# Local CCRA Next Review Requirements ## 1. Return Metadata - project_id: `the-mindscape-of-bro-tsong` - review_round: `round-04` - local_pass: `pilot-02` - turn_type: `planning_turn` - runtime_id: `ccra-local-review` - runtime_version: `0.1.1` - agent_id: `ccra-local-reviewer` - agent_version: `0.1.1` - invocation_mode: `prompt_path_reference` - carrier_type: `codex_thread` - child_session_id: `019edcde-24be-7593-99b2-5cf014467340` - same_child_session_as_review_turn: `yes` - first_turn_report_path: `local_ccra_reviews/round-04/pilot-02/04_LOCAL_CCRA_REVIEW_REPORT.md` - owner_decision_path: `local_ccra_reviews/round-04/pilot-02/owner-decision.md` - planning_output_path: `local_ccra_reviews/round-04/pilot-02/next-review-requirements.md` - generated_at: `2026-06-19` ## 2. Planning Scope - next_review_goal: define the submitted materials and verification evidence required for a Round 04.1 post-patch routing verification / regression pass. - target_round_or_pass: `round-04.1_selector-rule-patch_verification` - planning_basis: - `local_ccra_reviews/round-04/pilot-02/04_LOCAL_CCRA_REVIEW_REPORT.md` - `local_ccra_reviews/round-04/pilot-02/owner-decision.md` - `local_ccra_reviews/round-04/pilot-02/agent-invocation-packet.md` - explicit_non_goals: - Do not re-review the full initial Round 04 bundle in this planning turn. - Do not use this planning output as Owner approval. - Do not treat Round 04.1 verification as Round 05 acceptance. - Do not promote QPI or Intellectual Archaeology lifecycle status. - Do not modify selector rules, tests, knowledge assets, model files, or review bundles in this turn. Owner decision status is still `pending`. This plan therefore defines required review materials and recommended defaults; it does not convert the first-turn findings into accepted work. ## 3. Required Submitted Materials ```text - material_id: M-01-review-brief required_path_or_bundle_slot: ccra_review_bundle//00_OPEN_THIS_FIRST_*.md purpose: state the review goal, target failures, non-goals, and authority boundary. required: yes notes: Must say this is a post-patch verification pass, not a second blind test and not lifecycle promotion. - material_id: M-02-rule-patch-summary-and-diff required_path_or_bundle_slot: ccra_review_bundle//01_RULE_PATCH_SUMMARY_AND_DIFF_*.md purpose: show exactly what selector rule surfaces changed and why. required: yes notes: Must map changes to R04-BI-002, R04-BI-024, R04-BI-035, R04-BI-036, and any accepted same-class case such as R04-BI-022. - material_id: M-03-expected-routing-labels-or-owner-decisions required_path_or_bundle_slot: ccra_review_bundle//02_EXPECTED_ROUTING_LABELS_*.md or owner-decision-backed equivalent purpose: prevent Web CCRA from inferring target labels from prose. required: yes notes: Must separate must-fix, same-class boundary, calibration-only, and explicitly-not-patched cases. - material_id: M-04-post-patch-routing-verification-report required_path_or_bundle_slot: reports/Round04_1_post_patch_routing_verification_report_*.md and bundle copy purpose: provide full frozen-pool rerun after the patch. required: yes notes: Must include all 38 Round 04 inputs, not only target cases. - material_id: M-05-before-after-routing-diff required_path_or_bundle_slot: helper-outputs/routing-behavior-diff.md or a report section with equivalent table purpose: show target changes, non-target changes, accepted collateral behavior, and regressions. required: yes notes: If the helper cannot run, include a manual before/after table and explain the missing helper risk. - material_id: M-06-regression-and-command-log required_path_or_bundle_slot: ccra_review_bundle//04_REGRESSION_AND_COMMAND_LOG_*.md purpose: prove executed tests and validation gates, not dry-run declarations. required: yes notes: Include exact commands, exit codes, timestamps or run context, and changed-file impact. - material_id: M-07-existing-selector-regression required_path_or_bundle_slot: reports/selector_regression_report*.md or command log section purpose: prove existing no-call, model-exclusion, QPI-before-IA, and depth-limiting protections still pass. required: yes notes: Must not rely only on targeted target-case checks. - material_id: M-08-calibration-smoke required_path_or_bundle_slot: reports/selector_calibration_smoke_report*.md or command log section purpose: prove direct-execution, QPI, and IA boundaries remain sane after patch. required: yes notes: Especially watch translation/direct transformation and IA trigger over-expansion. - material_id: M-09-bundle-manifest required_path_or_bundle_slot: ccra_review_bundle//BUNDLE_FILE_MANIFEST_*.md purpose: let reviewers inspect package completeness without relying on upload order. required: yes notes: Use versioned filenames to avoid Web ChatGPT file confusion. - material_id: M-10-optional-raw-changed-files required_path_or_bundle_slot: ccra_review_bundle//optional_raw_changed_files_.zip purpose: allow exact inspection of selector, tests, scripts, reports, and relevant docs. required: yes notes: Preserve source-relative paths. Include zip validation/readback summary. - material_id: M-11-local-ccra-first-turn-report required_path_or_bundle_slot: