cdeust/Cortex

• release: v3.15.0 — E1 v3 verification campaign + arXiv-ready papers + BEAM-10M harness
• style: drop unused id_to_cond local in judge.py (CI fix)
• style: ruff check --fix unused imports in BEAM-10M harness (CI fix)
• style: ruff format BEAM-10M harness + tests (CI fix)
• feat(verif): wire BEAM-10M head-to-head live mode (smoke pending API keys)
• feat(verif): BEAM-10M LLM head-to-head harness scaffold (Stage 0)
• docs: profile README draft points AI Architect to website not archived repo
• docs(paper): arXiv submission readiness — refresh both papers + endorsement template
• fix(paper): recompile with bibtex pass — all 45 citations now resolve
• docs(verif): E1 v3 LoCoMo post-plasticity-fix integration — paper §6.3 third pass + README/CLAUDE.md sync
• docs: profile README draft for cdeust/cdeust (controls AI Overview narrative)
• style: ruff format run_e1_v3_locomo.py (CI fix)
• chore(verif): redirect LoCoMo driver to locomo_v3_post_plasticity_fix output
• docs(paper): add compiled thermodynamic memory paper PDF (26 pages)
• docs(paper): compile thermodynamic memory paper PDF (26 pages)
• docs(paper): §6.3 second pass — LoCoMo subsection + plasticity-fix narrative
• feat(verif): E1 v3 LoCoMo ablation results — 14-row two-baseline evidence
• fix(verif): apply_hebbian_update preserves result-shape contract on ablation
• style: ruff format + remove unused imports (CI fix)
• fix(backfill): discover_files walks all four session layouts (issue #15)
• docs(verif): integrate E1 v3 LME-S evidence into §6.3 + cadence-fix narrative
• feat(verif): E1 v3 LoCoMo driver — 14-row two-baseline sweep
• fix(verif): consolidation cadence uses ingested_at instead of wall-clock created_at
• feat(verif): --ablate + --with-consolidation flags for LoCoMo harness
• feat(verif): VADER -> user_mood EMA hook in remember (closes MOOD_CONGRUENT signal gap)
• feat(verif): E1 v3 LME-S per-category delta analysis (mechanism specialization)
• fix(verif): user_mood DDL comment semicolon + test uses dominant beta
• feat(verif): E1 v3 LongMemEval-S ablation results — 17-row evidence dataset
• feat(verif): wire PgMemoryStore.get_user_mood for MOOD_CONGRUENT_RERANK
• feat(verif): blend-weight calibration results — six engineering defaults stand
• feat(verif): --with-consolidation flag for LongMemEval-S harness
• feat(verif): wire RECONSOLIDATION into recall post-retrieval path (Nader 2000)
• chore(verif): harness dirty-check ignores submodule internal state
• chore(verif): finalize E1 v2 ablation archive + match harness dirty-check to pre-reg
• feat(verif): blend-weight calibration infrastructure + pre-registration
• feat(verif): wire EMOTIONAL_RETRIEVAL + MOOD_CONGRUENT_RERANK as live read-path stages
• fix(verif): extend query-entity resolution to natural-language tokens (engineer follow-up)
• fix(verif): batch Hopfield embeddings + real entity-set Jaccard for dendritic stage
• feat(verif): wire HOPFIELD/HDC/SPREADING_ACTIVATION/DENDRITIC_CLUSTERS into pg_recall pipeline
• fix(verif): handler-level read-path ablation guards in recall.py composition root
• fix(verif): wire CORTEX_ABLATE_ env-var reads into production hot-paths (23 mechanisms)
• fix(schema): remove '; ' from comment that broke ddl.split(';') statement extractor
• docs(paper): integrate full E2b Zipf curve into §6.4 — N=100k datapoint landed
• docs(reorg): rename arxiv/ -> arxiv-context-assembly/ + move paper-md into docs/papers/
• fix(ci): repair docstring boundary in cls.run_cls_cycle (broke in 3eab1ed)
• docs(paper): add §6.4 Operating regime — reframe falsifications as predicted boundaries
• feat(verif): rebuild E2 N-scan as real-benchmark subsample + Zipf synthetic
• fix(verif): wire E2 N-scan ablation env vars into production code path
• plan(bench): BEAM-10M LLM h2h v3 — Haiku 4.5 then Gemini 2.0 Flash sequenced
• plan(bench): BEAM-10M LLM head-to-head — pivot to lightweight generator panel
• docs(paper): prose polish (paper-writer) + BEAM 0.543->0.591 number fix
• fix(paper): BEAM Overall 0.543 -> 0.591 in CLAUDE.md + markdown source
• docs(paper): port thermodynamic-memory paper to LaTeX matching arxiv/main.tex style
• fix(ci): ruff format on verification harnesses + bump tool count to 47
• feat(verification): paper + reproducibility infrastructure for thermodynamic memory claims
• chore(format): ruff format on memories_page + memories_facets
• feat(viz): default landing = Knowledge; Graph tab gains warning banner
• revert(viz): restore Graph view to pre-d3-removal state
• feat(viz): paged Knowledge + Board with filter chips, lazy-load, graph banner
• fix(viz): tilemap auto-recovers when /api/quadtree returns no_layout
• test(viz): update open_visualization tests for tilemap pipeline
• feat(viz): drive prepare-then-render entirely from cortex:open_visualization
• feat(viz): server-tile + Datashader path for >1M-node graphs (CPU-only, no CUDA)
• style: ruff format on workflow_graph_source_ast.py
• fix(ast): uncap L6 symbol/edge ingestion; surface file-import chain

Full Changelog: https://github.com/cdeust/Cortex/compare/v3.14.12...v3.15.0