The Live Cross-Package Harness

The seam layer takes the sibling identities on trust; this harness verifies them live. An out-of-gate runner loads digestive and mind and confirms cross-volume substrate drift is exactly 0, digestive's LIVE IBD course reproduces the T23/T24 thresholds (induction 0.8849, maintenance 0.1151), and mind's module names the HPA and inflammatory endpoints. The digest hashes only the immune-side contract. Grade [V].

The harness runs OUTSIDE every gate and SKIPS cleanly when siblings are absent, so it can never break a sibling-free build. Its digest (2b07d4d5…) hashes ONLY the immune-side contract — the closed-form substrate identity (ok=true), the vendored IBD snapshot, the gut saddle-node identity (induction 0.8849, maintenance 0.1151) and the neuro-immune endpoint — computed with no sibling present, hence byte-identical with or without them. When run with siblings present, all five live checks pass: substrate drift 0 (immune↔digestive and immune↔mind), digestive's live ibd_relapsing_course() reproduces the vendored thresholds AND equals this volume's T23/T24 saddle-node, and mind's §27 module carries the HPA-cortisol (IN) and inflammatory (OUT) endpoints.

From trust to verification

The seam layer (chapter 16) had to take three things on trust because the gated package may never import a sibling: that digestive's substrate is byte-identical to this one, that digestive's LIVE inflammatory-bowel course actually produces the IBD thresholds this volume vendored, and that the mind volume really carries the endpoints the neuro–immune pointers land on. This harness discharges that trust. It loads THIS volume together with its sibling volumes IN ONE PROCESS and checks the identities against the LIVE sibling engines — not snapshots.

The three live checks

(1) Shared substrate, drift 0. The harness loads each sibling's spinodal-bearing engine by file path and compares its R19 spinodal 2(γ/3)^1.5 and barrier γ²/4 against this volume's, bit-for-bit, over several γ values. The measured cross-volume drift is exactly 0 for both immune↔ digestive and immune↔mind. This is the foundation: a zero-drift substrate is what lets the tolerance saddle-node derived HERE be read in digestive's mucosa and in mind's HPA cascade without any refit.

(2) Gut–immune latch identity, live. The harness calls digestive's LIVE ibd_relapsing_course() and confirms its induction threshold (0.8849), maintenance threshold (0.1151) and relapsing flag reproduce the snapshot this volume vendored — AND that those live thresholds equal this volume's own T23 saddle-node (antigen + spinodal) and T24 suppressor complement (antigen − spinodal). So the claim of chapter 16 — that the gut mucosal tolerance latch IS this volume's machinery localised — is verified against digestive's real code, not merely asserted from a copied number.

(3) Neuro–immune endpoint, live. The harness loads the mind engine (confirming the same spinodal) and reads mind's depression-chronification module (mind §27), confirming it names the HPA/cortisol cascade (the substrate for the IN direction) and the INFLAMMATORY contributor (where the OUT cytokine pointer lands). The seam endpoints are real reading points in the sibling volume, not dangling references.

Why this cannot break the build

The harness runs OUTSIDE every gate. The research gate (repro/run_all.py) and the canonical build (tools/build_docs.py) compute nothing here and never import a sibling; each volume still re-establishes its entire trusted state from its own archive with the siblings ABSENT — the verify-alone invariant. When the siblings are not discoverable on disk, this module SKIPS the live checks and exits cleanly. The sibling engines are loaded by FILE PATH under unique module names with the package-internal module table cleared between loads (so two volumes' identically-named vp_substrate cannot collide) — this is loading, not importing: no sibling code path is wired into this volume's engine.

Crucially, the harness digest hashes ONLY the immune-side contract — the closed-form substrate identity (ok=true), the vendored IBD snapshot, the gut saddle-node identity (0.8849 / 0.1151) and the neuro–immune endpoint descriptor. That contract is computed from THIS package's own modules with no sibling present, so the harness digest (2b07d4d5…) is byte-identical whether or not the siblings are on disk — exactly what a deterministic build requires. The engine circulate()/emit() 2×sha256 (e7a2a5b8…), the thirty-five-target stress battery, the six discipline gates and the seam digest are ALL unchanged by this layer; the harness carries its OWN 2×sha256.

What this is — and is not

The harness is a reproducibility instrument, not a new scientific claim. It adds no physics — it confirms that the four volumes share one substrate and that the seam identities hold against the live sibling code. It is run with python repro/run_harness.py when the sibling packages are present. It is NOT part of any gate, NOT a clinical claim, and NOT a validation of VP theory; it is the evidence that the cross-package identities of chapter 16 are real, live, and deterministic.