ahead-pi 0.2.0 → 0.3.0

package/generated/reference/docs/workflows/internal-improvement.md

@@ -0,0 +1,219 @@
+# Internal Improvement Workflow
+
+Status: pilot v0.1
+
+## Outcome
+
+Use this flow when externally required behavior should remain invariant while an internal quality improves: maintainability, comprehensibility, performance, reliability margin, operability, testability, cost, or preventive risk reduction.
+
+Use product change when externally meaningful behavior is intentionally changing, or corrective debugging when an observed defect is the reason for the work.
+
+## Lifecycle
+
+```text
+┌──────────────────────────────────────────────┐
+│ 1. DEFINE INVARIANTS                         │
+│                                              │
+│ HUMAN                                        │
+│ • Define behavior that must not change       │
+│ • Define scope and non-goals                 │
+│                                              │
+│ AI — ASSIST                                  │
+│ • Identify overlooked contracts/consumers    │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 2. BASELINE                                  │
+│                                              │
+│ HUMAN accepts method, evidence, uncertainty  │
+│ AI — ASSIST challenges noise / confounders   │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 3. TARGET                                    │
+│                                              │
+│ HUMAN defines quality, threshold, guardrails │
+│ AI — ASSIST exposes gaming / shifted cost    │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 4. OPTIONS                                   │
+│                                              │
+│ HUMAN — FIRST PASS                           │
+│ • Propose at least one improvement approach  │
+│                                              │
+│ AI — ASSIST                                  │
+│ • Expand alternatives • expose coupling      │
+│ • Challenge complexity and assumptions       │
+│                                              │
+│ HUMAN                                        │
+│ • Evaluate and refine options                │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 5. HUMAN DECISION GATE                       │
+│ • Select approach and accept tradeoffs       │
+│ • Record rationale                           │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 6. PLAN                                      │
+│                                              │
+│ HUMAN — FIRST PASS                           │
+│ • Sequence change, checks, and rollback      │
+│                                              │
+│ AI — ASSIST                                  │
+│ • Find affected boundaries and missing tests │
+│                                              │
+│ HUMAN finalizes and approves                 │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 7. IMPLEMENT                                 │
+│ ENGINEER owns and understands the change     │
+│ AI — ASSIST with bounded contributions       │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 8. AI REVIEW                                 │
+│ • Hidden behavior changes • complexity       │
+│ • Tests • coupling • plan compliance         │
+└──────────────────────┬───────────────────────┘
+                       ↓
+          HUMAN DISPOSITIONS EACH MATERIAL
+          AI FINDING WITH RATIONALE / EVIDENCE
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 9. INDEPENDENT HUMAN REVIEW                  │
+│ • Reviewer is not the implementer            │
+│ • Final judgment on preservation/improvement │
+└──────────────────────┬───────────────────────┘
+                       ↓
+       10. HUMAN AUTHORIZES DEPLOY / RELEASE
+                    WHEN APPLICABLE
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 11. VERIFY                                   │
+│                                              │
+│ HUMAN                                        │
+│ • Verify invariants                          │
+│ • Compare before/after target evidence       │
+│                                              │
+│ AI — ASSIST                                  │
+│ • Analyze measurements / suggest checks      │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 12. AI AUDIT                                 │
+│ • Compare invariants, target, plan, reviews, │
+│   deployment, and observed evidence          │
+└──────────────────────┬───────────────────────┘
+                       ↓
+          HUMAN DISPOSITIONS AUDIT FINDINGS
+                       ↓
+                ┌───────────────┐
+                │ INVARIANTS    │
+                │ PRESERVED?    │
+                └───────┬───────┘
+                   NO ↙   ↘ YES
+                     │     │
+                     │     ▼
+                     │  ┌────────────────┐
+                     │  │ TARGET QUALITY │
+                     │  │ IMPROVED?      │
+                     │  └───────┬────────┘
+                     │     NO ↙   ↘ YES
+                     │       │      └────► HUMAN OUTCOME GATE (13)
+                     │       ▼
+                     │  HUMAN CHOOSES
+                     │  ├─ Rework ─────────────↺ IMPLEMENT
+                     │  ├─ Roll back ─────────► HUMAN OUTCOME
+                     │  └─ Accept partial/no improvement
+                     │     with rationale ─────► HUMAN OUTCOME
+                     ▼
+              HUMAN CHOOSES
+              ├─ Rework ──────────────────────↺ IMPLEMENT
+              ├─ Roll back ──────────────────► HUMAN OUTCOME
+              └─ Reclassify ─────────────────► NEW PRODUCT CHANGE RUN
+                                                + HUMAN OUTCOME
+
+Review changes requested ─────────────↺ IMPLEMENT
+```
+
+Preservation and improvement are separate judgments. A failed invariant cannot be accepted as an internal improvement: the team must roll back, rework, or deliberately reclassify the work as a product change. If invariants hold but the target quality does not improve, a human may stop and record a partial or unsuccessful result rather than manufacture success.
