Pinned excerpts and command output behind the chapter's slides (guide
metis-review-authoring) and its brief (Eve document
37c7785e-6f3c-488f-9daf-ca31ec1b3a25). Every excerpt is the repository at
commit b6f1ccf1c111a1b364faa114524aad0181975ed5, copied by the command shown above it. The test run is of the
working tree on 7 October 2026. No review, signoff, export or authoring request
ran against a deployed Metis service, and no migration was applied.
Counts and constants read from the code#
Produced by check-data.py in
docs/audits/presentation-center-eve-redesign-metis-review-authoring/2026-10-07/,
which reads each value from the pinned source and compares it with the words on
the slides. All eighteen comparisons were equal. The values it read:
{
"dimensions": [
"sme_correctness",
"pedagogy_alignment",
"accessibility",
"fairness_cultural",
"legal_rights",
"publisher_target"
],
"verdicts": [
"approve",
"changes_requested",
"reject"
],
"changeKinds": [
"text_edit",
"answer_key_change",
"media_replaced",
"alt_text_edit",
"source_citation_change",
"rights_metadata_change",
"locale_change",
"catalogue_metadata_change"
],
"changeAffects": {
"text_edit": [
"sme_correctness",
"pedagogy_alignment",
"accessibility",
"fairness_cultural",
"publisher_target"
],
"answer_key_change": [
"sme_correctness",
"pedagogy_alignment",
"publisher_target"
],
"media_replaced": [
"accessibility",
"fairness_cultural",
"legal_rights",
"publisher_target"
],
"alt_text_edit": [
"accessibility",
"publisher_target"
],
"source_citation_change": [
"sme_correctness",
"legal_rights",
"publisher_target"
],
"rights_metadata_change": [
"legal_rights",
"publisher_target"
],
"locale_change": [
"pedagogy_alignment",
"accessibility",
"fairness_cultural",
"legal_rights",
"publisher_target"
],
"catalogue_metadata_change": [
"publisher_target"
]
},
"invalidationReasons": [
"affected_by_change",
"expired",
"template_superseded"
],
"publisherSignsAfterEveryOther": true,
"carriedSignoffNames": [
"original_signoff_id",
"reviewer_digest",
"from_revision_sha256",
"to_revision_sha256",
"change_kinds"
],
"signoffBinds": [
"revision_sha256",
"template_version",
"evidence_sha256s"
],
"anchorKinds": [
"entity",
"field",
"text",
"region",
"timecode",
"item_option",
"rubric_cell"
],
"subjectKinds": [
"course",
"lesson",
"block",
"objective",
"rubric",
"item",
"response",
"stimulus",
"media"
],
"anchorStatuses": [
"exact",
"shifted",
"ambiguous",
"orphaned"
],
"protectionReasons": [
"learner_response_excerpt",
"live_assessment_material",
"private_source_excerpt"
],
"commentVerbs": [
"create",
"reply",
"edit",
"resolve",
"reopen",
"navigate"
],
"conflictTypes": [
"financial_interest",
"employment",
"authorship",
"close_personal",
"institutional",
"competing_product",
"prior_involvement",
"supervisory"
],
"coverageGaps": [
"no_declaration",
"stale",
"does_not_cover_subject",
"unsure_unresolved"
],
"dispositions": [
"permitted",
"manageable_with_mitigation",
"disqualifying"
],
"conflictRefusals": [
"declaration_missing",
"declaration_stale",
"declaration_does_not_cover_subject",
"unsure_unresolved",
"disqualifying_conflict",
"mitigation_not_approved",
"recused"
],
"scopeLevels": [
"platform",
"region",
"institution",
"program",
"course"
],
"directions": [
"higher",
"lower",
"superset",
"subset",
"on",
"fixed"
],
"directionKinds": {
"higher": [
"int"
],
"lower": [
"int"
],
"superset": [
"set"
],
"subset": [
"set"
],
"on": [
"bool"
],
"fixed": [
"bool",
"int",
"enum",
"set"
]
},
"overlayConflictReasons": [
"loosens_floor",
"wrong_value_kind",
"unknown_key",
"fixed_by_platform"
],
"overlayStatuses": [
"draft",
"in_review",
"scheduled",
"active",
"superseded",
"rolled_back"
],
"activationRefusals": [
"not_reviewed",
"impact_not_the_one_approved",
"wrong_status",
"effective_in_the_past",
"resolution_conflicts"
],
"rollbackIsANewVersion": true,
"pauseReasons": [
"awaiting_learner",
"awaiting_evidence",
"awaiting_third_party",
"reviewer_absent"
],
"breachStates": [
"on_track",
"warned",
