@alphazede/bearing-lite 0.1.11 → 0.2.1

package/skills/map-the-route/references/artifact-grammar.md

@@ -3,18 +3,26 @@
 Apply this closed grammar while authoring or validating planning artifacts.
 Bearing rejects deviations with typed findings.
 
+The canonical Journey planning artifacts are exactly
+`<journey-topic>-technical-plan.md` with `type: technical-plan`, `design.md`,
+`seit.json`, `implementation.json`, and `review.html`. Markdown carries
+technical-plan and design reasoning; the JSON files are schema-validated
+machine authorities; `review.html` is their complete integrated human
+projection. There is no sixth canonical planning artifact. XLSX may be an
+optional derived export and is never authority.
+
 ## Shared artifact rules
 
-1. Frontmatter declares `type` as `plan-spec`, `design`, `seit`, or
-   `implementation`, and `status` as `complete` or `amended`.
+1. Frontmatter declares `type` as `technical-plan` or `design`, and `status` as
+   `complete` or `amended`.
 2. Journey-authored requirement IDs use `AC-*` or `RISK-*`; references to an
    existing requirements register keep that register's identities verbatim.
    Design IDs use `DES-*` or `CONTRACT-*`; SEIT rows use `SEIT-*`; commands
    use `CMD-*` or `PROC-*`. Suffixes contain only uppercase letters, digits,
    dots, or hyphens.
-3. `<journey-topic>-spec.md` declares requirements, Entry criteria, Exit criteria,
-   Rollback or repair, and Accountable controller. `design.md` declares design
-   IDs. `seit.md` declares SEIT rows and commands.
+3. `<journey-topic>-technical-plan.md` declares requirements, Entry criteria,
+   Exit criteria, Rollback or repair, and Accountable controller. `design.md`
+   declares design IDs. `seit.json` declares SEIT rows and commands.
 4. Preserve stable IDs. Every implementation reference resolves to a declared
    requirement, design contract, SEIT row, and command.
 5. Owner-supplied lineup identities and proposed `review_cadence: at-end` are
@@ -39,7 +47,7 @@ Bearing rejects deviations with typed findings.
    to registered identities (no restated text), derivations from them
    (`AC-X derives from REG-ID`), or Journey-local `AC-*`/`RISK-*` criteria about
    write sets, seams, gates, or concurrency. Do not restate registered content.
-4. `seit.md` references verification allocations where the register provides
+4. `seit.json` references verification allocations where the register provides
    them. Where the register provides none for a referenced requirement, author
    Journey-level proof for it or return `NEEDS_OWNER_DECISION` when register
    authority is unclear; do not silently drop coverage. Journey-level proof
@@ -48,6 +56,13 @@ Bearing rejects deviations with typed findings.
    Journey-local, so a reviewer can tell which artifact owns each statement.
 6. `design.md` is unaffected: it records how the work is built, which no
    requirements register covers.
+7. A specification-authoring Journey carries its requirement register (UID,
+   statement, rationale, verification method, allocation) as a planning
+   artifact: a draft `.sdoc` path (Markdown sections are not lint-checkable),
+   recorded as `implementation.json` `journey_settings.journey_type:
+   specification` plus `journey_settings.requirement_register`. The freeze
+   fails without an existing register; the Requirements Engineer gates it
+   before the integrated owner review; no Expedition wave re-gates it.
 
 ## Published standards
 
@@ -59,38 +74,55 @@ Bearing rejects deviations with typed findings.
 
