Python Repair Routing Implementation Plan

Status: proposed

Delivery state: partially delivered; retained as a historical work-item ledger for the open promotion track.

Delivery note (2026-08-14): the plan's early slices landed. The scoped Python repair-routing loop is governed usable alpha (#741), Python project detection can enable the adapter by default when config is absent, and the receipt/packet/first-pr bridges below are fixture-backed. Later sections are historical planning text, not the current queue. Every per-item Status, Blocked by, dependency, and progress field below is a snapshot of its original planning context, not a present-tense delivery verdict: current-state-inventory.md and #1160 carry the current state and the open promotion work.

Owner: language-adapter / swarm

Created: 2026-05-29

Plan ID: RIPR-PLAN-0017

Linked proposal:

• RIPR-PROP-0017: Python Repair Routing Lane

Linked specs:

• RIPR-SPEC-0026: Language Adapter Contract
• RIPR-SPEC-0028: Python Preview Static Facts
• RIPR-SPEC-0057: RIPR Swarm Repair Loop
• RIPR-SPEC-0058: RIPR Swarm External Agent Handoff
• RIPR-SPEC-0061: Lane 1 Canonical Actionability Contract

Linked ADRs:

• None.

Active goal:

• Not active. The former active execution manifest .ripr/goals/active.toml was deleted in #1701 PR 3. This plan does not supersede live GitHub issue/PR state unless a later activation explicitly selects it.

Support-tier impact:

• Delivered for the scoped loop only: the scoped support-tier review (#741) promotes Python repair routing to usable alpha. Broader Python static facts remain preview/advisory; this plan text does not widen that claim.

Policy impact:

• Register this plan and its proposal in policy/doc-artifacts.toml.

Required evidence for plan edits:

cargo xtask check-doc-artifacts
cargo xtask check-doc-index
cargo xtask markdown-links
cargo xtask check-static-language
cargo xtask check-support-tiers
cargo xtask check-pr-shape
git diff --check

Non-goals:

• No analyzer behavior changes in the charter PR.
• No output-schema changes in the charter PR.
• No default CI blocking or gate behavior changes.
• No source edits or generated tests.
• No provider or model calls.
• No runtime mutation execution.
• No release, publish, signing, marketplace, or source-repo authority changes.

Claim boundary:

• This plan makes the Python repair-routing lane reviewable and restartable. It does not claim Python is usable alpha, gate eligible, Rust parity, or runtime-confirmed.

Rollback:

• Revert the proposal, this plan, index links, support-tier clarifications, and document artifact ledger entries. No runtime behavior changes are involved.

Current state

Python preview evidence already exists under the language adapter preview contract. The current claim is syntax-first and advisory: Python can provide owner, test, assertion/oracle, probe, related-test, and static-limit facts when enabled, but support remains preview.

The lane target is higher than parser support. Python should become the first non-Rust proof that RIPR can turn changed behavior into a bounded repair task:

changed behavior
-> missing evidence
-> focused test repair
-> verify command
-> receipt

Milestones

Milestone	Work items	User value
A. Python is recognized	PR 1-5	RIPR can run on a Python repo without pretending it is Rust.
B. Python has real evidence	PR 6-12	RIPR can identify changed Python behavior and distinguish strong tests from weak tests.
C. Python produces repair cards	PR 13-15	RIPR gives the next test to add.
D. Python works in daily workflows	PR 16-19	CLI, PR, CI, and editor show the same guidance.
E. Python becomes application-useful	PR 20-23	Common API, CLI, field, and parameterized-test shapes become useful.
F. Swarm turns it into leverage	PR 24-26	RIPR creates safe parallel test-repair work and proves what closed.
G. Promotion	PR 27-30	Python support is honest, measured, and ready to promote if evidence supports it.

Work items

Work item: docs/python-repair-routing-charter

Status: done

Linked proposal:

• RIPR-PROP-0017

Linked specs:

• RIPR-SPEC-0026
• RIPR-SPEC-0028
• RIPR-SPEC-0057
• RIPR-SPEC-0058
• RIPR-SPEC-0061

Linked ADR:

• n/a

Blocks:

• docs/python-current-state-inventory

Blocked by:

• n/a

Branch:

• docs-python-repair-routing-charter

Issue:

• n/a

PR:

• #518

Goal

Define what success means for the Python repair-routing lane before behavior implementation spreads.

Production delta

• Add RIPR-PROP-0017 as the lane charter and support contract.
• Add this implementation plan.
• Register the proposal and plan in policy/doc-artifacts.toml.
• Link the charter from proposal, plan, documentation, support-tier, and language-preview surfaces.

Non-goals

• No Python analyzer behavior changes.
• No fixture or golden changes.
• No CLI, output-schema, LSP, generated-CI, swarm, receipt, or gate behavior changes.
• No active-goal manifest changes.
• No support-tier promotion.

Acceptance

• Every future Python repair-routing PR can point back to the charter.
• Docs say Python remains static/advisory preview until promoted.
• "Fully working Python" is defined as the repair loop, not parser existence.
• The plan preserves the PR-by-PR lane sequence without making it active.

Proof commands

cargo xtask check-doc-artifacts
cargo xtask check-doc-index
cargo xtask markdown-links
cargo xtask check-static-language
cargo xtask check-support-tiers
cargo xtask check-pr-shape
git diff --check

Rollback

• Revert the proposal, plan, doc index links, support-tier clarification, and document artifact ledger entries.

Work item: docs/python-current-state-inventory

Status: done

Inventory:

• Python repair routing current-state inventory

Branch:

• docs-python-current-state-inventory

PR:

• #521

Linked proposal:

• RIPR-PROP-0017

Linked specs:

• RIPR-SPEC-0026
• RIPR-SPEC-0028
• RIPR-SPEC-0057
• RIPR-SPEC-0058
• RIPR-SPEC-0061

Linked ADR:

• n/a

Blocks:

• analysis/python-project-detection

Blocked by:

• docs/python-repair-routing-charter

Goal

Inventory current Python preview code, fixtures, Rust/Cargo assumptions, output surfaces, and the first fixture matrix before changing behavior.