"breached"
],
"deliveryOutcomes": [
"sent",
"deferred_quiet_hours",
"sent_over_quiet_hours",
"suppressed"
],
"urgentBases": [
"integrity_incident",
"safety_incident"
],
"calendarHasHolidays": true,
"zonesFromTheDatabase": true,
"gateKinds": [
"publish_high_stakes_assessment",
"release_learner_grades",
"issue_credential",
"admit_provider_model",
"uphold_integrity_finding",
"release_learner_data_export"
],
"nonHumanActors": [
"system",
"service",
"agent",
"model",
"provider",
"import",
"migration",
"scheduled_job"
],
"authenticationMethods": [
"password_mfa",
"passkey",
"sso_mfa"
],
"materialChanges": [
"subject_revised",
"evidence_replaced",
"policy_changed",
"expired"
],
"gateRefusals": [
"actor_is_not_human",
"actor_is_delegated",
"actor_is_impersonated",
"not_interactive",
"step_up_required",
"role_not_eligible",
"conflict_not_declared",
"conflict_declared",
"subject_changed",
"evidence_changed",
"gate_expired",
"already_signed",
"quorum_not_met",
"signoff_copied",
"gate_policy_changed"
],
"gateHeaders": [
"x-metis-session-id",
"x-metis-tenant-id",
"x-metis-auth-method",
"x-metis-authenticated-at",
"x-metis-batch-context"
],
"exportSections": [
"subjects_and_revisions",
"templates_and_gates",
"measurements_and_evidence",
"comments",
"assignments_and_conflicts",
"decisions_appeals_signoffs",
"actors_and_roles",
"audit",
"redactions"
],
"exportPurposes": [
"institutional_audit",
"accreditation_review",
"regulatory_request",
"learner_subject_access",
"internal_quality_review"
],
"requesterRoles": [
"institution_auditor",
"registrar",
"data_protection_officer",
"quality_lead",
"learner"
],
"fieldClasses": [
"learner_identity",
"answer_content",
"integrity_detail",
"source_text",
"reviewer_identity",
"private_note"
],
"treatments": [
"kept",
"pseudonymised",
"redacted",
"omitted"
],
"validationFindings": [
"file_digest_mismatch",
"manifest_digest_mismatch",
"signature_invalid",
"signature_absent",
"row_count_mismatch",
"undeclared_file",
"missing_file",
"section_not_total",
"redaction_not_declared",
"unsafe_file_name",
"expired"
],
"signatureStates": [
"signed",
"not_configured"
],
"authorableKinds": [
"course",
"lesson",
"item",
"rubric",
"source",
"objective",
"standard",
"review",
"release"
],
"aggregateKinds": [
"course",
"lesson",
"item",
"rubric",
"source",
"objective",
"standard",
"review",
"release",
"claim",
"citation"
],
"aggregateStates": [
"draft",
"in_review",
"approved",
"published",
"withdrawn"
],
"storeRefusals": [
"revision_conflict",
"unknown_aggregate",
"unknown_revision",
"illegal_transition",
"cross_tenant_reference",
"duplicate_idempotency_key",
"not_authorized",
"scope_mismatch"
],
"writeIsOneTransaction": true,
"outcomeStates": [
"success",
"validation_failed",
"conflict",
"stale",
"forbidden",
"not_found",
"redacted",
"duplicate_key"
],
"historyChangeKinds": [
"added",
"removed",
"changed",
"moved",
"unchanged",
"redacted",
"unknown"
],
"maxDepth": 6,
"walkUsesAQueue": true,
"previewSaysTruncated": true,
"impactStates": [
"current",
"stale",
"blocked",
"unaffected"
],
"migrationPhases": [
"expand",
"backfill",
"verify",
"cutover",
"contract"
],
"uninformativeColumns": [
"content_json",
"created_at",
"id",
"revision",
"updated_at"
],
"rolloutPhases": [
[
"0",
"Baseline"
],
[
"1",
"Decide"
],
[
"2",
"Adapt"
],
[
"3",
"Rehearse"
],
[
"4",
"Cohort"
],
[
"5",
"Cut over"
],
[
"6",
"Deprecate"
]
],
"flagsCannotBypass": "authorization, rights, privacy, accessibility, integrity, or release gates",
"reviewRolesPending": [
"domain",
"security",
"privacy",
"accessibility",
"engineering",
"product"
],
"approvalsRecorded": 0,
"inventoryRow": [
"71",
"56"
]
}
Test run on the working tree#
apps/metis/service/.venv/bin/python -m pytest over 12 test files of the
Metis service: 377 passed, 14 skipped, 15 warnings in 458.81s (0:07:38). The skipped tests are the ones that need a test
database, which this run did not have. The brief records 343 passed and 22
skipped at the pin; the working tree is later than the pin.
