Publish GH-WP-0001-T06 conformance contracts

Assistant: codex
Assistant-Model: gpt-5.6-sol
Assistant-Session: 01a05e30-2884-71b0-98d7-7edd16ae737b
This commit is contained in:
tegwick 2026-09-01 20:46:34 +02:00
parent a4066aadb5
commit 8cb7250230
8 changed files with 716 additions and 1 deletions

View file

@ -0,0 +1,132 @@
# Conformance reporting contract
**Owner:** gate-house
**Executor:** whitehat-security
**Posture observer:** kings-guard
**Finding triage:** risk-nexus
**Status:** normative for GH-WP-0001-T06
**Version:** conformance-reporting.v1
**Date:** 2026-09-01
## 1. Purpose
This contract establishes how execution of a Gate House assurance
specification returns to doctrine review. It preserves four distinct judgments:
```text
Gate House specifies the invariant and executable target
whitehat-security attacks and reports whether the observed control held
risk-nexus owns finding severity, disclosure, and owner routing
Gate House decides whether the result requires doctrine change
```
Gate House does not grade its own specification by reinterpreting a whitehat
result, assign severity, close an implementation defect, or block delivery.
## 2. Report envelope
Whitehat-security continues to use `whitehat-run/v1` as its native report. For
Gate House conformance return, the report or a value-safe projection must also
make these fields expressible:
```yaml
schema_version: conformance-reporting.v1
report_id: <stable id>
specification:
id: asm-assurance-targets.v1
test_id: T-03
target:
component: <repo or protected system>
revision: <immutable revision>
environment: <fixture/build/production>
claim:
invariant_ids: [A-14, INV-05]
engagement_id: <whitehat engagement>
authorization_id: <recorded authorization>
run_id: <whitehat run>
outcome: pass | finding | inconclusive | aborted
known_bad_calibration_ref: <evidence ref>
evidence_refs: []
emission_guarantee_refs: []
limitations: []
assurance_statement: <bounded statement>
returned_at: <RFC 3339 UTC>
```
The projection contains no credential value, secret, unrelated response body,
severity, disclosure decision, or exploit detail unsafe for the recipient.
## 3. Outcome rules
- Every outcome returns, including `pass`, because target revision and run date
define the freshness of limited assurance.
- `inconclusive` and `aborted` do not count as conformance.
- A known-bad calibration that has never failed makes the target
`inconclusive`, even if the real target probe reports no effect.
- A changed target revision invalidates any claim that the earlier result is
current unless the executor records why the change is immaterial.
- Absence of a report is `not_run`, not `pass`.
- A passing report states only that the recorded attacks did not cross the
boundary under the recorded conditions.
## 4. Return transport
The durable coordination envelope is a State Hub inbox message to
`gate-house`:
```text
Subject: [GH-CONFORMANCE] <test-id> <outcome> <target>@<revision>
Body: report id, safe report/evidence references, specification version,
run and authorization ids, limitations, and requested Gate House action
```
Large or sensitive evidence remains under whitehat-security/risk-nexus custody;
the message carries stable references and a value-safe summary. The emitting
repository records the report in its own work structure before sending the
message.
## 5. Finding routes
| Result | Required route |
| --- | --- |
| Implementation control failed | whitehat-security → risk-nexus → implementation owner; Gate House receives the safe conformance report. |
| Doctrine is ambiguous, contradictory, unsafe, or impossible to test | whitehat-security or kings-guard → Gate House intake/message, citing test and evidence. |
| Target or execution harness gap | whitehat-security owns or promotes the gap in its repository; Gate House records the test as pending. |
| Posture, stream-silence, or live-observation gap | kings-guard owns or promotes the observation gap and returns the safe finding to Gate House. |
| Severity/disclosure question | risk-nexus only. |
One observation may take both the implementation and doctrine routes. The two
records cross-reference one another; Gate House does not absorb the owner's
remediation task.
## 6. Gate House review record
For each returned report, Gate House records:
- report, target, revision, test, and invariant identifiers;
- whitehat outcome without changing it;
- doctrine disposition: `no_change`, `clarify`, `revise`, `new_target`, or
`withdraw_claim`;
- any Gate House intake, decision, contract revision, or workplan produced;
- implementation/risk-nexus reference where a control failed;
- next review trigger: target revision, cadence date, or finding closure.
A conformance review is complete when the result has an explicit doctrine
disposition and every residual has a live owner. Closing a Gate House review
does not close the underlying security finding.
## 7. Cadence and aggregation
Cadence belongs to whitehat-security because the run interval is part of the
attacker model and exposure window. Gate House may request a minimum review
condition, but it must not label stale evidence current.
Gate House may publish an aggregate conformance view only if it preserves:
- target and revision specificity;
- last executed and next due times;
- outcome and limitations;
- tests that are pending, inapplicable, inconclusive, or not run;
- the audit completeness bound.
An aggregate must not turn missing reports into green cells.

