This document specifies the Operational Integrity Score (OIS), a composite metric that measures whether operational decisions within a governed system remain independently reconstructable, policy-bound, and cryptographically verifiable over time. The OIS is computed from six orthogonal integrity dimensions, each of which captures a distinct failure mode in operational governance.
The OIS is not a security score, a compliance checkbox, or a risk rating. It is a quantitative measure of verifiable operational state. An OIS value of 1.0 indicates that every operational decision in the measurement window can be independently replayed, that every governance policy was continuously attested, that every signature was produced by at least three independent post-quantum families, and that no vendor infrastructure is required to verify any of the above.
An OIS value below 1.0 indicates a specific, enumerable set of degradation events. The score composition is deterministic: given identical input telemetry, any conformant implementation MUST produce an identical OIS value.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119.
This document is a Draft specification. It has not been submitted to any standards body. The specification is published by H33.ai, Inc. for public review and implementer feedback. Normative requirements are binding for implementations that claim OIS conformance.
Comments and errata SHOULD be directed to standard@h33.ai.
| Term | Definition |
|---|---|
| Operational Integrity | The property of a system in which every operational decision is independently reconstructable from its governance chain, policy state, and cryptographic attestations, without reliance on the originating infrastructure. |
| Integrity Dimension | One of six orthogonal axes along which operational integrity is measured. Each dimension corresponds to a distinct class of degradation event. |
| Score Component | A normalized value in the range [0.0, 1.0] representing the measured integrity of a single dimension within the scoring window. |
| Degradation Event | An observable event that reduces one or more score components. Degradation events are enumerated in Section 7. |
| Recovery Event | An observable event that increases one or more score components. Recovery events are enumerated in Section 8. A recovery event does not erase the historical degradation; it restores the dimension to a higher value for subsequent measurement windows. |
| Measurement Window | The time interval over which telemetry is aggregated for a single OIS computation. The default measurement window is 24 hours. Implementations MAY support configurable windows with a minimum granularity of 1 hour. |
| Scoring Window | Synonym for measurement window. Used interchangeably in this document. |
The Operational Integrity Score is composed of six dimensions. Each dimension captures an independent failure mode. Dimensions are scored individually and combined via a weighted composite function defined in Section 5.
Measures the rate at which operational decisions can be deterministically replayed from their governance chain and attestation history. A decision is reproducible if and only if an independent verifier, given the governance chain state at the time of the decision, can derive an identical outcome without access to the originating system.
Computation:
DR = count(decisions_successfully_replayed) / count(total_decisions)Where decisions_successfully_replayed is the number of decisions in the measurement window for which deterministic replay produced a byte-identical governance receipt. Decisions with missing chain entries are counted as replay failures.
Measures the rate of scope violations relative to total agent actions. A scope violation occurs when an agent executes an action outside its delegated authority as defined by the governing scope constraint set.
Computation:
ASI = 1.0 - (count(scope_violations) / count(total_agent_actions))An environment with zero agent actions in the measurement window MUST report ASI as 1.0 (no violations possible). Escalation events that are handled within the escalation protocol defined in Agent Governance Attestation Model Section 8 are not counted as scope violations.
Measures the percentage of the governance chain that is independently verifiable at the time of measurement. A chain entry is independently verifiable if its signatures can be validated, its policy reference can be resolved, and its predecessor hash is consistent with the chain state, all without contacting the originating infrastructure.
Computation:
ES = count(independently_verifiable_entries) / count(total_chain_entries)Entries with expired signatures, unreachable policy references, or broken predecessor hashes are not independently verifiable. Entries relying on proprietary verification endpoints that are not publicly specified are not independently verifiable.
Measures the proportion of the measurement window during which policy enforcement attestation was continuously maintained. A gap in policy continuity occurs when the system operates without an active, attested policy for any duration.
Computation:
PC = total_attested_policy_seconds / total_measurement_window_secondsPolicy transitions (version N to version N+1) MUST NOT produce a gap if the new policy is attested before the old policy expires. Implementations MUST treat overlapping policy validity windows as continuous coverage.
Measures the post-quantum signature family coverage across attestations in the measurement window. Signer diversity quantifies the number of independent cryptographic hardness assumptions that must be simultaneously broken to forge any attestation in the window.
Computation:
SD = count(distinct_families_used) / count(supported_families)Where supported_families is the set of PQ signature families defined in the Multi-Family Post-Quantum Signature Framework. In the current specification, supported_families = 3 (ML-DSA-65, FALCON-512, SLH-DSA-SHA2-128f). An attestation signed by only one family contributes 1/3 to SD. An attestation signed by all three families contributes 3/3 = 1.0.
SD is computed as the minimum across all attestations in the window:
SD = min(SD_i) for all attestations i in the measurement windowMeasures the percentage of attestations in the measurement window that can be verified without any vendor-operated infrastructure. An attestation is independently verifiable if its verification algorithm is publicly specified, its verification inputs (public keys, domain parameters, transcripts) are available from non-vendor sources, and its verification can be performed by a third-party implementation.
