Fetching metadata from the HF Docker repository...
Apr 22 18:10 fix: trust summary resilience, Unicode in reason strings, full GitHub README 13:15 docs: rewrite README for v2 — edge reliability, updated eval numbers, clean authorship 12:38 tune STALE threshold to 0.50 — better Cat B recall (44% vs 2% at 0.40) 12:04 Phase 2: UI reliability badges, changelog, eval runner — Phase 2 complete 10:54 Phase 2: add v2 full eval runner script 10:54 fix: suppress OPENALEX_API_KEY warning when no DOI present 10:51 Phase 2.6: add per-paper trust summary to synthesizer output — addresses Armin's explainability request 10:47 Phase 2.5: wire reliability scorer into critic — edge reliability replaces mean-age STALE threshold 10:47 Phase 2.4b: fix dominant_signal — FOUNDATIONAL driven by centrality+coherence not age 10:43 Phase 2.4: add edge reliability scorer — three-signal formula, foundational paper test passes 10:39 Phase 2.2-2.3: add doi to Paper, populate from S2 externalIds, integrate OpenAlex into retriever 10:34 Phase 2.1: add OpenAlex API integration (search, DOI lookup, citation centrality) 10:28 Phase 1.3: eval results, test scripts, gap filter reverted — no improvement, changelog update 09:33 Phase 1.3: smoke test scripts, eval runner, check_progress, fix unicode in run_eval.py 09:02 Phase 1.2: fix STALE-before-CONTRADICTED bug — both checks now run, CONTRADICTED wins when both fire 08:56 Phase 1.1: archive patch_contradiction.py — research integrity fix 08:55 docs: add v2 master context document 08:54 Phase 0: branch setup for v2-edge-reliability 04:38 chore: harden .gitignore — .claude/, secrets, session artifacts; remove duplicate lines Apr 21 23:10 docs: commit eval summary; clarify critic as LLM-assisted-judge; fix test imports - eval/results/summary.csv: un-gitignored and committed. All numeric
claims in the README eval table (52% staleness catch, 43.9% position
accuracy, 17.1s avg latency) are now reproducible on clone.
- README: clarified that the critic uses deterministic threshold
routing for STALE/INSUFFICIENT/PASS verdicts and an LLM-as-a-Judge
prompt only for pairwise CONTRADICTED detection.
- Test_Files/test_setup.py: removed — dev-time environment check script
(Phase 1 setup) with imports no longer in requirements.txt. Mar 31 19:19 Phase 14: README 18:48 resolve merge: use local RECON codebase Mar 30 00:45 initial commit 00:41 Fix sdk_version quotes in README yaml 00:33 Fix: remove gradio from requirements.txt - managed by sdk_version in README Mar 29 20:59 Fix HF Spaces: tmp paths for cache and model, preload model at build 20:53 Fix: move demo.launch() outside __main__ guard for HF Spaces 20:48 Disable SSR mode for HF Spaces compatibility 20:42 Remove ragas/datasets/heavy deps - not needed at runtime 20:21 Fix Gradio 6 css parameter location 20:03 Prevent binary files from being tracked 20:01 Prevent binary files from being tracked 19:44 Pin Gradio to 6.10.0 matching local version 19:43 Remove cached API responses from tracking 19:28 Add HF Spaces metadata and project README 19:13 Phase 13: HF Spaces deploy ready - verdict logging, clean requirements Mar 28 23:41 Phase 9.1: Cleaned up directory 23:40 Phase 9 polish: structured output format, HTML UI components, citation highlighting 23:09 Phase 9: Gradio UI, critic debug panel, claim table, session sidebar, MD export 22:42 Phase 8: LangGraph state machine, full pipeline, retry loop, session loader 22:39 Phase 7: synthesizer agent, claim extraction, session write, MD export 22:36 Phase 6: critic agent, 4-verdict taxonomy, contradiction detection, question rewriter 09:11 Phase 5: retriever agent, S2 keyword conversion, hybrid scoring, citation graph 08:41 Phase 4: planner agent with temporal decomposition and session context 08:38 Phase 3: retriever utils, hybrid scoring, DDG+S2 wrappers, disk cache 08:28 Phase 2: state schema, SQLite session memory, data dir 07:49 Initial commit Mukul Rayana