+
+## Minimal phases
+
+| Phase | Human owns | AI may | Minimum record | Advance when |
+|---|---|---|---|---|
+| Invariants | Behavior that must remain unchanged, scope, and non-goals | Identify overlooked contracts and consumers | Explicit invariants and exclusions | Human accepts the preservation contract |
+| Baseline | Current quality evidence and measurement validity | Gather metrics and identify measurement gaps | Tests, measurements, structural evidence, or observations | Baseline is credible enough for comparison |
+| Target | Quality to improve, success threshold, and tradeoffs | Suggest measures and unintended effects | Target quality and acceptance signal | Improvement can be evaluated |
+| Options | Initial approach and evaluation of alternatives | After the human first pass, expand alternatives, expose coupling, and challenge abstraction | Human option, AI challenge when used, and evaluated options | Viable approaches and tradeoffs are understood |
+| Decision | Selection, risk, confidence, and reversibility | Check the option against baseline, target, and invariants | Decision and rationale | Human approves the approach |
+| Plan | First-pass sequence, verification, and rollback | Find affected boundaries, migration needs, and missing tests | Plan and invariant checks | Human approves the final plan |
+| Implement | Engineering change and scope control | Bounded refactoring, explanation, tests, and mechanical assistance | Linked changeset and deviations | Change is ready for review |
+| AI review | Validate and disposition every material AI finding | Review the exact snapshot for hidden behavior changes, incidental complexity, tests, and plan alignment without modifying it | Snapshot-bound AI findings; separate human disposition | Every material finding is fixed, invalid, accepted risk, or follow-up with rationale |
+| Human review | Independent final judgment by someone other than the implementer about simplicity, coupling, risk, and preservation | Answer targeted questions and retrieve evidence | Current independent human review | Independent human reviewer accepts the current change |
+| Deploy or release | Authorization and rollout decision | Analyze readiness evidence within policy | Version, environment, actor, time, and result | The intended version reaches the target environment or deployment is explicitly not applicable |
+| Verify | Invariant preservation and target-quality comparison | Analyze measurements and suggest negative checks | Before/after and deployed-outcome evidence when applicable, plus invariant results | Invariants are evaluated and target-quality results are measured; an invariant failure routes to rollback, rework, or product-change reclassification |
+| AI audit | Disposition of findings and required response | Compare invariants, target, plan, reviews, deployment, and observed evidence; identify divergence or weak proof | AI audit findings; separate human disposition | The human disposer accepts the audit gate or reopens work |
+| Outcome | Acceptance, rollback, partial result, or new work | Summarize learning | Outcome, remaining debt, and follow-ups | Human accepts closure |
+
+## Preservation and improvement chain
+
+```text
+HUMAN-OWNED INVARIANTS / NON-GOALS
+               │
+               ▼
+HUMAN-ACCEPTED BASELINE / TARGET
+               │
+               ▼
+HUMAN OPTION + AI-EXPANDED ALTERNATIVES
+               │
+               ▼
+HUMAN DECISION / FIRST-PASS PLAN
+               │
+               ▼
+ENGINEER-OWNED CHANGESET
+               │
+               ▼
+SNAPSHOT-BOUND AI REVIEW
+               │
+               ▼
+IMPLEMENTING-HUMAN DISPOSITION
+               │
+               ▼
+INDEPENDENT HUMAN REVIEW
+               │
+               ▼
+HUMAN-AUTHORIZED DEPLOYMENT WHEN APPLICABLE
+               │
+               ▼
+HUMAN-ACCEPTED INVARIANT VERIFICATION
+AND BEFORE / AFTER COMPARISON
+               │
+               ▼
+AI AUDIT / HUMAN DISPOSITION
+               │
+               ▼
+HUMAN OUTCOME / REMAINING DEBT
+```
+
+## Non-waivable pilot rules
+
+- Invariants are stated before the change.