Production delta

• Add a current-state inventory doc or plan section covering existing Python preview code, fixtures, CLI assumptions, and output surfaces.
• Define the first fixture matrix: basic_function, predicate_boundary, changed_return_value, changed_exception, dict_field_change, pytest_exact_assert, pytest_smoke_assert, unittest_assert_equal, fastapi_route_optional, cli_output_optional, and dynamic_unsupported.

Non-goals

• No behavior change.
• No support-tier promotion.
• No fixture implementation yet unless the inventory finds an existing fixture and only indexes it.

Acceptance

• Clear map of current state and remaining work.
• Every later PR has a fixture home.
• Rust/Cargo assumptions that block Python-only repos are listed with owners.

Proof commands

cargo xtask check-doc-index
cargo xtask markdown-links
cargo xtask check-static-language
cargo xtask check-pr-shape
git diff --check

Rollback

• Revert the inventory doc and index links.

Work item: analysis/python-project-detection

Status: done

Blocked by:

• docs/python-current-state-inventory

Goal

Let ripr pilot --root <python-repo> recognize Python repos without requiring a Cargo workspace.

Acceptance

• Detect pyproject.toml, setup.py, setup.cfg, requirements.txt, pytest.ini, tox.ini, noxfile.py, tests/, and src/.
• Exclude .venv/, venv/, .tox/, .nox/, site-packages/, .pytest_cache/, .mypy_cache/, dist/, build/, and detectable generated files.
• Python-only and mixed repos fail closed with named limitations rather than Cargo-specific errors.
• ripr pilot --root fixtures/python/basic works without Cargo.

Delivered:

• Missing ripr.toml now keeps Rust-only defaults unless Python project markers are present.
• Python project detection recognizes root marker files and Python files under src/ or tests/, while skipping virtualenv, cache, build, distribution, and generated Python files.
• Explicit ripr.toml remains authoritative, so [languages] enabled = ["rust"] still disables Python preview even in Python-shaped repos.
• fixtures/python/basic pins the no-config Python project path used by ripr pilot and diff-scoped ripr check.

Work item: analysis/python-source-facts

Status: complete

Blocked by:

• analysis/python-project-detection

Goal

Extract stable Python source facts with file, span, owner, and language metadata.

Acceptance

• Facts cover modules, classes, functions, methods, decorators, parameters, returns, raises, predicates, comparisons, boolean expressions, calls, assignments, attribute writes, dict/list/set literals, string literals, and print/log calls.
• Malformed Python produces a named limitation, not a crash.
• No repair recommendations yet.

Delivered:

• Python analysis now flows through an internal source-fact snapshot with stable file, span, owner, and language = "python" metadata.
• The snapshot records modules, classes, functions, methods, decorators, parameters, returns, raises, predicates, comparisons, boolean expressions, calls, assignments, attribute writes, dict/list/set literals, string literals, and print/log calls.
• Malformed Python records an unsupported_syntax source-fact limitation instead of silently returning empty facts.
• Existing owner/test extraction reuses the snapshot; no repair-card or recommendation surface changed in this slice.

Work item: analysis/python-diff-owner-mapping

Status: complete

Blocked by:

• analysis/python-source-facts

Goal

Map changed Python lines to stable, language-qualified owners.

Acceptance

• Owner IDs cover functions, methods, classes, and module-level code.
• Mixed Rust/Python repos do not collide.
• Changed Python owner is visible in JSON output.
• Unrelated line movement avoids unnecessary ID churn where possible.

Delivered

• Python preview findings now populate stable, language-qualified probe.owner IDs using python:<path>::<qualified_owner>.
• Changed-line owner selection prefers the narrowest matching owner, so function/method changes do not collapse to class or module owners while class-body and module-level changes still receive durable owners.
• JSON and human output surface Python preview probe.owner values; the existing owner_kind vocabulary remains unchanged, with class owners represented by probe.owner only.
• Focused unit tests pin function, method, class, module, line-movement, JSON, and human rendering behavior; python_owner_file_match pins fixture output.

Work item: analysis/python-pytest-oracles

Status: complete

Blocked by:

• analysis/python-diff-owner-mapping

Goal

Recognize common pytest tests and classify assertion strength.

Acceptance

• Detect test_*.py, *_test.py, def test_*, class Test*, plain assert, pytest.raises, pytest.mark.parametrize, fixture parameters, common client tests, capsys, caplog, and monkeypatch.
• Classify exact, boundary, exception, field, output, status-code, broad smoke, reach-only, and unknown helper oracles.
• Unknown helpers remain conservative.

Delivered

• Pytest test discovery now records fixture/parameter names and limits class-scoped pytest discovery to class Test* while preserving unittest.TestCase method discovery for the next slice.
• Python assertions now keep an internal pytest oracle shape for exact, boundary, exception, field, output, status-code, broad-smoke, reach-only, mock, and custom-helper evidence while preserving the shared OracleKind / OracleStrength output vocabulary.
• pytest.raises context managers, imported raises(...), caplog / capsys output observers, response.status_code / exit_code, dict/attribute field assertions, parametrized tests, and monkeypatch fixture parameters are represented as preview evidence.
• python_pytest_oracle_shapes pins output/log assertion evidence; existing Python preview fixtures now record fixture parameters and non-exact oracle shapes in JSON evidence without emitting repair cards yet.

Work item: analysis/python-unittest-oracles

Status: done

Blocked by:

• analysis/python-pytest-oracles

Goal

Support common unittest repos without a separate output model.

Acceptance

• Detect unittest.TestCase, def test_*, assertEqual, assertTrue, assertFalse, assertRaises, assertIn, assertRegex, and assertDictEqual.
• Verify commands can use pytest or python -m unittest when appropriate.
• Unittest facts enter the same oracle taxonomy as pytest.

Delivered

• Python test facts now preserve a class-qualified test name, so pytest class methods and unittest.TestCase methods can be addressed by stable static selectors.
• Preview evidence now records framework-shaped verify commands for related tests: pytest path::node for pytest and python -m unittest module.Class.test_method for unittest.
• Unittest assertion calls now use assertion arguments to preserve output, status-code, and dict/object field oracle shapes while keeping the shared OracleKind / OracleStrength vocabulary.
• python_unittest_oracle_shapes pins a unittest self.assertIn(...) output assertion and the generated python -m unittest verify command; existing Python preview fixtures record verify-command evidence without emitting repair cards yet.

