docs: GraphRAG-Bench Novel benchmark refresh + 256-dim embedder default

[galshubeli]

Summary

Cherry-picks the accurate pieces of #224 (which won't be merged as a whole).

• `docs/getting-started.md` and `docs/index.md`: replace "~85% / 100-question" wording with the GraphRAG-Bench Novel numbers (63.73 ACC / 2,010 questions)
• `graphrag_sdk/README.md`: refresh benchmark table with the Novel sub-category breakdown
• `graphrag_sdk/examples/01_quickstart.py`: switch default embedder to `text-embedding-3-large` with `dimensions=256` and pass matching `embedding_dimension=256` to `GraphRAG`

Skipped from #224: migration guide + FAQ (stale v0.8.2 framing and three inaccuracies — nonexistent `[semantic]` extra, fake `ChainedResolution` class, and `gpt-5.4` model).

Test plan

• CI passes
• `examples/01_quickstart.py` runs end-to-end with an OpenAI key

🤖 Generated with Claude Code

Summary by CodeRabbit

• Documentation

  • Updated benchmark documentation with specific GraphRAG-Bench Novel results (63.73 ACC over 2,010 questions)

• Chores

  • Updated quickstart example with revised embedder configuration

[coderabbitai]

📝 Walkthrough

Walkthrough

The pull request updates benchmark documentation across multiple files to reflect more specific results on the GraphRAG-Bench Novel dataset (2,010 questions over 20 novels), replacing generic accuracy claims. It also updates example code to use a larger embedding model with 256 dimensions.

Changes

Cohort / File(s)	Summary
Benchmark Documentation docs/getting-started.md, docs/index.md, graphrag_sdk/README.md	Updated benchmark claims from generic "~85% accuracy on 100-question benchmark" to specific "#1 on GraphRAG-Bench Novel" with 63.73 ACC over 2,010 questions. Expanded metrics table in README with per-dimension performance metrics.
Example Configuration graphrag_sdk/examples/01_quickstart.py	Changed embedder model from text-embedding-3-small to text-embedding-3-large with dimensions=256 and aligned embedding_dimension=256 in GraphRAG constructor.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

Poem

🐰 Huzzah! The benchmarks now shine bright and true,
With 63.73 ACC and dimensions anew,
From small embeddings we hop to large,
GraphRAG-Bench Novel takes the charge! 🌟

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 50.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately describes the two main changes: benchmark documentation updates referencing the GraphRAG-Bench Novel results and the embedder configuration change to 256 dimensions.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch docs/salvage-from-224

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@graphrag_sdk/README.md`:
- Around line 155-166: Update the README intro to remove or replace the outdated
“~85% accuracy ... 100-question benchmark” claim so it matches the refreshed
benchmark section; locate the intro text containing “~85% accuracy” and either
delete it or replace it with a concise statement reflecting the new
GraphRAG-Bench results (e.g., 63.73% Novel ACC on a 2,010-question, 20-novel
benchmark) and add a link to docs/benchmark.md for methodology—ensure the intro
and the new “#1 on GraphRAG-Bench” paragraph are consistent.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 4786b9a1-4e9e-4bbf-9cb0-f61dfcb06fb4

📥 Commits

Reviewing files that changed from the base of the PR and between 15e7e31 and 54e34d4.

📒 Files selected for processing (4)

• docs/getting-started.md
• docs/index.md
• graphrag_sdk/README.md
• graphrag_sdk/examples/01_quickstart.py

[Copilot]

Pull request overview

Updates public-facing docs/examples to reflect the GraphRAG-Bench Novel benchmark results and switches the quickstart example to a 256-dimension embedding configuration.

Changes:

• Refresh benchmark claims/numbers across docs and the SDK README to GraphRAG-Bench Novel (63.73 ACC / 2,010 questions).
• Update 01_quickstart.py to use text-embedding-3-large with dimensions=256 and set embedding_dimension=256 on GraphRAG.
• Adjust getting-started benchmark wording to reference the GraphRAG-Bench Novel corpus.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
graphrag_sdk/examples/01_quickstart.py	Switches default embedder to 256-dim and configures GraphRAG to match.
graphrag_sdk/README.md	Replaces older benchmark blurb/table with GraphRAG-Bench Novel results breakdown.
docs/index.md	Updates highlight bullet to GraphRAG-Bench Novel ACC and question count.
docs/getting-started.md	Updates benchmark link description to reference GraphRAG-Bench Novel corpus.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

embedding_dimension is hard-coded to 256, but get_providers() includes an Azure option that would typically use 1536-dim embeddings (e.g., text-embedding-ada-002). If a user switches to the Azure block and forgets to update embedding_dimension, the example will misconfigure the vector index / raise a config mismatch. Consider returning the dimension from get_providers() (or defining a single EMBEDDING_DIM constant used for both LiteLLMEmbedder(..., dimensions=...) and GraphRAG(..., embedding_dimension=...)), and adding a clear note for the Azure embedder’s expected dimension.

[Copilot]

This README now claims GraphRAG-Bench Novel results in the Benchmark section, but the README intro and code snippets above still reference the old "~85% accuracy on a 100-question benchmark" and use text-embedding-3-small. That makes the document internally inconsistent. Suggest updating/removing the earlier benchmark sentence and aligning the examples (or explicitly scoping which benchmark/config each claim refers to).