fix(config): sweep every remaining stale description of the intraday gate

config/phase_i5a_intraday_tail_framework.yaml

@@ -2,7 +2,8 @@
 #
 # Proves the shared event-driven backtest engine drives an IntradayTailEventModel
 # end-to-end on REAL SH/SZ minute data: decision at 14:50, execution at the first
-# valid 1min close in [14:51, 14:56:59], exec-to-exec holding returns. The score is
+# valid 1min bar in [14:51, 14:56:59] priced on the active execution_price_basis
+# (the bar VWAP by default), exec-to-exec holding returns. The score is
 # a single PIT-safe I3 feature (intraday_ret_0930_1450), not a research alpha.
 #
 # Minute bars are read from the EXISTING intraday cache only (read-only; a cache

config/phase_i5b_intraday_execution_feasibility.yaml

@@ -1,11 +1,12 @@
 # Phase I5b — intraday execution-time price-limit feasibility (hardening, NOT research).
 #
 # Same intraday tail-rebalance smoke as I5a (decision 14:50, execute at the first
-# valid 1min close in [14:51, 14:56:59], exec-to-exec holding returns, PIT-safe I3
+# valid 1min bar in [14:51, 14:56:59] priced on the active execution_price_basis, exec-to-exec holding returns, PIT-safe I3
 # score) PLUS execution-time price-limit feasibility: a buy is blocked at the raw
-# upper limit and a sell at the raw lower limit, comparing the selected
-# execution-minute RAW 1min close to the raw stk_limit band (NEVER qfq / daily
-# close / daily-close-derived flags).
+# upper limit and a sell at the raw lower limit, comparing the RAW price that
+# EXECUTES at the selected minute -- the bar VWAP under the default
+# execution_price_basis -- to the raw stk_limit band (NEVER qfq / daily close /
+# daily-close-derived flags).
 #
 # Minute bars are read from the EXISTING intraday cache only (read-only; a miss is
 # a hard blocker -> zero stk_mins live calls). Raw stk_limit flows through the same
@@ -42,8 +43,9 @@ universe:
   filters:
     # Selection-level daily tradability flags stay DISABLED: at the 14:50 decision
     # the daily-close-derived limit flag is an EOD value not known yet. The honest
-    # direction-aware blocking happens at EXECUTION via raw stk_limit vs the raw
-    # execution-minute close (intraday.price_limit_check below). Disclosed in the report.
+    # direction-aware blocking happens at EXECUTION via raw stk_limit vs the RAW
+    # price that EXECUTES at that minute (intraday.price_limit_check below).
+    # Disclosed in the report.
     missing_close: true
     suspended: false
     st: false
@@ -95,7 +97,8 @@ intraday:
   - '14:56:59'
   require_cache_coverage: true
   missing_execution: block
-  # I5b: execution-time raw price-limit feasibility (raw 1min close vs raw stk_limit).
+  # I5b: execution-time raw price-limit feasibility (the RAW price that EXECUTES
+  # at that minute vs raw stk_limit).
   price_limit_check: true
   require_price_limit_coverage: true
   limit_tolerance: 1.0e-06

config/phase_i5c_mmp_minute_factor.yaml

@@ -42,7 +42,8 @@ universe:
   filters:
     # Selection-level daily tradability flags stay DISABLED (as in I5b): the daily
     # close limit flag is an EOD value not known at 14:50. The honest direction-aware
-    # blocking is execution-time raw stk_limit vs the raw execution-minute close.
+    # blocking is execution-time raw stk_limit vs the RAW price that EXECUTES at that minute
+    # (the bar VWAP under the default execution_price_basis).
     missing_close: true
     suspended: false
     st: false

config/phase_i5d_mmp_quintile_5y.yaml

@@ -42,8 +42,9 @@ universe:
   filters:
     # Selection-level daily tradability flags stay DISABLED (as in I5b/I5c): the
     # daily close limit flag is an EOD value not known at 14:50. The honest
-    # direction-aware blocking is execution-time raw stk_limit vs the raw
-    # execution-minute close.
+    # direction-aware blocking is execution-time raw stk_limit vs the RAW price
+    # that EXECUTES at that minute (the bar VWAP under the default
+    # execution_price_basis).
     missing_close: true
     suspended: false
     st: false

config/phase_i5e_mmp_quintile_csi300_generalization.yaml

@@ -44,8 +44,9 @@ universe:
   filters:
     # Selection-level daily tradability flags stay DISABLED (as in I5b/I5c/I5d): the
     # daily close limit flag is an EOD value not known at 14:50. The honest
-    # direction-aware blocking is execution-time raw stk_limit vs the raw
-    # execution-minute close.
+    # direction-aware blocking is execution-time raw stk_limit vs the RAW price
+    # that EXECUTES at that minute (the bar VWAP under the default
+    # execution_price_basis).
     missing_close: true
     suspended: false
     st: false

