3:51 ET); no financial harm on paper; structural gap mirrors the now-fixed `_fri_ah_close_queued` (8845a24). **Fix:** Add `fri_flat_queued BOOLEAN DEFAULT FALSE` to `ensure_trade_state_columns`, persist with `_update_trade_field(t.trade_id, fri_flat_queued=True)` at bot.py:4615, restore in `_load_trades_from_db` alongside `fri_ah_close_queued`. --- ### RISK: `_bxt1m_last_bar` not persisted — BXt-1m dedup replays last bar on restart (CARRIED from 2026-05-22 → 2026-05-28) **File:** `bot.py:3915-3916`, `bot.py:4431` **Evidence:** ```python # bot.py:3915-3916 — always empty on restart if not hasattr(self, "_bxt1m_last_bar"): self._bxt1m_last_bar = {} # bot.py:4431 — in-memory only self._bxt1m_last_bar[sym] = idx[-1] # no Redis write ``` On restart the dedup guard resets, and the catch-up replay re-evaluates the last processed 1m bar as "new." If `desired_dir == 0` (1m stalled, 5m alignment gone) at restart, an open BXT1m position is force-closed prematurely. Risk is dormant if `BXT_1M_NAKED_SYMBOLS` is unset (empty by default on both environments); dev currently sets it. **Fix (unchanged from 2026-05-22):** `_regime_redis.setex(f"bxt1m_last_bar:{sym}", 120, str(last_ts))` at bot.py:4431; restore on init. --- ## OK (checked, working or recently fixed) **New since 2026-05-28 — commits 3b0d6ac (scale-in redesign), ec7ed26 (slot-advance-on-skip), 82c80de (RTH pre-market re-open), b2178a7 (multi-fill original, reverted by 3b0d6ac), 44aec76 (DISABLE_DAILY_LOSS_LIMIT):** - **`_tv_scale_add_shares` DB write** — mutates `parent.shares` and `parent.entry_price` via `_update_trade_field(parent.trade_id, shares=new_shares, entry_price=...)` (bot.py:4345-4347). These columns are restored in `_load_trades_from_db` at line 1102-1103. ✓ Parent trade correctly survives restart with accumulated shares regardless of `_tv_scale_state` loss. - **`_tv_block_logged` dict** — still purely cosmetic (dedup for one dashboard log line per signal change; bot.py:4159-4168). No trading impact on restart. ✓ - **`TV_BXT_RTH_ONLY` gate** — pure time check, no new in-memory state. On block outside session, `_tv_bxt_signal_clear(sym)` deletes both Redis keys (bot.py:4066). No signal survives the block. ✓ - **`TV_BXT_SCALE_IN=false` revert path** — falls through the `if TV_BXT_SCALE_IN` block at line 4075, reaches the flow-confirm path at line 4078. `_tv_confirm_state` code path correctly resumes. ✓ - **`DISABLE_DAILY_LOSS_LIMIT`** — pure env flag in `core/risk.py:106-112` wrapping the daily-loss check. No new state, no persistence concern. ✓ - **Single-record redesign audit (3b0d6ac commit message):** py_compile clean, 0 hits for `my_positions` (reverted), all close paths use `my_pos` single record, `_open_trade` GUARD bypass removed, `_tv_scale_add_shares` defined and called only by `_tv_scale_step`. ✓ - **Slot-advance-on-skip (ec7ed26):** `slots_done` now increments on every call regardless of fill outcome — dedup slots correctly (no retry). No new state beyond what `_tv_scale_state` already tracks. ✓ **From 2026-05-28 audit — all still OK:** - `_tv_bxt_signal_clear` correctly deletes both `tv_bxt_signal:{sym}` and `tv_bxt_signal_ts:{sym}` on close and entry (bot.py:3726). ✓ - Flow-confirm `_ratio` / `_flow_min` / `_vol_min` — computed live from WS overlay on each call; no per-trade state. ✓ - `_pending_manual_orders`, bar snapshots, OBV trail, all DB-persisted trade flags (`trimmed_1/2/3`, `fri_ah_close_queued`, etc.) — no regressions detected. ✓ **Long-standing — re-confirmed no regression:** - All 7 items from 8845a24 (oracle TTL, consec_losses, tick_peak, tick_fire_ts, fri_ah_close_queued, signal_price/time) remain correctly implemented. ✓ - `ensure_trade_state_columns()` idempotent migration — no new columns needed for the scale-in redesign (shares and entry_price are existing columns). ✓ - OBV divergence swing-point history — dormant (`OBV_DIVERGENCE_ENABLED=false`). ✓