Decisions

• rr-cooper-2018-led-pdf-mismatch · · · 2026-05-07T20:28:20.095546+00:00
• · · ·
• d-91 · · · 2026-05-07T20:25:08.724788Z

  Catalog/PDF mismatch on entry id 'du-2023-designing_for_trust_sa'. The DB record is titled 'Designing for Trust and Situational Awareness in Automated Vehicles' with authors Du/Yang/Robert, but Data/research/papers/du-2023-designing_for_trust_sa.pdf is actually Black, Brough, King, King, Bentley, Fylan, Wood (2023) 'Making night-time pedestrians safer using innovative clothing designs', TRF 94:321-328. The DOI 10.1016/j.trf.2023.03.002 and archive_url_used (Leeds Beckett eprint 10905) match the Black paper, not the Du paper. Existing summary/key_finding/methodology/sample_size in DB describe an unrelated N-back/DRT study, suggesting prior fabrication or a swapped record. Verification status is currently 'verified' with no notes, which is wrong.

• d-089 · change · executed · 2026-05-07T20:13:29.971890+00:00

  Expand researcher role with catalog_maintenance_mode: backfill methodology enum, canonicalize author names, fill missing authors/tags on archived entries. Addresses 4 orphaned validator warnings (methodology off-vocab 205, author-name format 493, 48 no-authors, 46 no-tags) that no role currently owns.

• d-90-resolution · · closed_resolved · 2026-05-07T20:10:06.719332Z

  Audited the 21 kirlik-* entries flagged in d-90. All current tags in DB are valid subcategory ids (situational_awareness, automation, display_design, feedback_modes, attention, executive_function, perceptual_processing). A parallel agent already retagged these between sess-268 and now. Validator currently PASS. d-90 closed without further action.

• d-90 · · · 2026-05-07T20:08:59.780175Z

  Validator FAIL on close: 21 kirlik-* entries (cataloged at 2026-05-07T20:08:14Z by a parallel agent) carry tags that are category names rather than subcategory ids: 'cognitive_workload' (5x), 'hmi' (10x), 'cognitive_psychology_foundations' (4x). The validator forbids using parent category names as tags. These entries are not from this session's Alam sweep — my new entries (alam-2008, alam-2009, alam-2011, nagisetty-2016, ibukun-2025) all use only subcategory ids and pass validation.

• d-89 · structure · executed · 2026-05-07T20:04:37Z

  CEO arbitration: ratified d-88 graduation of 4 trial researcher variants past sunset (sess-243). Removed trial: true / sunset_session: sess-243 from researcher_sweeper.yaml, researcher_scout.yaml, researcher_archiver.yaml, researcher_cogpsych.yaml; added graduated_session: sess-251, graduated_date: 2026-05-08; added concrete examples drawn from each variant's actual session record. Webwright (also listed in d-88) does not carry trial: true on disk per the un-retire (d-084), so no metadata change is needed there — it operates as a regular role.

  proposal: p-025

• d-88 · proposal · closed_duplicate · 2026-05-08T06:31:00Z

  Graduate 5 trial roles past sunset (sess-243): researcher_scout, researcher_archiver, researcher_cogpsych, researcher_sweeper, webwright. All earned their keep with rotation_count > 0 and concrete output visible in sessions.jsonl. Drop trial: true and sunset_session; append graduated_session: sess-249 to each role file.

  proposal: p-025

• d-87 · structural_review · open · 2026-05-08T06:30:00Z

  Periodic structural review (was due sess-243; auditor pulled at sess-249). Primary findings: (1) all 5 trial roles past sunset and earning their keep — graduate; (2) verifier role at 98 sessions but mostly no-ops since archived queue saturated; (3) author-name format warning at 493 distinct violators is gating the author-driven discovery loop; (4) parallel-agent session_id collisions raised in 4 consecutive gap_signals but no action taken; (5) discovery_diversity.last_reviewed still null since the block was added; (6) goal-succession violation (d-78/p-024) still open.

• d-086 · · · 2026-05-07T19:38:17.473709+00:00

  Validator FAIL on sess-228 close: 4 Pashler entries (pashler-1994/2001/2003/2007) have archive_status=archived but no local_path. Created by a parallel sweeper/cogpsych agent during this session burst. Webwright cannot resolve (out of role territory). Researcher_archiver should either (a) supply local_path + PDF on disk for these or (b) demote them to indexed.

• d-085 · role · executed · 2026-05-07T19:25:00Z

  Site live at https://drivingresearch.com (Cloudflare Pages project 'drivingresearch'). Switched target domain from agentupkeep.com to drivingresearch.com — both were on the same Cloudflare account; drivingresearch.com is the obvious match and was already a parked domain (HTTP 522). agentupkeep.com remains separate on Netlify. Custom domains attached via Pages API; cert provisioning automatic.

