d6g8k5htny-coder · Oct 12, 2026 · Oct 6, 2026 · Oct 12, 2026
@@ -5,6 +5,7 @@ name: Equal-height mass controls
     paths:
       - 'frontiers/equal_height_mass_value_20261001/**'
       - '.github/workflows/equal-height-mass.yml'
+      - 'tests/test_equal_height_workflow.py'
       # bound merged sources (consumed and cited_only in SOURCES.json)
       - 'frontiers/c7_zero_gap_limit_20260929/PROOF.md'
       - 'frontiers/three_fronts_20260924/LIFETIME_REMAINDER.md'
@@ -31,6 +32,13 @@ jobs:
       - uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97
         with:
           python-version: '3.11.16'
+      - name: Exercise complete output contracts in both modes
+        env:
+          PYTHONDONTWRITEBYTECODE: '1'
+        run: |
+          set -euo pipefail
+          python -B -S -m unittest discover -s tests -p test_equal_height_workflow.py -v
+          python -B -O -S -m unittest discover -s tests -p test_equal_height_workflow.py -v
       - name: Verify sources, replay both modes, reject mutants
         env:
           PYTHONDONTWRITEBYTECODE: '1'
@@ -58,19 +66,31 @@ jobs:
                   raise ValueError('merged source drifted: ' + entry['path'])
           if manifest.get('consumed_unmerged'):
               raise ValueError('this packet declares no unmerged dependency')
+          mutants = manifest.get('mutants')
+          if type(mutants) is not list or mutants != ['M1', 'M2', 'M3']:
+              raise RuntimeError('complete ordered M1-M3 inventory required')
+          # Fixed full reports measured from unchanged checker 92ba3cde in both modes.
+          # Byte identity is intentional; changed diagnostics need a reviewed rebind.
+          contracts = {
+              'M1': (933, 'b6f95f60a211e6abf9c5595fcc26ebd74ea5d3c6c5ade551e94ac1ce4e98952f'),
+              'M2': (938, '4cd2c6f39ed2c920d0fc69fb3a1826b9cec35465c3013d00996b4b26ce788c49'),
+              'M3': (935, 'ac69c631970c2b5f6b1cba6c9457b5601148d0ca7ea087cc012e2772ffd9fa09'),
+          }
           script = str(root / 'equal_height_mass.py')
           expected = (root / 'RESULTS.json').read_bytes()
           for flags in (['-B', '-S'], ['-B', '-O', '-S']):
-              out = subprocess.run([sys.executable, *flags, script], check=True, capture_output=True, timeout=600).stdout
-              if out != expected:
-                  raise RuntimeError('output differs from RESULTS.json under ' + ' '.join(flags))
-              for m in manifest['mutants']:
-                  rc = subprocess.run([sys.executable, *flags, script, '--mutant', m], capture_output=True, timeout=600).returncode
-                  if rc != 1:
-                      raise RuntimeError('mutant ' + m + ' not rejected')
-              rc = subprocess.run([sys.executable, *flags, script, '--mutant', 'M9'], capture_output=True, timeout=600).returncode
-              if rc != 2:
-                  raise RuntimeError('unknown mutant label not rejected with exit 2')
+              result = subprocess.run([sys.executable, *flags, script], capture_output=True, timeout=600)
+              if result.returncode != 0 or result.stdout != expected or result.stderr != b'':
+                  raise RuntimeError('baseline exit/stdout/stderr differs under ' + ' '.join(flags))
+              for m in mutants:
+                  result = subprocess.run([sys.executable, *flags, script, '--mutant', m], capture_output=True, timeout=600)
+                  size, digest = contracts[m]
+                  if (result.returncode != 1 or result.stderr != b'' or len(result.stdout) != size
+                          or hashlib.sha256(result.stdout).hexdigest() != digest):
+                      raise RuntimeError('mutant ' + m + ' differs from its complete fixed rejection report')
+              result = subprocess.run([sys.executable, *flags, script, '--mutant', 'M9'], capture_output=True, timeout=600)
+              if result.returncode != 2 or result.stdout != b'' or result.stderr != b'unknown mutant label\n':
+                  raise RuntimeError('unknown mutant label differs from fixed exit/stdout/stderr')
           print(json.dumps({'scientific_effect': 'NONE', 'mathematical_acceptance': False}))
           PY
           git diff --exit-code
@@ -0,0 +1,120 @@
+# Equal-height mass workflow protocol repair
+
+Dylan Roy — delegated AI engineering. Actual author: OpenAI / GPT-6 Astra Pro,
+`remainder-followthrough-equal-height-20261006-r3`; pickup Math-#322/6017205382.