Work item: analysis/python-related-test-linking

Status: done

Blocked by:

• analysis/python-unittest-oracles

Goal

Connect changed owners to likely tests using conservative static signals.

Acceptance

• Use imports, direct calls, class references, obvious route/client references, filename similarity, test naming similarity, and fixture names.
• Distinguish related strong tests, related weak tests, and no related test.
• Weak related tests are preferred repair locations.
• Uncertain links are marked uncertain.

Delivered

• Related-test ranking now keeps direct syntactic calls and import-alias calls ahead of heuristic links so weak directly related tests remain preferred repair locations.
• Same-stem file proximity, test-name similarity, and fixture-name proximity are treated as heuristic-only links: they keep weak reachability, do not promote assertions to strong revealability, and emit related_test_uncertain evidence.
• python_related_test_name_similarity and python_fixture_name_relation pin the new uncertain relation outputs, while existing same-stem and module-level fixtures were refreshed to preserve the same uncertainty boundary.

Work item: analysis/python-canonical-gap-identity

Status: done

Blocked by:

• analysis/python-related-test-linking

Goal

Create durable Python canonical gap IDs.

Acceptance

• Identity includes language, file, owner path, behavior kind, probe kind, and normalized expression, field, exception, or output.
• Duplicate raw signals collapse into one canonical finding.
• Line-number-only identity is avoided where possible.
• Same ID appears across CLI, JSON, SARIF, PR, LSP, and agent packet surfaces.

Delivered

• Python preview findings now carry an optional canonical_gap_id and typed canonical_gap identity made from language, file, owner path, behavior kind, probe kind, and normalized discriminator text.
• Canonical Python identities omit source line numbers, so line movement does not churn the ID when the changed owner and discriminator are stable.
• JSON output records canonical_gap_id, canonical_gap_group_size, and the structured identity parts; human, SARIF, GitHub annotation, LSP diagnostic, hover, and context-packet surfaces carry the same scalar ID.
• Static-limit Python findings keep static_limit_kind without a canonical repair-gap ID until typed non-actionable gap states land.
• Existing Python fixture goldens pin the identity on non-static-limit preview findings while static-limit fixtures stay unchanged.

Work item: analysis/python-ripr-evidence-model

Status: done

Blocked by:

• analysis/python-canonical-gap-identity

Goal

Express Python evidence using RIPR reachability, infection, propagation, and revealability concepts.

Acceptance

• Actionable gaps carry evidence for reachability, changed behavior, propagation, and revealability.
• Non-actionable cases carry stop reasons.
• Code changes alone do not produce recommendations.

Delivered

• Python preview findings now use family-specific RIPR infection and propagation evidence for predicates, return values, exception paths, field/object state, and call/output effects instead of placeholder unknown summaries.
• Static-limit Python findings fail closed as static_unknown, preserve the observed reach/oracle facts, carry typed stop reasons, and omit canonical repair-gap IDs and recommendations.
• Simple predicate-boundary findings can carry an activation-level missing discriminator such as amount == threshold; the value is visible in JSON, human evidence paths, and fixture goldens.
• Findings with no related Python test remain no_static_path evidence and do not emit repair recommendations until repair-card, verify-command, and receipt contracts exist.

Work item: analysis/python-repair-classes-v1

Status: done

Blocked by:

• analysis/python-ripr-evidence-model

Goal

Ship the first high-confidence Python repair classes.

Acceptance

• Predicate boundary, return value, exception path, dict/object/dataclass field, and output/log behavior each have positive and negative fixtures.
• Every actionable gap includes a missing discriminator.
• Dynamic or ambiguous cases remain non-actionable.

Delivered

• Direct weak Python preview findings now emit family-specific missing discriminators for predicate boundaries, return values, exception paths, field/object values, and output/log/call effects.
• Weak direct findings get repair-class next-step wording that names the missing discriminator without claiming a full repair card, verify command, or receipt.
• Strong-oracle, no-path, heuristic-only, and static-limit cases suppress repair guidance instead of being treated as repair-ready work.
• Existing Python fixture goldens pin positive and negative examples across the first repair classes while preserving Python's preview/support-tier boundary.
• python_dict_field_repair_gap pins returned-dict field discriminator extraction without requiring runtime dataclass or serializer semantics.
• python_model_field_repair_gap pins syntax-only returned constructor keyword field routing for model-like objects, recommends an object-field assertion such as assert result.active == True, and adds a stop condition for cases where the returned object does not expose the keyword as a public field or attribute.

Work item: output/python-ranking-noise-control

Status: done

Blocked by:

• analysis/python-repair-classes-v1

Goal

Make ranking-facing output show a curated small set of Python findings. The existing ripr pilot command still consumes Rust seam inventory; Python first-use projection remains in cli/python-first-use-path.

Acceptance

• Rank higher for public owners, related weak tests, concrete discriminators, available verify commands, clear assertion shape, and core changed behavior classes.
• Rank lower or stop for dynamic imports, opaque helpers, monkeypatch-only behavior, generated code, metaprogramming, missing test locations, and unclear discriminators.
• "No actionable Python gaps" is an honest supported result.

Delivered

• The shared finding sorter now assigns Python preview findings an actionability rank without changing non-Python file/line ordering.
• Direct weak Python findings with canonical gaps, concrete missing discriminators, public owners, direct related-test evidence, verify-command evidence, and core repair families sort ahead of lower-value Python preview findings.
• Already-observed, no-path, heuristic-only, unknown, and static-limit Python findings remain visible but sort after repairable direct weak gaps.
• python_ranking_noise_control pins the report order: a direct repairable predicate-boundary gap appears before observed, no-related-test, and dynamic-dispatch static-limit findings even when those noisy files sort earlier by path.

Work item: output/python-test-placement-verify

Status: done

Blocked by:

• output/python-ranking-noise-control

Goal

Recommend where and how to verify a Python repair.

Acceptance

• Actionable gaps include suggested test file, test name, node ID when possible, and pytest or unittest command.
• Command confidence is included.
• Commands do not assume dependencies that are not detected.

