evals/industry_demos: add run_all.py suite runner

ADR-0046 — Industry Demo Suite runner.

Adds `evals/industry_demos/run_all.py`, the single-entry-point script
that executes all three falsifiable demos in sequence, collects their
structured JSON evidence, and exits 0 iff every demo passes.

Design choices:
- Runs each demo in an isolated try/except so a crash in demo_01 does
  not suppress evidence from demo_02 and demo_03 (fail-open evidence
  collection, fail-closed exit code).
- Prints a human-readable banner + structured JSON evidence per demo.
- Prints a final machine-readable JSON summary `{"all_passed": bool,
  "results": [...]}` on stdout for CI consumption.
- Exits 0 when all_passed, 1 otherwise.
- Zero new dependencies: only stdlib + the same imports each individual
  demo already uses.

Also updates `evals/industry_demos/__init__.py` to document the new
runner in the module docstring.

Verification path:
  python -m evals.industry_demos.run_all
  echo $?   # 0 on full pass

evals/industry_demos/__init__.py

@@ -7,11 +7,40 @@ no transformer-LLM wrapper can reproduce.  Run individually:
     python -m evals.industry_demos.demo_02_geometry_drives_identity
     python -m evals.industry_demos.demo_03_deterministic_audit
 
-Each exits 0 on pass, 1 on fail, and prints structured JSON evidence
-to stdout.
+Or run the full suite via the suite runner:
+
+    python -m evals.industry_demos.run_all
+
+Each individual demo exits 0 on pass, 1 on fail, and prints structured
+JSON evidence to stdout.  The suite runner exits 0 iff all three pass
+and emits a final machine-readable JSON summary for CI consumption.
+
+Demos
+-----
+demo_01_forward_constraint
+    Claim: PropositionGraph constrains the generation walk via CGA
+    geometry (not a prompt filter or keyword list) BEFORE any tokens
+    are produced.  Evidence: allowed_indices < vocab_size; every index
+    scores positive cga_inner against at least one named node versor;
+    region label encodes the graph root node ID.
+
+demo_02_geometry_drives_identity
+    Claim: Swapping identity pack (precision_first_v1 vs
+    generosity_first_v1) changes the manifold's alignment_threshold
+    geometrically — not via a system prompt, different model, or
+    temperature setting.  Evidence: threshold_p (0.55) ≠ threshold_g
+    (0.40); identity_score.alignment ordered precision ≤ generosity
+    on the same input.
+
+demo_03_deterministic_audit
+    Claim: Three independent ChatRuntime instances on the same input
+    produce byte-identical JSONL audit records for the seven
+    architecturally-determined fields (versor_condition, vault_hits,
+    dialogue_role, stub_path, safety_upheld, ethics_upheld, flagged).
+    This is structural determinism, not seeded randomness.
 
 The exact-recall-at-scale claim (CGA vault recall at N up to 100k) is
-covered by ADR-0045 — measured on the actual vault path, with properly
+covered by ADR-0045 — measured on the actual vault path with properly
 constructed versors — and is not duplicated here under a weaker
 construction.  See ADR-0046, "Industry Demo Suite", for the rationale.
 """

evals/industry_demos/run_all.py

@@ -0,0 +1,101 @@
+"""
+evals/industry_demos/run_all.py — ADR-0046 suite runner.
+
+Runs all three falsifiable industry demos in sequence.  Each demo makes
+exactly one claim that a transformer-LLM wrapper cannot reproduce.  This
+runner collects structured JSON evidence from all three, prints a human-
+readable report, and exits 0 iff every demo passes.
+
+Usage::
+
+    python -m evals.industry_demos.run_all
+
+Output (stdout):
+    - Per-demo banner + JSON evidence block.
+    - Final structured JSON: {"all_passed": bool, "results": [...]}
+
+Exit code:
+    0  — all three demos passed.
+    1  — one or more demos failed or raised an exception.
+"""
+
+from __future__ import annotations
+
+import json
+import sys
+import traceback
+from typing import Any
+
+
+# ---------- demo registry ----------
+
+_DEMOS = [
+    ("demo_01_forward_constraint",      "evals.industry_demos.demo_01_forward_constraint"),
+    ("demo_02_geometry_drives_identity", "evals.industry_demos.demo_02_geometry_drives_identity"),
+    ("demo_03_deterministic_audit",      "evals.industry_demos.demo_03_deterministic_audit"),
+]
+
+_SEP = "=" * 72
+
+
+# ---------- runner ----------
+
+
+def _run_demo(name: str, module_path: str) -> dict[str, Any]:
+    """Import and run a single demo module.  Returns its result dict.
+
+    Wraps the run() call in a broad try/except so that a crash in one
+    demo does not prevent the others from executing.  A crashed demo
+    is recorded as failed with the traceback embedded in the evidence.
+    """
+    import importlib
+    try:
+        mod = importlib.import_module(module_path)
+        result: dict[str, Any] = mod.run()
+    except Exception:  # noqa: BLE001
+        tb = traceback.format_exc()
+        result = {
+            "demo": name,
+            "claim": "(exception — see evidence.traceback)",
+            "evidence": {"traceback": tb},
+            "passed": False,
+        }
+    return result
+
+
+def run_all() -> dict[str, Any]:
+    """Execute all registered demos and return a summary dict."""
+    results: list[dict[str, Any]] = []
+
+    for name, module_path in _DEMOS:
+        print(_SEP)
+        print(f"DEMO: {name}")
+        print(_SEP)
+
+        result = _run_demo(name, module_path)
+        results.append(result)
+
+        status = "PASS" if result.get("passed") else "FAIL"
+        print(f"Status : {status}")
+        print(f"Claim  : {result.get('claim', '')}")
+        print("Evidence:")
+        print(json.dumps(result.get("evidence", {}), indent=2))
+        print()
+
+    all_passed = all(r.get("passed") for r in results)
+
+    print(_SEP)
+    print(f"SUITE RESULT: {'ALL PASSED' if all_passed else 'ONE OR MORE FAILED'}")
+    print(_SEP)
+
+    summary = {
+        "all_passed": all_passed,
+        "results": results,
+    }
+    print(json.dumps(summary, indent=2))
+    return summary
+
+
+if __name__ == "__main__":
+    summary = run_all()
+    sys.exit(0 if summary["all_passed"] else 1)