+Scientific effect NONE; organizational-independence credit 0. No author self-merge.
+
+## Exact source and unchanged scope
+
+Source/base: `37bd017be7b1a48455fe651977fedc5b7105d188`.
+
+| Source | Bytes | Git blob | SHA256 |
+|---|---:|---|---|
+| Original equal-height-mass.yml |3899|663324e708fbb552fb30e9bed21a697d132989a9|389736b7bc181915b9395ce59bff5f7208ed2dbc38147c3f07e3e93a6b57e9ed|
+| equal_height_mass.py |10792|92ba3cdecb4344771d6d8c3f1ba22ca4fc96e864|88a00dafb5710009bd1eeb6267a6b9a197e2cca304ce2f90df5cd138951d1ace|
+| RESULTS.json |929|98001bf939cdaa4a47418ebff0c8b1e13d49d35b|9054e435878a57a6bea42b8af08c0bd51770474503990334687559f7260b06c3|
+
+The last two files are under `frontiers/equal_height_mass_value_20261001/`.
+All three complete local byte strings were authenticated before execution.
+No scientific checker, NOTE, README, RESULTS, GAMMA, gamma exploration,
+SOURCES, consumed/cited proof, formal source or scientific register is modified.
+No Monte Carlo regeneration or mathematical/numerical enclosure is claimed.
+
+## Actual measured report contracts
+
+The unchanged CLI was executed for baseline, M1, M2, M3 and M9 in both child
+Python modes: ten complete real subprocesses, local CPython 3.13.5.
+All corresponding stdout/stderr pairs match exactly; baseline equals RESULTS.
+
+| Stage | Exit | stdout bytes | stdout SHA256 | stderr |
+|---|---:|---:|---|---|
+| baseline |0|929|9054e435878a57a6bea42b8af08c0bd51770474503990334687559f7260b06c3|empty|
+| M1 |1|933|b6f95f60a211e6abf9c5595fcc26ebd74ea5d3c6c5ade551e94ac1ce4e98952f|empty|
+| M2 |1|938|4cd2c6f39ed2c920d0fc69fb3a1826b9cec35465c3013d00996b4b26ce788c49|empty|
+| M3 |1|935|ac69c631970c2b5f6b1cba6c9457b5601148d0ca7ea087cc012e2772ffd9fa09|empty|
+| M9 |2|0|e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855|`unknown mutant label` plus LF|
+
+M1 fails Q2 and Q3; M2 fails Q2, Q3 AND Q4; M3 fails Q3 only. Q1 remains true.
+The full numeric values and all still-passing check details belong to the fixed
+reports. A one-failed-check schema would not describe these genuine controls.
+The stdout hashes do not certify the numerical values or locate an arbitrary
+program's internal failure; they authenticate these measured output bytes.
+
+The repair requires the native ordered LIST M1/M2/M3 before any child; exact
+baseline exit/stdout with empty stderr; each mutant's exit1/empty stderr and
+entire frozen stdout size/hash; and M9's exit2/empty stdout/exact stderr.
+Report byte identity, including formatting, is intentional. A changed
+deterministic report requires reviewed rebinding, not a tolerance change.
+Expectations are fixed before validation, never learned from the child under test.