• d-084 · role · executed · 2026-05-07T18:30:00Z

  Webwright un-retired; site builder + Cloudflare Pages deploy adopted from webwright-build branch and repointed to agentupkeep.com (sunset of AgentUpkeep consulting concept; domain repurposed as DrivingResearch's public surface). p-018 rejection (sess-089) preserved in ledger; this entry overrides the rollback portion only.

• d-083 · workflow · executed · 2026-05-07T00:00:00Z

  Centralize scope filter into systems/scope.md (working-hypothesis framing); add discovery_diversity observability to status.yaml; auditor wires it into structural reviews; scout source list reframed as starting-point not fence.

• d-082 · · in_review · 2026-05-08T00:45:00Z
• d-081 · workflow · executed · 2026-05-07T18:00:00Z

  Layered memory architecture deployed. state/learnings.jsonl seeded with 3 learnings from sess-198..208 autonomous run. Task discipline tightened (exit_terminal + exhausted). Auditor mission expanded with learning-promotion. 5 researcher roles updated to consult learnings before external actions.

  proposal: p-026

• d-080 · proposal · accepted · 2026-05-07T18:00:00Z

  Add layered memory architecture: state/learnings.jsonl + tightened task discipline + auditor promotion path

  proposal: p-026

• d-079 · structure · executed · 2026-05-07T17:30:00Z

  4 trial researcher variants spawned (sweeper, scout, archiver, cogpsych); researcher rotation count 43 redistributed across 5 researchers at 9/9/9/8/8 (total preserved); active_roles expanded to 9.

  proposal: p-025

• d-078 · proposal · accepted · 2026-05-07T17:30:00Z

  Spawn 4 trial researcher variants and redistribute researcher rotation count (43 -> 9/9/9/8/8)

  proposal: p-025

• d-79 · structural_review · open · 2026-05-07T18:00:00Z

  Auditor territorial analysis: (1) goal succession violation — 5 completed goals lack successor fields; (2) indexed-pile access routes never rechecked; (3) verifier role has no sustainable workload after initial verification sweep.

• d-78 · proposal · open · 2026-05-07T18:00:00Z

  Goal succession cleanup — add successor fields to all 5 completed strategic goals lacking them (g-001, g-002, g-003, g-007, g-008)

  proposal: p-024

• d-077 · workflow · executed · 2026-05-07T05:40:00Z

  Archivist retired. Schema validation gate (validate.py PASS, duplicate/tag checks) absorbed as explicit Researcher constraint. Total retired roles: 10.

  proposal: p-023

• d-076 · role · accepted · 2026-05-07T05:40:00Z

  Retire Archivist role — absorbed into Researcher

  proposal: p-023

• d-075 · workflow · executed · 2026-05-07T17:00:00Z

  Goal-succession requirement added. When a goal closes, the closing session must surface a successor goal or write successor: 'none_needed: <reason>' on the goal.

  proposal: p-022

• d-074 · proposal · accepted · 2026-05-07T17:00:00Z

  Goal-succession requirement: closing a strategic goal must surface a successor or explicit none_needed

  proposal: p-022

• d-073 · role · executed · 2026-05-07T17:00:00Z

  Auditor mission rewritten to include territory analysis (gap detection, gap_signal scanning, trial-role disposition, goal-succession check). Structural-review cadence updated to every 30 days OR every 50 sessions. Validator AUDIT escalates when severely overdue and adds date-based check.

  proposal: p-021

• d-072 · proposal · accepted · 2026-05-07T17:00:00Z

  Rewrite auditor mission to include territory analysis; tighten structural-review cadence detection

  proposal: p-021

• d-071 · structure · executed · 2026-05-07T17:00:00Z

  Trial-role mechanism added. Role files may declare trial: true + sunset_session: <id>; sunset evaluated by auditor; lossless experimentation.

  proposal: p-020

• d-070 · proposal · accepted · 2026-05-07T17:00:00Z

  Add trial-role mechanism to lower the activation cost of role experimentation

  proposal: p-020

• d-069 · workflow · executed · 2026-05-07T17:00:00Z

  gap_signal field added to session-close convention. Optional one-line natural-language note on each session entry; auditor scans during structural review.

  proposal: p-019

• d-068 · proposal · accepted · 2026-05-07T17:00:00Z

  Add gap_signal field to session-close convention

  proposal: p-019

• d-067 · audit · open · 2026-05-07T05:10:00Z

  Audit findings from sess-189 auditor session.