+- Passing existing tests alone does not prove preservation when those tests do not cover the contract.
+- AI may propose a refactor; humans decide whether it reduces conceptual complexity.
+- File movement or abstraction count is not itself an improvement.
+- The outcome compares before and after evidence against the declared target.
+- An invariant failure cannot close as a successful internal improvement.
+- AI review and implementer self-review do not satisfy independent human review.
+- Implementation completion is not deployment or outcome verification.
+
+## Pilot questions
+
+- Were the invariants specific enough to catch unintended behavior changes?
+- Did the selected measure reflect the quality people actually wanted?
+- Did the work reduce conceptual complexity or merely rearrange it?
+- When did an internal improvement reveal that a product change or bug flow was needed?

package/generated/reference/docs/workflows/investigation.md

@@ -0,0 +1,161 @@
+# Investigation Workflow
+
+Status: pilot v0.1
+
+## Outcome
+
+Use this flow when the primary deliverable is knowledge or reduced uncertainty and no intervention has yet been selected. It may answer feasibility, causal, capacity, vendor, architectural, or technical questions.
+
+An investigation may end with no action, insufficient evidence, several viable interventions, or work routed to another flow.
+
+## Lifecycle
+
+```text
+┌──────────────────────────────────────────────┐
+│ 1. FRAME QUESTION                            │
+│                                              │
+│ HUMAN                                        │
+│ • State the question and why it matters      │
+│ • Record initial understanding               │
+│ • Name audience or dependent decision        │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 2. BOUND INVESTIGATION                       │
+│                                              │
+│ HUMAN                                        │
+│ • Define scope, exclusions, confidence need  │
+│ • Set time/evidence budget and stopping rule │
+│                                              │
+│ AI — ASSIST                                  │
+│ • Challenge whether it is answerable         │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 3. GATHER EVIDENCE                           │
+│                                              │
+│ AI — ASSIST                                  │
+│ • Locate, summarize, compare, and cite       │
+│ • Expose conflicting or missing evidence     │
+│                                              │
+│ HUMAN                                        │
+│ • Evaluate source quality and relevance      │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 4. HUMAN MODEL                               │
+│ • Build/update the explanation of the domain │
+│ • Record assumptions and unknowns            │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 5. EXPLORATION PATHS                         │
+│                                              │
+│ HUMAN                                        │
+│ • Contribute/select what is worth exploring  │
+│                                              │
+│ AI — ASSIST                                  │
+│ • Generate hypotheses and alternatives       │
+│ • Suggest analyses, tests, or prototypes     │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 6. HUMAN SELECTS AND AUTHORIZES              │
+│ • Test / analysis / disposable prototype     │
+│ • Prediction, safety, scope, and disposal    │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 7. EXECUTE EXPLORATION                       │
+│                                              │
+│ HUMAN / ENGINEER OWNS THE EXPERIMENT         │
+│ AI — ASSIST                                  │
+│ • May fully generate an isolated throwaway   │
+│   prototype under prototype policy           │
+│ HUMAN records prototype disposition, or      │
+│ explicitly records that none was used        │
+└──────────────────────┬───────────────────────┘
+                       ↓
+                ┌───────────────┐
+                │ STOPPING      │
+                │ CONDITION     │
+                │ REACHED?      │
+                └───────┬───────┘
+                   NO ↙   ↘ YES
+           ↺ MODEL / PATHS  │
+                            ▼
+┌──────────────────────────────────────────────┐
+│ 8. SYNTHESIZE                                │
+│                                              │
+│ HUMAN                                        │
+│ • Interpret evidence, contradictions, limits │
+│                                              │
+│ AI — ASSIST                                  │
+│ • Organize findings and challenge overclaims │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 9. HUMAN CONCLUSION / OUTCOME GATE           │
+│ • State conclusion, confidence, and unknowns │
+│ • Recommend no action or link another flow   │
+└──────────────────────────────────────────────┘
+```
+
+Full vibe coding is permitted for an isolated, explicitly disposable prototype when its purpose is learning. The learning is retained; prototype code is discarded or deliberately re-enters the appropriate production workflow.