config/phase_i5f_intraday_liquidity_diagnostics.yaml

@@ -1,7 +1,7 @@
 # Phase I5f — intraday execution liquidity diagnostics (REPORT-ONLY, NOT research).
 #
 # Same intraday tail-rebalance smoke as I5a/I5b (decision 14:50, execute at the
-# first valid 1min close in [14:51, 14:56:59], exec-to-exec holding returns,
+# first valid 1min bar in [14:51, 14:56:59] priced on the active execution_price_basis, exec-to-exec holding returns,
 # PIT-safe I3 score, raw stk_limit execution-time feasibility ON) PLUS an opt-in,
 # REPORT-ONLY execution liquidity diagnostic. For each desired rebalance trade
 # (|target_weight - current_weight|) it sizes the trade against the SELECTED
@@ -48,7 +48,8 @@ universe:
   filters:
     # Selection-level daily tradability flags stay DISABLED (as in I5b): the daily
     # close limit flag is an EOD value not known at 14:50. The honest direction-aware
-    # blocking is execution-time raw stk_limit vs the raw execution-minute close.
+    # blocking is execution-time raw stk_limit vs the RAW price that EXECUTES at that minute
+    # (the bar VWAP under the default execution_price_basis).
     missing_close: true
     suspended: false
     st: false

qt/config.py

@@ -340,8 +340,11 @@ class IntradayCfg(_Strict):
     # I5b execution-time price-limit feasibility. OFF by default, so every I5a /
     # daily config validates and behaves unchanged. When enabled, the intraday
     # tail model gates buys at the raw upper limit and sells at the raw lower
-    # limit, comparing the selected execution-minute RAW close to raw ``stk_limit``
-    # (never qfq / daily close). ``require_price_limit_coverage`` makes a missing
+    # limit, comparing the RAW price that EXECUTES at the selected minute -- the
+    # bar VWAP under the default ``execution_price_basis`` -- to raw ``stk_limit``
+    # (never qfq / daily close; see qt.intraday_tail_framework.limit_basis_phrase,
+    # the single authored statement of this, which this module cannot import
+    # without a cycle). ``require_price_limit_coverage`` makes a missing
     # required (anchor date, symbol) limit row a hard, pre-result failure instead
     # of a silent "checked" pass; ``limit_tolerance`` is the raw-price equality
     # band for the at-limit test.

qt/intraday_tail_framework.py

@@ -13,7 +13,8 @@
     -> deterministic PIT-safe score = the I3 ``intraday_ret_0930_1450`` feature
        (only bars with available_time <= 14:50 enter it)
     -> engine + IntradayTailEventModel: decision 14:50, execute at the first valid
-       1min close in [14:51, 14:56:59], exec-to-exec holding returns
+       1min bar in [14:51, 14:56:59] priced on the active execution_price_basis
+       (the bar VWAP by default), exec-to-exec holding returns
     -> NAV / feasibility / holdings / event logs -> markdown report.
 
 It is intentionally small: the score is a single already-PIT-safe feature solely

runtime/backtest/event_models.py

@@ -132,6 +132,15 @@ class IntradayTailEventModel:
     symbol/date's raw ``stk_limit`` band. A buy is blocked at the upper limit, a
     sell at the lower limit, directionally.
 
+    That sentence is authored once, in
+    ``qt.intraday_tail_framework.limit_basis_phrase``, and every report composes
+    it. This docstring cannot import it — ``qt.intraday_tail_framework`` already
+    imports this class, so the reverse would cycle — so it is one of the two
+    restatements that survive by necessity. It is covered instead by the repo-wide
+    scan in ``tests/test_i5b_execution_feasibility.py``, which fails if any file
+    claims this gate compares a CLOSE. Change the behaviour and that scan will
+    point here.
+
     Why the gate reads the VWAP and not the bar close. A limit-up execution minute
     has exactly two shapes, and they are the feasibility question:
 

tests/test_i5b_execution_feasibility.py

@@ -476,55 +476,67 @@ def test_group_report_feasibility_prose_names_the_active_execution_basis():
     assert "adj_factor" in text
 