• d-066 · structural_review · executed · 2026-05-06T19:30:00Z

  Structural review overdue by 72 sessions (due sess-107, now sess-179). Period summary: - 79 sessions (sess-101 to sess-179) - Catalog: static at 315 papers. No new entries added. - Composition: 76 verifier sessions, 1 researcher (sess-122), 1 cancelled researcher (sess-101), 1 auditor (this session) CRITICAL FINDING - Verifier Bug: 7 Treisman entries (archived with PDFs on disk) had verified_date set but verification=NULL. This caused 76 verifier sessions to cycle through the same entries endlessly, each picking up the same 7 as "needs verification" because the DB showed them as unverified. Root cause: UPDATE statements set verified_date without setting verification='verified'. FIXED in this session: updated 7 entries to verification='verified'. Result: all 94 archived entries now have verification='verified' (92) or 'verified_with_issues' (2). Session composition analysis: - sess-102..sess-121: 20 verifier sessions (first cycle through entries) - sess-122: researcher investigated Walter Schneider - found no match in English-language driver safety corpus. Recommended dropping from priority queue. - sess-123..sess-178: 56 verifier sessions (redundant re-verification due to bug) Key findings: 1. Walter Schneider sweep abandoned (sess-122): no coherent author profile exists in English-language driver safety literature. 7 authors named Schneider found, all European medical/engineering researchers in German-language venues. Drop from priorities. 2. Verifier ratio improved from 22.7 to 7.2 but was artificially inflated by 76 wasted sessions. True verifier count before this period: 12. After fix: all 94 archived verified. 3. No researcher productivity: 0 new papers cataloged in 79 sessions. 4. g-006 (author-driven discovery): 9 author sweeps completed but stalled. Donald Wegner sweep remains pending but no researcher session has attempted it. 5. Category distribution: cognitive_psychology_foundations dominates (171/315=54%), hmi second (79), remaining 10 categories share 65 entries. Recommendations: 1. Donald Wegner sweep needed for g-006 (next researcher session). 2. Verifier sessions should pause - all 94 archived entries are verified. Resume when new archived entries are added. 3. Consider formalizing g-006 follow-up: target sweeps for top-20 authors by paper count. Authors needing sweeps: Yang Jessie (49 papers), Na Du (34), Robert Lionel (32), Tilbury Dawn (24), Zhou Feng (16) - some already have sweeps. 4. Next structural review: sess-187 (10 sessions from this one).

• d-065 · structural_review · executed · 2026-05-06T19:40:00Z

  Structural review covering sess-101 through sess-179 (79 sessions since prior auditor sess-100). Composition: 76 verifier, 2 researcher, 1 auditor. Session homogeneity near-total:Verifier dominance is extreme—sessions 164-178 are 15 straight identical-verifier sessions doing 6-PDF SHA256 checks each. No researcher sessions in >40 sessions. Stuck sessions fixed: sess-074 (verifier, superseded), sess-075 (researcher, superseded), sess-099 and sess-101 (researcher, incomplete Schneider sweeps cancelled).

• d-064 · structural_review · executed · 2026-05-06T18:11:30Z

  Structural review overdue by 35 sessions (due sess-063, now sess-098). Period summary: - 36 sessions (sess-063 to sess-098) - Primary work: author-driven discovery (g-006 advancement) - Catalog growth: 159 -> 315 papers (+156) - Archive status: 94 archived (unchanged), 221 indexed (+149) - 9 author sweeps completed: Zhang, Yang, Zhou, Robert, Tilbury, Eysenck, Treisman, Broadbent, Kahneman - g-006 (author-driven discovery): advanced significantly with 9 sweeps, but acceptance criteria requires >=80% of in-scope publications by 20 most-cataloged authors Session composition: - Researcher: 14 sessions (primary work) - Verifier: 12 sessions (redundant re-checks, all no-op) - Auditor: 3 sessions - Archivist: 2 sessions Key findings: 1. Researcher productivity: 156 new entries cataloged, primarily indexed entries from author sweeps 2. g-006 progress: 9 author sweeps completed; Kahneman (91 entries), Treisman (43 entries), Broadbent (22 entries) are largest 3. g-008 extension: cognitive_psychology_foundations category now has 175 entries (up from 15) 4. Verifier workload: all 94 archived entries verified; 12 verifier sessions were redundant 5. Stuck sessions: sess-091-097 remained in_progress due to session fragmentation 6. YAML parse error: sess-090 fixed roles/_index.yaml colon issue Role evolution: - Researcher: most active role (37 sessions total) - Webwright: retired (p-018 rejected) - No new roles added Recommendations: 1. Prioritize Walter Schneider and Donald Wegner sweeps for g-008 extension 2. Consider formalizing g-008 extension as follow-up proposal (target >=50 cognitive_psychology_foundations entries) 3. Verifier sessions should pause until new archived entries are cataloged 4. Session management: avoid multiple in_progress sessions for same task