 ## SEIT rules
 
-1. Include non-empty `Required Commands`, `Traceability Matrix`, and
-   `Cross-cutting Checks` sections.
-2. Declare each command as `- **<CMD/PROC-id>** — description`.
-3. Use one flat traceability table with exactly these columns: SEIT row ID |
+1. `seit.json` is the canonical JSON Schema-validated tailored V&V plan.
+   Always-on sections: scope/baseline; responsibility/change authority;
+   applicable documents/precedence; requirements flowdown/architecture context;
+   V&V methods; verification and validation matrices; levels/integration
+   sequence; environments/fixtures/data/simulations/support;
+   procedures/commands; evidence/pass-fail; anomaly/corrective/closure.
+   System fields only where applicable.
+2. Include non-empty `Required Commands`, `Traceability Matrix`, and
+   `Cross-cutting Checks` sections or JSON equivalents.
+3. Declare each command with a `CMD-*` or `PROC-*` id and description.
+4. Use one flat traceability table with exactly these columns: SEIT row ID |
    Acceptance/risk ID | Design/contract ID | Boundary/test layer | Positive
    case | Negative/failure case | Command/procedure ID | Evidence.
-4. Every row carries exactly one SEIT row ID, requirement ID, design ID, and
+5. Every row carries exactly one SEIT row ID, requirement ID, design ID, and
    command ID, and names an observable failure.
-5. If procedure narratives are used, title each `### SEIT-<id> <title>` under
-   a procedures section and restate Command, Positive case, Negative case, and
-   Evidence exactly. Every PROC row then has exactly one matching narrative.
+6. For a specification wave, each case-authoring proof row cites the planning
+   register gate receipt.
+7. Bind a stable decision-baseline projection of confirmed decision identities
+   and open-item statuses rather than the whole-file `journey.json` digest.
 
 ## Implementation rules
 
-1. Slice headings are `### Slice <id>` and manifests are
-   `### <id> execution manifest`. IDs are a letter-run plus an integer or
-   dotted integer. Every slice has exactly one matching manifest.
-2. Every slice declares Goal, Requirement IDs, Design IDs, SEIT proof rows,
-   Type, Design lenses, named implementation role, owner-selected model route,
-   reasoning level, and review path. Goals are at most 512 characters.
-3. Every manifest declares Write set, Command IDs, Stop condition, and Human
-   decision. Optional fields are Shared interfaces (`path#Symbol`), Integration
-   boundary, Published standard (`doc#clause`) when applicable, and Parallel
-   safe (`yes` or `no` plus reason).
-4. Write sets use one line: `Write only `path``. Paths are bounded, normalized,
+1. Regular JSON is the nested execution authority for Journey settings, lineup
+   snapshots, waves, slices, dependencies, traceability, and manifests.
+2. Every slice declares one named role plus goal, type, requirement IDs, design
+   IDs, SEIT proof rows, exact design-lens names sourced from `design.md`,
+   owner-selected model route, reasoning, review path, write set, command IDs,
+   stop condition, human decision, and `authority_id`. Goals are at most 512
+   characters. Slice actions and write sets are subsets of current authority.
+3. `work_class` is `light` or `judgement` (default). A `light` slice names
+   the Light Implementer role, at least one command id, and a
+   `work_class_reason`; see `skills/light-implementer` for the criteria.
+4. For `journey_type: specification`, a specification-authoring Expedition
+   wave defaults to scaffold → author from the gated register → Test Engineer
+   verification cases → bind to the host → readback of the bound revision and
+   digest from the host → Park Ranger with both named in the review
+   request → owner decision. A deviation is a planning-review finding.
+5. Optional fields are Shared interfaces (`path#Symbol`), Integration
+   boundary, Published standard (`doc#clause`) when applicable, SysML and
+   integration fields when selected, and Parallel safe (`yes` or `no` plus
+   reason).
+6. Write sets use one line: `Write only `path``. Paths are bounded, normalized,
    repository-relative literals. Put prohibitions in prose, not the write set.
-5. Multi-slice plans declare consecutive `Wave <n>: <ids>` lines. Every slice
+7. Multi-slice plans declare consecutive `Wave <n>: <ids>` lines. Every slice
    belongs to one wave. Dependencies use acyclic `S1 --> S2` arrows.
-6. Optional phase graphs use Phase | Slices | Depends on phases | Integration
-   checkpoints. Optional Inputs and Produces name metric denominator, ledger
-   key, or contract field values in backticks.
-7. Plans may contain at most 128 slices, manifests, write paths, and commands.
+8. Ordered integration steps, resources, ownership, and rollback live here.
+   Owner-configured reviewer count `n`, repair bound `k`, and confirmation
+   count `c` are explicit fields with no assistant default integers.
+9. Plans may contain at most 128 slices, manifests, write paths, and commands.
    Aim for at most 500 estimated tokens per slice plus manifest; split larger
    packets when practical.
 
