test: add scaled N256 Taylor-Green diagnostic

tests/validation/taylor_green/assess_projected_pairs.py

@@ -1,6 +1,7 @@
 """New-source four-case repeat of 11901a7; frozen oracle and gates are unchanged."""
 
 import argparse
+import hashlib
 import json
 import math
 from pathlib import Path
@@ -8,16 +9,25 @@ from pathlib import Path
 from assess import BASELINE_FACTOR, FLOORS
 from oracle import LIMITS, TIMES, U, rectangles
 from run_case import (INITIALIZATION, sample_steps, NUMERICAL_REVISION, SOURCE_SHA256,
-                      BUILD_CAPTURE_SHA256, REFERENCE_HARNESS)
+                      BUILD_CAPTURE_SHA256, REFERENCE_HARNESS, SCALED_DIAGNOSTIC)
 
-# Host-generated snapshots of a135252 at the frozen two local configurations.
+# Host-generated snapshots of a135252 at the frozen local configurations.
 # These describe expected inputs; actual sidecars are captured before cleanup.
 BUILD_SOURCES = {
     128: "803177416b7599ab89cc5afd35cf4a28a673b2f4bb8e8a6a1abc0b6d77610873",
     256: "17ec5e60f87b840d3d13b8778bc8a3aecf485469c5a061af46c5cc19b73396cf",
+    512: "4e85e38098858ec95c137d1b1e96846a58d9cf3cc242b349a3ad7c8b8e3ce0aa",
 }
 CASES = ((100, 128, 1, "constant"), (100, 256, 1, "constant"),
          (100, 128, 2, "constant"), (100, 128, 2, "projected_linear"))
+REFERENCE_REVISION = "2cf9853952e52f59bea63cf5c1d3a1cd76e53eed"
+# Audited 2cf9853 scalar receipts, archived at evidence commit 90849a55cad4a0da9b72747da7bfe6eb75962a44.
+REFERENCE_RECEIPTS = {
+    "uniform128": "724efafc267eaa59dcc59e8d4c16177b67f6c6a08fd2732a481833237bb19cee",
+    "uniform256": "3c312a99ebf747a135674dd481b77455a3d2c9a406ca388f30524b37dad9bbe9",
+    "constant128": "2ee28efd1e81b2e28b7902b6c4f95bb2c1ea4e5f0ec484638fa5ae83dca34bba",
+    "projected128": "ee67b1d4d90a048c0d3d1a16447c6eab3d89c94d151f791306e373f96dd8b963",
+}
 NUMERICS = dict(lattice_model="D2Q9", collision="SRT", store_precision="FP32",
                 ddf_shifting=True, streaming="double_buffer", les_enabled=False,
                 boundary="periodic", block_shape=[8, 8],
@@ -73,7 +83,11 @@ def valid_builds(record, n, depth):
     return True
 
