fix(learn): withhold label-tracking signals from the v1 classifier; leak audit

Maintainer decision "Neutralise both for v1".

1. Phantom candidates. A pair side's candidates are now only the functions that
   side declares (plane/harvest.py `declares`, the excerpt builder's
   function_bounds); a labelled function absent from a side is listed under
   `absent_on_side` and never asked, and a side declaring none of its verify
   family's functions is no verify unit. The recorded harvest units held 58
   phantom verify candidates (53 fixed, 5 vulnerable sides). build_features.py
   skips label rows whose side does not declare the function and drops their
   earlier rows (MemoryStore.drop_rows); rebuilt the pair records from the
   recorded signals, no model call: 126 label rows absent_on_side (117 fixed,
   9 vulnerable) -- access_control 13, config_secrets 4, deserialization 20,
   injection 28, memory 4, output_encoding 16, path 15, untrusted_destination
   26 -- so 252 feature rows dropped (1712 -> 1460); no other row changed.

2. Engine withheld. Input profiles in learn/schema.py (`v1` withholds the
   engine instrument, `full` withholds nothing); ProgramSpec.inputs defaults to
   v1 and compile.toml records it. Neither the engine's state nor its values
   reach the candidate, examples, demonstrations or the instruction proposer's
   error summary, and the Gower retrieval distance leaves the engine block out.
   Stored records are unchanged. design.md Risks records the decision and that
   the engine returns only through a pre-registered A/B on assumed-benign and
   benign-control negatives.

3. `ousast learn audit-leaks` (learn/leaks.py): per feature and per instrument
   state, separation = (pairs the signal alone puts on the right side - on the
   wrong side) / pairs, over all pairs and per family; flagged at >= 0.20 over
   >= 10 pairs. Counts under
   benchmarks/measurements/2026-10-01-decision-engine-leak-audit/ (before and
   after the rebuild). After: facts.function_lines (0.80 untrusted_destination,
   0.75 config_secrets, 0.26 injection; higher on fixed), verify.turns_mean
   0.67 config_secrets, eng.* 0.35-0.40 output_encoding, verify.flag_a/votes/
   site_offset and agree.final 0.20-0.30 on 10-12 pair families. Before the
   rebuild the phantom fixed sides drove state.source/state.roles/ms.flagged
   to 1.0 and state.engine to 0.95 on deserialization.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01EmM1j46qgKMh3zqD1QMwEo

src/openultrasast/learn/retrieve.py

@@ -8,7 +8,8 @@ profile, computed per instrument block so that missing is explicit, never impute
   both is ``0``, in one is ``1``;
 - different states (``ran`` vs ``none``/``failed``/``not_applicable``, or ``failed`` vs ``none``): ``1``; the
   same non-``ran`` state: ``0``;
-- ``D`` is the uniform mean over the profile's blocks; prior features are dropped with ``priors = "off"``.
+- ``D`` is the uniform mean over the profile's blocks; prior features are dropped with ``priors = "off"``, and the blocks
+  of instruments the input profile withholds (``inputs``, default ``v1``: the engine) are left out entirely.
 
 Ties break by ``sha256(example id + seed)``; the first ``n1 = 40`` go to stage 2, a cosine re-rank of the code
 embeddings, ``r = lam (1 - D) + (1 - lam) cos``. Without the target's vector the whole retrieval is
@@ -34,7 +35,7 @@ from .embeddings import cosine
 from .examples import Example
 from .folds import Fold
 from .labels import repo_name
-from .schema import UNBOUNDED_FLOATS, FeatureSpec, features_for, instruments_for
+from .schema import DEFAULT_INPUTS, UNBOUNDED_FLOATS, FeatureSpec, features_for, instruments_for, withheld
 
 N1 = 40
 K_RET = 6
@@ -91,9 +92,12 @@ def gower(
     profile: str,
     *,
     priors: str = "off",
+    inputs: str = DEFAULT_INPUTS,
 ) -> float:
-    """The per-instrument Gower distance of two records of ``profile`` (module docstring), in [0, 1]."""
-    total, blocks = 0.0, _blocks(profile, priors)
+    """The per-instrument Gower distance of two records of ``profile`` (module docstring), in [0, 1]. The blocks of
+    instruments the input profile withholds (v1: the engine) are not compared, so neighbours cannot carry them."""
+    hidden = withheld(inputs)
+    total, blocks = 0.0, [(name, specs) for name, specs in _blocks(profile, priors) if name not in hidden]
     for name, specs in blocks:
         sa, sb = a_instruments.get(name, {}).get("state", "none"), b_instruments.get(name, {}).get("state", "none")
         if sa != sb:
@@ -192,11 +196,12 @@ def retrieve(
     lam: float = LAM,
     seed: int = 0,
     priors: str = "off",
+    inputs: str = DEFAULT_INPUTS,
 ) -> Retrieval:
     """The ``k`` balanced examples for ``target`` from the eligible part of ``pool`` (module docstring)."""
     candidates = [e for e in pool if e.profile == target.profile and eligible(e, target, fold)]
     scored = sorted(
-        ((gower(target.x, target.instruments, e.x, e.instruments, target.profile, priors=priors), e) for e in candidates),
+        ((gower(target.x, target.instruments, e.x, e.instruments, target.profile, priors=priors, inputs=inputs), e) for e in candidates),
         key=lambda pair: (pair[0], _tie(pair[1].id, seed)),
     )[:n1]
     mode = "signals+embeddings" if target.vector is not None and vectors else "signals_only"