tests/test_anchored_comment.py
tests/test_authoring_api.py
tests/test_content_review_template.py
tests/test_human_signoff_gate.py
tests/test_institution_overlay.py
tests/test_persistence_gap_map.py
tests/test_review_evidence_export.py
tests/test_review_sla.py
tests/test_reviewer_conflict.py
tests/test_revision_history_diff.py
tests/test_source_graph.py
tests/test_tenant_isolation.py
Which signoffs a change reopens#
git show b6f1ccf1c111:apps/metis/service/src/metis/schemas/content_review_template.py, lines 683 to 716:
683 #: Total. Every kind reaches the publisher, who signs the artifact as shipped.
684 CHANGE_AFFECTS: Mapping[ChangeKind, tuple[ContentReviewDimension, ...]] = {
685 "text_edit": (
686 "sme_correctness",
687 "pedagogy_alignment",
688 "accessibility",
689 "fairness_cultural",
690 "publisher_target",
691 ),
692 "answer_key_change": ("sme_correctness", "pedagogy_alignment", "publisher_target"),
693 "media_replaced": ("accessibility", "fairness_cultural", "legal_rights", "publisher_target"),
694 "alt_text_edit": ("accessibility", "publisher_target"),
695 "source_citation_change": ("sme_correctness", "legal_rights", "publisher_target"),
696 "rights_metadata_change": ("legal_rights", "publisher_target"),
697 "locale_change": (
698 "pedagogy_alignment",
699 "accessibility",
700 "fairness_cultural",
701 "legal_rights",
702 "publisher_target",
703 ),
704 "catalogue_metadata_change": ("publisher_target",),
705 }
706
707
708 def _validate_changes() -> None:
709 if set(CHANGE_AFFECTS) != set(CHANGE_KINDS):
710 raise ValueError("CHANGE_AFFECTS is total over the change kinds")
711 for kind, affected in CHANGE_AFFECTS.items():
712 if "publisher_target" not in affected:
713 raise ValueError(f"{kind} changes what ships; the publisher signs again")
714 if len(set(affected)) != len(affected):
715 raise ValueError(f"{kind} lists a dimension twice")
716
A carried signoff names the one it came from:
git show b6f1ccf1c111:apps/metis/service/src/metis/schemas/content_review_template.py, lines 750 to 768:
750 class CarriedSignoff(ImmutableIntakeModel):
751 """A signoff a change did not touch, re-bound to the new revision with its provenance.
752
753 The original is untouched; this record names it, the revisions and the change
754 kinds, so a reader can see the approval was carried, not made again.
755 """
756
757 carried_id: Identifier
758 original_signoff_id: Identifier
759 dimension: ContentReviewDimension
760 reviewer_digest: Sha256Hex
761 content_id: Identifier
762 from_revision_sha256: Sha256Hex
763 to_revision_sha256: Sha256Hex
764 change_kinds: tuple[ChangeKind, ...] = Field(min_length=1, max_length=8)
765 template_id: Identifier
766 template_version: SemVer
767 original_signed_at: datetime
768 carried_at: datetime
A comment keeps its revision#
git show b6f1ccf1c111:apps/metis/service/src/metis/schemas/anchored_comment.py, lines 1 to 15:
1 """Closed M7.7 contracts: comments anchored to an exact revision of nine kinds of subject.
2
3 An anchor is typed, so what it points at is part of its shape: an entity, a
4 named field, a quoted run of text, a region of an image, a span of time, an
5 item option, or a rubric cell. A thread names the exact revision it was written
6 against; when the subject moves, each anchor is re-resolved against the new
7 revision as exact, shifted, ambiguous or orphaned, and the resolution CARRIES
8 the original — a comment is never silently re-pointed.
9
10 Nothing here has a field for content, an answer, a grade or an approval: a
11 comment is a thing said about a subject, and the type gives it no way to change
12 one. Anchors into a learner's response, a live assessment item or a private
13 source are refused by a total policy over subject kind x anchor kind, so the
14 protection is a property of the shape rather than a rule someone must remember.