 
-def assess(records, expected_revision=None):
+def assess(records, expected_revision=None, *, diagnostic=None):
+    if diagnostic not in (None, SCALED_DIAGNOSTIC):
+        raise ValueError("Unknown diagnostic")
+    base_n = 256 if diagnostic else 128
+    cases = tuple((re, n * base_n // 128, depth, scheme) for re, n, depth, scheme in CASES)
     indexed, duplicates = {}, []
     for record in records:
         key = tuple(record.get(k) for k in ("re_L", "n", "depth", "explosion"))
@@ -81,9 +95,11 @@ def assess(records, expected_revision=None):
             duplicates.append(str(key))
         else:
             indexed[key] = record
-    missing = [str(k) for k in CASES if k not in indexed]
+    missing = [str(k) for k in cases if k not in indexed]
-    checks = dict(complete_matrix=not missing, no_extra_cases=set(indexed) <= set(CASES),
+    checks = dict(complete_matrix=not missing, no_extra_cases=set(indexed) <= set(cases),
                   no_duplicate_cases=not duplicates, finite_receipts=finite_tree(records))
+    if diagnostic or any("diagnostic" in r for r in records):
+        checks["diagnostic_identity"] = all(r.get("diagnostic") == diagnostic for r in records)
     sources = [r.get("source", {}) for r in records]
     checks["same_source_and_harness"] = bool(sources) and len({
         json.dumps(s, sort_keys=True) for s in sources}) == 1
@@ -106,9 +122,11 @@ def assess(records, expected_revision=None):
     ) and len(set(devices)) == 1
     enabled = [r.get("early_pressure", {}).get("enabled") for r in records]
     checks["same_early_sampling"] = bool(enabled) and all(type(v) is bool for v in enabled) and len(set(enabled)) == 1
+    if diagnostic:
+        checks["diagnostic_early_sampling"] = all(v is True for v in enabled)
     shared_valid = all(checks.values())
     complete, case_valid, early = {}, {}, []
-    for key in CASES:
+    for key in cases:
         if key not in indexed:
             continue
         r = indexed[key]
@@ -135,7 +153,7 @@ def assess(records, expected_revision=None):
             (("module", "__init__.py"), ("compiler", "cuda/compiler_v2.py"), ("kernel_root", "lbm/kernels")))
         expected_patch = [] if depth == 1 else rectangles(n, depth)
         checks[label + " patch"] = json.dumps(r.get("refinements")) == json.dumps(expected_patch)
-        cells = [n*n] if depth == 1 else [8704, 30720]
+        cells = [n*n] if depth == 1 else [17*n*n//32, 15*n*n//8]
         hops = [16*n * 2**l for l in range(depth)]
         work = r.get("workload", {})
         checks[label + " workload"] = (
@@ -168,19 +186,22 @@ def assess(records, expected_revision=None):
 
     comparisons, scheme_results = [], {}
     for scheme in ("constant", "projected_linear"):
-        key = (100, 128, 2, scheme)
+        key = (100, base_n, 2, scheme)
         for metric, floor in FLOORS.items():
-            amr, coarse, fine = (worst(k, metric) for k in (key, CASES[0], CASES[1]))
+            amr, coarse, fine = (worst(k, metric) for k in (key, cases[0], cases[1]))
             budget = None if coarse is None else max(BASELINE_FACTOR * coarse, floor)
             budget = budget if finite(budget) else None
             ratio = None if amr is None or fine is None else amr / max(fine, 1e-30)
             passed = amr is not None and budget is not None and amr <= budget
             checks[f"baseline {scheme}/{metric}"] = passed
             comparisons.append(dict(explosion=scheme, metric=metric, amr=amr,
-                                    uniform128=coarse, baseline_budget=budget, passed=passed,
+                                    baseline_budget=budget, passed=passed,
-                                    uniform256=fine, amr_to_uniform256=ratio if finite(ratio) else None))
+                                    **{f"uniform{base_n}": coarse, f"uniform{2*base_n}": fine,
+                                       f"amr_to_uniform{2*base_n}": ratio if finite(ratio) else None}))
+            if diagnostic:
+                comparisons[-1]["floor_dominates"] = None if coarse is None else floor >= BASELINE_FACTOR * coarse
         baseline_passed = all(checks[f"baseline {scheme}/{metric}"] for metric in FLOORS)
-        valid = shared_valid and all(case_valid.get(k, False) for k in (CASES[0], CASES[1], key))
+        valid = shared_valid and all(case_valid.get(k, False) for k in (cases[0], cases[1], key))
         scheme_results[scheme] = dict(passed=valid and baseline_passed,
                                      valid_case_and_controls=valid, baseline_passed=baseline_passed)
     return dict(passed=all(checks.values()), missing=missing, duplicates=duplicates, checks=checks,
@@ -190,19 +211,89 @@ def assess(records, expected_revision=None):
                 floors=FLOORS, baseline_factor=BASELINE_FACTOR,
                 main_statistic="maximum of four post-initial samples at t*=.12,.24,.36,.48",
                 early_statistic="sampled pressure maximum only, t*=k*.001875, k=0..64; not an all-step or continuous maximum",
-                finest_control="uniform256 and AMR/ uniform256 ratios are descriptive, not acceptance gates",
+                finest_control=f"uniform{2*base_n} and AMR/ uniform{2*base_n} ratios are descriptive, not acceptance gates",
                 scope="2D periodic Re100 smooth-decay diagnostic; no new algorithm, convergence-order or general AMR qualification")
 