Delivered

• Direct weak Python findings with a concrete missing discriminator now emit placement metadata for the nearest direct pytest or unittest related test: suggested test file, suggested test name, pytest node ID when applicable, verify command, and verify-command confidence.
• Human output renders a compact Repair placement block before the next-step wording; JSON output carries the same data as an additive repair_placement object.
• The command builder only emits placement when a related test framework is detected statically, so heuristic-only, no-path, static-limit, and already-observed findings do not get invented commands.
• python_test_placement_verify pins pytest and unittest placement output, while existing direct weak Python goldens now show the same placement fields.

Work item: output/python-repair-card-v1

Status: done

Blocked by:

• output/python-test-placement-verify

Goal

Produce copy-ready human Python repair cards.

Acceptance

• Cards include changed owner, changed behavior, current test evidence, missing discriminator, recommended test shape, suggested location, verify command, receipt status, and stop conditions.
• Cards are present in human CLI and JSON output.
• Cards do not edit files.
• Receipt commands remain deferred to outcome/python-gap-ledger; this v1 card exposes the receipt slot honestly as unavailable instead of inventing a before/after command before Python outcome records exist.

Delivered

• Direct weak Python findings that already have a canonical gap, concrete missing discriminator, related-test evidence, placement, and verify command now render a Python repair card in human output and an additive python_repair_card object in JSON.
• The v1 card names the changed owner and behavior, current weak test evidence, missing discriminator, recommended pytest/unittest-shaped test shape, suggested assertion, test location, verify command and confidence, preview/advisory authority boundary, deferred receipt status, stop conditions, and limits.
• Field/object repair cards now specialize common assertion shapes for returned mapping fields, response JSON fields, and response status-code fields so the suggested assertion is directly copyable while staying static/advisory.
• Static-limit, heuristic-only, no-path, and already-observed Python findings still do not get repair cards.

Work item: swarm/python-agent-packet-export

Status: done

Blocked by:

• output/python-repair-card-v1

Goal

Export deterministic, bounded Python repair packets for swarm use.

Acceptance

• Packet fields include canonical gap ID, language, allowed files, forbidden files, task, missing discriminator, test shape, verify command, receipt command, and stop conditions.
• Packets are suitable for parallel execution without overlapping edits where possible.

Delivered

• ripr reports gap-ledger --check-output <check.json> now derives PR-local Python GapRecord entries from actionable python_repair_card findings without rerunning analysis.
• The derived records preserve canonical Python gap IDs, preview language status, source anchors, suggested test files, suggested test names, verify commands, stop conditions, and preview/advisory authority boundaries.
• ripr agent packet --gap-ledger <ledger> --gap-id <id> --json can export the selected Python record through the existing agent packet envelope.
• Gap-ledger packets now carry explicit allowed_files, forbidden_files, conflict_group, receipt_command, and receipt_status fields so agents get bounded test-edit scope and same-file conflict grouping.
• Check-output-derived Python GapRecords now synthesize deterministic ripr outcome receipt commands from the supplied before check JSON, target/ripr/reports/after-check.json, and a gap-scoped receipt path, so the same records can render through ripr agent packet --gap-ledger.
• Python preview records remain advisory: gate and RIPR-zero projections stay ineligible until later policy and outcome-ledger work exists.

Work item: cli/python-first-use-path

Status: in progress

Blocked by:

• swarm/python-agent-packet-export

Goal

Make first Python runs useful in CLI.

Acceptance

• ripr pilot --root ., ripr first-pr --root . --base origin/main --head HEAD, and ripr check --root . --format json show detected Python project, supported/unsupported features, top repairable gap, limitation count, repair card, verify command, and receipt command when evidence supports it.

Progress

• ripr pilot --root . now projects the existing diff-scoped Python preview repair card into the terminal summary and pilot-summary.{json,md} when a Python project/diff yields a repairable gap. The card keeps preview/advisory boundaries, limitation counts, verify command, and deferred receipt status.
• ripr check --root . --format json already emits the underlying python_repair_card object.
• ripr first-pr --root . --base origin/main --head HEAD can now accept Python-only project roots and select preview Python GapRecords from the existing gap decision ledger into start-here.{json,md} with preview_limited output state, missing discriminator, verify command, and receipt command.
• Python-only first-pr gap-ledger recovery now points at the existing bridge: ripr check --json followed by ripr reports gap-ledger --check-output, rather than the Rust repo-exposure path.
• Python before/after outcome receipts can now compare check-output JSON by canonical gap ID, and the check-output gap-ledger bridge now supplies the packet receipt command.
• ripr first-pr --check-output <check.json> now accepts saved Python check JSON directly, materializes the check-output-derived gap-decision-ledger.{json,md}, and then selects the preview Python start-here repair through the normal GapRecord path.
• Raw ripr check and pilot repair cards now include receipt guidance that tells users to save check JSON and run ripr first-pr --check-output or ripr reports gap-ledger --check-output to materialize a gap ledger with a concrete receipt command.

Work item: output/python-surface-projection

Status: in progress

Blocked by:

• cli/python-first-use-path

Goal

Project Python repair cards consistently across output surfaces.

Acceptance

• JSON, Markdown, SARIF, PR comments, generated summaries, and output-contract tests share canonical IDs.
• Python findings are not Rust-shaped findings with Python labels.
• PR summary highlights top Python repair cards.

Progress

• Eligible python_repair_card findings now project into diff-scoped SARIF properties with the same advisory card fields as check JSON.
• Python preview static-limit findings now project explicit diff-scoped SARIF python_no_action properties with repairability = "analyzer_limitation", no verify command, no receipt command, no repair card, and no packet-ready authority.
• Python preview ordinary no-action findings now project explicit diff-scoped SARIF python_no_action properties for already_observed, no_related_test, and heuristic_only states with repairability = "no_action", no verify command, no receipt command, no repair card, and no packet-ready authority.
• GitHub annotation output now includes a concise Python repair-card sentence with the missing discriminator, suggested test target, verify command, and preview/advisory boundary.
• GitHub annotation output now also names Python preview ordinary no-action and static-limit states, explicitly preserving the no repair card / no agent packet boundary for already-observed, no-related-test, heuristic-only, and static-limit findings.
• cargo xtask pr-summary now highlights the top Python preview repair card from actionable-gaps.json with the canonical gap, changed owner, missing discriminator, suggested test target, verify command, receipt command, and stop conditions while preserving the static/advisory boundary.
• Check-output-derived Python already_observed, no_related_test, and heuristic_only findings now materialize as report-only gap-ledger records, and the PR review front panel surfaces the top typed Python no-action or static-limitation record with a no_repair_packet route instead of silently falling back to a generic clean state.