15 """
Silence is not a clean declaration#
git show b6f1ccf1c111:apps/metis/service/src/metis/schemas/reviewer_conflict.py, lines 1 to 20:
1 """Closed M7.8 contracts: reviewer conflict-of-interest declarations and what they block.
2
3 A declaration says what kind of conflict, with what, through what relationship,
4 from when to when, and how it came to light. Silence is not a nil return: a
5 reviewer who has nothing to declare says so, in a declaration, and an assignment
6 without a current declaration is refused rather than assumed clean. A reviewer
7 who is unsure says that too, and an unsure declaration can never resolve itself
8 — only a coordinator can.
9
10 A detected signal is an indication, not a conflict. It carries the detector, its
11 confidence and its evidence, and it says in its own type that it is not a
12 finding; a person confirms or dismisses it.
13
14 The policy is a total matrix over conflict type x review dimension giving
15 permitted, manageable-with-mitigation or disqualifying. Risk and an institution
16 overlay may only tighten it. A disqualifying conflict blocks assignment, blocks
17 signoff, and does not count toward a quorum; a manageable one counts only once a
18 mitigation has been approved by somebody else.
19
20 Conflict detail is confidential: each audience sees a different shape, every
Which way is stricter#
git show b6f1ccf1c111:apps/metis/service/src/metis/schemas/institution_overlay.py, lines 74 to 87:
74 Direction = Literal["higher", "lower", "superset", "subset", "on", "fixed"]
75 DIRECTIONS: tuple[Direction, ...] = get_args(Direction)
76 ValueKind = Literal["bool", "int", "enum", "set"]
77 VALUE_KINDS: tuple[ValueKind, ...] = get_args(ValueKind)
78 #: Which value kinds each direction can be applied to. A direction that cannot
79 #: describe a kind is a mis-specification, not a runtime surprise.
80 DIRECTION_KINDS: Mapping[Direction, tuple[ValueKind, ...]] = {
81 "higher": ("int",),
82 "lower": ("int",),
83 "superset": ("set",),
84 "subset": ("set",),
85 "on": ("bool",),
86 "fixed": ("bool", "int", "enum", "set"),
87 }
The human gate#
git show b6f1ccf1c111:apps/metis/service/src/metis/schemas/human_signoff_gate.py, lines 113 to 136:
113 GateKind = Literal[
114 "publish_high_stakes_assessment",
115 "release_learner_grades",
116 "issue_credential",
117 "admit_provider_model",
118 "uphold_integrity_finding",
119 "release_learner_data_export",
120 ]
121 GATE_KINDS: tuple[GateKind, ...] = get_args(GateKind)
122 SignoffRole = Literal[
123 "assessment_lead", "registrar", "integrity_lead", "data_protection_officer", "institution_admin"
124 ]
125 SIGNOFF_ROLES: tuple[SignoffRole, ...] = get_args(SignoffRole)
126 #: Total over the gate kinds: who may sign, how many distinct people are needed,
127 #: and whether the person must have authenticated again just now.
128 GATE_REQUIREMENTS: Mapping[GateKind, tuple[tuple[SignoffRole, ...], int, bool]] = {
129 "publish_high_stakes_assessment": (("assessment_lead", "institution_admin"), 2, True),
130 "release_learner_grades": (("registrar", "institution_admin"), 1, False),
131 "issue_credential": (("registrar",), 1, True),
132 "admit_provider_model": (("institution_admin", "integrity_lead"), 2, True),
133 "uphold_integrity_finding": (("integrity_lead",), 1, False),
134 "release_learner_data_export": (("data_protection_officer", "institution_admin"), 2, True),
135 }
136 #: How long a signoff speaks for before the gate must be walked again.
The API reads the actor from five request headers:
git show b6f1ccf1c111:apps/metis/service/src/metis/api/human_signoff_gate.py, lines 90 to 94:
90 SESSION_HEADER = "x-metis-session-id"
91 TENANT_HEADER = "x-metis-tenant-id"
92 AUTH_METHOD_HEADER = "x-metis-auth-method"
93 AUTH_AT_HEADER = "x-metis-authenticated-at"
94 BATCH_HEADER = "x-metis-batch-context"
The export's digests are over the bytes written#
git show b6f1ccf1c111:apps/metis/service/src/metis/schemas/review_evidence_export.py, lines 1 to 16:
1 """Closed M7.14 contracts: review evidence exports an institution can audit.
2
3 An export is a claim made to somebody outside the system, so every part of it is
4 built to be checked by somebody who does not trust us. Three things follow.