 
+def load_reference(paths):
+    """Use the four audited historical files verbatim, never relabel them fresh controls."""
+    blobs = [path.read_bytes() for path in paths]
+    hashes = [hashlib.sha256(blob).hexdigest() for blob in blobs]
+    if len(hashes) != 4 or set(hashes) != set(REFERENCE_RECEIPTS.values()):
+        raise ValueError("Historical N128 receipt hashes differ from the audited 2cf9853 evidence")
+    return [json.loads(blob) for blob in blobs]
+
+
+def resolution_trend(records, reference, expected_revision):
+    """Signed changes are descriptive; they never replace the frozen acceptance budgets."""
+    old = assess(reference, REFERENCE_REVISION)
+    new = assess(records, expected_revision, diagnostic=SCALED_DIAGNOSTIC)
+    # Numerical screen failures still have useful raw errors; only malformed inputs block trends.
+    valid = all(v for result in (old, new) for k, v in result["checks"].items()
+                if not k.startswith("baseline ") and not k.endswith(" absolute"))
+    valid = valid and all(maximum(r["rows"], m) is not None for r in reference + records for m in LIMITS)
+    oracle = reference[0]["source"]["harness_sha256"].get("oracle.py") if valid else None
+    valid = valid and bool(oracle) and records[0]["source"]["harness_sha256"].get("oracle.py") == oracle
+    if not valid:
+        return dict(valid=False, per_time=[], worst_post_initial=[], uniform256_bridge=[])
+    key = lambda r: (r["re_L"], r["n"], r["depth"], r["explosion"])
+    before, after = {key(r): r for r in reference}, {key(r): r for r in records}
+
+    def change(role, metric, low, high, **extra):
+        ratio = high / low if low else None
+        percent = 100*(ratio-1) if ratio is not None else None
+        return dict(role=role, metric=metric, n128_campaign_error=low, n256_campaign_error=high,
+                    signed_change=high-low, new_to_old=ratio if finite(ratio) else None,
+                    ratio_unavailable_reason=None if finite(ratio) else "zero historical error" if not low else "overflow",
+                    change_percent=percent if finite(percent) else None, **extra)
+
+    per_time, worst, bridge = [], [], []
+    for old_key, role in zip(CASES, ("uniform_root", "uniform_fine", "constant", "projected_linear")):
+        re, n, depth, scheme = old_key
+        low, high = before[old_key], after[(re, 2*n, depth, scheme)]
+        for metric in LIMITS:
+            per_time.extend(change(role, metric, a[metric], b[metric], tstar=a["tstar"])
+                            for a, b in zip(low["rows"], high["rows"]))
+            worst.append(change(role, metric, maximum(low["rows"][1:], metric),
+                                maximum(high["rows"][1:], metric)))
+    for a, b in zip(before[CASES[1]]["rows"], after[CASES[1]]["rows"]):
+        bridge.extend(dict(metric=m, tstar=a["tstar"], historical=a[m], fresh=b[m], signed_change=b[m]-a[m])
+                      for m in LIMITS)
+    low, high = before[CASES[1]], after[CASES[1]]
+    bridge_identity = dict(main_rows_identical=low["rows"] == high["rows"],
+                           early_pressure_rows_identical=low["early_pressure"]["rows"] == high["early_pressure"]["rows"])
+    same_runtime = reference[0]["runtime_identity"] == records[0]["runtime_identity"]
+    return dict(valid=True, reference_revision=REFERENCE_REVISION, reference_receipt_sha256=REFERENCE_RECEIPTS,
+                reference_scheme_results=old["scheme_results"], per_time=per_time, worst_post_initial=worst,
+                uniform256_bridge=bridge, uniform256_bridge_identity=bridge_identity,
+                same_recorded_runtime=same_runtime,
+                trend_interpretation_ready=all(bridge_identity.values()) and same_runtime,
+                scope="Historical same-production/oracle reference versus fresh controls; same-placement two-resolution trend only, not convergence order or phase robustness")
+
+
 if __name__ == "__main__":
     p = argparse.ArgumentParser(description=__doc__)
     p.add_argument("files", type=Path, nargs="+")
     p.add_argument("--output", type=Path, required=True)
     p.add_argument("--expected-revision", required=True)
+    p.add_argument("--diagnostic", choices=[SCALED_DIAGNOSTIC])
+    p.add_argument("--reference-files", type=Path, nargs=4)
     args = p.parse_args()
     if args.output.exists():
         p.error("Output already exists; choose a new path to retain prior evidence")
-    result = assess([json.loads(path.read_text()) for path in args.files], args.expected_revision)
+    if bool(args.diagnostic) != bool(args.reference_files):
+        p.error("--diagnostic scaled_n256 requires --reference-files; references are diagnostic-only")
+    records = [json.loads(path.read_text()) for path in args.files]
+    result = assess(records, args.expected_revision, diagnostic=args.diagnostic)
+    if args.diagnostic:
+        try:
+            reference = load_reference(args.reference_files)
+        except ValueError as exc:
+            p.error(str(exc))
+        result["diagnostic"] = args.diagnostic
+        result["resolution_trend"] = resolution_trend(records, reference, args.expected_revision)
+        result["checks"]["valid_historical_reference_and_trend"] = result["resolution_trend"]["valid"]
+        result["passed"] = all(result["checks"].values())
     args.output.write_text(json.dumps(result, indent=2, allow_nan=False) + "\n")
     print(json.dumps(result, indent=2, allow_nan=False))
     raise SystemExit(0 if result["passed"] else 1)