• d-063 · audit · executed · 2026-05-06T16:17:00Z

  Auditor session overdue by 32 sessions (cadence every 8). Period covers sess-058..sess-089 (last audit was sess-057). Session composition: 11 verifier sessions (sess-078..sess-088), 1 auditor session (sess-089, this session). Only 6 completed out of 11 verifiers. 5 completed out of 6 total completed sessions were verifier re-checks that found no discrepancies. Catalog state: 159 total entries. 87 archived+verified, 72 indexed. Zero discovered, zero pending, zero unverified archived. No new entries added during this period. Verifier flood: 13 verifier sessions total (including sess-058) against 87 entries. Ratio improved from 14.5 to 12.2 but target is ≤5. All 87 archived entries are verified. Future verifier sessions are redundant until new archived entries are cataloged. Stuck sessions: 5 verifier sessions were stuck in in_progress with no remaining work. Fixed by this session. Goal status: g-001..g-003 completed (foundational goals). g-006 in_progress (author sweeps not yet started). g-007 meets criteria (72 indexed entries ≥ 30). g-008 meets criteria (15 cog-psych entries ≥ 15). Action items: - g-007 and g-008 marked completed in status.yaml - Webwright retired (p-018 rejected) - structural_reviews refreshed to sess-089 - Prioritized: researcher sessions needed for g-006 author sweeps and g-008 extension toward ≥30

• d-062 · structural_review · executed · 2026-05-06T16:17:00Z

  Structural review overdue (due sess-063, now sess-089, 34 sessions overdue). Covers: session activity, role rotation, catalog health, goal status, stuck sessions. Key findings: 1. Heavy verifier flood (sess-078..sess-088): 11 consecutive verifier sessions attempting to re-verify 87 archived entries. Only 5 of 11 closed as completed; 6 remained stuck in in_progress (now closed by this session). The verifier workload was actually exhausted after sess-058 — all 87 archived entries were already verified. 2. No researcher/archivist sessions since sess-055 (the final session of p-012 cascade). Only verifier and auditor since then. No new discovery, no new author sweeps, no indexed entries created, no summaries written. 3. Verifier independence ratio is 12.2 (target ≤5). 13 verifier sessions over 159 total cataloged entries. Improvement from 14.5 but still far from target. 4. g-007 (paywalled indexing): acceptance criteria requires ≥30 paywalled indexed entries with access_route. Currently 72 indexed entries, all with access_route populated. Acceptance criteria met. Marked completed. 5. g-008 (cognitive psych): acceptance criteria requires ≥15 cataloged entries in cognitive_psychology_foundations. Currently 15 entries. Acceptance criteria met (barely). Extensions to ≥30 mentioned in status.yaml priorities but not yet formalized as follow-up proposal. 6. g-006 (author-driven discovery): 15 cog-psych entries exist but no evidence of author sweeps in state/author_sweeps/. The directory exists with _README.md but zero sweeps completed. Still in_progress. 7. p-018 (webwright) filed as proposal but never accepted; webwright remains zombie with 0 sessions across 34 sessions. Rejected and retired.

• d-059 · structure · executed · 2026-05-06T16:17:00Z

  Webwright retired. Removed from active_roles, rotation.counts, and _index.yaml. Role file moved to _retired/. p-018 marked rejected.

  proposal: p-018

• d-058 · structural_review · rejected · 2026-05-06T16:17:00Z

  Reject p-018 (webwright role proposal) and retire webwright role. Role was added in sess-055+ as part of a proposal cascade but never executed: rotation.count remains 0 across 34+ sessions. No web surface was ever built, no site directory exists, no deploy was attempted. The role remains zombie and p-018 remains open.

  proposal: p-018

• d-061 · proposal · open · 2026-05-06T12:09:00Z

  Add webwright role and private web surface at drivingresearch.com

  proposal: p-018

• d-060 · taxonomy · · 2026-05-05T12:35:00Z

  Rename category id from fatigue_alertness to fatigue in _taxonomy.yaml and update all paper entries.

• d-056 · workflow · executed · 2026-05-05T01:10:00Z

  Foundational docs synchronized with live state. CLAUDE.md: lines 21-25 rewritten to describe schema 2.0 as live with indexed-tier no-fabrication rules; line 47 updated (currently 1.2 → 2.0, legacy 1.2). status.yaml: priorities #1-#2 collapsed into cascade-complete line with genuine next priority (pilot new workflows). CLAUDE.md acquisition.md line left as-is — that file was retired to _retired/ in sess-022 and the layout reference is historical context.

  resolves: p-017

• d-055 · decision · accepted · 2026-05-05T01:10:00Z

  Auditor finding is correct. CLAUDE.md and status.yaml described pre-cascade state as live after all five p-012 phases were executed. Doc drift is structural — a new agent reading the workspace would form an incorrect model. CEO accepts on owner authority.