+
+Stopping conditions may be satisfied by enough evidence for a conclusion, exhaustion of the agreed time or evidence budget, discovery that the question is not currently answerable, or another human-approved boundary. Therefore “insufficient evidence” can advance to synthesis and an honest conclusion rather than forcing an endless loop.
+
+## Minimal phases
+
+| Phase | Human owns | AI may | Minimum record | Advance when |
+|---|---|---|---|---|
+| Frame | Question, decision relevance, audience, scope, and initial understanding | Clarify ambiguity and identify adjacent questions | Primary question and why it matters | Human accepts the question |
+| Bound | Confidence needed, exclusions, time or evidence budget, and stopping conditions | Challenge whether the question is answerable | Scope, limits, and stopping rule | Investigation can proceed without unbounded research |
+| Gather | Source evaluation and evidence access | Locate, summarize, compare, cite, and expose conflicts | Evidence links, source quality, and observations | Relevant available evidence is assembled |
+| Model and explore | Mental model, path selection, test authorization, interpretation, and prototype disposition | Generate hypotheses, analysis approaches, or an isolated disposable prototype | Model, predictions, tests, results, and human prototype disposition or no-prototype record | Evidence is sufficient for synthesis or budget is reached |
+| Synthesize | Meaning, confidence, contradictory evidence, and limitations | Organize findings and challenge overclaiming | Findings, confidence, limits, and unknowns | Human can state a bounded conclusion |
+| Conclude and route | Accepted conclusion, recommendation, and next workflow | Suggest follow-up questions or routes | Conclusion and linked decision/change/debug/improvement runs | Human accepts closure or explicitly extends the investigation |
+
+## Knowledge chain
+
+```text
+HUMAN QUESTION / RELEVANCE
+           │
+           ▼
+HUMAN SCOPE / STOPPING RULE
+           │
+           ▼
+AI-ASSISTED EVIDENCE GATHERING
+HUMAN SOURCE EVALUATION
+           │
+           ▼
+HUMAN MODEL / ASSUMPTIONS
+           │
+           ▼
+AI-EXPANDED EXPLORATION PATHS
+HUMAN-SELECTED TEST / ANALYSIS / PROTOTYPE
+           │
+           ▼
+RESULTS / CONTRADICTIONS
+           │
+           ▼
+HUMAN CONCLUSION / CONFIDENCE / LIMITS
+           │
+           ▼
+HUMAN RECOMMENDATION / LINKED RUN
+```
+
+## Non-waivable pilot rules
+
+- The question is human-owned; AI does not quietly redefine it.
+- Facts, retrieved claims, inference, and AI synthesis remain distinguishable.
+- Important citations are followed to authoritative sources.
+- A prototype demonstrates only what it actually tested.
+- “Insufficient evidence” is a valid result.
+
+## Pilot questions
+
+- Were the stopping conditions usable or did research expand indefinitely?
+- Did the conclusion state confidence and contradictory evidence honestly?
+- Did a disposable prototype answer the learning question without leaking into production?
+- Was the next workflow clear when the investigation ended?

package/generated/reference/docs/workflows/operational-stabilization.md

@@ -0,0 +1,187 @@
+# Operational Stabilization Workflow
+
+Status: pilot v0.1
+
+## Outcome
+
+Use this flow when a live system is outside an acceptable operating state and the immediate outcome is restoration, stabilization, or demonstrated convergence. It covers code and non-code causes such as reconciliation storms, configuration drift, capacity exhaustion, provider behavior, dependency failure, and emergent controller interaction.
+
+Incident mode is an urgency and coordination overlay. It does not assert a cause.
