Dashboard › institutional-transition-lab › Distillation
542fde47-679c-4461-9ad6-d136af5cf7b9["lore_tm_v1_LTRsH52Kt3BsYS8vwSt9j7tEB8G5IdMq-s5uv0FqYoI","lore_tm_v1_G4Ds6fZ_vcRntpCJbCmjTszGWP0h5Sd9tkcv-CUnc_0","lore_tm_v1_bP8Bhz25OFdcFSSQjFw1kyYIj3nhC8_pqBSqhbw6u2g","lore_tm_v1_o-408pcCE7B4ZOSPvYB59VWCQgeD-sWwimo0wJRtg7E","lore_tm_v1_XWRcpGlTupxM2Hfbnm4nwzv7nf5Va3ebPLFAmtfXj9s","lore_tm_v1_wF30cNf1v0AXF6hMb5SIXkww0wercpI7AK-ruDdrCjY","lore_tm_v1_jLGr7ir0sLSFORvU1e2KQC1rUTzUFCEwB_1PpTOyB6A","lore_tm_v1_50fqLqZujw2Ci_b1rSCyVRIg7djxhJYiHDI5iS7eDPQ"]
Date: Sep 8, 2026
/home/byk/Code/institutional-transition-lab: .git/, .github/, .gitignore, cases/, data/, docs/, package-lock.json, package.json, PROTOCOL.md, pyproject.toml, README.md, research/, schema/, scripts/, src/, tests/, and uv.lock.data/; only data/raw/.gitkeep and data/derived/.gitkeep were listed. Relevant retrieval/coding machinery exists in .github/workflows/retrieve-oss-governance.yml, .github/workflows/continue-governance-coding.yml, src/institution_lab/source_retrieval.py, src/institution_lab/source_enrichment.py, src/institution_lab/governance_coding.py, scripts/copilot-code-governance.mjs, and tests/test_governance_coding.py.#4, “Adjudicate the frozen 40-record governance corpus,” pins successful Actions run 33063470968, artifact 9642983578, and exact 40-document bundle SHA-256 d70f7f795847968339a645b6dbd6d4428af6bd6044de44fa2602d4c4343b0ee9; run 33086454759 is only a stochastic repeat and must not replace the frozen v1.2 codings.#4 requires explicit per-record decisions from accept, revise, reject, or abstain; source-body/patch review rather than model-summary-only review; preserved evidence references and source URLs for every accepted fact; reviewer type and assistance provenance; and no “human adjudication” label if an LLM performs any review.#4 acceptance criteria require: 1. a committed versioned adjudication schema and validator; 2. all 40 record IDs exactly once; 3. pinned and reverified artifact IDs and hashes; 4. grounded source evidence for every non-rejected institutional claim; 5. an audit summary covering class decisions, unresolved records, graph-edge revisions, and reviewer provenance; 6. a canonicalization gate that remains closed while any required record is unresolved.#4 explicitly excludes Gemini 3.7 Flash from canonical adjudication; it is authorized only for event triage.README.md describes an independence boundary: performance timeline A and institutional-event timeline B are produced separately and joined only after freezing, because searching for events after seeing transition points would create a narrative generator rather than a test.README.md lines 233-247 records that cases/model-selection/event-triage-v1.json selects gemini-3.7-flash for broad candidate discovery/triage only; its output is a review queue, not a canonical power graph or causal finding. The frozen governance-coding protocol uses two independent Luna coders plus Terra advisory routing.README.md lines 312-344 documents the source pipeline: .github/workflows/retrieve-oss-governance.yml builds the candidate corpus, caches GitHub search responses, deduplicates issues/PRs, emits a bounded corpus and retrieval audit, and validates Luna-medium’s one-to-one responses and quoted evidence spans. The enrichment stage creates a deterministic 40-document audit envelope containing merge state and bounded patches.README.md reports that V2 retrieval evaluated 206 identical valid documents, achieved 94.7% four-class agreement and 98.1% institutional-binary agreement, but missed merged Valkey foundational-governance PR #345 and misread another PR’s final rights direction; the narrow-query safety net retained Valkey #345.README.md lines 346-352 exposes frozen-corpus coding results—67.5% event-class agreement, 55.0% complete grounded power-edge-set agreement, 35 records routed to source adjudication, and five with provisional core agreement.README.md lines 361-370 exposes the initial Terraform/OpenTofu result—three raw OpenTofu issue-throughput