fix(executor): treat optimal zero-trade optimizer solve as a valid HOLD#196
Merged
Merged
Conversation
is_log_usable() returned False on empty would_be_trades, conflating the optimizer's legitimate "current portfolio already matches target, hold" verdict with genuine failures (None / non-ok / non-optimal diag). The 2026-05-19 weekday rerun solved optimal with turnover_one_way=0.0017 (0.17%, below the trade threshold) → would_be_trades=[]; the planner then logged a false `optimizer log is not usable … Operator must investigate` ERROR and framed a correct hold as a safety fallback. The optimizer is the authoritative trade driver (cutover live since 2026-05-13). This never suppressed real trades — on any day the optimizer wants trades, would_be_trades is non-empty so is_log_usable was already True — it only mislabeled valid hold-days and cried wolf. - is_log_usable: usability now depends solely on a clean ok/optimal (or optimal_inaccurate) solve. Emptiness is the caller's concern. - main.py: usable + zero entries/exits → INFO stating it is a valid HOLD (with turnover_one_way); the ERROR + "operator must investigate" is now reserved strictly for genuine non-usable logs. - tests: flip the (bug-encoding) empty-trades case to expect usable, add missing-key + apply-is-safe-noop coverage. Suite 914 passed. Separate concern from the 2026-05-19 SF-failure PR stack; kept its own PR in the sensitive optimizer-cutover path. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
3 tasks
cipher813
added a commit
that referenced
this pull request
May 24, 2026
…first half (#207) L1346 (c) — the executor-side retirement of SPY-special-cases that alpha-engine #185 added as a defensive measure while SPY was only in macro library (Close-only). Post-fix evidence: alpha-engine-data #245 (MERGED 2026-05-15) promoted SPY to a full `universe` ArcticDB member via `_UNIVERSE_EXTRA = frozenset({"SPY"})`, written by both backfill.py (Saturday) and daily_append.py (weekday). 2026-05-24 DataPhase1 SSM log confirms: `Backfill write complete: 904 ok` = 903 constituents + SPY. universe.SPY now carries the atr_14_pct feature column that load_atr_14_pct needs. The #185 comment block itself anticipated this retirement: "the right remedy for ATR is exclusion (macro lib has no ATR), not macro-lib dispatch" — that's now stale advice. universe.SPY has full OHLCV + features; ATR computation works. Implementation: - New `_MACRO_SYMBOLS_NO_OHLCV = _MACRO_SYMBOLS - {"SPY"}` constant at the ATR exclusion site. Subtracts SPY from the exclusion (because universe.SPY now has OHLCV) while preserving the exclusion for VIX, TNX, IRX, sector ETFs, etc. (still Close-only in macro lib). - Comment block updated to reference L1346 (c) closure + the gate (a) verification evidence. Tests: 2 new in `test_l1346_spy_atr_exclusion_retired.py`: - White-box: SPY not in `_MACRO_SYMBOLS - {"SPY"}` derived set; legacy Close-only symbols (VIX/TNX/IRX/XLK/XLF) still in exclusion. - Source-level pin: `executor/main.py` derivation uses `_MACRO_SYMBOLS_NO_OHLCV`, NOT the full `_MACRO_SYMBOLS`. Future refactor that drops the subset distinction silently re-introduces the bug L1346 (c) closed. Full executor suite 962 pass. OUT OF SCOPE — continuing arc: - `executor/price_cache.py:135` `_MACRO_SYMBOLS` routing flip (read SPY from universe first, fall back to macro) — mirrors the predictor pattern from alpha-engine-predictor #196 (L1346 (b)). Defer as follow-up because price_cache.load_price_histories has more callers than the single ATR site here. - `executor/eod_reconcile.py:655` `_MACRO_SYMBOLS` routing flip — same pattern. Defer for the same reason. Both can ship together as a follow-up PR mirroring the predictor's defensive macro-fallback shape. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2 tasks
cipher813
added a commit
that referenced
this pull request
May 25, 2026
…refer universe.SPY with macro fallback (#208) Completes the L1346 (c) retirement of SPY-special-cases across the executor. Pre-fix sites used `if ticker in _MACRO_SYMBOLS` to route SPY reads to macro library (Close-only); post-L1346 #245 universe.SPY now has full OHLCV via `_UNIVERSE_EXTRA`. Implementation (mirrors alpha-engine-predictor #196 defensive shape): - `executor/price_cache.py::load_price_histories` — gates on `_MACRO_SYMBOLS_NO_OHLCV = _MACRO_SYMBOLS - {"SPY"}`. SPY routes to universe first; on universe-read failure, defensive fallback to macro.SPY (clearly logged via dual-exception-class read_errors line). - `executor/eod_reconcile.py` — same routing flip + defensive fallback. Non-SPY macro symbols (VIX/TNX/IRX/sector ETFs) remain macro-routed (still Close-only in macro library; no universe entry). Defensive fallback can be removed once L1346 (b)+(c) soak clean for ≥1 Saturday cycle (operator-promoted soak; ~1-2 Saturdays out). Test plan: - Full executor suite 962 pass. - Pre-existing alpha-engine #185 ATR exclusion site already updated to `_MACRO_SYMBOLS_NO_OHLCV` (PR #207). Closes ROADMAP L1346 (c) second half. Combined with #207 (first half) + #196 (predictor (b) gate), the entire L1346 retirement arc is now shipped except the operator soak-watch. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Surfaced by
The 2026-05-19 weekday-SF rerun (after the #194/#264 stack fixed the MorningEnrich timeout). With the pipeline running to completion, the planner hit:
Root cause
is_log_usable()returnedFalseonnot would_be_trades. The shadow log was healthy:shadow_status=ok,diagnostics.status=optimal,target_weights≈current_weights,turnover_one_way=0.0017(0.17%, below the trade threshold) →would_be_trades=[]. That is the optimizer's legitimate "current portfolio already matches target — hold" verdict, conflated with genuine failures (None / non-ok / non-optimal diag).Severity: low / cosmetic-but-misleading. It never suppressed real trades — on any day the optimizer wants trades,
would_be_tradesis non-empty sois_log_usablewas alreadyTrue. It only mislabeled valid hold-days, emitting a false "operator must investigate" ERROR and framing a correct hold as a safety fallback. The optimizer remains the authoritative trade driver (cutover live since 2026-05-13, config #172).Fix
is_log_usable: usability depends solely on a cleanok+optimal/optimal_inaccuratesolve. Emptiness ofwould_be_tradesis the caller's concern, not a usability signal.main.py: usable + zero entries/exits →INFOexplicitly stating it is a valid HOLD (withturnover_one_way); theERROR+ "operator must investigate" + "leaving order book empty for safety" is now reserved strictly for genuine non-usable logs.test_is_log_usable_empty_tradesto expect usable (renamed + regression-documented), add missing-key and apply-is-safe-noop coverage. Full suite: 914 passed.Behaviorally inert for today (zero trades either way) — this stops the recurring false incident on every legitimate no-rebalance day. Kept as its own PR (sensitive optimizer-cutover path), separate from the 2026-05-19 SF-failure stack (#264/#233/#194/#265).
🤖 Generated with Claude Code