Skip to content

Surface pressure window evidence in Lab reports #378

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
hyeokjun32 merged 1 commit into from
Jun 23, 2026
Merged

Surface pressure window evidence in Lab reports #378

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 8 additions & 6 deletions docs/ci/runtime_intelligence_gitlab_artifacts.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -142,9 +142,10 @@ The initial gate is conservative:
the bundle gate verifies those requirements against the bundled
`guard_analysis.evidence` type set
- if the handoff declares `optional_aiguard_evidence_types`, the bundle gate
validates the known stale-drop labels (`stale_frame_risk` and
`edgeenv_orchestrator_stale_drop_summary`) as optional context that remains
separate from the required AIGuard evidence set
validates the known optional labels (`stale_frame_risk`,
`edgeenv_orchestrator_stale_drop_summary`, and
`edgeenv_orchestrator_pressure_window_summary`) as optional context that
remains separate from the required AIGuard evidence set
- if the handoff declares `optional_aiguard_source_traceability`, the bundle
gate validates the `read_only_optional_source_traceability` AIGuard source
artifact path and reproduction command as read-only metadata, without making
Expand Down Expand Up @@ -188,6 +189,7 @@ packaging runs. That marker set preserves `Runtime Intelligence Risk Summary`,
`AIGuard scheduler fairness evidence`,
`AIGuard worker health trend evidence`,
`AIGuard policy pressure evidence`,
`AIGuard pressure window evidence`,
`AIGuard runtime operation anomalies`,
`AIGuard remote dispatch event summary`,
`AIGuard remote event summary consistency`,
Expand Down Expand Up @@ -275,9 +277,9 @@ marker enforcement to Lab's bundle/report gates.
The same alignment artifact also preserves
`optional_aiguard_evidence_types` as `read_only_optional_guard_context`. The CI
artifact gate checks `aiguard_validates_optional_evidence_as_required=false`,
records which optional stale-drop evidence labels are absent from the bundled
guard artifact, and keeps those absences out of the required evidence failure
path. A companion
records which optional stale-drop and pressure-window evidence labels are
absent from the bundled guard artifact, and keeps those absences out of the
required evidence failure path. A companion
`aiguard_edgeenv_handoff_alignment_optional_present.json/.md` fixture exercises
the same optional labels when they are present, proving the gate can record
`optional_guard_evidence_types_present` without promoting those labels to
Expand Down
11 changes: 6 additions & 5 deletions docs/portfolio/edgeenv_runtime_regression_lab_handoff.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -148,10 +148,11 @@ This second smoke uses committed lightweight artifacts to represent the cross-re
- `examples/runtime_intelligence_chain/edgeenv_lab_handoff_manifest.json` mirrors the EdgeEnv producer-side handoff manifest and its `lab_bundle_alignment` metadata, so Lab can verify EdgeEnv-produced file keys separately from external AIGuard evidence.
- The same handoff declares `external_aiguard_required_evidence_types`; Lab's bundle gate checks that those types are present in the external `guard_analysis.evidence` list without making AIGuard the final decision owner.
- If the handoff also declares `optional_aiguard_evidence_types`, Lab validates
the known optional stale-drop evidence labels (`stale_frame_risk` and
`edgeenv_orchestrator_stale_drop_summary`) as optional context only. Those
labels must remain separate from the required Guard evidence set and do not
become a Lab deployment-decision requirement.
the known optional evidence labels (`stale_frame_risk`,
`edgeenv_orchestrator_stale_drop_summary`, and
`edgeenv_orchestrator_pressure_window_summary`) as optional context only.
Those labels must remain separate from the required Guard evidence set and do
not become a Lab deployment-decision requirement.
- If the handoff also declares `optional_aiguard_source_traceability`, Lab
validates the `read_only_optional_source_traceability` source artifact path
and AIGuard reproduction command as read-only metadata. This links the
Expand Down Expand Up @@ -192,7 +193,7 @@ Expected Lab behavior:
- The same gate requires EdgeEnv-preserved Orchestrator producer markers to carry `source_repository=InferEdgeOrchestrator`, `artifact_role=orchestrator-supplemental-operation-context`, and `producer_contract=inferedge-orchestrator-edgeenv-runtime-telemetry-feed-v1`.
- When EdgeEnv preservation context is present, Lab renders `Lab EdgeEnv preservation context` with `lab_report_preservation_context_present=True`, `lab_preservation=present`, and `lab_context=present` so the Runtime Intelligence report gate and entrypoint evidence index use the same Lab-owned marker vocabulary.
- When an EdgeEnv handoff manifest is provided, the bundle gate requires EdgeEnv-produced file keys, external AIGuard file keys, source repository mapping, artifact roles, producer contracts, and boundary flags to match Lab's Runtime Intelligence bundle contract.
