GC-IR Reference Implementation -- Results

Governance-to-control compiler: measured evidence for the current release

release v1.1.0 freeze preregister-tier0-v4.1 commit efe4e165619e generated 2026-10-02 07:39 UTC

Case bundles

manifest/tooling correction only; scientific state byte-identical to preregister-tier0-v4

f5cbc3a8016159d2…
Case A hash
match: True
0d8b602a4c8af888…
Case B v1.1 hash
match: True
6
Case B v1.1 risks
9
Approved Control Specs
9
Risk-derived predicates
3
Compiler invariants
12
Total predicates
446 passed
Tests
0 failed, 0 skipped

Full hashes (for verification)

Case A payload SHA-256f5cbc3a8016159d2074005fa021bf76ca04fb7aed1408f5ec845418460a43536
Case B v1.1 payload SHA-2560d8b602a4c8af888beb27058b7217893eff92d2f9df7f2944d35347c1031cfc1
Freeze commitefe4e165619ecb5d781fb795113f33720794a2e5
Manifest root SHA-2560640ad1e7a703879d4b9083a088f5b562b638173590292a2c1aed1625102e0e6

Canonical audit queries (Q1-Q10)

Empty result set = conformant. Each case checked independently.

case a
10/10
case b
10/10
case b v1 1
10/10
case c
10/10

Negative fixtures

19 targeted fixtures
19/19
Each fixture is a deliberately corrupted copy of a real compiled bundle, detected by its intended query -- 19 fixtures covering the 10 queries above, not 10 additional queries.

Case B v1.1 injection scenarios

13 adversarial scenarios, each targeting one predicate or invariant. Two independent decision axes -- release_decision is never a fourth peer authorization_decision.

authorization_decision

PERMIT (1)DENY (8)HOLD (4)

release_decision (meaningful only when authorization_decision = PERMIT)

ALLOW_RELEASE (12)BLOCK_RELEASE (1)
ScenarioTargetAuthorizationReleasesafe_stateExternalizedResult
missing_predicate ACS-B01-01 HOLD ALLOW_RELEASE True False PASS
corrupted_input ACS-B01-02 HOLD ALLOW_RELEASE True False PASS
network_partition_or_delay ACS-B04-01 HOLD ALLOW_RELEASE True False PASS
session_intent_compromise ACS-B01-01 HOLD ALLOW_RELEASE True False PASS
toctou ACS-B06-01 PERMIT BLOCK_RELEASE False False PASS
replay_attack ACS-B05-01 DENY ALLOW_RELEASE True False PASS
payload_mutation ACS-B01-01 DENY ALLOW_RELEASE True False PASS
concurrency_conflict ACS-B01-03 DENY ALLOW_RELEASE True False PASS
adaptive_attacker ACS-B01-02 DENY ALLOW_RELEASE True False PASS
identity_provenance_deception ACS-B04-01 DENY ALLOW_RELEASE True False PASS
runtime_infrastructure_drift INV-VERSION DENY ALLOW_RELEASE True False PASS
economic_logic_fragility ACS-B02-02 DENY ALLOW_RELEASE True False PASS
cross_entity_fraud_propagation ACS-B02-01 DENY ALLOW_RELEASE True False PASS

Determinism

Translation Determinism (TD)
62/62
31 runs per case (10 repeat, 10 row-shuffle, 5 key-shuffle, 3 locale, 3 timezone) x 2 cases = 62.
1.000
Case A TD
1.000
Case B v1.1 TD

Monte Carlo rating-robustness

K = 250,000 draws, seed = 20260201. Maximum per-risk heat-map gate-flip probability.

