docs: improve README visual hierarchy - #95
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: bed2f7a0d1
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
|
||
| <p align="center"> | ||
| <img src="docs/images/evidence-backed-agent-examples.png" alt="Three evidence-backed examples: focused context holds Recall at 5 while reducing retrieved tokens, answerable questions return supported evidence, and off-topic questions abstain" width="100%"> | ||
| <img src="docs/images/evidence-backed-agent-examples.png" alt="Three evidence-backed examples: focused context keeps Recall at 5 while reducing returned content, answerable questions return cited support, and unsupported questions explicitly abstain" width="100%"> |
There was a problem hiding this comment.
The replacement graphic no longer names the long-document versus grounded fixture, or each fixture's document/question scope, on the individual cards; it provides only a global pair of commands. This contradicts the adjacent caption claiming that each card names its deterministic fixture and test scope, so readers cannot tell which fixture supports each displayed result. Restore the per-card labels or revise the caption to describe the graphic accurately.
AGENTS.md reference: AGENTS.md:L220-L222
Useful? React with 👍 / 👎.
| <svg xmlns="http://www.w3.org/2000/svg" width="1693" height="929" viewBox="0 0 1693 929" role="img" aria-labelledby="title desc"> | ||
| <title id="title">Three evidence-backed Engraphis agent behaviors</title> | ||
| <desc id="desc">A clear three-card summary of deterministic offline fixtures. Focused context returns 808.8 → 218.4 tokens while retaining Recall at 5 of 1.000. A grounded answer returns support for 5/5 answerable questions grounded. An unsupported question safely abstains for 5/5 off-topic questions abstained. Reproduce with eval.chunking_eval and eval.grounded.</desc> | ||
| <image href="evidence-backed-agent-examples.png" width="1693" height="929"/> |
There was a problem hiding this comment.
When this PNG is regenerated with different visible numbers, test_example_visual_uses_the_checked_in_offline_fixture_results will still pass: it searches the SVG source and now finds every expected result only in the manually duplicated <desc>, while this line renders the opaque PNG and test_readme_visual_pngs_match_their_svg_canvas verifies dimensions only. Previously those assertions matched rendered SVG text elements, so the evaluation gate no longer validates the public-facing claims; keep the values in rendered SVG content or add a generation/verification mechanism tied to the PNG.
AGENTS.md reference: AGENTS.md:L154-L155
Useful? React with 👍 / 👎.
1 participant
Summary
Validation
python -m pytest -o addopts='' tests/test_benchmark_evidence.py tests/test_commercial_hardening.py tests/test_licensing_boundary_docs.py tests/test_pro_cta.py -qruff check .