tests/validation/taylor_green/run_case.py

@@ -18,6 +18,7 @@ SOURCE_SHA256 = "a5dbfd804befce868067d79a23e5fd47dce7f505b3f9b1bb2145fb57cb5f217
 BUILD_CAPTURE_SHA256 = "3dbaa6547cb0a4cb1541054f3b64ad8b6e18cc70527ac6166e56b944e9bd464a"
 REFERENCE_HARNESS = "11901a7f77557a901e017a171eeae1758d584fa5"
 INITIALIZATION = "physical FP64 equilibrium G at actual cell centers; rho=1+3p"
+SCALED_DIAGNOSTIC = "scaled_n256"
 
 
 def sample_steps(n, early_pressure=False):
@@ -145,10 +146,12 @@ def main():
     p.add_argument("--run-gpu", action="store_true")
     p.add_argument("--dry-run", action="store_true")
     p.add_argument("--re", type=int, choices=[100, 1000], required=True)
-    p.add_argument("--n", type=int, choices=[32, 64, 128, 256], required=True)
+    p.add_argument("--n", type=int, choices=[32, 64, 128, 256, 512], required=True)
     p.add_argument("--depth", type=int, choices=[1, 2, 3], required=True)
     p.add_argument("--output", type=Path, required=True)
     p.add_argument("--explosion", choices=["constant", "projected_linear"], default="constant")
+    p.add_argument("--diagnostic", choices=[SCALED_DIAGNOSTIC],
+                   help="explicit four-case N256/fine512 same-placement diagnostic")
     p.add_argument("--early-pressure", action="store_true",
                    help="also save 65 scalar pressure samples at t*=k*.001875, k=0..64")
     args = p.parse_args()
@@ -156,6 +159,13 @@ def main():
         p.error("Output already exists; choose a new path to retain prior evidence")
     if args.explosion != "constant" and args.depth != 2:
         p.error("projected_linear requires depth 2")
+    if args.diagnostic:
+        cases = {(256, 1, "constant"), (512, 1, "constant"),
+                 (256, 2, "constant"), (256, 2, "projected_linear")}
+        if args.re != 100 or (args.n, args.depth, args.explosion) not in cases or not args.early_pressure:
+            p.error("scaled_n256 requires one of its four Re100 cases and --early-pressure")
+    elif args.n == 512:
+        p.error("N512 requires --diagnostic scaled_n256")
     selected = os.environ.get("CUDA_VISIBLE_DEVICES", "")
     if not args.dry_run and (not args.run_gpu or not selected or "," in selected or selected == "-1"):
         p.error("--run-gpu and exactly one explicit CUDA_VISIBLE_DEVICES are required")
@@ -164,7 +174,8 @@ def main():
     source = verify_source(repo, args.expected_revision)
     runtime = runtime_source(repo)
     if args.dry_run:
-        print(json.dumps(dict(source=source, runtime_source=runtime, dry_run=True)))
+        print(json.dumps(dict(source=source, runtime_source=runtime, dry_run=True,
+                              **({"diagnostic": args.diagnostic} if args.diagnostic else {}))))
         return 0
     from CelerisLab.cuda.context import CudaContext
     from CelerisLab.lbm.blocks import BlockLayout2D
@@ -212,6 +223,8 @@ def main():
                             sampled_max_pressure_l2=None),
         completed=False,
     )
+    if args.diagnostic:
+        report["diagnostic"] = args.diagnostic
     args.output.parent.mkdir(parents=True, exist_ok=True)
     def save():
         temporary = args.output.with_suffix(args.output.suffix + ".tmp")

tests/validation/taylor_green/test_projected_pairs.py

@@ -17,6 +17,7 @@ from unittest.mock import patch
 import numpy as np
 
 import run_case
+import assess_projected_pairs as assessor
 from assess_projected_pairs import (BUILD_SOURCES, CASES, NUMERICS, NUMERICAL_REVISION,
                                     SOURCE_SHA256, assess)
 from oracle import LIMITS, TIMES, U, rectangles
@@ -31,7 +32,7 @@ RUNTIME = dict(device=dict(name="host mock", pci_bus_id="mock", capability=[7, 0
 
 def source():
     return dict(revision=MOCK_REVISION, dirty=False, source_tree_sha256=SOURCE_SHA256,
-                harness_sha256={"run_case.py": "a" * 64},
+                harness_sha256={"run_case.py": "a" * 64, "oracle.py": "c" * 64},
                 build_capture_sha256=run_case.BUILD_CAPTURE_SHA256)
 
 
@@ -42,12 +43,14 @@ def build_metadata(size):
                 arch="sm_70", flags=["--use_fast_math"], ptx_sha256="b" * 64)
 
 
-def records(early=False):
+def records(early=False, diagnostic=None):
     from CelerisLab.lbm.hierarchy import build_hierarchy
     from CelerisLab.lbm.refinement import ACTIVE
 
     out = []
     for re, n, depth, scheme in CASES:
+        if diagnostic:
+            n *= 2
         cfg = run_case.config(n, re)
         refinements = [] if depth == 1 else rectangles(n, depth)
         h = build_hierarchy(cfg, refinements)
@@ -66,6 +69,8 @@ def records(early=False):
             early_pressure=dict(enabled=early, rows=[
                 dict(step=s, tstar=s*U/n, pressure_l2=.01 if s not in main_steps else 0.0) for s in early_steps],
                 sampled_max_pressure_l2=.01 if early else None)))
+        if diagnostic:
+            out[-1]["diagnostic"] = diagnostic
     return out
 
