Conversation
Per CS3321 final-deliverables roundtable convergence, sweep all gradable artifacts to remove drift between completed materials and the merged codebase. No code changes; this commit edits and annotates existing material so it can be reused unchanged in the final report. Document fixes: - docs/07: full rewrite against current backend/app/api + services + frontend/src/pages + CLI + evaluation framework + Neo4j optional. - docs/03: enum sweep — memory_type 12 values, status 7 values; add embedding caches; expand MemoryItem/Wiki/Forget/AuditLog rows. - docs/02: source-import audit claim corrected to imported_at / imported_by_user_id; recall DFD expanded to hybrid + retrieval_info. - docs/06: v_agent_visible_memory description requires caller-side WHERE agent_id; SQLite reclassified as design-only. - docs/normalization §1.4: trg_memory_before_update increments, trg_memory_after_update writes revision/audit. - docs/00: title broadened to organization/team memory database; P0/P1/P2 -> delivered-core / delivered-extension / future. - docs/01: admin role rewritten; Entity/Scene/Forget reclassified P1; Graph/Embedding/CLI/Eval/QA added. - docs/04: ER adds WorkspaceMember, MemoryEmbedding, SourceChunkEmbedding; conflict edges split left/right. - docs/05/08/09/10/11: edits per gap 9 review findings; docs/10 rebuilt as coverage matrix against actual backend/tests/test_*.py. - docs/index-rationale: 8 categories -> 9. - docs/explain-analyze: all four cases re-run on fresh demo DB (memory_item=25, audit_log=39, source_chunk=20, wiki_page=3); §3.2 re-classified as B-tree competing path; §3.3 BRIN path confirmed. SQL annotation pass (teacher requirement "具有注释的源程序"): - database/01_schema_core.sql, 02_schema_memory.sql, 03_schema_governance.sql, 05_views.sql, 06_triggers.sql each gain business-meaning comments above CREATE TABLE / VIEW / TRIGGER blocks. SQL fixes: - database/08_demo_queries.sql query #5: replace stale app.agent_id set_config with explicit WHERE agent_id and explanatory comment. - database/10_governance_demo_fixture.sql: drop stale "(如有)" hedge now that the walkthrough doc exists. New supporting documents: - docs/gap9-audit-findings.md: codex audit (26 files reviewed). - docs/gap9-audit-review.md: claude review approving findings. - docs/governance-demo-walkthrough.md: walkthrough for the governance fixture (db:setup, verification SQL, extraction audit note). Validation: - pytest backend/tests: 166 passed - ruff check backend/app backend/tests evaluation: pass - cd frontend && npm run lint && npm run build: pass - npm run db:setup: pass; governance fixture counts and 3 wiki pages verified
Six diagrams under docs/final-assets/diagrams/ covering core ER, full ER, business sequences, and the memory_item status state machine. Mermaid sources (.mmd) are the source of truth; SVG outputs are committed for direct PDF/PPT embedding. - 01-er-core: 8-entity lifecycle ER with attribute boxes (workspace, source_document, source_chunk, memory_item, memory_evidence, memory_revision, wiki_page, audit_log) for report §3. - 02-er-full: 25-entity full ER with all relationships (semantic, governance, runtime) for the E-R appendix. - 03-source-to-wiki: end-to-end business sequence — import, chunk, rule-based candidate extraction, approve, revision, wiki export; trigger-driven current_revision_no and audit_log writes shown. - 04-recall: recall path with v_agent_visible_memory permission filter, GIN lexical + optional JSONB vector retrieval, context pack assembly, recall_log with retrieval_info and fallback_reason. - 05-governance: three governance flows — conflict, forget, revision — showing trigger paths and audit_log writes. - 06-memory-status: stateDiagram-v2 (direction LR) for the 7-state memory_item.status lifecycle enforced by the CHECK constraint; labels kept short (service paths explained in report text). README.md documents re-render instructions and a per-file map linking each diagram to its target chapter in the final report. Rendered with @mermaid-js/mermaid-cli 11.15.0, theme=neutral.
Rename conflict edges left_conflict/right_conflict -> left/right and wiki edges projects_to/projects_to -> scene_page/memory_page so the auto-layout has more room near MEMORY_ITEM. Diagram remains the full 25-entity ER; only edge labels changed.
Audit against backend/app/api, services, and database/01-06. Fixed
contract errors and added missing foreign-key edges.
- 05-governance: replace fictitious /api/governance/* paths with real
/api/conflicts and /api/forget-requests; rename service methods to
create_conflict / update_conflict / create_forget_request /
update_forget_request; show real trigger flow
(trg_conflict_after_insert / trg_conflict_after_update,
fn_conflict_mark_memory_conflicted /
fn_conflict_restore_memory_if_clear); use real audit action names
(conflict.create, forget_request.update, etc.); add alt branch for
"no remaining open conflicts" vs "still conflicted elsewhere".
- 03-source-to-wiki: candidate approve endpoint corrected to
/api/memory-candidates/{id}/approve; evidence_role changed from
"supporting" to "source" (matches CHECK enum); response types
corrected; add wiki.revision.insert audit via
trg_wiki_revision_after_insert and trg_memory_after_insert for
candidates.
- 06-memory-status: remove illegal archived -> active restore (no
service code supports it); rename conflicted -> superseded edge to
"manual merge" (supersede is a separate editor action, not a
conflict-trigger output).
