feat(models): Model Governance Registry — champion/challenger + historic + DataClass classifiers

docs/contract-additions/inference-gateway-intersection.md

@@ -47,3 +47,20 @@ Shared `InferenceGateway` request/response + `GatewayCallAudit` on every call; a
 (agent-machine) and the cloud adapter (model-fabric) both conform; consent-gated + receipted;
 the catalog/leaderboard reads one board across planes. Passes the **inference seam** purple-team
 (no un-consented or un-audited call returns output).
+
+## The governed model plane — beyond foundation models
+The catalog is not only LLMs. Every **business target** (fraud, churn, credit, AML, propensity)
+is a governed model with:
+- **Champion / challenger** — the production champion and shadow challenger(s), each with a
+  metric (AUC / PR-AUC / Gini / recall). A challenger is promoted only through the eval gate
+  (SHACL + eval, fail-closed); a promotion is a `RunReceipt` — replayable, examiner-ready.
+- **Historic comparison** — metric by version, champion vs challenger over time; model lineage.
+- **Drift** — PSI per target; a breach flags the target and can trip the automaton.
+- **DataClass classifiers** — the [trained+registered](./2026-05-agent-machine-model-carry-contracts.md)
+  data-catalog classifiers governing each target's data (transaction / merchant / PII / bureau /
+  protected-attribute / jurisdiction …), each with coverage + registration status.
+
+Surface: [Model Governance Registry](../surfaces/model-governance.html) — the champion/challenger +
+historic + classifier view. Regulated targets (credit, AML) require examiner sign-off before
+promotion. This is model risk management (APRA CPS 230 / OCC model-risk) rendered as a witness
+surface — the governs-agents thesis made examinable.

docs/surfaces/README.md

@@ -43,6 +43,9 @@ the real SociOS/GNOME chrome (top bar, launcher card) so they read as in-shell,
 - **`lampstand-launcher.html`** — Tier 2, *actionable*. The Spotlight replacement: natural
   language → typed intents/entities/relations (the annotation tree) → **governed actions**
   (purpose-bound, consent-gated, receipted), ranked by sherlock (IR). Not web search — typed acts.
+- **`model-governance.html`** — Tier 2, *regulated model registry*. Business targets (fraud/churn/
+  credit/AML/propensity) with champion-vs-challenger, historic version comparison, PSI drift, and the
+  DataClass classifiers per target. Model risk management as a witness surface.
 - **`model-catalog.html`** — Tier 2, *intersection cockpit*. Every foundation model + OS
   capability, cloud∩local placement, a sovereignty×governance-weighted leaderboard, and the
   SP-vs-Claude-Enterprise positioning (marked DRAFT/VERIFY). Backed by `inference-gateway-intersection.md`.

docs/surfaces/data/model-governance.json

@@ -0,0 +1,105 @@
+{
+  "provenance": "sample",
+  "generated_at": "2026-08-03T09:41:30Z",
+  "source": "sample seed (governed model plane ledger + DataClass classifier registry)",
+  "targets": [
+    {
+      "id": "fraud",
+      "name": "Card fraud",
+      "industry": "payments · real-time",
+      "champion": {
+        "name": "GBM-fraud",
+        "ver": "v4",
+        "met": "AUC",
+        "val": 0.962
+      },
+      "challenger": {
+        "name": "GNN-fraud",
+        "ver": "v1",
+        "met": "AUC",
+        "val": 0.971
+      },
+      "verdict": "promote",
+      "drift": 0.08,
+      "driftWarn": false,
+      "history": [
+        {
+          "l": "v3",
+          "c": 0.955,
+          "h": 0.958
+        },
+        {
+          "l": "v4",
+          "c": 0.962,
+          "h": 0.971
+        }
+      ],
+      "classifiers": [
+        {
+          "n": "transaction-class",
+          "d": "txn type · MCC",
+          "st": "registered",
+          "cov": "99.8%"
+        },
+        {
+          "n": "merchant-risk-class",
+          "d": "reputation tier",
+          "st": "registered",
+          "cov": "97.2%"
+        },
+        {
+          "n": "pii-class",
+          "d": "PII tagging",
+          "st": "registered",
+          "cov": "100%"
+        }
+      ]
+    },
+    {
+      "id": "credit",
+      "name": "Credit risk",
+      "industry": "lending · regulated",
+      "champion": {
+        "name": "Scorecard",
+        "ver": "v12",
+        "met": "Gini",
+        "val": 0.58
+      },
+      "challenger": {
+        "name": "LGBM-credit",
+        "ver": "v3",
+        "met": "Gini",
+        "val": 0.61
+      },
+      "verdict": "promote",
+      "drift": 0.11,
+      "driftWarn": true,
+      "history": [
+        {
+          "l": "v11",
+          "c": 0.56,
+          "h": 0.59
+        },
+        {
+          "l": "v12",
+          "c": 0.58,
+          "h": 0.61
+        }
+      ],
+      "classifiers": [
+        {
+          "n": "bureau-class",
+          "d": "bureau tier",
+          "st": "registered",
+          "cov": "99.4%"
+        },
+        {
+          "n": "protected-attr-class",
+          "d": "fairness guard",
+          "st": "registered",
+          "cov": "100%"
+        }
+      ]
+    }
+  ]
+}