- The same manifest gate requires `expected_report_markers` to preserve these exact Lab-owned report markers: `Runtime Intelligence Risk Summary`, `Runtime replay duration scope`, `Orchestrator operation feed context`, `EdgeEnv fixture matrix coverage`, `Reviewer operation quick scan`, `Orchestrator task event rollup`, `Lab EdgeEnv preservation context`, `AIGuard operation risk rollup evidence`, `AIGuard task event rollup evidence`, `AIGuard operation timeline evidence`, `AIGuard scheduler fairness evidence`, `AIGuard worker health trend evidence`, `AIGuard policy pressure evidence`, `AIGuard runtime operation anomalies`, `AIGuard remote dispatch event summary`, `AIGuard remote event summary consistency`, `Remote fallback starter evidence`, `lab=Remote fallback starter evidence; evidence=remote_execution_recovered_by_fallback`, `AIGuard producer-lineage guard alignment`, and `Lab remains the final deployment decision owner.`.
- The same manifest gate requires `expected_report_markers` to preserve these exact Lab-owned report markers: `Runtime Intelligence Risk Summary`, `Runtime replay duration scope`, `Orchestrator operation feed context`, `EdgeEnv fixture matrix coverage`, `Reviewer operation quick scan`, `Orchestrator task event rollup`, `Lab EdgeEnv preservation context`, `AIGuard operation risk rollup evidence`, `AIGuard task event rollup evidence`, `AIGuard operation timeline evidence`, `AIGuard scheduler fairness evidence`, `AIGuard worker health trend evidence`, `AIGuard policy pressure evidence`, `AIGuard pressure window evidence`, `AIGuard runtime operation anomalies`, `AIGuard remote dispatch event summary`, `AIGuard remote event summary consistency`, `Remote fallback starter evidence`, `lab=Remote fallback starter evidence; evidence=remote_execution_recovered_by_fallback`, `AIGuard producer-lineage guard alignment`, and `Lab remains the final deployment decision owner.`.
- Reviewer navigation can trace Orchestrator curated samples through AIGuard evidence into Lab report rows: `agent_scheduler_delay_sample.json` -> `scheduler_delay_pattern` -> `Reviewer operation quick scan` / `Orchestrator queue/deadline/fallback markers`, and `remote_fallback_recovery_sample.json` -> `remote_execution_recovered_by_fallback` -> `Remote fallback starter evidence`. Lab does not treat those curated samples as benchmark outputs, production retry proof, or deployment policy inputs.
- The external AIGuard artifact gate also requires `runtime_queue_overload`
raw context to preserve
Expand Down
36 changes: 24 additions & 12 deletions examples/runtime_intelligence_chain/aiguard_edgeenv_handoff_alignment.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,10 +6,10 @@
"diagnosis_owner": "aiguard",
"handoff_schema_version": "edgeenv.runtime-intelligence-lab-handoff.v1",
"guard_analysis_schema_version": "inferedge-aiguard-diagnosis-v1",
"required_evidence_type_count": 10,
"optional_evidence_type_count": 2,
"guard_evidence_type_count": 11,
"lab_expected_report_marker_count": 19,
"required_evidence_type_count": 12,
"optional_evidence_type_count": 3,
"guard_evidence_type_count": 14,
"lab_expected_report_marker_count": 21,
"lab_expected_report_markers": [
"Runtime Intelligence Risk Summary",
"Runtime replay duration scope",
Expand All @@ -22,7 +22,9 @@
"AIGuard task event rollup evidence",
"AIGuard operation timeline evidence",
"AIGuard scheduler fairness evidence",
"AIGuard worker health trend evidence",
"AIGuard policy pressure evidence",
"AIGuard pressure window evidence",
"AIGuard runtime operation anomalies",
"AIGuard remote dispatch event summary",
"AIGuard remote event summary consistency",
Expand Down Expand Up @@ -51,11 +53,13 @@
"runtime_history_seed_run_config_traceability",
"runtime_queue_overload",
"runtime_thermal_instability",
"remote_execution_recovered_by_fallback"
"remote_execution_recovered_by_fallback",
"edgeenv_orchestrator_worker_health_trend"
],
"optional_aiguard_evidence_types": [
"stale_frame_risk",
"edgeenv_orchestrator_stale_drop_summary"
"edgeenv_orchestrator_stale_drop_summary",
"edgeenv_orchestrator_pressure_window_summary"
],
"guard_analysis_evidence_types": [
"runtime_telemetry_context_coverage",
Expand All @@ -65,20 +69,25 @@
"edgeenv_orchestrator_task_event_rollup",
"edgeenv_orchestrator_operation_timeline_summary",
"edgeenv_orchestrator_scheduler_fairness_summary",
"edgeenv_orchestrator_worker_health_trend",
"edgeenv_orchestrator_policy_pressure_summary",
"edgeenv_orchestrator_pressure_window_summary",
"runtime_history_seed_run_config_traceability",
"runtime_thermal_instability",
"runtime_queue_overload",
"remote_execution_recovered_by_fallback"
],
"missing_required_evidence_types": [],
"optional_guard_evidence_types_present": [],
"optional_guard_evidence_types_present": [
"edgeenv_orchestrator_pressure_window_summary"
],
"missing_optional_evidence_types": [
"edgeenv_orchestrator_stale_drop_summary",
"stale_frame_risk"
],
"supplemental_guard_evidence_types": [
"edgeenv_orchestrator_operation_risk_summary"
"edgeenv_orchestrator_operation_risk_summary",
"edgeenv_orchestrator_pressure_window_summary"
],
"invalid_required_evidence_types": [],
"invalid_optional_evidence_types": [],
Expand All @@ -99,13 +108,16 @@
"edgeenv-smoke-candidate",
"edgeenv-smoke-missing"
],
"handoff_policy_pressure_summary_run_ids": [
"guard_alignment_summary_errors": [],
"handoff_policy_pressure_summary_run_ids": [],
"guard_analysis_policy_pressure_summary_run_ids": [],
"policy_pressure_summary_errors": [],
"handoff_pressure_window_summary_run_ids": [
"edgeenv-smoke-candidate"
],
"guard_analysis_policy_pressure_summary_run_ids": [
"guard_analysis_pressure_window_summary_run_ids": [
"edgeenv-smoke-candidate"
],
"policy_pressure_summary_errors": [],
"guard_alignment_summary_errors": [],
"pressure_window_summary_errors": [],
"errors": []
}
Loading
Loading