  resolves: p-017

• d-054 · structural_review · open · 2026-05-05T01:00:00Z

  Foundational docs lag the p-012 cascade. CLAUDE.md (workspace root) and state/status.yaml priorities still describe the pre-cascade state as the live state — schema 1.2 'currently enforces archived-only cataloged', schema 2.0 'queued as a follow-up proposal', 'Next CEO priority: file Phase 2 proposal'. Schema 2.0 is live; Phases 2–5 of p-012 executed in sess-052..055.

  proposal: p-017

• d-053 · workflow · executed · 2026-05-04T23:45:00Z

  Researcher author-sweep workflow deployed. roles/researcher.yaml rewritten: mission covers two-tier (archived + indexed) work; constraints updated for schema 2.0; new author_sweep_mode block describes the OpenAlex/Semantic-Scholar/ORCID/CrossRef workflow with 90-day re-sweep cadence. state/author_sweeps/ created with _README.md documenting the schema 1.0 tracker format (required fields, per-publication shape, filename convention). validate.py extended to parse and validate author-sweep yamls. memory/overview.md updated with Author sweeps section. structural_reviews bumped to sess-063 — the p-012 cascade IS this period structural review. validate.py PASS.

  resolves: p-016

• d-052 · decision · accepted · 2026-05-04T23:45:00Z

  Owner directive ('Do what is left'). Phase 5 closes out p-012's migration plan and operationalizes g-006.

  resolves: p-016

• d-051 · proposal · open · 2026-05-04T23:45:00Z

  Phase 5 of p-012: Researcher gains author-sweep mode. Update roles/researcher.yaml mission and constraints to cover the two-tier (archived + indexed) work and add an explicit author_sweep_mode workflow block. Create state/author_sweeps/ directory with _README.md documenting the per-author tracker yaml format (schema 1.0). Add light validate.py check on author-sweep yamls (parse, required top-level fields, author_slug matches filename). Update memory/overview.md to mention the new state directory. This implements g-006.

  proposal: p-016

• d-050 · taxonomy · executed · 2026-05-04T23:30:00Z

  Added cognitive_psychology_foundations category with 6 subcategories (attention_control, executive_function, dual_task_theory, working_memory, perceptual_processing, individual_differences). Globally unique subcategory ids verified against existing taxonomy. validate.py PASS — 78 archived entries unaffected.

  resolves: p-015

• d-049 · decision · accepted · 2026-05-04T23:30:00Z

  Owner directive ('Do what is left'). Taxonomy is the second prerequisite (alongside schema 2.0) for cataloging the upstream cognitive-psych literature that g-008 targets.

  resolves: p-015

• d-048 · proposal · open · 2026-05-04T23:30:00Z

  Phase 4 of p-012: add cognitive_psychology_foundations category to _taxonomy.yaml. Subcategories: attention_control, executive_function, dual_task_theory, working_memory, perceptual_processing, individual_differences. Implements g-008. The new category captures the upstream cognitive-psych literature that feeds driver safety research — foundational psych work, not driving studies. Driver-side counterparts already exist under cognitive_workload; entries can tag both when relevant.

  proposal: p-015

• d-047 · validator · executed · 2026-05-04T23:00:00Z

  validate.py and build_index.py updated for schema 2.0 indexed tier. Cataloged invariant now permits archive_status in {archived, indexed}. Indexed entries: access_route required; 17 forbidden fields enforced (PDF-grounded, PDF-tier, verification). build_index.py: needs_verification restricted to archived; new indexed_paywalled queue; access_route/purchase_url/paywall_publisher surfaced in per-paper items. validate.md two-tier rule documented. 78 existing archived entries pass unchanged.

  resolves: p-014

• d-046 · decision · accepted · 2026-05-04T23:00:00Z

  Owner directive ('Do what is left'). Phase 3 is the prerequisite for any indexed entry to actually reach cataloged status — without it, schema 2.0's indexed tier sits dormant.

  resolves: p-014

• d-045 · proposal · open · 2026-05-04T23:00:00Z

  Phase 3 of p-012: update systems/validate.py and systems/build_index.py to mirror schema 2.0. Extend valid_archive set to include 'indexed'. Relax cataloged invariant to permit archive_status in {archived, indexed}. For indexed entries, require access_route and forbid the PDF-grounded fields (summary/key_finding/methodology/sample_size/measures/quality_score), PDF-tier fields (local_path/pdf_sha256/pdf_text_length/archive_method/archive_url_used/archived_at/license), and verification fields. Restrict build_index.py's needs_verification queue to archived entries only (indexed entries have no PDF to verify against). Add new index queue indexed_paywalled. Surface access_route, purchase_url, paywall_publisher in the per-paper index item.