- 04-recall: split permission filter into "agent_id provided" vs "no
agent_id" branches (the latter falls back to access_level filter
rather than v_agent_visible_memory); clarify lexical retrieval
always runs even in vector/hybrid mode.
- 02-er-full: add 11 missing FK-backed edges (owner_user_id /
owner_agent_id / created_from_doc_id / superseded_by_memory_id on
memory_item; started_by_user_id on agent_session; reply_to_message_id
on message; session_id and imported_by_user_id on source_document;
memory_id and doc_id on timeline_entry; agent_id and user_id on
recall_log; requester_user_id and reviewed_by_user_id on
forget_request). File comment changed to "main persisted/logical
relationships" rather than claiming exhaustive polymorphic coverage.
- README.md: refresh per-file descriptions to match corrected
contracts.
Verification: API paths, service methods, trigger names, audit action
names, evidence_role enum, and FK targets all verified via grep
against backend/app and database/. All 5 modified .svg files
re-rendered with mermaid-cli 11.15.0; visual PNG check on 02/05/06
confirmed layout and content.
UI captures (21 PNG), SQL/EXPLAIN evidence (4 PNG + 4 SVG with full text-wrapped output, no thumbnail clipping), pytest/Vite build (2 PNG + 2 SVG), and raw command logs (6 .txt) for the final report and PPT. The optional QA path is verified with a controlled DeepSeek configuration (LLM_PROVIDER=deepseek, deepseek-v4-flash); no API key is stored in the repo. README inventories every artifact with URL/action, what-it-proves, and where to use it. Two-version captures (graph-explorer, runtime-hybrid-search) retain the pre-click state alongside the loaded-data state.
The earlier gap4 commit silently dropped 6 raw .txt logs because the global `logs/` ignore rule (intended for runtime app logs) matched them. README references these files as the canonical source the SQL/test PNG screenshots are rendered from; without them the reproducibility chain breaks. Add a negation rule `!docs/final-assets/screenshots/logs/` so future doc-evidence logs are tracked while runtime logs remain ignored.
Cross-branch commit audit of three identified members (hopecommon, dzx0902, lywzc0419 = "huiyijian") plus a placeholder for the unidentified fourth member. Each member has a raw commit TSV (sha/date/kind/branches/file_count/top_level/ subject) and a Chinese report-ready draft. The recall/search path is collaborative and has an explicit boundary note: hopecommon owns lexical search + recall context-pack formatter + CLI dogfood; dzx0902 owns hybrid recall, embedding, QA, and structured context-pack metadata. Graph hardening is attributed to hopecommon over lywzc0419's initial Neo4j integration. Evaluation framework + LoCoMo Bench credit goes primarily to dzx0902. Cautions: 9 local-only commits not yet on origin/jflin; merge commits are process not implementation evidence; GitHub issue state is not a reliable completion source.
feat(evaluation): add resumable benchmark judging and temporal memory evaluation
Slides cover: project positioning, research landscape, core data link, ER + normalization, 8 index types + EXPLAIN evidence, views/triggers, system architecture, three demo flows (Source→Memory→Evidence, Recall + Context Pack, Governance + Graph), test coverage, LongMemEval 500-case results (35.2% / 58.4%, framed as engineering validation only), eight innovation points, team contribution + future work. Built from slides-weasy template via Chrome headless print-to-pdf (280×158mm, 16:9). Humanizer pass removed AI-writing patterns (triadic lists, "不是 X，是 Y" parallels, AI vocab). Codex review fixes applied: source_chunk.search_text_zh field name, 25 core tables / 7 business triggers clarification, p11 callout shortened, footer-mark removed to avoid overlap with callouts.
docs: final report, defense slides and per-gap deliverables (gap 1-9)
- rename gap-prefixed deliverables to clean names (research-landscape, innovation-analysis, contribution-ledger, contribution-audit-*, source-sql-appendix, final-defense slides) - move internal planning/roadmap/audit docs into docs/process/ - add docs/README.md material guide; update README docs section - update all cross-references; regenerate final-report.pdf and final-defense.pdf; make export script find Chrome on macOS/Linux
…esh ledger - retitle deliverable docs to delivery-oriented headings (drop 'Gap N —') - fix literal paths pointing to docs/ that moved into docs/process/ - refresh contribution-ledger baseline: HEAD b038e1b, Cofstars 5 commits now on origin/dev (was stale '2 local commits'), regen Cofstars audit TSV - drop 'Gap 4 evidence' watermark text from source SQL/test SVGs
- rename 'P0 MVP' section to 功能模块 with the complete delivered feature set (governance, agent visibility, graph explorer, CLI, evaluation, etc.) - surface optional LLM extraction/QA and hybrid recall in 技术栈
dzx0902
approved these changes
Jun 15, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
概述
将 dev 上的全部最终交付内容合并到 main,作为正式提交态。
包含的工作
docs/process/docs/README.md材料导览(对照作业七项要求),README 文档段改为项目介绍视角llm_analysis.py补注释/docstringfinal-report.pdf(26 页)与final-defense.pdf(17 页,含四位成员)对照作业要求(全部满足)
需求分析 / 概念(ER) / 逻辑 / 物理设计 · 系统演示截图 · 流程图+注释源程序(SQL+高级语言) · 分工与个人任务说明 —— 均有对应材料,详见
docs/README.md。合并安全性
已实测
git merge-tree origin/main dev无冲突,且合并结果树 == dev 树(main 仅多出 4 个历史 PR-merge 节点,无独有内容)。