Work item: ci/python-advisory-mode

Status: done

Blocked by:

• output/python-surface-projection

Goal

Let teams run Python repair-routing in PRs safely.

Acceptance

• Advisory GitHub Actions support uploads report artifacts and normalized result checks.
• Fork/untrusted behavior is clear.
• No provider calls, mutation execution, default self-hosted runner use, or default CI blocking.

Work item: lsp/python-repair-card-projection

Status: in progress

Blocked by:

• output/python-surface-projection

Goal

Bring Python repair cards into editor surfaces.

Acceptance

• Diagnostics, hovers, and code actions match CLI reports.
• Code actions can copy repair card, pytest skeleton, agent packet, and open related test file.
• Stale state is obvious and no hidden code edits occur.

Progress

• LSP GapRecord command validation now accepts bounded Python verify commands for pytest ... and python -m unittest ... while preserving shell metacharacter and parent-directory rejection.
• Python preview GapRecord diagnostics with a safe pytest verify command now expose a Write Python test: copy pytest skeleton code action. The copied skeleton includes the canonical gap, suggested file, missing discriminator, changed behavior, verify command, stop conditions, and a fail-fast NotImplementedError placeholder rather than silently generating a passing test.
• Python preview GapRecord diagnostics with a safe verify command now expose a Copy Python repair card code action. The copied card is marked as current validated GapRecord evidence and includes the changed owner, changed behavior, current weak test evidence, missing discriminator, suggested assertion/location, verify command, receipt command when available, stop conditions, and preview/advisory limits.
• Current actionable and repairable Python GapRecord diagnostics now expose an Agent handoff: copy Python packet code action. The action reuses the existing GapRecord-backed ripr.collectContext packet path and fails closed without safe gap-ledger paths, repair-route paths, verify commands, and receipt commands.
• GapRecord code actions now fall back from a bare Python test name to repair_route.target_file when opening the related test file, matching the check-output-derived Python repair-card shape.

Work item: analysis/python-http-api-pack-v1

Status: done

Blocked by:

• output/python-surface-projection

Goal

Support simple FastAPI/Flask-shaped repair cards.

Acceptance

• Simple route decorators, returns, status codes, JSON fields, and obvious client tests can produce framework-shaped repair cards.
• Dynamic routing remains a named limitation.

Progress

• fixtures/python_api_route_decorator_repair_gap now proves a simple FastAPI/Flask-shaped @api.post(...) route decorator can remain syntax-first route metadata instead of a decorator-indirection limit when a changed response.status_code assignment has weak pytest evidence.
• fixtures/python_api_json_field_repair_gap now proves a literal client route call such as client.post("/checkout") can link to the route owner and produce a framework-shaped response.json()["detail"] repair card.
• fixtures/python_dynamic_route_registration_limit now proves dynamic route registration fails closed as a named dynamic_route_registration limitation without producing a repair card or agent-packet-eligible canonical gap.
• Arbitrary decorators remain fail-closed through python_decorator_indirection_limit.

Work item: analysis/python-cli-output-pack-v1

Status: done

Blocked by:

• output/python-surface-projection

Goal

Support Python CLI/output repair cards.

Acceptance

• Simple Click, Typer, argparse, print, stdout/stderr, and exit-code shapes can produce output assertion cards.
• Ambiguous command construction remains non-actionable.

Progress

• fixtures/python_cli_output_repair_gap now proves a simple Click @click.command() owner with changed click.echo(...) output routes to a bounded pytest repair card with output contains ..., a suggested existing test, and a focused verify command.
• fixtures/python_argparse_output_repair_gap now proves an argparse-shaped command can route a changed static print(...) output to a bounded pytest CLI output repair card without importing argparse or executing tests.
• Unit coverage pins Typer app.command transparency only when a typer import is present, keeps custom app.command decorators fail-closed, and recognizes click.echo, typer.echo, sys.stdout.write, sys.stderr.write, and simple literal sys.exit / SystemExit exit-code discriminators.
• Python repair-card copy now specializes output/call-effect side effects into CLI output or CLI exit-code assertion guidance when the missing discriminator is output text or a literal exit code.

Work item: analysis/python-parametrized-boundaries

Status: done

Blocked by:

• analysis/python-repair-classes-v1

Goal

Suggest native pytest parameterization for clear boundary predicates.

Acceptance

• Suggest parameterization only when candidate values are explainable.
• Simpler one-case test remains available.
• Expected values are not invented without uncertainty labeling.

Progress

• Predicate-boundary repair cards now keep the equality discriminator as the minimum repair and offer optional pytest below/equal/above parameterized rows only for simple identifier or integer boundaries.
• Suggested assertion copy labels parameterized expected values as domain-specific placeholders instead of inventing outputs.
• Boundary cards add a stop condition telling agents and humans to keep only the equality assertion when below/above expected values are unclear.
• fixtures/python_parametrized_boundary_repair_gap pins the human and JSON output contract for this guidance.

Work item: analysis/python-existing-test-strengthening

Status: done

Blocked by:

• analysis/python-repair-classes-v1

Goal

Prefer strengthening weak related tests over adding redundant tests.

Acceptance

• Cards can distinguish "strengthen existing test" from "add new test".
• Agent packets can restrict edits to one existing test.
• Outcome receipt shows broad oracle becoming more exact.

Delivered

• Direct weak pytest and unittest placements now emit suggested_repair_action: strengthen_existing_test, target the existing weak related test name/node, and verify that test instead of proposing a redundant new test.
• Python repair cards expose repair_action, render "strengthen existing" guidance in human, JSON, pilot, SARIF, and GitHub-projected card payloads, and keep the preview/advisory receipt boundary.
• Check-output-derived Python GapRecords map strengthening cards to StrengthenExistingTest, so ripr agent packet --gap-ledger ... emits task = "strengthen_targeted_test" with the existing test file as the allowed edit surface and production Python files forbidden.
• Python fixture goldens now pin the stronger routing across predicate, return, exception, field/object, output/log/call-effect, pytest, and unittest examples.