+
+## Lifecycle
+
+```text
+             LIVE SYSTEM OUTSIDE
+           ACCEPTABLE OPERATING STATE
+                       │
+                       ▼
+┌──────────────────────────────────────────────┐
+│ 1. ASSESS                                    │
+│                                              │
+│ HUMAN                                        │
+│ • Assess impact, urgency, and scope          │
+│ • Define desired versus actual state         │
+│ • Assign accountable ownership               │
+│                                              │
+│ AI — ASSIST                                  │
+│ • Correlate authorized signals               │
+│ • Identify missing information               │
+└──────────────────────┬───────────────────────┘
+                       ↓
+                ┌───────────────┐
+                │ HUMAN SELECTS │
+                │ RESPONSE MODE │
+                │ Normal /      │
+                │ incident /    │
+                │ emergency     │
+                └───────┬───────┘
+                        ↓
+        ┌──────── PARALLEL — NO JOIN BARRIER ────────┐
+        │                                            │
+        ▼                                            ▼
+┌──────────────────────┐          ┌──────────────────────┐
+│ 2A. RESPOND:         │          │ 2B. RESPOND:         │
+│ INVESTIGATE          │          │ STABILIZE            │
+│                      │          │                      │
+│ HUMAN LEADS          │          │ HUMAN LEADS          │
+│ • Model system       │          │ • Set priorities     │
+│ • Select tests       │          │ • Select candidate   │
+│ • Interpret evidence│          │   intervention       │
+│                      │          │ • Assess risk        │
+│ AI — ASSIST          │          │                      │
+│ • Organize telemetry│          │ AI — ASSIST          │
+│ • Suggest hypotheses│          │ • Compare actions    │
+│ • Find conflicts     │          │ • Find side effects  │
+└──────────┬───────────┘          └──────────┬───────────┘
+           │ CONTINUES                         │
+           │ WHILE NEEDED                      ▼
+           │                      ┌─────────────────────────┐
+           │                      │ 2C. HUMAN ACTION GATE   │
+           │ MAY INFORM           │                         │
+           ├─ - - - - - - - - - ►│ • Authorize actor/scope │
+           │   NO BARRIER         │ • Blast radius          │
+           │                      │ • Rollback/containment  │
+           │                      │ • Accepted uncertainty  │
+           │                      └────────────┬────────────┘
+           │                                   ↓
+┌──────────────────────────────────────────────┐
+│ 3. EXECUTE AND OBSERVE                       │
+│                                              │
+│ HUMAN / PREAUTHORIZED AUTOMATION             │
+│ • Performs the operational action            │
+│                                              │
+│ AI — ASSIST                                  │
+│ • Interprets authorized observations         │
+│ • Does not execute the intervention          │
+│ • Workflow phase grants no production access │
+└──────────────────────┬───────────────────────┘
+                       ↓
+                ┌───────────────┐
+                │ CONVERGING?   │
+                └───────┬───────┘
+                   NO ↙   ↘ YES
+          ↺ INVESTIGATE     │
+            / STABILIZE     ▼
+┌──────────────────────────────────────────────┐
+│ 4. VERIFY RECOVERY                           │
+│                                              │
+│ HUMAN                                        │
+│ • Verify convergence and user-visible state │
+│                                              │
+│ AI — ASSIST                                  │
+│ • Analyze authorized recovery indicators     │
+└──────────────────────┬───────────────────────┘
+                       ↓
+┌──────────────────────────────────────────────┐
+│ 5. MONITOR RECURRENCE                        │
+│ HUMAN selects duration and signals           │
+│ AI — ASSIST                                  │
+│ • Summarize authorized telemetry             │
+└──────────────────────┬───────────────────────┘
+                       ↓
+                ┌───────────────┐
+                │ STABLE?       │
+                └───────┬───────┘
+                   NO ↙   ↘ YES
+          ↺ INVESTIGATE     │
+            / STABILIZE     ▼
+┌──────────────────────────────────────────────┐
+│ 6. HUMAN OUTCOME                             │
+│ • Accept closure and remaining risk          │
+│ • Link corrective, investigation, decision, │
+│   or improvement follow-up runs              │
+└──────────────────────────────────────────────┘
+```
+
+Investigation and stabilization proceed independently: there is no join gate. Investigation may inform an intervention and continue after it, but service restoration does not wait for a complete causal explanation when a human authorizes a proportionate intervention.
