docs(readme): neutral reframe — drop competitor editorial, move comparisons to docs/benchmarks.md

[jaylfc]

Summary

• Removes competitive/dismissive framing of other memory systems from the README
• Keeps all benchmark numbers; moves the competitor-comparison table + Judge-vs-Recall@5 methodology notes to docs/benchmarks.md
• Repositions around what taosmd is (library metaphor, zero-loss archive) rather than what others aren't

What changes

README.md

• "Why this exists": drop the opening jab about "recreating human thinking"/"dressed up as cognition" and the side-commentary on MemPalace being "one person's mind dressed up". Library metaphor reads the same without them.
• Consolidation paragraph: reframe as a neutral architectural note — no named competitor, no "I can never remember my dreams" aside, same point about verbatim-first.
• Benchmark Results section: drop the competitor-comparison table. Keep the per-category breakdown and fusion-strategy comparison (both are taosmd-internal). Link to docs/benchmarks.md for the cross-system view.

docs/benchmarks.md (new)

• Carries the full competitor table, methodology explainer (end-to-end Judge vs Recall@5), Librarian axis-C numbers, and reproduction instructions.

Not changing

• Benchmark numbers: unchanged, same harness, same dataset, same 97.0% on master.
• Technical architecture sections: untouched.
• Install / quickstart / API / hardware sections: untouched.

Test plan

• README renders cleanly on GitHub with the new intro + shorter benchmark section
• docs/benchmarks.md link resolves
• No stale references to "we didn't dress up a vector database" anywhere else in the repo

Summary by CodeRabbit

• Documentation
  • Streamlined README with condensed benchmark claims and improved consolidation explanation, linking to detailed documentation
  • Added comprehensive benchmark documentation including end-to-end evaluation results, per-category breakdowns, system comparisons, and reproducibility commands

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: f863f9ce-aef3-4bad-b5a8-ede2aa6fe6ca

📥 Commits

Reviewing files that changed from the base of the PR and between 7d2c780 and 4b2ad9e.

📒 Files selected for processing (2)

• README.md
• docs/benchmarks.md

---

📝 Walkthrough

Walkthrough

Two documentation files were updated: README.md was condensed by removing detailed benchmark comparisons and simplifying the consolidation explanation, while docs/benchmarks.md was added as comprehensive benchmark documentation for the LongMemEval-S suite, including end-to-end Judge results, fusion-strategy comparisons, and reproduction instructions.

Changes

Cohort / File(s)	Summary
README Condensing README.md	Removed detailed system comparisons and retrieval-only metric justifications; replaced with single-sentence benchmark claim and links to benchmarks documentation. Simplified "dreaming" explanation to emphasize verbatim archive preservation with coexisting summaries. Removed MemPalace column and reformatted one comparison table to two columns while retaining numerical results.
Benchmark Documentation docs/benchmarks.md	New file documenting LongMemEval-S end-to-end Judge results across 500 questions, including per-category breakdown, fusion-strategy comparisons, system comparisons, Librarian layer vocabulary-gap benchmark (recall@lag metrics), and shell commands for result reproduction with commit SHA and model pinning.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~5 minutes

Possibly related PRs

• docs(readme): propagate Judge accuracy framing across all LongMemEval sections #32: Both PRs modify the README's benchmark narrative and tables to emphasize end-to-end Judge accuracy as the primary metric, adjusting labels, table columns, and explanatory text accordingly.

Poem

🐰 Benchmarks once sprawling, now neat and concise,
A archive preserved without sacrifice,
Summaries rest where the transcripts take flight,
Judge scores shine bright, documentation done right! ✨

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch docs/readme-neutral-reframe

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Files Reviewed (2 files)

• README.md
• docs/benchmarks.md

---

_{Reviewed by seed-2-0-pro-260328 · 118,232 tokens}