Work item: swarm/python-gap-work-queue

Status: complete

Blocked by:

• swarm/python-agent-packet-export

Goal

Make multiple Python repair cards shardable.

Acceptance

• Queue entries include canonical gap ID, priority, owner, allowed edit files, verify command, expected receipt, and conflict group.
• Same-file conflicts and stale entries are visible.

Progress

• ripr swarm queue --language python ranks packetable Python GapRecords into conflict-grouped advisory work and excludes no-action, static-limit, and non-packetable records.
• Queue rendering now fails closed when a gap ledger omits root provenance or declares a different root from the selected --root, returning a blocked queue with no packets instead of assigning rootless, stale, or wrong-workspace repair work.
• Queue packets now surface explicit stale receipt movement from GapRecords as queue_state = "blocked_stale", staleness_status = "stale", and summary.stale_total, so a closed or stale Python repair packet is visible but not silently assignable.

Work item: swarm/python-agent-result-ingestion

Status: done

Blocked by:

• swarm/python-gap-work-queue

Goal

Classify agent repair attempts without trusting them blindly.

Acceptance

• Ingested results classify closed, partially improved, verify failed, edited forbidden file, uncertain, and stale outcomes.
• Production-code edits are flagged.
• Verify result and before/after movement are attached.

Delivered

• ripr swarm ingest --result <agent-result.json> reads one external agent result artifact, validates that the result path stays under the selected root, and emits an advisory swarm-ingest JSON envelope without rerunning tests, writing receipts, calling providers, generating tests, or editing files.
• Ingest classification now distinguishes closed, partially_improved, verify_failed, edited_forbidden_file, stopped_by_agent, stale_packet, and uncertain; missing verify evidence stays uncertain, and forbidden production-code edits are flagged before any reported success claim.
• The Python preview first-PR fixture now includes an agent-result input and expected ingest output proving that a test-only edit with passing verify evidence and resolved receipt movement becomes closed / attempt_outcome = "resolved" while keeping trusted_success = false for operator review.

Work item: outcome/python-gap-ledger

Status: in progress

Blocked by:

• swarm/python-agent-result-ingestion

Goal

Make Python gap improvement durable.

Acceptance

• Receipts show closed, new, unchanged, weakened, and strengthened Python gaps.
• Canonical Python gaps can open and close across runs.
• PR summary can report scoped Python gap movement without claiming correctness beyond static evidence movement.

Progress

• ripr outcome can compare Python check-output JSON snapshots by canonical gap ID and report weak-to-strong evidence movement as closed.
• fixtures/first_successful_pr/python-preview-gap now pins the same before/after check-output path with expected ripr outcome JSON and Markdown receipts for closed, unchanged, opened, strengthened, and weakened movement, proving the first-PR Python preview gap can close, remain weak, partially improve, weaken, or reopen without a Python-only receipt command.
• Strengthened-but-still-weak rows now stay visible in receipt remaining_weak_or_unknown output instead of being mistaken for closure.
• fixtures/first_successful_pr/python-return-gap pins a non-boundary return-value receipt where broad pytest evidence strengthens to an exact return assertion and closes the canonical Python gap.
• fixtures/first_successful_pr/python-exception-gap pins exception-path receipt movement where broad exception evidence strengthens to exact pytest.raises(..., match=...) message evidence and closes the canonical Python gap.
• fixtures/first_successful_pr/python-field-gap pins field/object receipt movement where broad object truthiness strengthens to exact returned-field evidence and closes the canonical Python gap.
• fixtures/first_successful_pr/python-output-gap pins output/log receipt movement where broad output smoke strengthens to exact output text evidence and closes the canonical Python gap.
• ripr reports gap-ledger --check-output now carries the corresponding receipt command into repairable Python GapRecords, which makes bounded packet delegation receipt-ready.
• Outcome, review-receipt, and agent-verify JSON now include summary.gap_movement counts for closed, opened, strengthened, weakened, unchanged, new, removed, and changed canonical gaps, so Python repair-loop receipts expose closure without requiring row-by-row inspection.

Work item: fixtures/python-false-positive-corpus

Status: done

Blocked by:

• outcome/python-gap-ledger

Goal

Prevent Python support from becoming noisy.

Acceptance

• Fixtures cover dynamic imports, monkeypatch-only behavior, generated files, metaclass/decorator magic, unresolved pytest fixtures, property-based tests with opaque discriminators, custom assertion helpers, async tests, broad smoke tests, reach-without-observe, and duplicate raw signals.
• Unsupported cases produce named limitations and do not enter the repair queue.

Delivered

• Check-output-derived Python static-limit findings now become report-only StaticLimitation GapRecords with repairability = "analyzer_limitation". The swarm queue excludes those records instead of turning preview limitations into agent repair packets.
• python_decorator_indirection_limit pins decorated owners as decorator_indirection static limitations, so decorator-modified call semantics are named instead of treated as hidden analyzer truth.
• python_opaque_custom_helper_limit pins custom assertion helpers as opaque_custom_assertion_helper static limitations so the adapter does not route a repair packet when the helper body might already observe the changed discriminator.
• python_property_based_limit pins Hypothesis-style property-based tests as property_based_test static limitations so the adapter does not infer that generated inputs cover a concrete missing discriminator.
• python_unresolved_fixture_limit pins pytest fixture-sourced inputs and expected values as unresolved_pytest_fixture static limitations so the adapter does not turn opaque fixture data into a repair packet or a discriminator claim.
• Static-limit findings now keep revealability/discriminator evidence unknown, even when a related test has an exact-looking oracle, because the named limitation prevents a safe discriminator claim.
• python_monkeypatch_module_limit pins pytest monkeypatch.setattr(...) substitution as a mocked_module static limitation so monkeypatch-only related tests stay visible but do not become repair cards, canonical gaps, or swarm packets.
• python_generated_file_excluded pins detectable generated Python file diffs such as *_pb2.py as excluded from preview diff analysis, so generated-code edits do not produce repair cards, canonical repair gaps, or swarm packets.
• python_dynamic_import_limit pins runtime import calls such as importlib.import_module(...) as missing_import_graph static limitations, so exact-looking related tests stay visible but do not become repair cards, canonical gaps, or swarm packets.
• python_metaclass_limit pins class ...(metaclass=...) declarations as metaprogramming static limitations, so class-level magic is named and kept out of repair cards, canonical gaps, and swarm packets.
• python_async_owner pins async owner and async pytest-style test discovery without executing an event loop or treating async syntax as runtime proof.
• python_broad_boolean_assertion and python_boundary_gap pin broad-smoke and reach-only evidence as weak repair-routing inputs: they can become strengthen-existing-test cards only when a concrete missing discriminator, suggested test target, verify command, stop conditions, and advisory limits are available.
• python_same_line_duplicate_collapse pins a returned dict line containing return, field, and string literal signals as one user-facing canonical repair gap with canonical_gap_group_size = 1, preventing same-line raw-signal noise from inflating Python repair work.
• Unsupported cases produce named limitations and are excluded from swarm queues, while supported weak direct evidence remains repairable only when it can carry a bounded repair card and verify command.