@@ -113,12 +145,15 @@ placeholders, bare negations, or deferral language.
 1. Select review-oriented feature diagrams (flow diagrams, state-machine diagrams, or process/sequence diagrams) when they materially clarify architecture or lifecycle behavior for owner review.
 2. Major features select a justified subset or full set of diagrams based on architectural complexity. Trivial features or minor fixes do not require unnecessary diagrams.
 3. Every feature diagram must include canonical, reviewable source (such as Mermaid code blocks), render visually inside generated `review.html`, and retain nearby authoritative explanatory text.
+4. Non-trivial technical plans include words plus context/boundary, use-case/outcome, and operational-flow views. In MBSE mode record model revision/digest; never silently substitute Mermaid for mandated SysML. Trivial work may record `diagram_not_required` with a concrete rationale.
 
 ## Completion boundary
 
-Author specification, design, SEIT, implementation, and review HTML in that
+Author technical-plan, design, SEIT, implementation, and review HTML in that
 dependency order with internal prospective checks; generate implementation and
-HTML together after their stable inputs. The single owner review gate requires
+HTML together after their stable inputs. `review.html` has two states:
+`planning-review` before implementation and `final-closeout` after
+implementation. Each records its stage and exact input digests. The single owner review gate requires
 the complete five-artifact package. Do not insert a lineup, route, or
 specification-only owner gate unless the owner explicitly requests staged
 approvals. The HTML becomes authoritative only after integrated owner approval.

package/skills/park-ranger/SKILL.md

@@ -15,16 +15,20 @@ Independent defect assurance, outside the mutation-authority ladder.
 ## Inputs and match
 
 - **Inputs:** approved baseline, exact candidate ref and diff, author identity,
-  relevant evidence, at-end boundary, review focus, and compact return schema.
-- **Match:** Park Ranger is declared and the final integrated candidate is
-  stable at-end.
+  relevant evidence, declared phase or wave end boundary, review focus, and
+  compact return schema.
+- **Match:** Park Ranger is declared and the declared phase or wave's
+  integrated candidate is stable at that end.
 - **Non-match:** slice or round boundary, candidate is unstable/unchanged, or
-  Validator/Surveyor work is requested.
+  Surveyor/Test Engineer work is requested.
 
 ## Algorithm
 
 1. Start a fresh session; reject author identity, author ancestry, candidate
-   discontinuity, or any boundary other than at-end.
+   discontinuity, or any boundary other than the declared phase or wave end.
+   Consume the Assurance
+   Test Engineer receipt. Do not routinely invoke Test Engineering; use it
+   only to adjudicate a specific suspected test defect.
 2. Review only introduced correctness, security, performance, and meaningful
    maintainability defects plus applicable plan drift. When the candidate
    implements a published standard, compare the change against the cited text
@@ -40,9 +44,10 @@ verdict, candidate_ref, changed_paths, tests, findings, and blocker.
 `ACCEPT`, `ACCEPT_WITH_FINDINGS`, and `BLOCK` are terminal. `REPAIR_REQUIRED`
 permits bounded correction. `ACCEPT_WITH_FINDINGS` accepts residual findings;
 do not follow it with another repair. Coordinators enforce
-`max_assurance_rounds` of 1. A repairable verdict permits one repair; the
-coordinator then runs deterministic verification and closes the gate. Do not
-review or repair that Journey again. A failed repair or scope change
-returns to Owner Authority.
+`max_assurance_rounds` of 1 per declared phase or wave. A repairable verdict
+permits one repair; the coordinator then runs deterministic verification and
+closes the gate without another review. Do not review or repair that declared
+unit again. The next distinct declared phase or wave carries its own budget. A
+failed repair or scope change returns to Owner Authority.
 
