# README artwork

Both README images were updated with the built-in image-generation tool, retaining
the original ivory, ink and vermilion visual style. They illustrate retrieval;
the example code and file tree are illustrative.

The headline rounds the observed 28.6% cost reduction to **~30%**. Both arms
solved 8/10 tasks; total Sol cost was $7.62 without Jevgrep and $5.44 with it.
The [benchmark report](../evals/results/relevance-threshold-2026-09-27.md) owns the
methodology, exact costs and limitations. Jev cost is excluded. These are observed
results on a tuned ten-task Python subset, not a speed claim or a universal guarantee.

## Final edit prompts

### cover

Edit this wide README hero in its existing warm ivory, black serif, vermilion editorial style. Preserve the large "jevgrep" wordmark, jg terminal pill, and elegant highlighted folder tree flowing into source excerpt cards. Replace the subtitle "Find the context. Start coding." with exactly "Same solve rate. ~30% lower cost." Make this subtitle clearly legible and comfortably fit the left area. Add one small but readable footnote along bottom left exactly "10-task SWE-bench subset · Sol cost · Jev excluded". Keep abundant whitespace and crisp illustration. Only these text strings plus "jevgrep" and "jg" should appear. Wide landscape about 2.4:1. The ~ symbol must be present before 30% to indicate rounding. No speed claim.

### explainer

Update this README infographic, preserving the warm ivory/black/vermilion editorial aesthetic and clear three-step workflow. Replace the main headline with exactly "~30% lower coding-agent cost". Below it put exactly "Same solve rate. Less spend." Preserve the three headings "Ask a repo question", "Jev finds relevant code", "Agent implements + tests", and their illustrations, keeping the example pool size consistent at 10. Update the bottom comparison to "Baseline $7.62" and "With jg $5.44". Replace the task-solves line with exactly "8/10 solves with and without jg". Footer two readable lines: "10-task Python SWE-bench subset. Observed reduction: 28.6%." and "Full Sol task cost, including failures. Jev excluded." Remove every old 40%, $4.52, and 7/10 claim. Ensure all type fits cleanly, no tiny stray marks, no speed claim. Landscape about 1.6:1. Use exact text; ~30% is an honest rounded headline, 28.6% in footer is precise.

### Final wording refinement

Edit only the subtitle in this image. Replace "Same solve rate. ~30% lower cost." with exactly "Same intelligence. ~30% lower cost." Preserve every other element, all illustration, wordmark, jg pill, colors, margins, aspect ratio, and footnote. Fit the new phrase legibly in the existing subtitle space. No other text changes.

Edit only two text lines in this image. Replace subtitle "Same solve rate. Less spend." with exactly "Same intelligence. Less spend." Replace "8/10 solves with and without jg" with exactly "8 of 10 tasks solved — with and without jg". Preserve everything else including headline "~30% lower coding-agent cost", workflow, example code, prices Baseline $7.62 and With jg $5.44, and both benchmark footer lines. Keep typography crisp and fit the text with generous margins, no other changes.

## Visual review

A fresh visual review found no clipping, overlap, misspellings or conflicting
benchmark numbers. The cover footnote is secondary at README scale; the README
and explainer also state its scope. A small punctuation speck remains inside the
illustrative question bubble and does not affect the text or comparison.