Longitudinal operational evidence assessment (experimental)
This optional, informative assessment is tracked under #98. It consumes the observation evidence profile introduced in #97. It is not authoritative ARPA status, assurance, reputation or permission, and does not amend the Candidate Specification or IETF Internet-Draft.
Processing flow
flowchart TD
A["Declared endpoint, method and bounded time window"] --> B["Schema-validate source observations"]
B --> C{"Subject and method match?"}
C -->|No| X["Reject incompatible input"]
C -->|Yes| D["Select half-open time window"]
D --> E["Detect duplicates and identifier collisions"]
E --> F{"Conflicting evidence?"}
F -->|Yes| G["conflicting_evidence"]
F -->|No| H{"Coverage established?"}
H -->|No| I["insufficient_evidence"]
H -->|Yes| J["observations_present"]
G --> K["Derived evidence summary"]
I --> K
J --> K
K -. "No implied change" .-> L["Separate current-status and authority evaluation"]
Boundaries and interpretation
Window selection uses start <= observed_at < end, with offset-aware timestamps. Duplicate observation identifiers with identical bytes are deduplicated; conflicting representations using one identifier cause rejection. Observations must concern the same subject and method; incompatible methods are not aggregated.
The evidence count and outcome counts describe observations only. observations_present never means uptime, readiness or authorization. The declared expected count is a caller-provided sampling assumption, not evidence that probes actually occurred. coverage.complete is therefore an input-based coverage claim, not an independently verified continuous-monitoring guarantee. A missing sampling schedule, insufficient samples, duplicate submissions, indeterminate outcomes or unperformed probes cannot yield complete coverage under this experimental profile.
Conflicts are detected conservatively for opposite success/failure outcomes at the same timestamp from different evidence items; different observation times may reveal operational change rather than disagreement. A different observer perspective can explain apparent differences, and no automatic observer trust ranking occurs. The evaluator does not authenticate observers, retrieve referenced evidence, validate digests, apply expiry, or incorporate later security or revocation events. A generated summary must not override those independent decisions.
Freshness and coverage qualification
The optional as_of and max_observation_age_seconds inputs bound the freshness of each observed timestamp. An observation after as_of is rejected; stale observations are retained as historical evidence but cannot satisfy the declared sampling-coverage condition. The resulting freshness metadata reports what was evaluated, not current service availability. Without explicit freshness inputs, freshness.evaluated is false and downstream reliance must not infer freshness.
coverage.basis is declared_count_only: the value coverage.complete refers only to satisfaction of a declared count under the modeled input rules. It does not establish uninterrupted monitoring or independently verified observation coverage. A production relying party must establish the sampling schedule and evidence integrity independently.
flowchart TD
A["Validated observations"] --> B{"as_of and max age supplied?"}
B -->|No| C["Freshness unassessed"]
B -->|Yes| D{"Any future or stale observations?"}
D -->|Future| E["Reject invalid as-of assessment"]
D -->|Stale| F["Historical only; count coverage insufficient"]
D -->|All within bounds| G["Freshness condition satisfied for reported samples"]
C --> H["Derived evidence, never current authority"]
F --> H
G --> H
Outcome diagram
stateDiagram-v2
[*] --> Input
Input --> Rejected: incompatible or invalid
Input --> Conflicting: conflicting evidence
Input --> Insufficient: incomplete sampling basis
Input --> ObservationsPresent: evidence present under declared sampling
Rejected --> [*]
Conflicting --> [*]
Insufficient --> [*]
ObservationsPresent --> [*]
These are evaluation results, not operational agent lifecycle transitions.
Validate
Run python3 scripts/validate_operational_assessment.py. For a report: python3 scripts/validate_operational_assessment.py --output artifacts/operational-assessment/validation-report.json.
The modeled fixture set verifies ordering, duplicate handling, missing coverage, conflicting observers, wrong methods, invalid windows and observation-identifier collisions. It does not provide independent runtime or production assurance.