pilot: we will help minimize five real Maven/Gradle or Python failures

[fly1d]

Decision

The real-failure pilot remains open. We need sanitized, repeatable Maven/Gradle or Python failures from non-maintainer users; no ReproMin installation is required to offer a public case.

This body is the canonical current status. New comments should add user evidence, a decision, a blocker, or a handoff rather than routine release or CI updates.

Reason

The project has strong technical feasibility evidence but not enough independent adoption evidence. The current north star is five non-maintainer workflows. A workflow counts when a user tries ReproMin and reports a useful, inconclusive, or blocked result. Stars, downloads, demo runs, CI clones, maintainer-created examples, and contribution PRs do not count.

After five trials, we continue this direction only if at least three users report concrete value.

Next step

For a public case, reply with:

• repository URL, exact revision, and open-source license;
• deterministic failure command; and
• target exit code, exception type, or stable output signature.

The command must not need credentials, private services, special hardware, or production data. A maintainer will check fit, run the read-only Doctor preflight, propose a strict oracle, and perform a bounded first reduction.

For a private workflow, keep the source local and use Show and tell to share aggregate, sanitized evidence.

Owner / done when

• Participant: provide the three public-case fields above, or reviewed private-run feedback.
• Maintainer: state the fit decision and reason, then name the next action, owner, and completion condition.
• Pilot complete: five independent trials are recorded with their outcome, value or blocker, and known limits.

Current evidence

• The maintainer-led tsdown pilot reduced 14 files to 8 and passed 3/3 fresh-copy replays.
• The maintainer-led Gradle composite-build pilot reduced 854 files / 3,817,947 bytes to 11 files / 87,862 bytes and passed 5/5 fresh-copy replays plus an empty-cache cold start.

These are technical evidence only; neither counts as independent adoption.

Safety and community boundary

Do not share secrets, credentials, tokens, private URLs, customer data, proprietary source, raw logs, commands containing secrets, or environment values. The host backend executes the supplied command directly and is not a sandbox.

Project maintenance and pilot preparation materially use OpenAI Codex. Contact with another project requires prior opt-in, a human policy check, and disclosure of material agent involvement in the first message. Stop immediately if the recipient declines or asks for no further contact.

[fly1d]

Maintainer update: v0.1.0.dev3 is now available with the GitHub Action artifact fix and richer payload summaries. The repository now also has two scoped starter issues for documentation contributions: #12 (Java/Gradle workflow) and #13 (report validation CI gate).

The pilot remains open for sanitized real-world workflows. A useful report can be small: language/build system, failure signature shape, options used, retained/removed payload, approximate duration, and limitations. Please do not share credentials, proprietary source, or confidential logs.

[fly1d]

Maintainer update: v0.1.0.dev4 is now released: https://github.com/fly1d/repomin/releases/tag/v0.1.0.dev4. CI passed on the main push and release tag, including action artifact/output validation and offline benchmarks. The pilot remains open for sanitized real-world workflows; a small report with language/build system, failure signature, options, retained/removed payload, duration, and limitations is enough.

[fly1d]

To make the pilot easier to use, the repository now has a step-by-step real failure pilot guide with a sanitization checklist, report validation command, and copyable summary template.

[fly1d]

Maintainer update (2026-08-27):

The pilot path is now easier to run and safer to review. On main (commit b54e354), the GitHub Action supports match, exact exit-code, or learned process-failure oracles; it can apply newline-separated ignore/ignore-path rules and root/nested .gitignore files before uploading the artifact. The action validates report.json and the payload fingerprint, then exposes reduction-size, attempt, mutation, and holdout outputs for downstream CI gates.

The repository CI run 33040022143 passed across Linux, macOS, Windows, Python 3.9/3.13, JDK 11/21, offline benchmarks, and isolated wheel/sdist tests.

If you can share a sanitized CI or dependency failure, please use the real failure template. A short report is enough: language/build system, failure contract, ReproMin version, retained/removed payload, duration, and what was confusing. Remove credentials, private URLs, proprietary source, and confidential logs; the host backend is not a sandbox.

[fly1d]

Follow-up status (2026-08-27): the latest main commit 11e2145 passed the full CI run 33042070352, including Action smoke/output/report validation, offline benchmarks, the Python/JDK matrix, and isolated wheel/source-distribution checks. The real CI or dependency failure template remains the recommended path for a new sanitized pilot report.

[fly1d]

Maintainer update (2026-08-27):

report replay is now ready for a real CI/dependency-failure pilot on main at 6662b1f. Each requested replay run starts from an independent fresh payload copy, validates the report and payload before and after execution, and keeps command output and environment values out of JSON evidence. Downloaded artifacts can use the content-fingerprint fallback when archive transport changes filesystem timestamps; the result labels that evidence clearly.

The full CI run 33061326981 passed the Python/JDK matrix, Windows, macOS, Linux, wheel and source-distribution installs, Action smoke test, and offline benchmarks.

A sanitized real failure is still the most useful next input. Please use the pilot template and omit credentials, private URLs, proprietary source, and confidential logs. The host backend is not a sandbox; replay is evidence that the configured oracle matched in the current environment, not a correctness, root-cause, production-reliability, or security guarantee.