 Never edit, self-review, duplicate general review, or grant publication rights.

package/skills/plan-integrator/SKILL.md

@@ -0,0 +1,46 @@
+---
+name: plan-integrator
+description: >
+  Reconcile specialist outputs and generate implementation.json and
+  review.html. Use for Plan Integrator. Do not invent owner intent,
+  requirements, models, or V&V, or author V&V.
+---
+
+# Plan Integrator
+
+Consumes stable specialist outputs, invokes Map the Route, cross-validates
+the five artifacts, and generates `implementation.json` and two-state
+`review.html`.
+
+## Inputs and match
+
+- **Inputs:** settled decisions, technical-plan, design.md, seit.json
+  drafts, lineup freeze, authority envelope.
+- **Match:** specialist outputs are stable enough to reconcile.
+- **Non-match:** unresolved owner intent; V&V authorship.
+
+## Algorithm
+
+1. Cross-validate the five canonical artifacts. Copy design-lens names
+   from design.md; never invent lens IDs.
+2. For `journey_type: specification`, default an Expedition wave to scaffold →
+   author from the gated register → Test Engineer verification cases → bind to the host
+   → readback of the bound revision and digest from the host → Park
+   Ranger with both named in the review request → owner decision. A deviation
+   is a planning-review finding.
+3. Generate `implementation.json` and `review.html` (`planning-review`,
+   then `final-closeout`) when inputs are stable.
+4. Classify every slice `work_class: light` or `judgement` with a
+   `work_class_reason`, using the five criteria in `skills/light-implementer`.
+   Light slices carry the Light Implementer role and at least one
+   `command_id`; the freeze rejects any other light slice.
+5. Request a Planning Test Engineer delta after relevant decision,
+   requirement, or design changes. Do not author V&V.
+
+## Return and recovery
+
+Return `PLAN_REVIEW_READY` or `NEEDS_OWNER_DECISION` with verdict,
+candidate_ref, changed_paths, tests, findings, and blocker.
+
+Never invent owner intent, requirements, models, integration strategy,
+or schemas.

package/skills/requirements-engineer/SKILL.md

@@ -0,0 +1,57 @@
+---
+name: requirements-engineer
+description: >
+  One Requirements Engineer role, planning stage only. Use for the
+  requirement register quality gate after Gather Supplies and before the
+  Systems Modeler finalizes mappings. Do not use inside an Expedition wave,
+  or for V&V, SysML modeling, implementation, defect review, or publication.
+---
+
+# Requirements Engineer
+
+One role, one planning session. It owns requirement quality: every statement
+is precise, measurable, traceable, allocated, and verification-ready.
+Requirements are fixed at plan approval; an Expedition wave stores, binds,
+adds verification cases, and publishes approved statements and never re-gates
+them (the Assurance Test Engineer re-verifies at wave end). It never persists
+SDoc, publishes, selects models or lineups, or writes tests.
+
+## Inputs and match
+
+- **Inputs:** settled owner decisions, the requirement register as a planning
+  artifact (UID, statement, rationale, verification method, allocation: a
+  draft `.sdoc`, since Markdown rows are invisible to the lint), the plan's `AC-*`
+  and `RISK-*` rows, published-standard citations, the
+  `requirements-engineering` method skill, `lint-sdoc.py --profile library`
+  output, compact return schema.
+- **Match:** a planning package whose requirement register needs a
+  quality-gate verdict before the integrated owner review.
+- **Non-match:** any Expedition wave, design, SEIT, implementation, Park
+  Ranger, Surveyor.
+
+## Algorithm
+
+1. Run after Gather Supplies and before the Systems Modeler finalizes
+   mappings, on the requirement statements themselves: gate every register
+   row's statement, rationale, verification method, and allocation, and the
+   `AC-*`/`RISK-*` rows that cite them, against the NASA-adapted checklist;
+   reject escape clauses and undefined terms; every row cites its register
+   identity or is marked Journey-local.
+2. Cite the mechanical output (`lint-sdoc.py --profile library` over the
+   register: EARS, banned terms, glossary references) and judge only what the
+   tool cannot decide. Missing tool output is `NEEDS_MORE_EVIDENCE`.
+3. When a published standard is cited, verify the document and clause.
+4. Return the smallest set of failing rows. Never rewrite silently: propose
+   the corrected statement and let the author apply it.
+
+## Return and recovery
+
+Return `PASS`, `REPAIRABLE_FAILURE`, `NEEDS_MORE_EVIDENCE`, or
+`NEEDS_OWNER_DECISION` with verdict, candidate_ref, changed_paths, per-row
+findings, and blocker. Rerun the gate on each corrected register within Map the
+Route's correction rounds. Exhaustion with fixable rows returns
+`NEEDS_OWNER_DECISION` listing those rows, never silent acceptance. Missing
+`requirements-engineering` method skill is a typed capability gap.
+
+Never implement, model, self-certify, persist or publish SDoc, or grant
+owner-only approval.