Case A (risk R-06)
0.321268
Case B v1.1 (risk B-01)
0.317980
0.0
Case B v1.1 FP_C*
4
Case B v1.1 GD_approved
0.0
Case B v1.1 GD_min_mean

L-DREA / ULB downstream reproduction

Pinned: https://github.com/AGLakhowal/Gamma-Permit-Package.git @ 40fa8f046ad3

284,807
Rows
492
Fraud-labelled
0
False permits
0
False denials

GC-IR ↔ downstream predicate-family correspondence

exact (4)family (3)not_established (5)
Claim boundary. The original 284,807-row experiment was reproduced fresh and unmodified this generation. The new Case B v1.1 twelve-predicate bundle was not executed against that corpus (new_v1_1_replay_executed = False). No fraud-detection, classifier-performance, or production-deployment claim is made anywhere in this artifact.

Machine-readable checks

PASSED
ieee-check-v4
3 stages
PASS
Freeze verification
45 checks, 0 failures

Historical campaign -- v1.0.5 / results/final_v2/

Superseded by the current release above; unchanged and reported here only for comparison. Governed by the unmodified preregister-tier0-v2.2 freeze.

f5cbc3a80161…
Bundle A
2850155a2ee7…
Bundle B (6 ACS)
1.000
TD
62/62 runs
280/280
Tests
18/18
Negative controls
A: 3/3 · B: 4/0
GD(15) / GD_min

Paper 2 v1.2 — development phase (not yet frozen)

Not a reportable campaign. reportable = False; freeze status = NOT CREATED. Every figure below is development-phase evidence from results/development_v12/, gathered ahead of a prospective preregister-tier0-v5 freeze that has not been created. No v1.2.0 tag exists.
8/8
Surface A
8 paired semantic families: nominal accepted
8/8
Surface A negatives
correct reason: 8/8
10/10
Case D matrix
D1–D10
16/16
Validator matrix
16 positive/negative pairs
19/19
Audit: historical
Q1–Q10 negatives still detected
9/9
Audit: new v1.2
new audit-regression negatives
155/155
Local determinism
31 runs x 5 cases (A, B v1.1, D CCS0/CCS1/CCS2)
21/21
6-leg CI matrix
Ubuntu/Windows/macOS x Py 3.11/3.12
155/155
Env-7 container (aarch64)
pinned Linux/aarch64, hashes match
14/14
Case B integration
13 adverse + 1 clean
2/2
Observation diagnostics
history-scoped + contract-completeness

Monte Carlo (GD/GDmin regression)

K = 250,000, seed = 20260201. Status: REPRODUCES historical baseline.

3 / 3.226356
Case A GD_approved / GD_min
4 / 0.0
Case B v1.1 GD_approved / GD_min

Compile timing (10 warmup + 100 measured, single host)

Casenmedianp95
case_a10012.45 ms12.80 ms
case_b_v1_11007.80 ms7.88 ms
case_d_ccs11003.83 ms3.87 ms

Single reportable reference machine only (macOS-26.6.2-arm64-arm-64bit, arm64, Python 3.11.15); hosts never pooled; not platform-general latency (spec section 12).

Full campaign history

Every frozen, reportable campaign this repository has produced, oldest first. Case A's payload hash (f5cbc3a8…) and Case B's historical payload hash (2850155a…) have never changed across any of these. Only v1.1.0 (results/final_v4_1/, detailed above) is the current reportable release.

ReleaseFreeze tagCommitDeterminismTestsStatus
v1.0.0-era preregister-tier0-v1 196820b23a 60/60 275/275 Superseded -- never pushed publicly
v1.0.5 preregister-tier0-v2.2 c44f25d6fd 62/62 280/280 Superseded -- first public freeze
v1.0.4 target preregister-tier0-v3 8f85d9b90d 62/62 320/320 Superseded by v3.1 (PERMIT/DENY/HOLD semantics correction)
v1.0.5 (corrected) preregister-tier0-v3.1 158c0bd378 62/62 320/320 Superseded by v4 (adds Case B v1.1 + Case C)
v1.1.0 pre-release preregister-tier0-v4 48791c720b 62/62 438/438 Superseded by v4.1 -- MANIFEST tooling fix only, science byte-identical
v1.1.0 (current release) preregister-tier0-v4.1 efe4e16561 62/62 446/446 Current reportable release -- shown in full above