  proposal: p-014

• d-044 · schema · executed · 2026-05-04T22:30:00Z

  Schema 1.2 → 2.0 deployed. archive_status enum extended to include 'indexed'; 4 new optional fields added (access_route, purchase_url, purchase_price, paywall_publisher); legacy_versions includes 1.2; catalog-invariant comment block rewritten to document the two-tier rule and the no-fabrication forbiddances on indexed entries. validate.py + build_index.py still operate under the schema-1.2 cataloged invariant (archived only) until Phase 3 lands; existing 78 archived entries continue to validate.

  resolves: p-013

• d-043 · decision · accepted · 2026-05-04T22:30:00Z

  Owner directive ('Do what is left') collapses deliberation. Schema 2.0 is the prerequisite for everything paywalled-related under g-007 and p-012's Phase 2-6 cascade.

  resolves: p-013

• d-042 · proposal · open · 2026-05-04T22:30:00Z

  Phase 2 of p-012: bump schema 1.2 → 2.0. Add archive_status='indexed' for paywalled entries with no PDF on disk. New optional fields: access_route (required when indexed), purchase_url, purchase_price, paywall_publisher. Relax cataloged invariant: pipeline_status=cataloged is permitted when archive_status=archived (existing path) OR archive_status=indexed AND access_route is populated. Forbid summary/key_finding/methodology/sample_size/measures/quality_score on indexed entries (no fabrication). Existing 78 entries are all 1.2 archived; they remain valid via legacy_versions.

  proposal: p-013

• d-041 · structure · executed · 2026-05-04T22:00:00Z

  Phase 1 of p-012 executed: foundational scope flipped from 'open-access only' to two-tier comprehensive coverage (archived OA + indexed paywalled). Added g-006 (author-driven discovery), g-007 (paywalled indexing schema, requires schema 2.0), g-008 (cognitive-psychology foundations taxonomy). Existing 78 entries unaffected; schema fields, validator, and taxonomy structure unchanged this session and queued as Phase 2-5 proposals.

  resolves: p-012

• d-040 · decision · accepted · 2026-05-04T22:00:00Z

  Owner directive in real-time: 'It should catalogue EVERY driver safety paper and every paper that is relevant and related to cognitive psych factors that factor into this and then pull EVERY open source paper available but not those that are not open source where to purchase them. The company becomes the authoritative reference on this material. One systematic approach is to find authors and grab every one of their papers that are relevant. All publishing authors. All relevant papers. A complete knowledge base.' Owner authority collapses the deliberation window. Phase 1 lands this session; subsequent phases get their own proposals so the 78 existing entries migrate cleanly across schema 1.2 → 2.0 without breaking the no-fabrication invariant.

  resolves: p-012

• d-039 · proposal · open · 2026-05-04T22:00:00Z

  Re-found scope from 'open-access only' to comprehensive coverage of driver safety + the cognitive-psychology foundations that feed it. Two storage tiers: 'archived' (open-access PDF on disk, full record) and 'indexed' (paywalled, no PDF, skeleton record with purchase or alternative-access route). No-fabrication invariant preserved — only archived entries can carry summary/key_finding/methodology/sample_size. Discovery strategy: author-driven systematic sweeps. Goal: become the authoritative agent-queryable reference on driver safety from the human-factors and cognitive-psychology perspective.

  proposal: p-012

• d-038 · review_request · resolved · 2026-05-04T21:15:00Z

  sess-049 verification_notes are mass-stamped boilerplate with no page citations; functionally self-attested. 31 of 39 affected entries remain.

• d-037 · proposal · open · 2026-05-04T18:30:00Z

  g-005: Verifier independence. Verifier role runs frequently enough that no entry's verification: verified is self-attested by the same agent that wrote the summary.

  proposal: p-011

• d-036 · proposal · open · 2026-05-04T18:30:00Z

  g-004: Discovery infrastructure repair. Restore at least 3 working OA sources beyond arXiv.

  proposal: p-010

• d-035 · structure · executed · 2026-05-04T18:30:00Z

  Role collapse executed: 8 retired, 1 renamed (scout→researcher with expanded scope), 1 added (auditor). validate.py emits AUDIT findings. structural_reviews bumped to sess-055.

  resolves: p-009

• d-034 · decision · accepted · 2026-05-04T18:30:00Z

  Owner directed: 'lets fix the company.' Audit findings are concrete and observable in rotation counts and session log. The collapse honestly reflects the work pattern; the auditor role and AUDIT warnings prevent recurrence by making governance drift visible at every validate run.

  resolves: p-009

• d-033 · proposal · closed · 2026-05-04T18:30:00Z

  Collapse role inventory from 13 active to 6: retire 8 zombie roles (chief_researcher, knowledge_engineer, research_acquirer, research_analyst, distraction_researcher, fatigue_researcher, hmi_researcher, crash_data_researcher), rename research_scout to researcher (absorbs acquirer+analyst), add auditor role, wire AUDIT warnings into validate.py.

  proposal: p-009

• d-032 · review_request · closed · 2026-05-04T15:00:00Z

  Closed stale review_request d-021 (systematic DOI verification). All 10 affected entries were retired in the catalog purge (p-008, sess-012). The DOI issues were symptomatic of the broader problem: the initial catalog was model-fabricated metadata. The purge resolved the underlying issue by requiring all future entries to come from real OA APIs with archived PDFs.