[fly1d]

Maintainer update (2026-08-31): v0.1.0.dev5 is now available as a GitHub pre-release: https://github.com/fly1d/repomin/releases/tag/v0.1.0.dev5

It includes repomin doctor, fresh-copy report replay, exact/content payload fingerprints, and the documented privacy and host/Docker boundaries. Main CI is green across the Python/JDK matrix, Windows/macOS/Linux, Action smoke, offline benchmarks, and isolated wheel/sdist installs (run 33354557317). The tag CI is also green. Please use the pilot guide and share only sanitized workflow details; no private source, credentials, or raw logs.

[fly1d]

Maintainer update (2026-09-01): v0.1.0.dev6 is now available as a GitHub pre-release. It includes the end-to-end report-replay benchmark and clearer, privacy-preserving mismatch diagnostics.

For a quick local sanity check, run:

python3 benchmarks/run_offline.py --only report-replay

The real-failure pilot remains open for a sanitized CI or dependency failure. A short summary of the language/build system, failure contract, ReproMin version, retained/removed payload, duration, and limitations is enough; please omit credentials, private URLs, proprietary source, and raw logs. The host backend is not a sandbox, and replay/holdout results are current-environment oracle evidence rather than correctness or production-reliability guarantees.

[fly1d]

Maintainer update (2026-09-01): v0.1.0.dev7 is now available as a GitHub pre-release: https://github.com/fly1d/repomin/releases/tag/v0.1.0.dev7

This release adds a privacy-safe adoption summary to repomin report validate --json, a workflow-feedback issue template for successful/inconclusive/blocked trials, and scalar Action outputs for oracle mode, retention ratios, and payload-fingerprint evidence. Replay mismatch JSON now includes expected and actual exit codes without exposing commands, match expressions, logs, or environment values.

The release assets were checked with twine, installed in isolated wheel and source-distribution environments, and tested with 560 passing tests and 16/16 offline benchmark checks. The tag CI is running here: https://github.com/fly1d/repomin/actions/runs/33466996058

For a pilot, use the feedback template for a successful, inconclusive, or blocked trial, or the real-failure template when a sanitized failure is available. Please omit credentials, private URLs, proprietary source, and raw logs. Results remain evidence for the configured oracle in the current environment, not a correctness, production-reliability, or security guarantee.

[fly1d]

Follow-up (2026-09-01): the v0.1.0.dev7 tag CI has completed successfully across the Python/JDK matrix, Windows/macOS/Linux, Action smoke, offline benchmarks, and isolated wheel/source-distribution checks: https://github.com/fly1d/repomin/actions/runs/33466996058

[fly1d]

Maintainer update (2026-09-01): v0.1.0.dev8 is now available as a GitHub pre-release: https://github.com/fly1d/repomin/releases/tag/v0.1.0.dev8\n\nThis release adds a deterministic, privacy-safe Markdown export for repomin report validate --format markdown, an opt-in GitHub Action Step Summary with a validated workflow-run link, and shell completion for the new format values. Validation summaries now use schema v2 and reject malformed budget_exhausted values; commands, expressions, logs, paths, and environment values remain outside the shareable summary.\n\nTag CI passed across the Python/JDK/OS matrix, Action smoke, offline benchmarks, coverage, and isolated wheel/source-distribution checks: https://github.com/fly1d/repomin/actions/runs/33474516876\n\nFor the pilot, install the wheel from the release page and share a sanitized CI or dependency-failure workflow using the real-failure template. A short report with the language/build system, failure contract, options, retained/removed payload, duration, and limitations is enough. Please omit credentials, private URLs, proprietary source, and raw logs; the host backend is not a sandbox.

[fly1d]

Maintainer update (2026-09-01): v0.1.0.dev9 is now available as a GitHub pre-release: https://github.com/fly1d/repomin/releases/tag/v0.1.0.dev9

The release adds repomin report compare, a read-only comparison for two or more validated report.json files. It reports aggregate reduction evidence and adjacent deltas, and warns when execution, budget, holdout, phase, input-selection, or oracle context drift limits a direct comparison. It does not read payloads, execute commands, access the network, or make performance/correctness claims.

For a pilot, install the wheel from the release page and try:

repomin report validate /path/to/before/report.json
repomin report validate /path/to/after/report.json
repomin report compare /path/to/before/report.json --label before /path/to/after/report.json --label after --format markdown

Please share only sanitized details: language/build system, failure contract shape, version/options, what was retained or removed, approximate duration, compare warnings, and any confusing step. A public fixture is welcome but not required. Do not include credentials, private URLs, proprietary source, raw logs, commands, or environment values.

[fly1d]

Follow-up (2026-09-01): the v0.1.0.dev9 tag CI has now completed successfully after a failed Windows-only attempt caused by runner startup timing. The rerun passed the Python 3.9/3.13 and JDK 11/21 matrix, Windows/macOS/Linux, GitHub Action smoke test, offline benchmarks, packaging, and isolated wheel/source-distribution checks: https://github.com/fly1d/repomin/actions/runs/33486832842

No source changes were needed; the initial failure was a timing-sensitive Windows test startup. The release remains a pre-alpha, descriptive-only pilot build.