Work item: dogfood/python-real-repo-evals

Status: done

Blocked by:

• fixtures/python-false-positive-corpus

Goal

Prove usefulness outside fixtures.

Acceptance

• Dogfood runs cover a tiny controlled Python repo, normal pytest app repo, API repo, CLI/tooling repo, and mixed repo when relevant.
• Each run records command, runtime, top finding, repair card, verify command, usability, before/after receipt, false-positive notes, and limitation notes.
• At least one gap closes with receipt before promotion is considered.

Progress

• fixtures/real-repair-attempts/corpus.json now includes a checked repo-local Python preview receipt where a bounded packet edits only tests/test_pricing.py, keeps app/pricing.py forbidden, verifies with a focused pytest command, and closes the canonical amount >= threshold predicate-boundary gap through ripr outcome.
• cargo xtask dogfood requires that Python receipt row as part of the durable repair-attempt corpus, so the first closed Python packet/receipt loop is visible in the same advisory dogfood report as other swarm repair attempts.
• fixtures/python-real-repo-evals/corpus.json now records a tiny controlled pytest scratch-repo eval where RIPR emits a predicate-boundary repair card, a human-run focused pytest command passes, and ripr outcome closes the canonical Python gap while preserving the preview/advisory claim boundary.
• The same corpus now records a normal pyproject-based pytest app eval where a free_shipping_offer threshold-boundary change routes to a strengthen-existing-test repair card, the focused pytest verify command passes, and ripr outcome closes the canonical Python gap.
• The same corpus now records a native pytest parametrized-boundary eval where a changed amount >= threshold predicate routes to a strengthen-existing-test card with optional below/equal/above row guidance, the bounded packet edits only tests/test_tax.py, the focused pytest verify command passes, and ripr outcome closes the canonical Python gap.
• The same corpus now records a CLI/output-style pytest eval where a changed print(...) side effect routes to a strengthen-existing-test repair card, the focused capsys pytest verify command passes, and ripr outcome closes the canonical Python output/call-effect gap.
• The same corpus now records a Click-shaped CLI output pytest eval where a changed click.echo(...) side effect routes to a strengthen-existing-test repair card, exports a bounded test-only packet, verifies with focused pytest against a local Click-shaped stub, and ripr outcome closes the canonical Python output/call-effect gap.
• The same corpus now records a Typer-shaped CLI output pytest eval where a changed typer.echo(...) side effect under @app.command() routes to a strengthen-existing-test repair card, exports a bounded test-only packet, verifies with focused pytest against a local Typer-shaped stub, and ripr outcome closes the canonical Python output/call-effect gap.
• The same corpus now records a lightweight API-handler pytest eval where a changed response.status_code assignment routes to a field/object repair card, the focused status-code pytest verify command passes, and ripr outcome closes the canonical Python API status gap.
• The same corpus now records a mixed Rust/Python pytest eval where a Python amount >= threshold predicate-boundary change routes to a repair card despite Cargo metadata, the focused pytest verify command passes, and ripr outcome closes the canonical Python gap.
• The same corpus now records a decorated route pytest eval where a simple @api.post(...) route handler changes response.status_code, RIPR emits a field/object repair card with missing discriminator response.status_code == 422, the focused pytest verify command passes, and ripr outcome closes the canonical Python gap.
• The same corpus now records a dataclass/model-field pytest eval where a changed returned constructor field routes to a strengthen-existing-test card with missing discriminator result.active == True, the bounded packet edits only tests/test_users.py, the focused pytest verify command passes, and ripr outcome closes the canonical Python gap.
• cargo xtask dogfood projects the Python real-repo eval corpus into the dogfood report as receipt-backed eval evidence separate from analyzer fixture goldens.
• The corpus now supplies the receipt-backed dogfood evidence consumed by the route-quality metrics and scoped support-tier review.

Work item: metrics/python-repair-routing-quality

Status: done

Blocked by:

• none; dogfood real-repo eval receipts are now fixture-backed.

Goal

Measure Python quality by repair usefulness, not finding volume.

Acceptance

• Metrics include time to first useful finding, top-1/top-3 actionable precision, verify-command validity, agent-packet boundary validity, concrete-discriminator rate, related-test-location rate, false-actionable rate, crash rate, unsupported limitation distribution, and receipt closure rate.
• Noisy changes fail quality gates.

Progress

• cargo xtask dogfood now derives Python repair-routing quality metrics from fixtures/python-real-repo-evals/corpus.json: top-1 actionable usefulness, top-3 actionable precision over captured ranked repair-card findings, verify-command validity, agent-packet boundary validity, concrete-discriminator coverage, suggested test-location coverage, false-actionable rate, crash rate, receipt closure rate, and unsupported limitation distribution.
• The Python eval corpus now records structured unsupported limitation kinds, and the decorated-route eval contributes dynamic_route_registration to the limitation distribution while keeping the support-tier boundary explicit.
• The Python eval corpus now records ranked top-3 repair-card findings for each dogfood case. Cases with fewer than three ranked repair cards must explain the capture limit so top-3 precision is measured without hiding sparse output.
• Corpus validation fails if the checked top Python repair cards become noisy: unusable top-1 card, invalid verify command, missing concrete discriminator, missing suggested test location, false actionability, crash/contract error, or no closed receipt. Validation also fails when ranked top-3 finding capture is missing, malformed, or not usable, concrete, placed, verifiable, and false-positive clean.
• This metric slice landed before support-tier review, so promotion evidence is based on top-finding usefulness and closure movement rather than raw finding volume.