5
6 The manifest is *versioned* and *total*: it names every section it could have
7 carried, including the ones it deliberately left out, so a reader can tell an
8 empty section from a withheld one from a section this build does not know about.
9
10 Nothing is minimised by omission alone. Every field class the scope touched is
11 listed with the treatment it received, and a redaction is a named entry in the
12 manifest rather than a hole in the data.
13
14 And the digests are over *the bytes that were written*, not over the objects
15 they came from — a checksum computed from the model rather than the file cannot
16 detect the one thing it exists to detect.
Eleven kinds, nine authorable#
git show b6f1ccf1c111:apps/metis/service/src/metis/schemas/aggregate_store.py, lines 47 to 80:
47 AuthorableKind = Literal[
48 "course",
49 "lesson",
50 "item",
51 "rubric",
52 "source",
53 "objective",
54 "standard",
55 "review",
56 "release",
57 ]
58 AUTHORABLE_KINDS: tuple[AuthorableKind, ...] = get_args(AuthorableKind)
59 #: Everything the store holds: the nine, and the two M10.4 added (migration
60 #: 071) — claims decomposed from a stored source revision and the citations
61 #: binding each to its span. The source pipeline writes those; no author does,
62 #: and the authoring API does not accept them, because a claim becomes
63 #: canonical only through a recorded human decision (M8.5).
64 AggregateKind = Literal[
65 "course",
66 "lesson",
67 "item",
68 "rubric",
69 "source",
70 "objective",
71 "standard",
72 "review",
73 "release",
74 "claim",
75 "citation",
76 ]
77 AGGREGATE_KINDS: tuple[AggregateKind, ...] = get_args(AggregateKind)
78
79 #: The operational states every aggregate moves through. Transitions are checked
80 #: against a total table, so "which states can this reach" has one answer.
One transaction:
git show b6f1ccf1c111:apps/metis/service/src/metis/services/aggregate_store.py, lines 1 to 8:
1 """M8.3: one durable store for the nine ratified aggregates.
2
3 The transaction boundary is the whole design. `write_revision` opens one
4 transaction and closes it around the revision, the aggregate row, the audit
5 entry, the outbox event and every reference — and returns an acknowledgement
6 only after the commit returns. Nothing acknowledges before it is durable,
7 because the acknowledgement is constructed after `commit()` and there is no
8 other path out of the function that produces one.
Eight outcomes#
git show b6f1ccf1c111:apps/metis/service/src/metis/schemas/authoring_api.py, lines 42 to 57:
42 OutcomeState = Literal[
43 "success",
44 "validation_failed",
45 "conflict",
46 "stale",
47 "forbidden",
48 "not_found",
49 "redacted",
50 #: The caller reused an idempotency key for a different intent. Not a
51 #: validation failure — the request is well formed — and not a revision
52 #: race, because there is no base revision it read and lost.
53 "duplicate_key",
54 ]
55 OUTCOME_STATES: tuple[OutcomeState, ...] = get_args(OutcomeState)
56 #: Outcomes that carry the thing the caller asked about.
57 CARRYING_STATES: tuple[OutcomeState, ...] = ("success", "redacted")
Seven kinds of difference#
git show b6f1ccf1c111:apps/metis/service/src/metis/schemas/revision_history.py, lines 49 to 52:
49 ChangeKind = Literal["added", "removed", "changed", "moved", "unchanged", "redacted", "unknown"]
50 CHANGE_KINDS: tuple[ChangeKind, ...] = get_args(ChangeKind)
51 #: Kinds that mean the two sides hold something different.
52 DIFFERING_KINDS: tuple[ChangeKind, ...] = ("added", "removed", "changed", "moved")
The impact walk#
git show b6f1ccf1c111:apps/metis/service/src/metis/services/source_graph.py, lines 1 to 19:
1 """M8.5: what a source supports, and what breaks when it stops supporting it.
2
3 The traversal is breadth-first over `store_references`, which M8.3 keyed by
4 exact revision. Three things make the result a preview somebody can act on
5 rather than a pile of ids.
6
7 It is deterministic. Neighbours are visited in a fixed order, so the shortest
8 path reported for a node is the same shortest path on every run, and a cursor
9 into the result means the same thing on the next call.
10
11 It stops honestly. A walk that hits the depth limit reports the depth it stopped
12 at, because a truncated answer that does not say it is truncated is a wrong
13 answer rather than a partial one.
14
15 And a retraction writes new state beside the history rather than into it. What a
16 published lesson said about a source it cited is what it said; the retraction
17 marks the lesson stale or blocked and queues it for its owner.
