Dogfooding

Dogfooding means using ripr on this repository to keep the product honest. It should produce focused evidence, not broad self-analysis dashboards.

Current Useful Commands

cargo xtask dogfood
cargo run -p ripr -- --version
cargo run -p ripr -- doctor
cargo run -p ripr -- check --diff crates/ripr/examples/sample/example.diff
cargo run -p ripr -- check --diff crates/ripr/examples/sample/example.diff --json
cargo run -p ripr -- explain --diff crates/ripr/examples/sample/example.diff probe:crates_ripr_examples_sample_src_lib.rs:error_path:c1a03250
cargo run -p ripr -- context --diff crates/ripr/examples/sample/example.diff --at probe:crates_ripr_examples_sample_src_lib.rs:error_path:c1a03250 --json

cargo xtask dogfood is the stable advisory loop. It runs ripr check --mode fast against checked fixture diffs, writes actual outputs under target/ripr/dogfood/, and writes target/ripr/reports/dogfood.md plus target/ripr/reports/dogfood.json. It also checks repo-local finding-alignment receipts under fixtures/finding-alignment-dogfood/ so real RIPR PR examples preserve the Lane 1 split between raw findings, canonical evidence items, and actionable canonical gaps. Python repair-routing eval receipts live under fixtures/python-real-repo-evals/; they record curated scratch or real-repo repair-card, verify, and outcome evidence without promoting Python beyond its preview/advisory boundary. Bun UB cross-language witness receipts live under fixtures/bun-ub-cross-language-dogfood/; they record calibrated Blob / ArrayBuffer TypeScript preview receipts without running Bun or producing repair packets.

Dogfooding Rules

• Prefer sample diffs and fixtures over broad repository scans.
• Treat repo-wide RIPR refreshes as build-heavy on this repo. Use repo-badge-json, generated receipts, or an explicit gap ledger for ordinary summary counts. Use cargo xtask repo-exposure-summary-report when local planning needs bounded repo exposure counts, and run only one no-ledger repo scan at a time. If it emits basis: "limited_runtime_status" with runtime_status.downstream_consumable: false, do not use that artifact as a repair queue or badge basis.
• Do not use full repo-exposure-json as the normal badge, receipt, top-file, or packet-queue input. Full exposure dumps require explicit operator intent and cleanup after inspection.
• Raise RIPR_COMPACT_REPO_SEAM_CACHE_MAX_SEAMS only for the command that needs a large-repo cache write, after checking disk headroom.
• When ripr finds a real gap in its own code, add a fixture or regression test before changing the analyzer.
• Do not use ripr findings as blocking CI until the SARIF policy and calibration work lands.
• Record useful findings in Learnings when they change how the project should be built.

Planned Dogfood Loop

After the fixture lab and evidence output exist:

make code change
-> run ripr against the diff
-> inspect finding evidence
-> add targeted test or document static_unknown stop reason
-> keep fixture/golden output aligned

The goal is to keep the analyzer grounded in real developer workflows while still respecting the product boundary: static evidence guides, real mutation confirms later.