Computation:
VI = count(vendor_independent_attestations) / count(total_attestations)Attestations that require contacting a vendor API for key resolution, nonce verification, or transcript retrieval are not vendor-independent.
The composite OIS value is computed as a weighted sum of the six dimension scores:
OIS = w_DR * DR + w_ASI * ASI + w_ES * ES + w_PC * PC + w_SD * SD + w_VI * VINormalization rules:
Determinism requirement: given identical telemetry inputs and identical weights, any conformant implementation MUST produce a bit-identical OIS value.
The following default weights are defined. These weights reflect the relative importance of each dimension for general-purpose operational integrity assessment.
| Dimension | Identifier | Default Weight |
|---|---|---|
| Decision Reproducibility | w_DR | 0.25 |
| Agent Scope Integrity | w_ASI | 0.15 |
| Evidence Survivability | w_ES | 0.20 |
| Policy Continuity | w_PC | 0.15 |
| Signer Diversity | w_SD | 0.10 |
| Verification Independence | w_VI | 0.15 |
Customization rules:
weights field (see Section 9).The following events cause one or more dimension scores to decrease. Each event is tagged with the affected dimension(s).
| Event | Dimension | Effect |
|---|---|---|
| Replay failure (non-deterministic outcome) | DR | Reduces DR by 1/N where N = total decisions |
| Missing chain entry for a decision | DR, ES | Reduces DR (replay impossible) and ES (chain gap) |
| Agent scope violation | ASI | Reduces ASI by 1/N where N = total agent actions |
| Broken predecessor hash in chain | ES | Reduces ES for that entry and all subsequent entries until chain integrity is restored |
| Expired signature on chain entry | ES | Reduces ES for that entry |
| Policy gap (no active attested policy) | PC | Reduces PC proportionally to gap duration |
| Single-family attestation | SD | SD floored at 1/3 for that window |
| Two-family attestation | SD | SD floored at 2/3 for that window |
| Vendor-dependent verification required | VI | Reduces VI by 1/N where N = total attestations |
| Proprietary key resolution dependency | VI | Reduces VI for all attestations requiring that key |
A single missing chain entry simultaneously degrades both Decision Reproducibility and Evidence Survivability. Implementations MUST account for cross-dimension degradation.
The following events cause one or more dimension scores to increase. Recovery events apply to subsequent measurement windows only; they do not retroactively alter historical OIS values.
| Event | Dimension | Effect |
|---|---|---|
| Chain entry backfilled with valid attestation | DR, ES | Entry becomes replayable and independently verifiable |
| Scope constraint tightened (reducing violation surface) | ASI | Future violations less likely; measured in subsequent windows |
| Predecessor hash repaired via re-attestation | ES | Chain integrity restored from that entry forward |
| Signature renewed before expiry | ES | Entry remains independently verifiable |
| Policy activated (closing a gap) | PC | Subsequent seconds count as attested |
| Additional PQ family added to signing | SD | SD floor raised for subsequent attestations |
| Public key registry published | VI | Attestations using those keys become vendor-independent |
Implementations MUST NOT modify historical OIS values. Each measurement window's OIS is immutable once computed. Recovery events improve future windows only.
Implementations MUST support the following JSON reporting format for OIS values. All fields are REQUIRED unless marked optional.
{
"specification": "https://h33.ai/specifications/ois/",
"version": "1.0.0",
"timestamp": "2026-05-22T00:00:00Z",
"measurement_window": {
"start": "2026-05-21T00:00:00Z",
"end": "2026-05-22T00:00:00Z",
"duration_seconds": 86400
},
"ois": 0.937500,
"dimensions": {
"decision_reproducibility": 0.980000,
"agent_scope_integrity": 1.000000,
"evidence_survivability": 0.950000,
"policy_continuity": 0.880000,
"signer_diversity": 0.666667,
"verification_independence": 0.950000
},
"weights": {
"w_DR": 0.25,
"w_ASI": 0.15,
"w_ES": 0.20,
"w_PC": 0.15,
"w_SD": 0.10,
"w_VI": 0.15
},
"degradation_events": [
{
"event_type": "policy_gap",
"dimension": "policy_continuity",
"timestamp": "2026-05-21T14:32:00Z",
"duration_seconds": 10368,
"detail": "Policy v7 expired; v8 activated 2.88 hours later"
}
],
"telemetry_hash": "sha3-256:9f86d08..."
