Report viewer
terrain-demo-shop
ac9ffc2390 · 1,466 LOC · model claude-opus-5 · adapter claude-cli
Showing the bundled sample run. Drop a file anywhere on this page to load your own — it is read in your browser and never uploaded.
Projected cost to fix 1000 issues
$230.84
90% interval $168.97–$292.92 · 2.5 agent-hours · 85.3M tokens
Measured sample
issues in repo
21
sampled
5
fix rate
100%
regressions
0
cost per fix
$0.231
tokens per fix
85.3K
Where the money goes
| issue class | attempts | fix rate | cost / attempt | total |
|---|---|---|---|---|
| eslint:error:autofix | 4 | 100% | $0.246 | $0.985 |
| eslint:error:manual | 1 | 100% | $0.169 | $0.169 |
Static friction signals
| signal | measured | score | evidence | |
|---|---|---|---|---|
| Median source file size | 22 LOC | 100 | Median 22 LOC across 5 source files; p90 is 25 LOC. | |
| Files over 1000 LOC | 0 files | 100 | No source file exceeds 1000 LOC. | |
| Type safety and escape hatches | 0 per KLOC | 40 | Strict type checking is off; 0 escape hatch(es) (any, ts-ignore, eslint-disable, noqa) — 0.0 per KLOC. | |
| Test-to-source ratio | 0.2 ratio | 40 | 1 test file(s) against 5 source file(s). | |
| Agent brief (AGENTS.md / CLAUDE.md) | 0 | 0 | No AGENTS.md or CLAUDE.md — every agent rediscovers the build, the test command and the conventions from scratch, on every task. | |
| Lint and format configuration | 0 | 0 | Missing lint or format configuration — style is decided per-file, so agents guess and reviewers argue. | |
| Dependency lockfile | 1 | 100 | Dependencies are pinned by a lockfile. | |
| Continuous integration | 0 | 0 | No CI configuration found — nothing mechanically checks a change before a human does. | |
| Documentation | 50 % | 50 | README, no CONTRIBUTING, no docs/ | |
| Directory structure | 2 levels | 100 | Median path depth 2.0; largest directory holds 3 source files. | |
| TODO / FIXME density | 0 per KLOC | 100 | 0 TODO/FIXME/HACK marker(s) — 0.0 per KLOC. |
Agent brief (AGENTS.md / CLAUDE.md) 0/100
Add an AGENTS.md with the build command, the test command, the directory map and three project-specific conventions. It is the cheapest point on this entire report and usually the largest single win.
Lint and format configuration 0/100
Commit an ESLint (or Biome/ruff) config and a formatter config. A mechanical style answer removes an entire category of judgement from every change.
Continuous integration 0/100
Add a CI workflow that runs lint, types and tests. It doubles as the specification an agent can run locally.
Documentation 50/100
Document how to run and change the project. An agent that cannot find the entry point reads the whole repo to guess it.
Every attempt
| task | rule | file | outcome | cost | time | diff |
|---|---|---|---|---|---|---|
| t001-10da818d | dot-notation | src/cart.js:17 | fixed | $0.323 | 8s | +1/-1 |
| t002-19beea3b | no-else-return | src/util/format.js:13 | fixed | $0.324 | 10s | +1/-2 |
| t003-50ee64ab | no-useless-return | src/user.js:27 | fixed | $0.169 | 9s | +0/-1 |
| t004-4fe66b56 | no-var | src/util/format.js:2 | fixed | $0.168 | 8s | +1/-1 |
| t005-66be4569 | eqeqeq | src/util/format.js:3 | fixed | $0.169 | 9s | +1/-1 |