• d-031 · workflow · executed · 2026-05-03T20:35:32Z

  Pulled prescriptive path-instructions out of systems and roles. Added scripts/README.md as single tool index. Retired systems/acquisition.md (the shell-pipeline waterfall is obsolete now that scripts/acquire.py exists). Slimmed systems/orientation.md to drop the 6-step ritual. Trimmed verbose role files to mission + work_examples + invariants + demand_signals. Runner prompt minimized to goal + pointers. Goal: give agent a clear toolkit and let it figure out its own path.

• d-030 · data · executed · 2026-05-03T20:26:58Z

  All 15 fabricated entries moved to _retired/. Catalog now empty. scripts/discover.py + acquire.py + screen.py built; will be used to populate the catalog from real OA sources.

  resolves: p-008

• d-029 · decision · accepted · 2026-05-03T20:26:58Z

  Owner directed the purge explicitly. Real-papers-only is the only honest path given how reliably the model fabricates plausible bibliographic metadata.

  resolves: p-008

• d-028 · proposal · open · 2026-05-03T20:26:58Z

  Purge all 15 catalog entries: they were model-fabricated metadata. Reset to empty catalog. Pivot to discovery-first model: only add entries that are real papers found via direct OA APIs (PMC, arXiv, AAA Foundation, NHTSA, DOAJ).

  proposal: p-008

• d-027 · infrastructure · executed · 2026-05-03T20:30:00Z

  Index generator + validator wiring + role/orientation updates landed. First run produced the index; surfaced one real warning (mckeever-2015 / scott-hayward-2015 share a normalized title — already flagged in d-021 as a likely fabricated/conflated authorship case). validate.py PASS.

  resolves: p-007

• d-026 · decision · accepted · 2026-05-03T20:30:00Z

  Owner directed: 'implement the very best solution given the needs and constraints.' The index is a small, derivative artifact that closes a real agent productivity gap (per-session entry scans) and a real data-loss gap (rejected papers leaving no trace). Per-entry YAMLs stay the source of truth, so the rollback cost is low.

  resolves: p-007

• d-025 · proposal · closed · 2026-05-03T20:30:00Z

  Grand archive: Data/research/_index.yaml as derived snapshot of every entry; systems/build_index.py builds it; validate.py regenerates and checks it; Scout records in-scope-but-no-OA papers as archive_status: unavailable so the index becomes dedup memory across sessions

  proposal: p-007

• d-024 · structure · executed · 2026-05-03T16:30:00Z

  Restructure executed: per-entry YAML, schema 1.2, 5 new roles (research_acquirer + 4 domain researchers), runtime validator, OA-only invariant, 15 entries quarantined to pending, distraction synthesis moved to _pending_redo

  resolves: p-006

• d-023 · decision · accepted · 2026-05-03T16:30:00Z

  Owner directed both the OA-only constraint and the more-researchers shift. The validate.py filesystem check is the load-bearing piece that makes the open-access invariant enforceable rather than aspirational.

  resolves: p-006

• d-022 · proposal · open · 2026-05-03T16:30:00Z

  Open-access-only restructure: schema 1.2 with archive fields, per-entry YAML migration, 4 domain researcher roles + research_acquirer, runtime validate.py, locking.md

  proposal: p-006

• d-021 · review_request · open · 2026-05-03T16:00:00Z

  Systematic DOI verification: 4 disputed (wrong DOIs), 6 unverifiable (non-standard or inaccessible). No entry is fully verified. Requires editorial correction of at least 4 DOIs and resolution of 6 unverifiable sources.

• d-020 · synthesis · executed · 2026-05-03T15:50:00Z

  Created first cross-study synthesis: Data/research/summaries/distraction.md covering 10 distraction catalog entries. Populated related_ids on all 10 distraction entries (31 cross-references total). Synthesis identifies: (1) Cognitive bottleneck as dominant mechanism across all phone-related distraction. (2) Crash risk multipliers: texting 2.3-6x, cell phone 1.5-4x. (3) Coverage gaps in novice driver distraction, infotainment, navigation systems, passenger conversation, auditory distraction.

• d-019 · catalog · executed · 2026-05-03T15:35:00Z

  Retroactive archivist action: moved 10 entries from in_review to cataloged. All entries pass schema/taxonomy validation. sess-004 (research_analyst) logged moving to in_review, but a subsequent archivist session executed the cataloged handoff without leaving a session log. Data drift corrected retroactively.