package/skills/scribe/SKILL.md

@@ -0,0 +1,34 @@
+---
+name: scribe
+description: >
+  Transcribe Journey decisions, authority events, ledger snapshots, and
+  lineup selections as an event side lane. Use for Scribe. Do not use to
+  activate authority, invent choices, or save lineups unless the owner
+  asks.
+---
+
+# Scribe
+
+Event side lane. Transcribes; cannot activate authority.
+
+## Inputs and match
+
+- **Inputs:** owner decisions, authority events, Gather ledger snapshots,
+  lineup selections, configuration digest, `journey.json`.
+- **Match:** a visible Journey event must be appended to history.
+- **Non-match:** planning dispatch, specialist engineering, execution.
+
+## Algorithm
+
+1. Append owner decisions, authority events, ledger snapshots, lineup
+   selections, and configuration digest to `journey.json` history.
+2. Do not activate authority or invent unresolved choices.
+3. Do not write Journey selections into `lineups.json` unless the owner
+   asks to save a reusable profile.
+
+## Return and recovery
+
+Return `READY` or `OWNER_DECISION_REQUIRED` with verdict,
+candidate_ref, changed_paths, tests, findings, and blocker.
+
+Never implement, self-assure, or select models.

package/skills/set-bearings/SKILL.md

@@ -15,19 +15,28 @@ Fresh planning node. The Router announces `Setting Our Bearings in <repo>.`
 
 - **Inputs:** confirmed repository root, owner-confirmed plan directory, Journey
   title, visible existing artifacts, repository rules, and return schema.
-- **Match:** the workspace or current-state repository map is missing or stale.
+- **Match:** the workspace or repository map is missing or stale.
 - **Non-match:** both are current and usable by Gather Supplies or Map the Route.
 
 ## Algorithm
 
 1. Re-read the exact confirmed root and plan directory; never derive a different
    root, slug, suffix, or sibling workspace.
-2. Preserve every existing artifact and unrelated edit. Resume rather than
-   replace an existing Journey.
-3. Create only the missing plan-directory stub and bounded repository map.
-4. Record observed systems, relevant paths, constraints, Git boundaries,
-   validation commands, and unknowns as evidence—not invented decisions.
-5. Verify that all written paths remain inside authority and are human-readable.
+2. If `workspace.md` exists in the plan directory, verify recorded root and plan
+   directory match confirmed inputs; a mismatch returns `BLOCKED`. Run `git
+   status --porcelain` scoped to mapped inputs; if unmodified, return
+   `WORKSPACE_RESUMED` (1 read, 1 status).
+3. Bound discovery: depth 2, max 40 paths, max 64 KiB, read-only. Anchors are
+   root manifest, task runner, CI entrypoint, top-level instructions, and root
+   test configs. Strictly prohibit discovery traversal into repo-relative
+   `src/`, `lib/`, `vendor/`, and `docs/`.
+4. If bounds are exhausted before required anchors are found, return
+   `NEEDS_EVIDENCE` naming the missing anchor. Never compose unobserved commands.
+5. Create missing plan directory or update `workspace.md` in place, preserving
+   existing recorded evidence and unrelated edits. Initialize a missing file
+   from `templates/workspace.md`. Quote validation commands with source and mark
+   `[observed, not run]`. Drop detail before Constraints and Unknowns.
+6. Verify written paths remain inside authority and return `WORKSPACE_READY`.
 