 
@@ -251,7 +256,7 @@ class ProjectedPairsTests(unittest.TestCase):
     def test_pinned_build_sources_match_host_generated_inputs(self):
         from CelerisLab.cuda import compiler_v2
 
-        for size in (128, 256):
+        for size in (128, 256, 512):
             cfg = replace(run_case.config(size, 100), nx=10, ny=10)
             cfg.validate()
             root = compiler_v2.generate_config(cfg)
@@ -263,6 +268,111 @@ class ProjectedPairsTests(unittest.TestCase):
             finally:
                 compiler_v2.cleanup_build_dir(root)
 
+    def test_scaled_geometry_work_and_frozen_gates(self):
+        data = records(True, run_case.SCALED_DIAGNOSTIC)
+        self.assertEqual(sum(r["workload"]["active_collision_updates"] for r in data), 4714397696)
+        self.assertEqual([r["viscosity"] for r in data], [.0768, .1536, .0768, .0768])
+        self.assertEqual(data[-1]["tau_by_level"], [.7303999999999999, .9608])
+        for r in data:
+            self.assertEqual([row["tstar"] for row in r["rows"]], TIMES)
+            self.assertEqual(len(r["early_pressure"]["rows"]), 65)
+            if r["depth"] == 2:
+                self.assertEqual(r["refinements"], [[(32, 32, 224, 192)]])
+                self.assertEqual(r["workload"]["active_cells_by_level"], [34816, 122880])
+                self.assertEqual(r["workload"]["level_hops"], [4096, 8192])
+        result = assess(data, MOCK_REVISION, diagnostic=run_case.SCALED_DIAGNOSTIC)
+        self.assertTrue(result["passed"])
+        self.assertEqual(result["floors"], dict(velocity_l2=.002, pressure_l2=.02, energy_relative=.004))
+        self.assertTrue(all(row["floor_dominates"] for row in result["comparisons"]))
+        data[-1]["rows"][-1]["velocity_l2"] = .00200001
+        self.assertFalse(assess(data, diagnostic=run_case.SCALED_DIAGNOSTIC)["passed"])
+        self.assertFalse(assess(data)["passed"])
+        data[-1]["rows"][-1]["velocity_l2"] = 0.0
+        for mutate in (lambda r: r.pop("diagnostic"), lambda r: r.update(diagnostic="wrong"),
+                       lambda r: r["source"].update(revision="wrong"),
+                       lambda r: r["early_pressure"].update(enabled=False),
+                       lambda r: r.update(refinements=[[(32, 32, 224, 200)]])):
+            bad = deepcopy(data)
+            mutate(bad[-1])
+            self.assertFalse(assess(bad, MOCK_REVISION, diagnostic=run_case.SCALED_DIAGNOSTIC)["passed"])
+
+    def test_raw_resolution_changes_and_reference_identity(self):
+        old, new = records(True), records(True, run_case.SCALED_DIAGNOSTIC)
+        for r in old:
+            r["source"]["revision"] = assessor.REFERENCE_REVISION
+        old[-2]["rows"][-1]["velocity_l2"] = .004  # historical constant failure remains usable
+        old[-1]["rows"][-1]["velocity_l2"] = .001
+        new[-1]["rows"][-1]["velocity_l2"] = .0015  # floor pass, but worse raw error
+        result = assessor.resolution_trend(new, old, MOCK_REVISION)
+        self.assertTrue(result["valid"])
+        self.assertTrue(result["trend_interpretation_ready"])
+        self.assertFalse(result["reference_scheme_results"]["constant"]["passed"])
+        self.assertEqual((len(result["per_time"]), len(result["worst_post_initial"]),
+                          len(result["uniform256_bridge"])), (80, 16, 20))
+        row = next(r for r in result["per_time"] if r["role"] == "projected_linear"
+                   and r["metric"] == "velocity_l2" and r["tstar"] == .48)
+        self.assertEqual((row["signed_change"], row["new_to_old"], row["change_percent"]), (.0005, 1.5, 50.))
+        self.assertTrue(assess(new, diagnostic=run_case.SCALED_DIAGNOSTIC)["passed"])
+        self.assertIsNone(result["per_time"][0]["new_to_old"])  # zero denominator is explicit
+        self.assertEqual(result["per_time"][0]["ratio_unavailable_reason"], "zero historical error")
+        new[-1]["rows"][-1]["velocity_l2"] = .06  # retain raw errors even when absolute screen fails