18 """
19
git show b6f1ccf1c111:apps/metis/service/src/metis/services/source_graph.py, lines 57 to 57:
57 MAX_DEPTH = 6
Migration phases#
git show b6f1ccf1c111:apps/metis/service/src/metis/schemas/persistence_migration.py, lines 30 to 40:
30 MigrationPhase = Literal["expand", "backfill", "verify", "cutover", "contract"]
31 MIGRATION_PHASES: tuple[MigrationPhase, ...] = get_args(MigrationPhase)
32 #: The order is the safety property, so it is written down once and asserted.
33 PHASE_ORDER: tuple[MigrationPhase, ...] = ("expand", "backfill", "verify", "cutover", "contract")
34
35 #: Phases that only add. These are safe to run while the old code is serving.
36 ADDITIVE_PHASES: tuple[MigrationPhase, ...] = ("expand", "backfill", "verify")
37 #: Phases that change something, and therefore owe an account of undoing it.
38 #: `verify` is deliberately absent: it reads, so there is nothing to undo, and a
39 #: rollback written for it would be a statement nobody could ever need to run.
40 WRITING_PHASES: tuple[MigrationPhase, ...] = ("expand", "backfill", "cutover", "contract")
The rollout sequence and the design record's status#
git show b6f1ccf1c111:V1/METIS_COURSEWARE_WORKBENCH_DESIGN_2026-07-23.md, lines 343 to 359:
343 ### 5.1 Migration and rollout sequence
344
345 | Phase | Forward action | Gate | Rollback |
346 | ------------ | ------------------------------------------------------------------------------------------------------ | ----------------------------------------------------------------------------------------- | ---------------------------------------------------------------------- |
347 | 0. Baseline | Freeze inventories, source hashes, owners, routes, contracts, and launch gaps. | All deterministic evidence checks pass. | Regenerate from the last trusted commit. |
348 | 1. Decide | Ratify hosts, roles, data ownership, persistence, standards, and experience boundaries. | Named approvals bind one document revision. | Keep current authorities and deny unexpressible duties. |
349 | 2. Adapt | Introduce generated clients, owner-qualified IDs, projections, route facades, and versioned envelopes. | Contract, authorization, tenancy, replay, and negative-leakage tests pass. | Disable adapters; current owner remains authoritative. |
350 | 3. Rehearse | Backfill and shadow-read by tenant without dual-authoritative writes. | Counts, hashes, relationships, semantics, accessibility, and restore drills agree. | Discard projections and restart from recorded cursor. |
351 | 4. Cohort | Enable capability and audience flags for internal then named tenant cohorts. | Quality, safety, accessibility, privacy, interop, SLO, and support gates pass. | Disable cohort flag and retain prior trusted route/release. |
352 | 5. Cut over | Switch canonical routes/stores together; monitor receipts and reconciliation. | No orphan, version regression, leakage, stale approval, or unexplained metric regression. | Restore prior authority and replay durable outbox from cutover cursor. |
353 | 6. Deprecate | Time-box redirects, remove facades/shadow paths, and archive decision evidence. | Compatibility window ends and consumers acknowledge migration. | Extend only through a reviewed ADR with owner and expiry. |
354
355 Feature flags are tenant-, audience-, capability-, and profile-scoped; default
356 is off. Flags cannot bypass authorization, rights, privacy, accessibility,
357 integrity, or release gates. Deprecation records require owner, replacement,
358 affected consumers, notice date, migration date, removal date, telemetry, and
359 rollback window.
git show b6f1ccf1c111:V1/METIS_COURSEWARE_WORKBENCH_DESIGN_2026-07-23.md, lines 471 to 485:
471 ## 6. Review and ratification (M0.13.f)
472
473 **Decision:** pending named ratification. This generated document cannot approve
474 itself.
475
476 | Required review role | Named reviewer | Decision | Reviewed version/date |
477 | -------------------- | -------------- | -------- | --------------------- |
478 | domain | — | pending | — |
479 | security | — | pending | — |
480 | privacy | — | pending | — |
481 | accessibility | — | pending | — |
482 | engineering | — | pending | — |
483 | product | — | pending | — |
484
485 Approvals recorded: 0. Review comments recorded: 0. Effective date: none.
What was not kept#
Nothing ran against a deployed service, so there is no log of one. The tests that need a real database were skipped. The gateway that would set and protect the signoff gate's identity headers was not examined. The draft render's measurements and the axe results are in the record directory named above, not in this document.