Every input and data source this artifact consumes, classified. Nothing here is an observational dataset, and the artifact never presents anything as one.
| Class | Meaning |
|---|---|
| synthetic fixture | Authored for this work. Fully specified, not observed. |
| forensic reconstruction | Retrospectively reconstructed from a real prior system’s published artifacts. Not a contemporaneous record. |
| generated perturbation | Produced by a declared stochastic model from a frozen seed. |
| prior-work external artifact | Belongs to cited prior work; referenced, not reproduced. |
| real observed dataset | Collected from a running system or real-world process. |
| human data | Produced by human participants. |
| deferred / unavailable | Named in the evaluation plan, not yet in existence. |
| # | Input | Path | Class | Notes |
|---|---|---|---|---|
| 1 | Case A governance assessment (M, S, O, R, A) | cases/case_a/inputs/assessment.json |
synthetic fixture | 16 register rows, 16 risk-analysis rows, 12 obligations. Investment-research RAG summarizer. Synthetic — not a deployed system. |
| 2 | Case A control derivation catalog | cases/case_a/inputs/control_derivation_catalog.json |
synthetic fixture | 16 approved catalog entries. |
| 3 | Case A C* profile | cases/case_a/inputs/cstar_profile.json |
synthetic fixture | Declared consequence-class rules and materiality order. |
| 4 | Case A invariant register, thresholds, compile parameters, policy metadata | cases/case_a/inputs/ |
synthetic fixture | Includes the declared heat-map threshold T_H = 15. |
| 5 | Case A judgment record, dispositions, ACS | cases/case_a/{judgment,dispositions,acs}/ |
synthetic fixture | The signed human refinement act, authored for the case. |
| 6 | Case A lifecycle registry, receipts, actuations | cases/case_a/lifecycle/ |
synthetic fixture | Plus negative/ fault fixtures used as traceability negative controls. |
| 7 | Case A non-runtime residual ratings (R-14, R-15, R-16) | inside input 1 | synthetic fixture, explicitly flagged | R-14 = 12, R-15 = 6, R-16 = 8. Not derivable from the case narrative. docs/FIXTURE_PROVENANCE.md FP-014. A published sweep shows GD_min would range 2–5 over all plausible values. |
| 8 | Case B governance assessment | cases/case_b/inputs/assessment.json |
forensic reconstruction | 6 register rows, 6 obligations. Transaction-authorization agent. Reconstructed retrospectively from the published enforcement artifact; not a contemporaneous governance record, and the reconstruction is stated openly in §X and §XII. |
| 9 | Case B catalog, C* profile, invariants, thresholds | cases/case_b/inputs/ |
forensic reconstruction | Same provenance as input 8. |
| 10 | Case B judgment record, dispositions, ACS, lifecycle | cases/case_b/{judgment,dispositions,acs,lifecycle}/ |
forensic reconstruction | |
| 11 | Case B → L-DREA predicate-family mapping | cases/case_b/ldrea_traceability.json, ldrea_predicate_family.json |
derived from a prior-work external artifact | Machine-verified against github.com/AGLakhowal/Gamma-Permit-Package @ 40fa8f04…. 13-member predicate family recovered from that artifact’s source; 9 mapping rows (4 exact, 3 family, 2 declared gaps). Establishes continuity of predicate family only. |
| 12 | Monte Carlo rating distributions | preregistration/monte_carlo_distributions_v1.json |
generated perturbation from an author-specified model | Symmetric ±1 ordinal rule: 0.6 on the approved rating, 0.2 on each adjacent value, out-of-scale mass reassigned to the approved rating. Author-specified and prospectively frozen. NOT panel-adjudicated and NOT an empirical estimate of rating uncertainty. docs/FIXTURE_PROVENANCE.md FP-020; monte_carlo_summary.json carries a provenance_warning saying the same. Seed 20260201, K = 250,000. |
| 13 | Determinism perturbation matrix | generated by experiments/run_determinism.py |
generated perturbation | Deterministic, seeded permutations of the committed inputs: 10 repeats, 10 row shuffles, 5 key shuffles, 3 locales, 3 time zones per case. |
| 14 | Adversarial corpus | tests/adversarial/, applied to inputs 1–10 |
synthetic fixture | 62 cases (54 negative, 8 positive controls), 26 structural checks, 3 seeded validation rows. |
| 15 | Case B injection scenarios | derived from the published enforcement artifact | synthetic fixture over prior-work families | The 8 adversarial attack families and 5 ASB scenario families that artifact enumerates, executed against the compiled Case B bundle. |
| 16 | Research signing keys | keys/*.ed25519.json |
synthetic fixture | TEST ONLY / NOT FOR PRODUCTION. Committed so signatures reproduce byte-identically from a fresh clone. They protect nothing and no real key is present. |
| 17 | Cross-environment determinism records | results/final_v2/cross_environment/ |
real observed execution records | Genuine GitHub Actions run artifacts (run 34261657261) plus a local host and a pinned container. These are real observations — of this artifact executing, not of any external process. |
| 18 | 284,807-event transaction corpus | not in this repository | prior-work external artifact | Belongs to [15]. The corpus is not shipped, not reproduced, and the compiled Case B bundle was never executed against it. No figure from that run is reproduced here. |
| 19 | Expert-study participant data | — | deferred / unavailable | RQ5 is preregistered and deferred. No panel convened, no practitioner recruited, no participant data exists. No script generates, simulates or approximates any. |
| 20 | Held-out expert-study register | — | deferred / unavailable | Not authored. Committed in a Tier-1 freeze before recruitment opens; preregistration/HELD_OUT_REGISTER.md records the commitment as openly outstanding rather than hashing a placeholder. |
| Total committed case input bytes | ~282 KB across both cases |
| Real observed datasets of an external process | 0 |
| Human-subject data | 0 |
| External datasets shipped | 0 |
| Documented fixtures with recorded provenance | 29 entries in docs/FIXTURE_PROVENANCE.md |
Case A is synthetic. Case B is a forensic reconstruction. The Monte Carlo distributions are an author-specified, prospectively frozen sensitivity model, not an empirical estimate. The 284,807-event corpus belongs to prior work and this paper claims no execution against it. No human-subject data exists.
All four statements appear in the manuscript itself — in §IX, §X, §XI-G and §XII — not only here.