+        self.assertTrue(assessor.resolution_trend(new, old, MOCK_REVISION)["valid"])
+        self.assertFalse(assess(new, diagnostic=run_case.SCALED_DIAGNOSTIC)["passed"])
+        new[-1]["rows"][-1]["velocity_l2"] = .0015
+        for sample in ("rows", "early_pressure"):
+            bad = deepcopy(new)
+            if sample == "rows":
+                bad[0]["rows"][-1]["velocity_l2"] = .0001
+            else:
+                bad[0]["early_pressure"]["rows"][1]["pressure_l2"] = .009
+            result = assessor.resolution_trend(bad, old, MOCK_REVISION)
+            self.assertTrue(result["valid"])
+            self.assertFalse(result["trend_interpretation_ready"])
+        for mutate in (lambda r: r["source"].update(revision=MOCK_REVISION),
+                       lambda r: r["source"].update(source_tree_sha256="wrong"),
+                       lambda r: r["source"]["harness_sha256"].update(**{"oracle.py": "wrong"}),
+                       lambda r: r.update(diagnostic=run_case.SCALED_DIAGNOSTIC),
+                       lambda r: r.update(refinements=[]), lambda r: r["rows"].pop()):
+            bad = deepcopy(old)
+            mutate(bad[-1])
+            self.assertFalse(assessor.resolution_trend(new, bad, MOCK_REVISION)["valid"])
+        with tempfile.TemporaryDirectory() as tmp:
+            paths = [Path(tmp) / f"{i}.json" for i in range(4)]
+            for path, receipt in zip(paths, old):
+                path.write_text(json.dumps(receipt))
+            with self.assertRaisesRegex(ValueError, "receipt hashes"):
+                assessor.load_reference(paths)
+            witnesses = {p.name: hashlib.sha256(p.read_bytes()).hexdigest() for p in paths}
+            with patch.object(assessor, "REFERENCE_RECEIPTS", witnesses):
+                self.assertEqual(assessor.load_reference(paths), json.loads(json.dumps(old)))
+                with self.assertRaises(ValueError):
+                    assessor.load_reference(paths[:-1] + paths[:1])
+                paths[-1].write_text(paths[-1].read_text() + " ")
+                with self.assertRaises(ValueError):
+                    assessor.load_reference(paths)
+
+    def test_scaled_runner_requires_explicit_bounded_diagnostic(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            base = ["run_case.py", "--repo", str(REPO), "--expected-revision", MOCK_REVISION,
+                    "--output", str(Path(tmp) / "case.json"), "--dry-run"]
+            diagnostic = ["--diagnostic", run_case.SCALED_DIAGNOSTIC, "--early-pressure"]
+            for n, depth, re, extra, valid in (
+                    (512, 1, 100, [], False), (512, 1, 100, diagnostic, True),
+                    (256, 2, 100, diagnostic, True), (128, 2, 100, diagnostic, False),
+                    (256, 3, 100, diagnostic, False), (256, 2, 1000, diagnostic, False),
+                    (256, 2, 100, diagnostic[:2], False), (512, 2, 100, diagnostic, False)):
+                argv = base + ["--n", str(n), "--depth", str(depth), "--re", str(re)] + extra
+                with self.subTest(argv=argv), patch.object(sys, "argv", argv), \
+                     patch.object(run_case, "verify_source", return_value=source()) as verify, \
+                     redirect_stdout(io.StringIO()), patch("sys.stderr", io.StringIO()):
+                    if valid:
+                        self.assertEqual(run_case.main(), 0)
+                    else:
+                        with self.assertRaises(SystemExit) as caught:
+                            run_case.main()
+                        self.assertEqual(caught.exception.code, 2)
+                        verify.assert_not_called()
+
     def test_source_preflight_requires_clean_exact_revision_and_tree(self):
         import oracle
         from tests.validation.multigrid import check_projected_pairs as gate
@@ -349,9 +459,15 @@ class ProjectedPairsTests(unittest.TestCase):
 
