feat(model-grounded-detection): group 1, subtract the noise architecture

99 source modules -> 81. 832 tests -> 646. The whole learning/ package (16
modules, ~4,100 lines) and six semantic/ modules removed, with the three gates
byte-identical to 2026-09-06-gate-baseline.txt at every step.

The sequence is what made it safe: vocabularies ported first (1.1), importers
detached second (1.2-1.3), deletion third (1.4). Every import break during the
deletion was therefore a detachment miss in the importer and was fixed there --
no module was restored to satisfy one. Both misses were the same shape and were
fixed as a class: operations.py and shapes.py each imported trailing_name from
the deleted variants.py, so it moved to semantic/ir.py, its natural home, and
both importers changed in one edit.

Ported before deleting: the closed guard vocabulary (verbatim, nothing else
holds it) into model/specs.py, which derives TaintSpec and DominanceSpec from
the retained ruleset facts rather than copying them. Re-homed: families ->
model/taxonomy, candidates -> model/candidates, endpoint -> model/endpoint,
plus Region, labeled_family and labeled_spans. labeled_family no longer falls
back to the LLM classifier -- an unlabeled pair is honestly unknown.

Four deviations, each recorded in the spec notes with its reason. improve/
evolve.py and validator.py were stripped, not deleted: each mixes the mechanism
lever (gone) with the deterministic rule-status loop `ousast improve` still
runs, which has no LLM in it; a test now asserts the lever's absence by symbol
so it cannot creep back. The learning CLI group became `model candidates`,
because deleting it outright would have orphaned the retained enumerator by
removing its only entry point. thinking moved from [learning] to [models].

A gap named rather than papered over: no retained sink fact routes by CWE to
path, output_encoding, untrusted_destination or prototype -- the same hole the
entailment ceiling diagnosed. The test asserts the seeded set and the gap set
exactly, so task 4.2 must update it when it authors those sinks.

The audit found exactly one genuine orphan: vectorstore.py, a bakeoff harness
with no production caller and no CLI, alive only through its own test. The five
named subsystems each carry an explicit decision. Manifest committed, with a
test asserting the tree still matches it.

Group 2 (the Joern go/no-go) has not started.

Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Claude-Session: https://claude.ai/code/session_01T7XD7x5KonMdmGRkp63EQG

src/openultrasast/config.py

@@ -36,6 +36,11 @@ class ModelConfig:
     judge: str | None = None  # second, independent judgement before anything is published as proven
     chat_base_url: str | None = None  # chat endpoint; independent of the embedding endpoint
     chat_api_key_env: str | None = None  # environment variable holding that endpoint's key
+    # Whether the chat model reasons before answering. Off by default because that is what the committed
+    # baseline was measured under: the provider silently ignores `temperature` while thinking, so thinking was
+    # disabled to make `temperature: 0` mean something. It did not — the measured run-to-run disagreement came
+    # from the agentic path, not the decoder — which is why the model layer arbitrates instead of averaging.
+    thinking: bool = False
 
 
 @dataclass(frozen=True)
@@ -138,26 +143,6 @@ class VariantsConfig:
     max_mechanisms: int = 500
 
 
-@dataclass(frozen=True)
-class LearningConfig:
-    # learning-harness: the classified detector set and its rounds.
-    enabled: bool = False  # off until round zero has written the family configurations this reads
-    families_path: str | None = None  # taxonomy file; the shipped ruleset when unset
-    configs_dir: str = ".openultrasast/learning/configs"
-    k_runs: int = 5  # runs per pair; a single run misses most real changes
-    round_cost_cap_usd: float = 10.0
-    minibatch: int = 8  # train pairs in a round's first stage
-    detector_model: str | None = None
-    judge_model: str | None = None
-    cutoff_date: str | None = None  # rows fixed after this date are reported as a separate slice
-    # Whether the detector model reasons before answering. Off by default because that is what round zero was
-    # measured under, not because it is known to be better: the provider silently ignores `temperature` while
-    # thinking, and thinking was disabled to make `temperature: 0` mean something. It did not — the measured
-    # disagreement between runs comes from the agentic path, not the decoder — so which mode wins is a question
-    # for a second baseline rather than for a default (Req 8.3).
-    thinking: bool = False
-
-
 @dataclass(frozen=True)
 class ObligationsConfig:
     # authorization-obligations: the obligation checker in MAP (standard/deep only).
@@ -183,7 +168,6 @@ class ResolvedConfig:
     regress: RegressConfig = RegressConfig()
     variants: VariantsConfig = VariantsConfig()
     obligations: ObligationsConfig = ObligationsConfig()
-    learning: LearningConfig = LearningConfig()
     runs_dir: str = ".openultrasast/runs"
 
 
@@ -210,7 +194,6 @@ def load_config(config_path: Path | None = None) -> ResolvedConfig:
         regress=_load_regress(data.get("regress", {})),
         variants=_load_variants(data.get("variants", {})),
         obligations=_load_obligations(data.get("obligations", {})),
-        learning=_load_learning(data.get("learning", {})),
         runs_dir=os.environ.get("OPENULTRASAST_RUNS_DIR", ".openultrasast/runs"),
     )
 
@@ -238,6 +221,7 @@ def _load_models(value: object) -> ModelConfig:
         judge=_string(data.get("judge")),
         chat_base_url=_string(data.get("chat_base_url")),
         chat_api_key_env=_string(data.get("chat_api_key_env")),
+        thinking=bool(data.get("thinking", False)),
     )
 
 
@@ -385,22 +369,6 @@ def _load_variants(value: object) -> VariantsConfig:
     )
 
 
-def _load_learning(value: object) -> LearningConfig:
-    data = _section(value)
-    return LearningConfig(
-        enabled=bool(data.get("enabled", LearningConfig.enabled)),
-        families_path=_string(data.get("families_path")),
-        configs_dir=_string(data.get("configs_dir")) or LearningConfig.configs_dir,
-        k_runs=_int_value(data.get("k_runs"), LearningConfig.k_runs),
-        round_cost_cap_usd=_float_value(data.get("round_cost_cap_usd"), LearningConfig.round_cost_cap_usd),
-        minibatch=_int_value(data.get("minibatch"), LearningConfig.minibatch),
-        detector_model=_string(data.get("detector_model")),
-        judge_model=_string(data.get("judge_model")),
-        cutoff_date=_string(data.get("cutoff_date")),
-        thinking=bool(data.get("thinking", False)),
-    )
-
-
 def _load_obligations(value: object) -> ObligationsConfig:
     data = _section(value)
     return ObligationsConfig(