An open documentary assurance method for evaluating whether human authority retained practical force in AI-mediated institutional decisions.
HIT evaluates what contemporaneous records establish about human access, judgment, authority, correction, repair, and reform. It does not infer intention, certify legal compliance, perform runtime enforcement, or treat a signature as proof of substantive judgment.
| Layer | Status | Meaning |
|---|---|---|
| Published repository release | 0.6.4 |
DOI-bearing software archive and metadata synchronization |
| Human-result release | 0.6.0 |
First bounded independent human result |
| Active empirical package | 0.7.0 candidate |
Three-case current-contract protocol and packet construction |
| Stable release candidate | 0.9.0 planned |
Clean-room implementation audit and release-candidate freeze |
| Stable public contract | 1.0.0 gated candidate |
Compatibility and independent-implementation commitment after all gates pass |
Current release: 0.6.4
Human-result release: 0.6.0
Conformance engine version: 0.5.0
Specification version: 0.4.0
Assessment schema version: 0.4.0
Dimension catalog version: 0.4.0
Current maturity: Level 2, Applicable
Human reliability claim: Supported for one frozen packet under HIT-IRP-CIGNA-001
Active replication protocol: HIT-IRP-HIT040-002, candidate, scoring prohibited
Stable target: 1.0.0, release prohibited until the published gates pass
Concept DOI, all software versions: 10.5281/zenodo.21204892
Version DOI, exact v0.6.4 release: 10.5281/zenodo.21446142
The GitHub Releases page contains published releases. Candidate and future-version documents in the repository are planning and release-control artifacts. They are not published releases.
Version 1.0.0 will declare that the public HIT assessment contract is stable enough for independent implementation, versioned compatibility commitments, and reproducible conformance checking from public artifacts.
It will not declare population-wide inter-rater reliability, causal effectiveness, legal conformity, certification, truthful reasoning, or independent institutional adoption. Semantic stability and research maturity remain separate claims.
The governing materials are:
ROADMAP.md, for the chronological release path;docs/v1-readiness-plan.md, for the stable-contract gates;docs/releases/v1.0.0-candidate.md, for the candidate release boundary;release/v1.0.0/contract-freeze.candidate.json, for the machine-readable gate ledger;implementation/v1.0.0-candidate/, for the clean-room implementation packet candidate.
| Version | State | Release function |
|---|---|---|
0.4.0 |
Complete | Stabilized the normative assessment contract |
0.5.0 |
Complete | Added complete-record executable conformance |
0.6.0 |
Published | Added the first bounded independent human result and advanced maturity to Level 2 |
0.6.4 |
Published | Established the standalone software archive and version-specific Zenodo DOI |
0.7.0 |
Active candidate | Freeze three current-contract packets and the multi-case replication protocol |
0.8.0 |
Pending | Publish current-contract applications and the empirical result or declared recruitment disposition |
0.9.0 |
Pending | Publish the stable release candidate after clean-room implementation review |
1.0.0 |
Gated | Promote synchronized stable components after every release gate passes |
Current blocking gates are:
- a signed human selection of one exercise-rich, one constraint-rich, and one evidence-limited case;
- three frozen packet boundaries, source manifests, identifiers, and digests;
- locked
v0.7.0scorer, coordinator, comparison, and workbook materials; - three current-contract public applications or documented migration exceptions;
- a complete standalone implementation packet;
- a clean-room implementation audit with original outputs preserved;
- a public
v0.9.0release candidate; - synchronized
1.0.0component promotion, breaking-change review, and exact-release validation.
Draft manual workbooks exist for Scorers A, B, and C. They remain marked DRAFT - SCORING PROHIBITED until the 0.7.0 protocol and all three packets are locked.
Can observable records distinguish substantive human influence from ceremonial human presence in AI-mediated institutional decisions?
- Counsel: Did a named human authority have actual pre-decision access to relevant underlying evidence?
- Judgment: Did the authority independently evaluate reasons, alternatives, uncertainty, and context?
- Command: Could the authority practically approve, reject, modify, stop, or escalate?
- Correction: Could the decision be contested, interrupted, reconsidered, modified, reversed, or appealed in practice?
- Repair: After qualifying harm, did a named actor own and deliver remediation to affected persons?
- Reform: Did a named authority have and exercise power to change the decision architecture?
- Telemetry Integrity: Can the institutional records and assessment packet be trusted as bounded audit evidence?
0: absent; requires affirmative evidence of absence;1: present but ceremonial; requires process-specific formal presence;2: present and substantively exercised; requires observed exercise or directly demonstrated operational capability;IE: insufficient evidence; records an unresolved evidentiary state.
IE is not converted to zero and is not averaged into an ordinal total.
Release 0.6.0 publishes the first completed locked human exercise.
Two eligible independent scorers applied frozen packet HIT-IR-CIGNA-PXDX-001 under the preserved 0.1.0 scorer contract. The pre-adjudication comparison produced:
- 7 of 7 exact agreements;
- exact-agreement proportion
1.0000; - zero critical disagreements;
- advancement threshold met.
Both scorers assigned 1 to Counsel, Judgment, Command, Correction, Repair, and Reform. Both assigned limited to Telemetry Integrity.
Supplementary Cohen's kappa is null because all six substantive ratings fell in one category. The data contain no category variance for chance-corrected estimation. The primary exact-agreement measure remains defined and passed.
See validation/results/, RESEARCH.md, and ADR-0004.
Release 0.4.0 stabilized the normative assessment contract. It introduced evidence states, explicit finding thresholds, dimension-specific rules, Repair triggers, split Telemetry Integrity, sampling and aggregation declarations, actor-authority attribution, contradiction handling, structured evidence propositions, precise locators, and 48 executable boundary fixtures.