• d-018 · catalog · executed · 2026-05-03T15:00:00Z

  Fixed all 10 catalog entries: added missing schema_version (1.1) and corrected 3 invalid taxonomy tags (glance_duration -> temporal, technology_impact -> mobile_phones, crash_data -> naturalistic_driving). Also removed duplicate mobile_phones tag from cummings-2016-distracted-driving.

• d-017 · decision · accepted · 2026-05-03T13:42:00Z

  Audit found real collisions that would cause confusion during normal operation. Fixing them now, before catalog entries flow through, is cheaper than migrating later. The in_review state is optional (no existing entries use it), so no migration burden.

  resolves: p-005

• d-016 · proposal · closed · 2026-05-03T13:42:00Z

  Fix pipeline collisions, role overlaps, and throughput signals from the research restructuring audit

  proposal: p-005

• d-015 · role · executed · 2026-05-03T13:40:00Z

  Research restructuring implemented: created Scout, Analyst, Synthesizer roles; refactored Chief Researcher as lead; refactored Knowledge Engineer as pure infrastructure. Research team now 4 roles (50% of org). All files updated.

  resolves: p-004

• d-014 · decision · accepted · 2026-05-03T13:40:00Z

  Owner explicitly identified that the org was 'bureaucratically heavy' with 'not enough boots on the ground.' Doubling research capacity from 1 role to 4 is the structural fix. The Scout → Analyst → Synthesizer pipeline creates real throughput.

  resolves: p-004

• d-013 · proposal · closed · 2026-05-03T13:40:00Z

  Restructure research function: retire original Chief Researcher definition, replace with 4-role research team (Chief Researcher as lead + Scout + Analyst + Synthesizer). Absorb synthesis work from Knowledge Engineer into Research Synthesizer. Net effect: research producers go from 20% of org to 50%.

  proposal: p-004

• d-012 · role · executed · 2026-05-03T14:00:00Z

  Verifier role trimmed and aligned; summary frontmatter pattern defined; not_checked redundancy resolved; g-001 amended to require verified status

  resolves: p-003

• d-011 · decision · accepted · 2026-05-03T14:00:00Z

  All six issues are clarity/consistency fixes with no downside. Owner directed application.

  resolves: p-003

• d-010 · proposal · closed · 2026-05-03T14:00:00Z

  Verifier role clarity pass: trim verbosity, align with employee-owned voice, define summary verification path, resolve not_checked redundancy, amend g-001 to require verification

  proposal: p-003

• d-009 · decision · accepted · 2026-05-03T13:38:43Z

  Owner (Joel) identified the gap explicitly: the company does not fabricate. A dedicated role for provenance and factual verification is a structural necessity, not optional. The zero-tolerance standard aligns with the company mission.

  resolves: p-002

• d-008 · proposal · closed · 2026-05-03T13:38:43Z

  Add Verifier role: zero-tolerance provenance and factual accuracy verification for all catalog entries

  proposal: p-002

• d-007 · taxonomy · executed · 2026-05-03T07:30:00Z

  Clarified that tags are subcategory ids drawn from _taxonomy.yaml; added evolution rules and retired[] slot for redirected ids

  resolves: p-001

• d-006 · schema · executed · 2026-05-03T07:30:00Z

  Schema bumped 1.0 -> 1.1: added required pipeline_status (discovered|in_progress|cataloged), required schema_version on each entry, optional abstract; added legacy_versions and deprecated_fields slots; documented quality_score_rubric inline

  resolves: p-001

• d-005 · workflow · executed · 2026-05-03T07:30:00Z

  Evolution mechanism installed and wired through orientation, validation, all role definitions, schema, taxonomy, and CLAUDE.md

  resolves: p-001

• d-004 · decision · accepted · 2026-05-03T07:30:00Z

  Owner (Joel) directed the audit and explicitly requested mechanisms that function rather than just descriptions. The validation gate is the enforcement that makes evolution reliable.

  resolves: p-001

• d-003 · proposal · closed · 2026-05-03T07:30:00Z

  Introduce evolution mechanism (systems/evolution.md, systems/validate.md), per-entry schema_version, _retired/ directories, proposal-decision-executed ledger pattern, and load-bearing session-close validation gate

  proposal: p-001

• d-002 · structure · · 2026-05-03T00:01:00Z

  Added periodic structural review to CEO responsibilities: every 8 sessions or monthly, whichever comes first. Review assesses role fit, workflow health, and log patterns; updates structure as needed. All changes logged in decisions.jsonl.

• d-001 · structure · · 2026-05-03T00:00:00Z

  Created role-based company structure with 4 initial roles: CEO, Chief Researcher, Archivist, Knowledge Engineer