W1 forensics 3: cap_150k adjacency decomposition + hh_size residual quantification (registered 4959668253) (#179)

Report-only (publishes regardless; gates nothing). Registered at issue #42
comment 4959668253 before execution. Instrumentation bit-identity vs the
committed machinery is 0.0 for both questions: Q10 re-runs the committed #117
ledger on the transported representative careers and reproduces the cube c2
exhaustion deltas bit-for-bit; Q11 re-runs regenerate_person_frame_v3 and
reproduces the cube hh_size cells bit-for-bit.

Q10 (cap_150k adjacency): the deployed 16.7y-vs-Smith-1y gap is dominated
(share 0.9925) by (a) the frame's above-cap payroll share; (b)+(c)
config+vintage are 0.0075. The entailment HOLDS: the same above-wage-base
compression correction that certifies the elim<->+2pp swap (A/B past the
2pp/12.4% breakeven) mechanically lifts cap_150k past +1pp/+2pp to rank-2, so
the 4-element ordering is internally inconsistent on a representative frame. No
contract-permitted lever restores the adjacency (construction empty). The
pair-scoped C2 (elim<->+2pp only, realised 3/3, anchor-supported) is the
anchor-consistent re-spec.

Q11 (hh_size residual post-c3): REFINEMENT of the registration. The
size-1<->size-3 mirror is a clean coresidence-composition signature (ratio
0.98), and coresidence dominates size-3, but coresidence does NOT own >50% of
every failing cell -- the fertility window loads the size-4/5plus deficits.
Mixed-but-structured attribution. Candidate-independence: levers EXHAUSTED (the
roster ceiling is hazard-capped at terminal coresidence 0.111 and does not
close size-1; the fertility window is at the support floor); the residual needs
a non-entry-state coresidence model (an amendment).

Artifact runs/gate_w1_forensics3_v1.json (+ env sidecar), one-shot via
write_new. Tests: tests/test_gate_w1_forensics3.py (artifact tier, 21 bindings,
mutation-checked) + reproduction pins (integration_psid, skips off-machine);
tier_counts artifact 928->949, integration_psid 797->800.

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

Author: MaxGhenis

runs/gate_w1_forensics3_v1.json

@@ -0,0 +1,15 @@
+{
+  "environment": {
+    "python": "3.14.4",
+    "numpy": "2.5.1",
+    "pandas": "3.0.3",
+    "sklearn": "1.8.0",
+    "scipy": "1.18.0",
+    "platform": "macOS-26.5.1-arm64-arm-64bit-Mach-O"
+  },
+  "contract": {
+    "blob_sha": "c43b2be2f7aeceafb2aae3181b3718e6e6145aa6",
+    "head_sha": "64ebf97118addb19e1a8d0623e6a3321b1dea4fe",
+    "path": "gates.yaml"
+  }
+}