fix(verif): wire E2 N-scan ablation env vars into production code path

The E2 verification campaign defines two conditions (cortex_full,
cortex_flat) by setting CORTEX_DECAY_DISABLED=1, CORTEX_HEAT_CONSTANT=0.5,
CORTEX_CONSOLIDATION_DISABLED=1. n_scan_runner.py:_apply_condition sets
them, but no production code was reading them — the first E2 run came
back with cortex_full and cortex_flat metrics IDENTICAL at N=1k and
N=10k, making E2 a placebo run.

Fix:
- mcp_server/core/pg_recall.py compute_pg_weights() honours
  CORTEX_DECAY_DISABLED=1 OR CORTEX_HEAT_CONSTANT (any value): heat
  weight forced to 0.0 in WRRF fusion. Heat at constant value cannot
  discriminate by ranking; equivalent to flat-importance baseline for
  retrieval purposes.
- mcp_server/handlers/consolidation/cls.py run_cls_cycle() returns the
  empty-stats dict immediately when CORTEX_CONSOLIDATION_DISABLED=1,
  skipping episodic-to-semantic abstraction.

Smoke: baseline heat weight = 0.3, both env vars set → heat = 0.0.

Source: tasks/verification-protocol.md E2; benchmarks/lib/n_scan_runner.py.

Author: cdeust

mcp_server/core/pg_recall.py

@@ -131,7 +131,21 @@ def compute_pg_weights(
 
     Derives base weights from core_weights (from query_intent) when available,
     then applies intent-specific PG overrides.
+
+    Verification ablation hooks (Popper C2 — operator-disablable mechanism):
+    - ``CORTEX_DECAY_DISABLED=1``: forces heat weight to 0.0 so the
+      thermodynamic decay signal cannot enter the WRRF fusion. Disabling
+      heat is equivalent to "flat heat" for ranking purposes — Cortex
+      degenerates to vector + FTS + ngram, the flat-importance baseline.
+    - ``CORTEX_HEAT_CONSTANT=<float>``: same effect on the weight (heat
+      cannot discriminate when constant), kept as a separate var so the
+      n_scan harness can force a specific constant heat at write time and
+      confirm the ranker reproduces flat baseline at read time.
+    Source: tasks/verification-protocol.md E2 (N-scan); env vars defined
+    by benchmarks/lib/n_scan_runner.py:_apply_condition.
     """
+    import os as _os
+
     cw = core_weights or {}
     # Vector is always 1.0 in the PG path — it's the primary discovery signal.
     # Other signals derived from core_weights (intent system) when available,
@@ -146,6 +160,10 @@ def compute_pg_weights(
     overrides = _PG_INTENT_OVERRIDES.get(intent)
     if overrides:
         base.update(overrides)
+    if _os.environ.get("CORTEX_DECAY_DISABLED") == "1" or _os.environ.get(
+        "CORTEX_HEAT_CONSTANT"
+    ):
+        base["heat"] = 0.0
     return base
 
 

mcp_server/handlers/consolidation/cls.py

@@ -52,6 +52,18 @@ def run_cls_cycle(
 ) -> dict:
     """Run CLS consolidation: episodic → semantic pattern extraction.
 
+    Verification ablation hook: when ``CORTEX_CONSOLIDATION_DISABLED=1``
+    is set (E2 N-scan condition `cortex_flat`), this returns the zero
+    state immediately. No episodic-to-semantic abstraction runs; the
+    flat-importance store is never enriched with patterns. Source:
+    tasks/verification-protocol.md E2; benchmarks/lib/n_scan_runner.py.
+    """
+    import os as _os
+
+    if _os.environ.get("CORTEX_CONSOLIDATION_DISABLED") == "1":
+        return dict(_EMPTY_CLS_STATS)
+
+
     Pattern extraction (`plan_cls_consolidation`) and causal-edge
     discovery (`_discover_causal_edges`) sample up to 2000 episodic
     memories each — raised from 500 after Feynman's audit of darval's