 ## Return and recovery
 

package/skills/set-bearings/templates/workspace.md

@@ -0,0 +1,28 @@
+# Workspace Environment: <journey-topic>
+
+## Repository Identity
+- Root: `<repository-root>`
+- Plan directory: `<plan-directory>`
+
+## Mapped Inputs
+- `<relative-path>`: `<what it establishes>`
+
+## Observed Systems
+- `<relative-directory>`: `<subsystem>`
+
+## Constraints and Git Boundaries
+- Branch / Worktree: `<branch>` (clean | dirty)
+- Rules: `<applicable-rules>`
+
+## Validation Commands [observed, not run]
+- Test: `<command>` (source: `<anchor-file>`) [observed, not run]
+- Lint: `<command>` (source: `<anchor-file>`) [observed, not run]
+- Build: `<command>` (source: `<anchor-file>`) [observed, not run]
+
+## Unknowns
+- `<unknown or none observed>`
+
+## Map Freshness
+- Tier 1: Recorded root and plan directory match confirmed inputs.
+- Tier 2: `git status --porcelain` scoped to mapped inputs shows no modifications.
+- Note: Does not detect committed changes postdating this observation.

package/skills/surveyor/SKILL.md

@@ -14,18 +14,22 @@ Independent outcome assurance, outside the mutation-authority ladder.
 ## Inputs and match
 
 - **Inputs:** approved specification/review baseline, exact integrated candidate,
-  author identities, user-facing evidence, at-end boundary, prior required
-  assurance, and compact return schema.
-- **Match:** Surveyor is declared and the final integrated candidate is ready
-  for user-facing comparison at-end.
+  author identities, user-facing evidence, declared phase or wave end boundary,
+  prior required assurance, and compact return schema.
+- **Match:** Surveyor is declared and the declared phase or wave's integrated
+  candidate is ready for user-facing comparison at that end.
 - **Non-match:** unfinished packet, slice or round boundary, unstable candidate,
-  Validator sufficiency, Park Ranger defect review, or owner-only release
-  decision.
+  Assurance Test Engineer sufficiency, Park Ranger defect review, or owner-only
+  release decision.
 
 ## Algorithm
 
 1. Start a fresh session; reject author ancestry; verify candidate continuity,
-   independence, completed prerequisite assurance, and the at-end boundary.
+   independence, completed prerequisite assurance, and the declared phase or
+   wave end boundary.
+   Consume Requirements Engineering, SysML Modeling, and Test Engineering
+   read-only. Do not rewrite requirements, repair models, author tests, or
+   repeat assurance.
 2. Exercise or inspect every observable approved outcome, including failure and
    recovery behavior relevant to the Journey.
 3. Map each gap to an exact requirement and evidence location. Separate observed
@@ -36,7 +40,9 @@ Independent outcome assurance, outside the mutation-authority ladder.
 
 Return `ACCEPT`, `GAPS`, or `OWNER_DECISION_REQUIRED` with verdict,
 candidate_ref, changed_paths, tests, findings, and blocker.
-A gap may receive one repair followed by coordinator verification; do not
-reassess the same Journey.
+Coordinators enforce `max_assurance_rounds` of 1 per declared phase or wave. A
+gap may receive one repair followed by deterministic coordinator verification
+without another review; do not reassess that declared unit.
+The next distinct declared phase or wave carries its own budget.
 
