STD-08 evidence pack
What to show for STD-08 delegation at three tiers: grants and renewals, policy records, intervention specifications, and correction capacity, with freshness rules.
Tier 0 (minimum): Proof the objects exist and are populated. Evidence: grant register export with state and state_history, policy register export with review_triggers and expires_at, one intervention specification per confirm-mode delegation.
Tier 1 (operational): Proof the states move. Evidence: renewal records naming what was examined, expansion transitions with their own evidence, fired policy triggers with the status changes they produced, measured approval rates and approval times.
Tier 2 (resilient): Proof the terms hold under pressure. Evidence: an expansion refused or narrowed for want of correction capacity, a reconsideration opened by near-unanimous approval, a grant moved to review_required by a policy expiry with no incident to prompt it.
Grant, policy, and intervention records are expected to be emitted as STD-07 records at Level 2 or better, so the pack is an export rather than a second bookkeeping system.
Grant register export: Every live grant with grant_id, scope, state, until, renewal_basis, policy_refs, intervention_ref, and the full state_history including renewal and expansion transitions.
Policy register export: Every policy a live grant cites, with policy_id, version, provenance, assumptions, review_triggers, expires_at, and status, plus the grants that depend on each.
Intervention roster: One row per intervention specification: spec_id, owner, mechanism, actions_preventable, states_alterable, on_disagreement, cost_to_exercise, reach_time_target, and the delegations that cite it.
Expansion decisions log: Every widening in the period, with the evidence produced for it, the correction-capacity evidence produced alongside it, the deciding authority, and the outcome including refusals and narrowings.
Capability catalog snapshot: What the assembled system could do at the date of the snapshot, so that grant scope can be read against capability rather than against intent.
- Grant register export: no older than the shortest renewal interval in it, and never older than 90 days.
- Policy register export: refreshed on any status change, and in full at each safety-case refresh.
- Intervention roster: refreshed whenever a specification changes and at each measurement period boundary.
- Approval-rate and approval-time measurements: per review period, with the period length stated.
- Expansion decisions log: appended at each decision; an empty period is recorded as empty, not omitted.
- Capability catalog snapshot: no older than the last material change to tools, models, integrations, or data access.
- Any artifact past its freshness rule is presented as stale. A stale artifact is not evidence of the current state.
Part A, authority as lease. Renewal records naming what was examined and who looked (§1.1, §1.2). State_history entries with reason expansion and their evidence references (§1.3, §1.4).
Part B, policy validity. Policy records with all five required fields (§2.1). Grants with version-pinned policy_refs (§2.2). Trigger-fire log with the resulting status change and the elapsed time against the trigger clock (§2.3). Dependent-grant transitions and the reconsiderations opened (§2.4). Expiry report showing no live grant resting on an expired policy (§2.5).
Part C, intervention. One specification per oversight claim, with all eight fields (§3.1). Confirm-mode grants matched to intervention_ref, and the list of confirm-mode grants with none (§3.2). Specifications with empty states_alterable, and how each is labeled (§3.3). Approval rate, median and tail time per approval, share reaching the owner inside reach_time_target (§3.4). Reconsiderations opened on near-unanimous approval (§3.5).
Part D, correction symmetry. Correction-capacity statements across the seven components with their evidence (§4.1). Expansion decisions paired with capacity evidence, including refusals (§4.2). Renewal-time capacity re-examinations against actual volume (§4.3). Institutional reversibility and correction_latency drawn from the dependency record (§4.4).
- The grant register is generated from the permission system, so it reports configuration rather than authorization decisions and every grant looks current.
- Renewal is a calendar event with a checkbox, and renewal_basis is copied forward unchanged from the original grant.
- Expansion is recorded as a renewal, so state_history shows a stable grant that has quietly quadrupled in scope.
- Policy records exist, but review_triggers name conditions nobody measures, so no trigger has ever fired.
- Intervention specifications are written for the pilot roster and never updated when volume rises by an order of magnitude.
- Approval rate is reported as a reliability metric, with 99% approval offered as evidence the system is safe.
- Correction capacity is a list of tools the institution owns rather than a route anyone has traveled end to end.
- Show the last three renewals and, for each, what was examined and what would have been visible had the grant been failing.
- Show every expansion in the period and the correction-capacity evidence produced at the same decision.
- Pick a live grant: name its policies, their versions, and their expiry dates, then show the grant's state.
- Show a fired policy trigger and the time between the fire and the status change on every dependent grant.
- Pick a control described as human review: show its specification and the states its owner can alter.
- Show approval rates and time-per-approval for the last two periods, and what happened where approval was near-unanimous.
- Show one expansion that was refused or narrowed, and why.
Referenced by
The standards, bindings, and validators this pack supports.
Standard: STD-08 Delegation
Record format: STD-07 Revisable delegation record
Crosswalk controls: CTRL-01, CTRL-05, CTRL-06, CTRL-07
Companion pack: STD-06 evidence pack (dependency ledger and withdrawal rehearsal)