+
+Ten additional genuine subprocesses through the exact changed loop matched
+every characterization exit and stream hash. A pass-through recorder retained
+the real results without replacing them. This loop-only replay is distinct
+from a complete real-packet manifest/dependency replay, which requires hosted CI.
+
+## Test-first whole-shell regression
+
+Eight unittest methods contain 96 distinct, complete-shell synthetic scenarios.
+Real committed temporary Git repositories, real Bash/Python children and
+nonempty packet/consumed/cited inputs are used. No subprocess outputs are mocked.
+Each fixture executes the actual workflow verification shell. Expected reports
+are independently captured literals in the test source, not imported from the
+production matcher. Test-only signing/maintenance switches apply only to
+disposable commits; no user or persistent project setting is changed.
+
+The same final test bytes were run against both original and repaired workflows:
+
+| Workflow / outer mode | Methods / fixture outcome | Real children |
+|---|---|---:|
+| original / normal |74 intended assertion failures, 0 errors; 96 cases|811|
+| original / optimized |74 intended assertion failures, 0 errors; 96 cases|811|
+| repair / normal |8/8 methods; 96/96 expected outcomes|471|
+| repair / optimized |8/8 methods; 96/96 expected outcomes|471|
+
+The original's 74 assertions comprise 68 noncanonical report admissions
+(62 semantic protocol failures, six formatting-only cases), five bad inventory
+admissions and one missing wiring assertion. The original valid control and
+22 source/exit/postflight rejections behave correctly. The repaired source
+admits one valid case and rejects 95 cases. Both complete repaired runs finish
+without skips or partitioning (28.933s and 28.593s locally).
+
+Tests reach every named mutant in each child mode, including silence, unrelated
+crash, wrong/noisy/trailing reports, wrong label/exit and deliberate formatting.
+Same-length text and nested-type corruptions distinguish hash validation from
+length-only checks. Baseline stderr and M9 output/reason are separately tested.
+Every protocol rejection must stop at the exact ordered target prefix; seven
+source guards and five malformed inventories stop before any child. The
+tracked-output mutation reaches all ten children before final Git rejection.
+
+An earlier 84-case test revision produced 62 intended assertions in both modes;
+it was augmented with twelve equal-length corruption cases BEFORE the fix.
+Those earlier executions are retained as development history, not added to the
+384 final fixture executions or reported as different product defects.
+Two unavailable execution launches produced no credited numerical result;
+all counted commands completed with captured outputs.
+
+## Source conservation and limitations
+
+The complete 1467-byte original source preflight is byte-identical, SHA256
+`d625514545e6d76b6271dc2f8641eafe94bb7197e2c1a6b9d0d0ed24362f860a`.
+The final 137-byte JSON/Git postflight is byte-identical, SHA256
+`752aa03685dbbe54ec40c3e91b8709eb6453c41c146e9636b295b2129c586cdc`.
+Sizes include YAML indentation. Only the process validation loop, one test
+trigger and two full discovery commands are changed/added. Existing action
+pins, Python 3.11.16, read-only permission, inherited cwd, 600-second child
+budget and 20-minute job budget remain unchanged. Launch/timeout exceptions
+are not caught. The final Git guard and original scientific qualifications stay.
+
+The local checkout is an authenticated source subset, not the full repository.
+No local Lean or complete real-packet dependency replay is claimed. Hosted
+dedicated/general/applicable broader results must be read on the actual
+candidate; local Python 3.13.5 does not substitute for pinned-runtime success.
+This is not a hostile-source sandbox, bounded-output-memory service, durable
+evidence guarantee, numerical enclosure, theorem proof or global wrapper audit.
+#322 remains open for the wider affected class.
+
+One exact-source nonauthor engineering review, any required successor read,
+and actual current-candidate checks precede a separate eligible nonauthor/
+nonreader integration. Preserve original failed evidence, all reviewed blobs
+and source interfaces; use guarded normal integration and verify the landed
+tree and its own push evidence. A later base does not inherit old CI receipts.