 Never implement, substitute for another assurance role, or approve publication.

package/skills/systems-modeler/SKILL.md

@@ -0,0 +1,35 @@
+---
+name: systems-modeler
+description: >
+  Select modeling mode and attach view metadata after Requirements
+  Engineer work and before design finalization. Use for Systems Modeler.
+  Do not finalize requirement mappings against unstable requirements.
+---
+
+# Systems Modeler
+
+Runs after the Requirements Engineer and before design finalization.
+
+## Inputs and match
+
+- **Inputs:** requirements, view need, modeling-mode metadata.
+- **Match:** engineering views or SysML mode selection is required.
+- **Non-match:** requirement authorship, V&V, implementation.
+
+## Algorithm
+
+1. Select `sysml-v2`, `diagram-assisted`, or `not-applicable`. Attach
+   view metadata. In `sysml-v2`, record model revision/digest and native
+   or derived views.
+2. Contextual work may start before requirements stabilize. Do not
+   finalize requirement relationship mappings until those requirements
+   are stable.
+3. Do not silently substitute Mermaid for mandated SysML. Missing SysML
+   Modeling skill is a typed capability gap, not invented behavior.
+
+## Return and recovery
+
+Return `READY` or `OWNER_DECISION_REQUIRED` with verdict,
+candidate_ref, changed_paths, tests, findings, and blocker.
+
+Never invent SysML coverage or claim local normative authority without a pin.

package/skills/test-engineer/SKILL.md

@@ -0,0 +1,48 @@
+---
+name: test-engineer
+description: >
+  One Test Engineer role with Planning and Assurance sessions. Use for
+  Planning Test Engineer, Assurance Test Engineer, SEIT authorship, or
+  declared phase or wave end V&V. Do not use for implementation, defect review, user
+  acceptance, or slice/round boundaries.
+---
+
+# Test Engineer
+
+One role retires Validator; Validator is absent from active roles.
+Planning Test Engineer and Assurance Test Engineer are sessions, not
+separate catalog roles.
+
+## Inputs and match
+
+- **Inputs:** approved baseline, candidate, evidence, SEIT, published
+  standard citations, declared phase or wave end boundary, compact return schema.
+- **Match:** V&V planning or declared phase or wave end assurance is declared.
+- **Non-match:** product implementation, Park Ranger defects, Surveyor
+  acceptance, slice or round boundary.
+
+## Algorithm
+
+1. Planning Test Engineer finalizes `seit.json` after Plan Integrator
+   reconciliation and performs delta reconciliation after relevant
+   decision, requirement, or design changes. Do not invent missing
+   method-skill behavior.
+   Expedition in-document verification cases are authored only against the
+   approved, planning-gated register.
+2. Assurance Test Engineer starts a fresh session; reject author ancestry;
+   evaluate the exact stable candidate at the declared phase or wave end only.
+   VALIDATING is owned here.
+3. When a published standard is cited, verify the document and clause.
+4. Return the smallest missing proof. Never mutate the candidate.
+
+## Return and recovery
+
+Return `PASS`, `NEEDS_MORE_EVIDENCE`, or `FAIL` with verdict,
+candidate_ref, changed_paths, tests, findings, and blocker.
+
+Coordinators enforce `max_assurance_rounds` of 1 per declared phase or wave:
+one aggregate repair, then deterministic closure without another review. The
+next distinct declared phase or wave carries its own budget.
+
+Never implement, self-certify, replace Park Ranger or Surveyor, or grant
+owner-only approval.

package/skills/validator/SKILL.md

@@ -1,44 +1,37 @@
 ---
 name: validator
 description: >
-  Independently validate evidence sufficiency and optional rubric scoring for
-  one exact stable candidate in a fresh session at the end. Use for Validator
-  or evidence validation. Do not use for implementation, defect review, user
-  acceptance, automatic reviews, author self-checks, or slice/round boundaries.
+  Compatibility diagnostic for plans that still name Validator. Use when
+  an existing plan assigns Validator. Do not use for new Journeys,
+  evidence validation, or as an active assurance role.
 ---
 
 # Validator
 
-Independent assurance responsibility, outside the mutation-authority ladder.
+Compatibility only. Not an active assurance role. Assurance Test Engineer
+owns VALIDATING. Router reroutes Validator assignments.
 