local_ccra_reviews/round-04/pilot-02/04_LOCAL_CCRA_REVIEW_REPORT.md purpose: optional context for Web CCRA on why these verification requirements exist. required: no notes: Include in Web upload only if Owner explicitly asks Web CCRA to review Local CCRA findings. - material_id: M-12-owner-decision-record required_path_or_bundle_slot: local_ccra_reviews/round-04/pilot-02/owner-decision.md or round-specific decision note purpose: show which findings and same-class boundaries Owner accepted, rejected, or deferred. required: yes before final acceptance claims notes: Current status is pending; absence blocks acceptance language, not evidence collection. ``` ## 4. Required Verification Evidence ```text targeted_regression: required: yes cases: R04-BI-002, R04-BI-024, R04-BI-035, R04-BI-036 conditional_cases: R04-BI-022 if Owner accepts it as same-class QPI-limited intent expected: target failures change in the intended direction only. full_frozen_pool_rerun_or_before_after_comparison: required: yes scope: all 38 Round 04 inputs expected: selected_models, rejected_models, no_call, score, reasons, penalties, and routing_notes before/after. existing_regression_or_unit_tests: required: yes minimum: round04 runner tests, selector regression, any post-patch verification tests, and touched selector helper tests. expected: executed, not dry-run. calibration_smoke: required: yes expected: no-call, direct execution, QPI problem-definition, IA refusal, QPI-before-IA, and depth-limiting smoke coverage. validation_report: required: yes minimum: index check and model library validation if package claims model-library consistency. expected: no lifecycle upgrade from validation pass. command_log: required: yes minimum: command, working directory, exit code, output file path, and whether files changed. expected: dry-run must be labeled dry-run and cannot support pass claims. dry_run_acceptable: no for final verification; yes only for planning/preflight manifest readability. ``` ## 5. Expected Before / After Comparisons ```text - comparison_id: C-01-target-fixes before_source: reports/Round04_blind_routing_evaluation_report_2026-06-18.md after_source: reports/Round04_1_post_patch_routing_verification_report_.md expected_fields: input_id, input_text, selected_models, rejected_models, no_call, score, reasons, penalties, routing_notes target_cases: R04-BI-002, R04-BI-024, R04-BI-035, R04-BI-036 non_target_cases_to_watch: R04-BI-001, R04-BI-003, R04-BI-004, R04-BI-005, R04-BI-021, R04-BI-025, R04-BI-031, R04-BI-032, R04-BI-033, R04-BI-034 - comparison_id: C-02-same-class-boundary before_source: reports/Round04_blind_routing_evaluation_report_2026-06-18.md after_source: post-patch frozen-pool rerun expected_fields: no_call, QPI selected/rejected, IA selected/rejected, exclusion reason target_cases: R04-BI-022 if accepted into patch scope non_target_cases_to_watch: all depth-limiting and IA-refusal cases, especially R04-BI-021 and R04-BI-023 - comparison_id: C-03-direct-execution-collateral before_source: initial Round 04 report plus existing selector regression after_source: post-patch report plus regression output expected_fields: no_call status and reasons for fact lookup, translation, rewrite, formatting, direct execution, how-to, dictionary judgment target_cases: R04-BI-002 non_target_cases_to_watch: R04-BI-001, R04-BI-003, R04-BI-004, R04-BI-005, R04-BI-031, R04-BI-032, R04-BI-033, R04-BI-034 - comparison_id: C-04-calibration-not-patched before_source: first-turn report section 12 after_source: post-patch report calibration section expected_fields: unchanged/deferred status, reason, Owner/Web question target_cases: none non_target_cases_to_watch: R04-BI-010 through R04-BI-017, R04-BI-018, R04-BI-026, R04-BI-027, R04-BI-030, R04-BI-037, R04-BI-038 ``` ## 6. Open Owner Decisions ```text - decision_id: OD-R04-001 decision_needed: accept the four must-fix initial Round 04 target failures as the patch target. options: accept_all | accept_subset | request_web_baseline_comparison_first recommended_default: accept_all blocks_next_review: no for evidence collection; yes for final acceptance language. - decision_id: OD-R04-002 decision_needed: classify R04-BI-022 as must-fix same-class, calibration-only, or acceptable conservative no-call. options: include_in_patch | calibration_only | leave_no_call recommended_default: include_in_patch blocks_next_review: yes if the next package claims R04-BI-022 as accepted collateral or target behavior. - decision_id: OD-R04-003 decision_needed: define policy for descriptive governance/capacity statements without explicit requests. options: provisional_qpi | ask_clarifying_question | no_call_until_request recommended_default: ask_clarifying_question or provisional_qpi with low confidence. blocks_next_review: no for narrow patch verification; yes for broader selector maturity claims. - decision_id: OD-R04-004 decision_needed: define direct IA versus QPI-first route for natural model-extraction/deep-mechanism inputs. options: direct_ia_when_depth_intent_clear | qpi_first_always | qpi_plus_ia_when_both_signals_present recommended_default: qpi_plus_ia_when both clear depth intent and model-extraction