}Field definitions:
| Field | Type | Description |
|---|---|---|
specification | string (URI) | Canonical URL of this specification. MUST be https://h33.ai/specifications/ois/. |
version | string (semver) | Specification version the report conforms to. |
timestamp | string (ISO 8601) | Time at which the OIS was computed. MUST be UTC. |
measurement_window | object | The time interval over which telemetry was aggregated. |
ois | number | The composite score. Range [0.0, 1.0]. 6 decimal places. |
dimensions | object | Per-dimension scores. Each value in [0.0, 1.0], 6 decimal places. |
weights | object | Weight vector used for computation. REQUIRED even when using defaults. |
degradation_events | array | List of degradation events in the window. MAY be empty. |
telemetry_hash | string | SHA3-256 hash of the input telemetry used to compute the score. Enables deterministic verification by third parties. |
The following operational thresholds are RECOMMENDED for deployment monitoring. These thresholds are advisory; conformant implementations MAY define alternative thresholds appropriate to their operational context.
| OIS Range | Classification | Operational Guidance |
|---|---|---|
> 0.95 | Healthy | All dimensions within acceptable bounds. No action required. Governance state is independently reconstructable. |
0.80 – 0.95 | Attention | One or more dimensions degraded. Review degradation events. Remediation SHOULD be scheduled within the next measurement window. |
< 0.80 | Critical | Significant integrity degradation. Independent reconstructability of operational decisions is compromised. Immediate remediation RECOMMENDED. |
Per-dimension thresholds:
This section enumerates conditions under which OIS computation is undefined or degraded.
| Failure Mode | Behavior | Requirement |
|---|---|---|
| Incomplete telemetry | One or more dimensions cannot be computed because telemetry data is missing. | Implementations MUST report the affected dimension(s) as null in the dimensions object and MUST NOT compute a composite OIS. The ois field MUST be null. |
| Stale attestations | Attestations in the chain have timestamps older than the measurement window. | Attestations outside the measurement window MUST NOT be included in the computation. If all attestations are stale, Evidence Survivability is 0.0 for that window. |
| Missing dimensions | A deployment does not use agents (ASI inapplicable). | Dimensions that are structurally inapplicable MUST be scored as 1.0 (no violations possible). The report MUST include a dimension_exclusions array listing excluded dimensions and their reason. |
| Clock skew | Telemetry timestamps are inconsistent across sources. | Implementations MUST reject telemetry with clock skew exceeding 5 seconds relative to the measurement window boundaries. Rejected telemetry entries MUST be logged. |
| Weight sum violation | Custom weights do not sum to 1.0. | Implementations MUST reject the computation and return an error. No partial OIS MAY be reported. |
An implementation conforms to this specification if and only if all of the following hold:
telemetry_hash field uses SHA3-256 and is computed over the canonical serialization of the input telemetry.Implementations that satisfy requirements 1-8 MAY claim conformance to "OIS Specification v1.0.0" and MUST reference the canonical URL https://h33.ai/specifications/ois/ in their reporting output.
Conformance is testable. A reference test vector set will be published alongside the first stable release. Implementations MUST pass all reference vectors to claim conformance.
Given the following telemetry for a 24-hour measurement window:
| Metric | Value |
|---|---|
| Total decisions | 1,000 |
| Decisions successfully replayed | 980 |
| Total agent actions | 500 |
| Scope violations | 0 |
| Total chain entries | 1,200 |
| Independently verifiable entries | 1,140 |
| Total window seconds | 86,400 |
| Attested policy seconds | 76,032 |
| Distinct PQ families used (min across attestations) | 2 |
| Total attestations | 1,200 |
| Vendor-independent attestations | 1,140 |
Dimension scores:
DR = 980 / 1000 = 0.980000
ASI = 1.0 - (0/500) = 1.000000
ES = 1140 / 1200 = 0.950000
PC = 76032 / 86400 = 0.880000
SD = 2 / 3 = 0.666667
VI = 1140 / 1200 = 0.950000Composite OIS (default weights):
OIS = (0.25 * 0.980000) + (0.15 * 1.000000) + (0.20 * 0.950000)
+ (0.15 * 0.880000) + (0.10 * 0.666667) + (0.15 * 0.950000)
= 0.245000 + 0.150000 + 0.190000 + 0.132000 + 0.066667 + 0.142500
= 0.926167| Version | Date | Changes |
|---|---|---|
1.0.0 | 2026-05-22 | Initial draft specification. |
| Reference | Title | URL |
|---|---|---|
| [RFC 2119] | Key words for use in RFCs to Indicate Requirement Levels | rfc-editor.org/rfc/rfc2119 |
| [IEEE 754] | IEEE Standard for Floating-Point Arithmetic | standards.ieee.org |
| [SHA3-256] | NIST FIPS 202: SHA-3 Standard | csrc.nist.gov |
| [Multi-Family PQ] | Multi-Family Post-Quantum Signature Framework v1.0.0 | h33.ai/specifications/multi-family-pq/ |
| [Agent Governance] | AI Agent Governance Attestation Model v1.0.0 | h33.ai/specifications/agent-governance/ |