Release 0.5.0 added executable complete-record conformance without changing the 0.4.0 contract.
Release 0.6.0 publishes the bounded human agreement result. It changes research maturity and claim status. It does not change the 0.4.0 specification, schema, catalog, handbook, or scoring semantics, and it does not change the 0.5.0 conformance engine.
Release 0.6.4 synchronizes the published software archive and DOI metadata. It does not change the normative contract, conformance engine, human result, H3 decision, or research maturity.
Release candidates 0.7.0, 0.9.0, and 1.0.0 remain governed by prospective gates. Their presence in main does not authorize tagging or publication.
See the repository release index for the distinction between published releases and candidate release documents.
git clone https://github.com/mj3b/human-influence-telemetry.git
cd human-influence-telemetry
python -m pip install --requirement requirements-dev.txt
python -m src conformance --all
python -m src conformance --path assessment.json
python -m src migration-plan --path historical-assessment.jsonThe complete-record engine checks cross-record references, evidence relations, actor attribution, finding and evidence-state consistency, Repair triggers, Telemetry Integrity derivation, aggregation scope, citation precision, and protected migration behavior.
A passing conformance report establishes assessment-contract conformance for the supplied record. It does not establish evidentiary truth, legal sufficiency, certification, or institutional effectiveness.
HIT remains a research instrument, not a certified standard.
| Capability | Status |
|---|---|
| Normative 0.4.0 rubric | Released |
| Machine-readable 0.4.0 assessment schema and catalog | Released |
| 48 accepted, rejected, and boundary fixtures | Released |
| Complete-record executable conformance | Released in 0.5.0 |
| Standalone software archive and DOI | Published in 0.6.4 |
| Historical public case studies | Four 0.1.0 assessments preserved |
| Public-case 0.4.0 migrations | Not claimed; explicit exceptions recorded |
| Locked human inter-rater protocol | Completed under the 0.1.0 scorer contract |
| Independent human submissions | Two verified submissions published |
| Pre-adjudication result | 7 / 7 exact agreement; zero critical disagreements |
| H3 | Supported for one frozen packet |
| Research maturity | Level 2, Applicable |
| Current-contract replication | Candidate protocol; cases and packets not yet locked |
| Clean-room implementation audit | Candidate protocol; audit prohibited until packet freeze |
| Stable public contract | 1.0.0 target; release prohibited |
| Prospective institutional validation | Pending |
| Legal or standards conformity determination | Not claimed |
The completed exercise supports reproducibility for one frozen packet. General reliability across sectors, evidence conditions, scorer populations, and contract versions remains unresolved.
The case-studies/ directory contains three retrospective narratives and four actor-specific machine-readable assessments released under specification and schema 0.1.0.
They remain immutable historical artifacts. No 0.4.0 public-case finding is claimed. The Cigna protocol is complete, but a current-contract Cigna reassessment would require a separate fresh source review.
src/ -- Public CLI and conformance implementation
compatibility/ -- Supported contracts and migration rules
schema/ -- Canonical 0.4.0 schema and dimension catalog
fixtures/ -- Boundary and complete-record conformance vectors
case-studies/ -- Historical public cases and migration dispositions
validation/results/ -- Published human result, decisions, and hashes
validation/v0.7.0/ -- Current-contract replication candidate
validation/v0.7.0/manual-workbooks/ -- Draft workbook controls and asset manifest
implementation/v1.0.0-candidate/ -- Clean-room implementation packet candidate
release/v1.0.0/ -- Machine-readable stable-contract gate ledger
coordinator/ -- Submission preservation and comparison procedures
recruitment/ -- Neutral human-scorer materials
docs/releases/ -- Published and candidate release records
docs/ -- Handbook, decisions, migration, and readiness plans
archive/ -- Superseded public contracts preserved for reproducibility
scripts/ -- Validation, comparison, and CLI smoke tests
- Microsoft Agent Governance Toolkit enforces and audits runtime actions. Its events may supply evidence to HIT; HIT does not enforce actions.
- ScopeBlind/Acta verifies signed-receipt integrity and interoperability. Verified receipts may strengthen provenance; HIT does not define or verify that protocol.
- Credo AI provides policy packs, risk and control mappings, governance workflows, evidence collection, and runtime governance. HIT does not provide policy packs, compliance automation, or universal control mappings.
- Governed Decision Intelligence structures consequential decisions in Governed Decision Records.
- Decision Evidence Applicability Specification evaluates what evidence may support under a specified requirement and where equivalence fails.
HIT does not intercept, allow, deny, block, sandbox, or terminate agent actions. It does not compile policy into runtime configuration, verify agent identity or privileges, create signed receipts, provide cryptographic receipt conformance, supply policy packs, automate legal compliance, declare evidentiary admissibility, or make governance evidence portable across regimes.
A high HIT finding means the available records satisfy the published documentary rubric. It does not establish that the decision was correct, lawful, fair, accurate, harmless, or morally justified.
The 0.6.0 result shows that two independent reviewers reached the declared threshold on one frozen packet. It does not estimate population reliability.
The 0.6.4 archive and DOI release changes discoverability and citation metadata. It does not add empirical evidence or alter the assessment contract.
The 1.0.0 target concerns contract stability and public implementability. It does not upgrade the empirical claim unless new eligible independent human evidence separately satisfies the active protocol.
Use CITATION.cff and identify the exact published release. Cite 10.5281/zenodo.21446142 when referring to the exact v0.6.4 artifact. Cite the all-versions concept DOI 10.5281/zenodo.21446141 when referring to Human Influence Telemetry as an evolving software project.