     def test_mocked_runner_default_early_and_failure(self):
         # Replace the GPU context and steppers before import. No CUDA module or flow is used.
-        for early, depth, fail in ((False, 1, None), (False, 2, None), (True, 2, None),
+        cases = [(early, depth, fail, 128, None, "projected_linear" if early else "constant")
+                 for early, depth, fail in
+                 ((False, 1, None), (False, 2, None), (True, 2, None),
                                    (True, 2, "resource"), (True, 2, "nonfinite"),
-                                   (True, 2, "probe_nonfinite")):
+                                   (True, 2, "probe_nonfinite"))]
+        cases += [(True, depth, None, n, run_case.SCALED_DIAGNOSTIC, scheme)
+                  for n, depth, scheme in ((256, 1, "constant"), (512, 1, "constant"),
+                                          (256, 2, "constant"), (256, 2, "projected_linear"))]
+        for early, depth, fail, n, diagnostic, scheme in cases:
             calls, selectors = [], []
 
             class FakeRun:
@@ -395,11 +511,14 @@ class ProjectedPairsTests(unittest.TestCase):
 
             with tempfile.TemporaryDirectory() as tmp:
                 output = Path(tmp) / "case.json"
-                argv = ["run_case.py", "--repo", str(REPO), "--re", "100", "--n", "128",
+                argv = ["run_case.py", "--repo", str(REPO), "--re", "100", "--n", str(n),
                         "--depth", str(depth), "--output", str(output),
                         "--expected-revision", MOCK_REVISION, "--run-gpu"]
+                argv += ["--explosion", scheme]
                 if early:
-                    argv += ["--early-pressure", "--explosion", "projected_linear"]
+                    argv += ["--early-pressure"]
+                if diagnostic:
+                    argv += ["--diagnostic", diagnostic]
                 with patch.dict(sys.modules, modules), patch.object(sys, "argv", argv), \
                      patch.dict("os.environ", CUDA_VISIBLE_DEVICES="mock"), \
                      patch.object(run_case, "verify_source", return_value=source()) as verify, \
@@ -413,15 +532,16 @@ class ProjectedPairsTests(unittest.TestCase):
                         self.assertEqual(run_case.main(), 0)
                 receipt = json.loads(output.read_text())
                 verify.assert_called_once_with(REPO, MOCK_REVISION)
-                self.assertEqual(selectors, ["projected_linear" if early else "constant"] if depth == 2 else [])
+                self.assertEqual(selectors, [scheme] if depth == 2 else [])
                 self.assertEqual(receipt["source"], source())
                 self.assertEqual(receipt["runtime_identity"], RUNTIME)
                 self.assertEqual(receipt["completed"], not fail)
+                self.assertEqual(receipt.get("diagnostic"), diagnostic)
                 self.assertEqual(calls[:2], [0, 1])  # READY probe is part of this trajectory
                 if not fail:
                     self.assertEqual([r["tstar"] for r in receipt["rows"]], TIMES)
                     self.assertEqual(len(receipt["early_pressure"]["rows"]), 65 if early else 0)
-                    self.assertEqual(calls[-1], 2048)
+                    self.assertEqual(calls[-1], 16*n)
                 else:
                     self.assertEqual(len(receipt["rows"]), 1)
                     self.assertEqual(len(receipt["early_pressure"]["rows"]),