 
-def test_no_module_repeats_the_pre_pr75_close_based_limit_claim():
-    """Guard the DEFECT CLASS, not just the two places already fixed.
-
-    The gate input moved from the bar close to the executed price in PR #75. The
-    first correction pass fixed the prose in the two obvious spots and MISSED a
-    third copy in `_write_report`'s "Limitations" section, so a run would have
-    emitted two contradictory descriptions of the same check in one document.
-
-    Review found it, and found why the other tests could not: none of them render
-    `_write_report` at all, so that location was structurally unreachable.
-
-    SCOPE, stated honestly because the first version of this docstring overstated
-    it. This catches a LITERAL revert -- a bad merge or rebase restoring the old
-    sentence, which is the failure that actually happened. It does NOT catch a
-    reworded restatement: review wrote seven plausible ones ("its last print",
-    "the final tick", "1-minute" for "1min") and all seven escape this regex. No
-    lexical guard can assert "no other sentence makes this claim"; English has too
-    many ways to make it. The durable fix is structural and is the reason
-    `limit_basis_phrase` exists -- with one authored sentence there is no second
-    one to reword. This test guards that one sentence against being un-written.
+def test_no_file_anywhere_claims_the_gate_compares_a_close():
+    """Repo-wide: NO file may state that the I5b gate compares a close.
+
+    The first version of this scanned only the two report modules. Review then
+    found SIX more live instances it could not reach even in principle -- one in
+    `qt/config.py` and five in the I5b/c/d/e/f config YAMLs, which are exactly what
+    an operator reads before enabling the feature, a higher-stakes readership than
+    a docstring. Scoping a guard to the files you already fixed guarantees it only
+    ever confirms what you already know.
+
+    Two restatements survive by necessity and are covered by this scan rather than
+    by composition: `runtime/backtest/event_models.py` and `qt/config.py` both sit
+    UPSTREAM of `qt.intraday_tail_framework` in the import graph (it imports them),
+    so neither can call `limit_basis_phrase` without a cycle.
+
+    Deliberately NOT matched: `bar_close` the basis name, "that bar's single
+    closing tick" the basis description, and the DAILY universe tradability filter
+    (`universe/filters.py`, `data/clean/tradability.py`), which is a separate and
+    older mechanism that genuinely does compare a raw daily close -- see
+    RUNBOOK.md. Those are correct and must not be swept up.
+
+    Scope, stated honestly: this catches a claim written with the word "close"
+    near "execution minute". It does NOT catch a reworded synonym ("its last
+    print", "the final tick") -- review demonstrated seven such escapes. No lexical
+    guard can assert "no sentence anywhere makes this claim". Composition is the
+    real answer where the import graph allows it; this scan is the net under the
+    places where it does not.
     """
     import re
     from pathlib import Path
 
-    import qt.intraday_group_report as gr
-    import qt.intraday_tail_framework as tf
-
-    # Phrasings that assert the LIMIT COMPARISON reads a close. "bar_close" the
-    # basis name and "that bar's single closing tick" the basis description are
-    # both legitimate and deliberately not matched.
+    root = Path(__file__).resolve().parent.parent
     forbidden = re.compile(
-        r"execution-minute\s+(\*\*raw\*\*\s+)?(1min\s+)?close"
-        r"|1min\s+close\s+to\s+the\s+raw"
-        r"|raw\s+execution-minute\s+close",
+        r"execution[- ]minute[^.\n]{0,40}\bclose\b"
+        r"|\b1min close\b"
+        r"|raw\s+close\s+to\s+(the\s+)?raw\s+`?`?stk_limit",
         re.IGNORECASE,
     )
-    for mod in (tf, gr):
-        src = Path(mod.__file__).read_text()
-        hits = [
-            f"line {i}: {ln.strip()}"
-            for i, ln in enumerate(src.splitlines(), 1)
-            if forbidden.search(ln)
-        ]
-        assert not hits, (
-            f"{Path(mod.__file__).name} still claims the price-limit gate compares "
-            f"a CLOSE. The gate reads the price that executes (the bar VWAP under "
-            f"the default basis). Derive the wording from execution_price_basis "
-            f"instead of restating it:\n  " + "\n  ".join(hits)
-        )
+    targets = [root / "qt" / "config.py", *sorted((root / "config").glob("*.yaml"))]
+    targets += sorted((root / "qt").glob("intraday*.py"))
+    targets += [root / "runtime" / "backtest" / "event_models.py",
+                root / "runtime" / "intraday_execution.py"]
+    # The exec-basis modules describe the same execution anchor from the factor
+    # side. They are accurate today; they are listed BEFORE they go stale, which
+    # is the whole lesson of how this guard had to be widened in the first place.
+    targets += sorted((root / "qt").glob("exec_basis*.py"))
+    targets += [root / "qt" / "exec_forward_returns.py"]
+
+    hits = []
+    for f in targets:
+        if not f.exists():
+            continue
+        for i, line in enumerate(f.read_text().splitlines(), 1):
+            if forbidden.search(line):
+                hits.append(f"{f.relative_to(root)}:{i}: {line.strip()}")
+    assert not hits, (
+        "these files claim the I5b price-limit gate compares a CLOSE. It compares "
+        "the price that EXECUTES (the bar VWAP under the default basis). Where the "
+        "import graph allows it, compose limit_basis_phrase; where it does not, "
+        "point at it:\n  " + "\n  ".join(hits)
+    )
 
 
 def test_every_gate_description_composes_the_single_phrase():