intent appear. blocks_next_review: no for narrow patch verification; yes for IA route maturity claims. - decision_id: OD-R04-005 decision_needed: decide whether helper profile tuning is required before formal Round 05 Local CCRA use. options: tune_now | document_limitations | defer recommended_default: document_limitations for this pilot, tune before repeated formal use. blocks_next_review: no, if limitations are explicit. ``` ## 7. Explicit Non-Goals ```text - non_goal: no lifecycle status upgrade reason: QPI and Intellectual Archaeology remain draft / B / pending until Owner / CCRA lifecycle authority changes that. - non_goal: no Round 05 acceptance claim reason: Round 04.1 verification only proves a scoped selector patch if evidence passes. - non_goal: no second blind test claim reason: after expected labels and target cases exist, this is a post-patch verification pass, not blind evaluation. - non_goal: no broad governance/capacity keyword patch reason: those cases are calibration-only until Owner/Web policy is explicit. - non_goal: no IA trigger broadening without calibration reason: IA is a deep model and should not become default because of words like structure, mechanism, long-term, or model. - non_goal: no LLM selector, RAG, database, frontend, backend, user system, or third model reason: outside model_library_mvp selector patch scope. - non_goal: no helper exit-code absolutism reason: helper output is evidence only; profile mismatch, dry-run, and broad scans must be judged by the reviewer. ``` ## 8. Recommended Web CCRA Review Questions 1. Do the Round 04.1 submitted materials prove the four target cases were fixed without creating non-target regressions across the full frozen 38-case pool? 2. Should `R04-BI-022` be treated as an accepted same-class QPI-limited override, calibration-only, or conservative no-call? 3. Are governance/capacity descriptive cases (`R04-BI-010` through `R04-BI-017`) better handled as provisional QPI, clarification, or no-call until the user asks for analysis? 4. Should natural model-extraction/deep-mechanism cases (`R04-BI-018`, `026`, `027`, `038`) directly select IA, select QPI first, or select both when depth intent is explicit? 5. Is the verification package sufficient to avoid a further small patch round, or should remaining boundary issues be deferred into a broader calibration round? 6. Does any wording in the package imply lifecycle promotion, Round 05 acceptance, or Owner/Web approval beyond the evidence? ## 9. Material Package Checklist ```text review_brief: required: yes pass_condition: states target scope, non-goals, and post-patch verification identity. rule_or_model_diff: required: yes pass_condition: selector rule changes mapped to target cases; no model content/lifecycle changes unless separately justified. expected_labels_or_decisions: required: yes pass_condition: target, same-class, calibration-only, and explicitly-not-patched cases are separated. post_patch_verification_report: required: yes pass_condition: all 38 frozen inputs rerun with score/reason details. regression_and_command_log: required: yes pass_condition: commands executed with exit codes; no dry-run used as pass evidence. bundle_manifest: required: yes pass_condition: versioned manifest names recognized by human reviewer even if generic helper profile is noisy. optional_raw_changed_files: required: yes pass_condition: source-relative zip entries and readback/validation summary included. local_ccra_report: required: optional pass_condition: included only with Owner approval; otherwise referenced locally, not uploaded as formal Web evidence. owner_decision_record: required: yes before acceptance claims pass_condition: records accepted, rejected, deferred, or escalated findings; current `pending` is not acceptance. ``` ## 10. Limits And Carry-Forward Risks - Owner decision remains pending. The next package may gather evidence, but it should not claim accepted fixes without a decision record. - The first-turn report identified clear `fix_now` targets but did not inspect any post-patch evidence. The next review must not cite this plan as proof that the patch worked. - `routing-behavior-diff-audit` was skipped in review_turn for good reason, but before/after routing comparison is required for the post-patch pass. - `regression-validation-gate-runner` was dry-run only. The next package needs executed gates if it claims verification. - `review-bundle-audit` used a generic profile that did not recognize versioned bundle manifests. Human-readable bundle manifest evidence is still required, and helper profile limitations should be documented. - `lifecycle-status-guard-scan` was broad and noisy. The next review should include a lifecycle/scope guard section focused on the submitted package, not a whole-repository false-positive dump. - Calibration-only cases must stay explicitly out of the narrow patch acceptance claim. They can be listed as next-round policy questions, not evidence that the patch is incomplete. - Current project docs may contain later Round 04.1 language. The next review should clearly distinguish source evidence, patch evidence, closeout language, and lifecycle claims.