candidates and zero robust transitions—and states that the first event inventory was assembled after pilot outcomes were inspected and is labeled exploratory_seed.research/oss-governance-coding-results-v1.2.md, research/luna-event-extraction-development-benchmark.md, research/luna-terra-hard-case-comparison.md, generated docs/terraform-opentofu.html, and docs/index.html; these are contamination risks if adjudicators browse the repository unrestricted.tests/test_governance_coding.py enforce ontology parity between schema/governance-coding-v1.schema.json and validator constants RESPONSE_KEYS, RECORD_CLASSES, EVENT_KINDS, BODY_PATCH_RELATIONS, RIGHT_KINDS, DIRECTIONS, and CHANGE_STATUSES.test_prepare_bundle_excludes_triage_and_retrieval_metadata() verifies that prepared model documents contain only record_id, source_url, and MODEL_INPUT_FIELDS; it ensures model_review_queue, retrieval_score, and source_context are absent and asserts transition_dates_excluded is True and outcome_data_used is False.test_validates_evidence_against_named_source_component() verifies exact quote grounding against a named source component; replacing the patch quote with "invented" must produce a validation error containing "not grounded".test_comparison_routes_edge_disagreement_and_bounded_evidence() verifies that actor disagreement plus patch_selection_truncated=True yields route reasons edge_disagreement and bounded_evidence, adjudication_count == 1, canonical_ledger_records == 0, a route coding ID ending in -terra-route, canonical_ledger_eligible is False, and provisional_core_agreement is False.test_routes_scope_evidence_and_body_patch_disagreements() verifies that changing body_patch_relation from consistent to patch_supersedes_body and edge scope from project releases to all project decisions yields both body_patch_relation_disagreement and edge_disagreement, with provisional_core_agreement_count == 0.schema/governance-coding-v1.schema.json is a JSON Schema draft 2020-12 coding-response schema titled “Performance-blinded governance coding v1”; it is not yet the issue-required adjudication decision schema because its top-level fields describe coder output rather than explicit accept/revise/reject/abstain adjudication and reviewer provenance.schema/governance-coding-v1.schema.json requires exactly these top-level research fields, with additional properties forbidden: record_class, event_kind, title, announced_on, effective_on, affected_scope, body_patch_relation, power_changes, confidence, and ambiguity.record_class values are ordered as: effective_institutional_change, announced_institutional_change, proposal_only, control_event, no_event, abstain. event_kind permits leadership, board_or_steering, control_rights, reorganization, foundation_transfer, license, fork, reunification, strategy, product, external, or null.body_patch_relation permits consistent, patch_supersedes_body, body_only, patch_only, conflict, not_applicable, or unclear.actor, right_kind, target, direction, change_status, scope, and at least one evidence_refs entry containing non-empty source_id and quote. right_kind permits appoint, remove, elect, vote, delegate, override, approve, merge, write, release, veto, own, license, steward, fund, set_budget, set_strategy, set_policy, set_membership, set_terms, and inform; direction permits added, removed, or modified; change_status permits effective, announced, proposed, rejected, or unclear.research/oss-governance-coding-protocol-v1.2.md was frozen on 2026-08-27 and is labeled “pre-adjudication evaluator correction; model-informed development protocol.”provisional_exact_agreement despite material differences.body_patch_relation disagreement as an explicit route reason, renaming the remaining non-routed state to provisional_core_agreement, and reporting full-response agreement separately because titles and numeric confidence are not canonical research fields.33060493337; only the Terra advisory pass was repeated over the corrected route to isolate evaluator changes from model sampling variance.