 ## Inputs and match
 
-- **Inputs:** approved baseline, exact candidate ref, author identity, scope,
-  evidence, acceptance, at-end boundary, optional rubric, and compact return
-  schema.
-- **Match:** Validator is declared and the final integrated candidate is ready
-  at-end.
-- **Non-match:** candidate is unstable, slice or round boundary, evidence alone
-  needs author self-check, or another assurance responsibility applies.
+- **Inputs:** a plan that still assigns `Validator`, plus the recorded
+  Journey snapshot.
+- **Match:** an existing plan names Validator as an assigned role.
+- **Non-match:** new Journeys, Test Engineer sessions, Park Ranger,
+  Surveyor.
 
 ## Algorithm
 
-1. Start a fresh session; reject author identity, author ancestry, candidate
-   discontinuity, or any boundary other than at-end.
-2. Check scope continuity, labeled evidence, required commands, positive and
-   negative cases, and acceptance support. Where the candidate cites a published
-   standard, verify against that text rather than neighbouring agreement.
-3. Apply `references/grading-rubric.md` only when explicit scoring is requested.
-4. Return the smallest missing proof or exact failing criterion. Never mutate the
-   candidate or manufacture evidence.
+1. Do not validate evidence, score rubrics, or run assurance. Read
+   identities only from the recorded Journey snapshot, never from the
+   current global defaults file.
+2. Return `REROUTED` to the Router with a diagnostic that Validator is
+   compatibility-only and is not an active assurance role; treat the
+   assignment as unused and continue under Assurance Test Engineer.
+3. Preserve the checkout lease. Do not write planning state.
 
 ## Return and recovery
 
-Return `PASS`, `NEEDS_MORE_EVIDENCE`, or `FAIL` with verdict, candidate_ref,
-changed_paths, tests, findings, and blocker. `PASS` is terminal.
-`NEEDS_MORE_EVIDENCE` and `FAIL` permit bounded correction. Coordinators
-enforce `max_assurance_rounds` of 1; this role does not redispatch or re-evaluate
-the Journey's repair. The coordinator verifies it deterministically.
+Return `REROUTED` with verdict, candidate_ref, changed_paths, tests,
+findings, and blocker. Findings name the compatibility path.
 
-Never implement, repeat unchanged review, replace Park Ranger/Surveyor, or
-grant owner-only approval.
+Never implement, self-assure, select models, or mutate remotes.

package/skills/validator/references/grading-rubric.md

@@ -1,8 +1,9 @@
-# Grading rubric (Validator scoring)
+# Grading rubric (optional Assurance Test Engineer scoring)
 
 Use only when the task, owner, or phase gate requests **explicit rubric scoring**.
-Sufficiency checks always run when Validator is active. Scoring is optional and
-never makes independent assurance automatic.
+Sufficiency checks always run when Assurance Test Engineer is active. Scoring is
+optional and never makes independent assurance automatic. Validator is
+compatibility-only and does not score active Journeys.
 
 ## Dimensions (equal weight unless the plan overrides)
 
@@ -27,7 +28,7 @@ Sum dimension scores (0–12).
 
 1. Run sufficiency on the exact candidate first.
 2. If scoring is requested, score the same candidate with this rubric.
-3. Final Validator outcome:
+3. Final Assurance Test Engineer outcome:
    - `FAIL` if either check fails
    - `NEEDS_MORE_EVIDENCE` if either needs more proof
    - `PASS` only when sufficiency passes and (scoring was not requested, or scoring supports PASS)