+
+The executable `respond` phase contains both parallel tracks and the human action gate shown as 2A–2C. Combining their evidence in one phase visit preserves the no-join rule: a bounded intervention may be authorized while causal investigation remains incomplete.
+
+## Minimal phases
+
+| Phase | Human owns | AI may | Minimum record | Advance when |
+|---|---|---|---|---|
+| Assess | Response-mode declaration, impact, scope, urgency, desired state, ownership, and recovery signals | Correlate authorized signals and expose missing information after the human assessment begins | Assessment and response-mode record | Response mode, ownership, boundaries, and recovery criteria are clear |
+| Respond | System model, investigation direction, stabilization priorities, and selected intervention | Organize telemetry, generate hypotheses, compare interventions, and expose contradictions or side effects | Human model, optional investigation ledger, and bounded intervention with rollback and stop conditions | Human authorizes the consequential action; proven root cause is not required |
+| Execute and observe | Action execution, stop or rollback decisions, and interpretation | Analyze observations; AI has no execute capability in this phase | Actor, command or change, time, result, and new evidence | Result is known and desired state is approached |
+| Verify recovery | Recovery criteria and user-visible validation | Analyze convergence and recurrence indicators | Health, convergence, and external behavior evidence | Recovery is demonstrated, not merely assumed |
+| Monitor | Duration and signals sufficient to detect recurrence | Summarize authorized telemetry | Monitoring window and result | Human accepts stability or reopens response |
+| Outcome and follow-up | Closure, remaining risk, causal confidence, and routed work | Summarize timeline and proposed follow-ups | Outcome, unknowns, deferred records, linked runs | Human accepts closure and follow-up disposition |
+
+## Parallel evidence and action chains
+
+```text
+OBSERVED OPERATING CONDITION
+              │
+              ▼
+HUMAN-OWNED IMPACT / SCOPE / DESIRED STATE
+              │
+       ┌──────┴────────┐
+       ▼               ▼
+EVIDENCE / MODEL   STABILIZATION OPTIONS
+       │ CONTINUES         │
+       ▼                   ▼
+HUMAN-SELECTED      HUMAN AUTHORIZATION
+HYPOTHESES / TESTS         │
+       │ MAY INFORM        ▼
+       ├─ - - - - - ► ACTION / OBSERVED RESULT
+       │                   │
+       └──── updates ◄─────┘
+                           │
+                           ▼
+            HUMAN-ACCEPTED RECOVERY / CONVERGENCE
+                           │
+                           ▼
+REMAINING RISK / LINKED FOLLOW-UP RUNS
+
+The investigation chain has no join barrier. AI assists
+evidence gathering, option generation, hypothesis
+generation, and analysis; it owns no gate.
+```
+
+## Incident and emergency overlay
+
+During incident or emergency mode, maintain at minimum:
+
+- current owner and impact;
+- timestamped consequential actions and decisions;
+- actor, rationale, scope, blast radius, rollback, and result;
+- current recovery criteria and communications owner;
+- deferred records that must be reconstructed after stabilization.
+
+## Non-waivable pilot rules
+
+- Workflow permission is not production authorization.
+- A live intervention requires an accountable human unless existing automation was previously authorized for that action.
+- Any lasting engineering change produced during stabilization still requires a human first-pass plan and independent human review, either in this record or a linked change run; emergency policy may defer but not erase those gates.
+- Kubernetes readiness, controller status, or a successful command is not automatically user-visible recovery.
+- The conclusion may include mechanism, trigger, enabling conditions, and detection or containment gaps rather than one root cause.
+- Closure does not require every follow-up to remain inside the operational run.
+
+## Pilot questions
+
+- What threshold should activate incident mode?
+- Which records were feasible during response versus reconstructed later?
+- When was unknown-cause remediation justified?
+- What evidence demonstrated convergence and user-visible recovery?
+- Did linked follow-up runs prevent the incident from remaining permanently open?