View file

@ -0,0 +1,121 @@
# Posture and findings return-path contract
**Doctrine owner:** gate-house
**Posture publisher:** kings-guard
**Decision point:** access-engine
**Assurance executor:** whitehat-security
**Finding triage:** risk-nexus
**Status:** normative for GH-WP-0001-T06
**Version:** posture-findings-return.v1
**Date:** 2026-09-01
**Agreement basis:** kings-guard `KG-DEC-2026-001` and `KG-WP-0003`
## 1. Agreed boundary
The existing agreement is:
```text
kings-guard publishes non-deterministic posture
Gate House defines posture's authority meaning and restrictive asymmetry
access-engine deterministically renders a decision
the responsible Engine or ops-warden performs an authorized response
```
Posture may reduce authority, require stronger authorization, or request
containment. It never creates or widens authority. A containment proposal is
not an authorization and kings-guard never bypasses an Engine to act on
Tooling.
## 2. Posture envelope
A posture input offered for authority use must make these facts expressible:
```yaml
posture_id: <stable id>
subject:
kind: actor | runtime | resource | tenant | environment
id: <canonical id>
assessment: HEALTHY | DEGRADED | SUSPICIOUS | CONTAIN | BLOCKED
source: kings-guard
observed_at: <RFC 3339 UTC>
expires_at: <RFC 3339 UTC>
evidence_refs: []
stream_completeness:
status: complete | degraded | unknown
reason: <value-safe explanation>
rationale: <value-safe explanation>
```
This is a logical contract, not a kings-guard wire schema. Confidence or record
richness may accompany it, but cannot override unknown or degraded stream
completeness and cannot grant authority.
Access-engine records the posture identifier, freshness, and policy version
that interpreted it. Unknown, missing, expired, or incomplete posture follows a
declared restrictive stance; it does not fall back to extra authority.
## 3. Observation and proposal linkage
A kings-guard finding or containment proposal carries:
- originating observation identifier;
- emitted signal/posture identifier;
- affected subject and target;
- evidence-stream completeness state;
- requested restrictive direction;
- no credential, secret value, or direct actuation instruction.
The eventual access-engine decision and authorized response retain those
origin references. This makes a containment decision reconstructable without
turning the proposal into authority.
## 4. Findings return
King's Guard returns these classes to Gate House:
- an invariant does not match observed behavior;
- an implementation supplies posture in a form that cannot be interpreted
without widening or ambiguity;
- a load-bearing evidence stream is silent, divergent, stale, or incomplete;
- a required live observation lane does not exist;
- a doctrine claim is recomputable and therefore misclassified as posture;
- a conformance result exposes a missing or untestable doctrine rule.
The durable transport is a State Hub inbox message:
```text
Subject: [GH-POSTURE] <finding-class> <subject>@<revision-or-time>
Body: posture/signal/observation ids, safe evidence references, completeness,
affected invariant, current restriction, and requested doctrine action
```
## 5. Gate House disposition
Gate House classifies the returned item without regrading security severity:
| Finding shape | Gate House action |
| --- | --- |
| Doctrine correct; implementation failed | Cross-reference risk-nexus/owner record; no foreign task is created here. |
| Doctrine ambiguous or contradictory | Create or promote a Gate House intake, decision, contract revision, or workplan. |
| Posture lane or observation mechanism missing | Return to kings-guard as its owning gap; mark affected conformance claim pending. |
| Deterministic/recomputable property called posture | Move the criterion toward its owning Engine; do not create a second grader. |
| Assurance specification incomplete | Revise the Gate House target and return the new version to whitehat-security. |
Every disposition replies with the resulting Gate House record identifiers or
`no_change` plus rationale. Silence is not a disposition.
## 6. Loop closure
The loop is considered staffed, not necessarily green, when:
1. whitehat-security has an executable target and reports every outcome;
2. kings-guard can observe at least one live lane and reports stream
completeness separately from record richness;
3. access-engine can bind posture to the decision it influenced;
4. Gate House records doctrine disposition and returns identifiers;
5. implementation remediation remains with the owning repository via
risk-nexus.
`KG-WP-0003-T06` and T07 are the kings-guard work that makes origin linkage and
live observation real. Until those tasks report completion, the corresponding
lanes are declared gaps rather than assumed capabilities.