Work item: campaign/python-usable-alpha-promotion

Status: done

Blocked by:

• none; metrics and dogfood receipt evidence are checked.

Goal

Promote Python only when the repair loop has receipt-backed evidence.

Acceptance

• Support docs, README claims, examples, capability matrix, traceability, and closeout evidence are updated by a dedicated promotion PR.
• Promotion target is at most usable alpha unless stronger evidence exists.
• Docs state that Python support provides static repair-routing for common pytest/unittest workflows and does not prove correctness, execute arbitrary code, or guarantee mutation adequacy.
• Source ripr remains the release/distribution authority.

Delivered

• docs/status/SUPPORT_TIERS.md now promotes only the scoped Python repair-routing loop to usable alpha: selected pytest/unittest repair cards, verify commands, bounded agent packets, queue/ingest handling, and before/after receipts.
• Root README, Quickstart, the language-adapter workflow, the capability matrix, and traceability now keep the same claim boundary: broader Python static facts and static limits remain preview/advisory.
• docs/handoffs/2026-05-31-python-repair-routing-usable-alpha-closeout.md records the proof commands, usable-alpha scope, remaining limits, policy non-claims, and next work.

Work item: dogfood/python-stability-evals-v1

Status: in progress

Blocked by:

• campaign/python-usable-alpha-promotion

Goal

Extend Python repair-routing evidence after usable alpha before any broader support-tier consideration.

Acceptance

• Add or refresh real or external-repo-style Python repair-routing evals beyond the promotion corpus.
• Each eval records command, runtime, top finding, repair card, agent packet, verify command, receipt or no-receipt reason, false-positive notes, and limitation notes.
• Route-quality metrics continue to emphasize top-1 usefulness, top-3 precision, verify-command validity, agent-packet boundary validity, concrete discriminators, suggested test location, false-actionable rate, crash rate, receipt closure, and limitation distribution.
• No support-tier promotion, gate eligibility, badge authority, baseline/RIPR Zero inclusion, provider calls, generated tests, arbitrary imports, mutation execution, or production-code edit authority changes.

Proof commands

cargo xtask dogfood
cargo xtask metrics
cargo xtask check-capabilities
cargo xtask check-traceability
cargo xtask check-output-contracts
cargo xtask check-pr
git diff --check

Rollback

• Revert the added eval entries, generated dogfood/metric expectations, and any docs or capability links. The scoped usable-alpha support claim remains unchanged.

Progress

• fixtures/python-real-repo-evals/corpus.json now records bounded agent packet fields for every checked Python dogfood eval: packet command, allowed test files, forbidden production files, and stop conditions.
• The dogfood checker now requires at least one full top-3 Python repair-card capture, and the corpus records multi_card_pytest_top3_receipt where three usable repair cards are ranked while only the top packet is delegated and receipted.
• The corpus now adds unittest_return_value_receipt as a post-promotion stability eval where a unittest return-value repair routes to one existing test method, verifies with python -m unittest, exports a bounded test-only packet, and closes the canonical Python gap through ripr outcome.
• Returned Python dict field discriminators now prefer a literal-valued field over a pass-through field in multi-field return literals, so {"name": name, "status": "active"} routes to the missing discriminator status == "active" instead of the non-discriminating name == name.
• The corpus now adds unittest_dict_field_receipt as a post-promotion stability eval where a unittest returned-dict field repair routes to one existing test method, verifies with python -m unittest, exports a bounded test-only packet, and closes the canonical Python field/object gap through ripr outcome.
• The corpus now adds src_layout_pytest_boundary_receipt as a post-promotion external-repo-style stability eval where a src/ package pytest app routes a boundary repair to one existing test file, exports a bounded packet that forbids the package production module, verifies with a focused pytest command, and closes the canonical Python gap through ripr outcome.
• The corpus now adds no_config_pyproject_boundary_receipt as a post-promotion project-detection stability eval where a pyproject-based Python repo without ripr.toml still routes a boundary repair to one existing pytest file, exports a bounded packet that forbids the production module, verifies with a focused pytest command, and closes the canonical Python gap through ripr outcome.
• The corpus now adds api_json_detail_pytest_receipt as a post-promotion stability eval where an API response JSON detail repair routes to one existing pytest method, verifies with a focused pytest command, exports a bounded test-only packet, and closes the canonical Python gap through ripr outcome.
• The corpus now adds flask_route_json_detail_pytest_receipt as a post-promotion stability eval where a Flask-style @app.route("/checkout", methods=["POST"]) response JSON detail repair routes to one existing pytest method, verifies with focused pytest against a local Flask-shaped stub, exports a bounded test-only packet, and closes the canonical Python gap through ripr outcome.
• The corpus now adds fastapi_route_json_detail_pytest_receipt as a post-promotion stability eval where a FastAPI-style @app.post("/checkout") response JSON detail repair routes to one existing pytest method, verifies with focused pytest against a local FastAPI-shaped stub, exports a bounded test-only packet, and closes the canonical Python gap through ripr outcome.
• The corpus now adds async_return_pytest_receipt as a post-promotion stability eval where an async def return-value repair routes to one existing pytest method that calls the owner through asyncio.run, verifies with focused pytest without requiring pytest-asyncio, exports a bounded test-only packet, and closes the canonical Python gap through ripr outcome.
• The corpus now adds exception_path_pytest_receipt as a post-promotion stability eval where a broad pytest.raises(ValueError) observer strengthens to exact pytest.raises(..., match=...) evidence, verifies with a focused pytest command, exports a bounded test-only packet, and closes the canonical Python exception gap through ripr outcome.
• The corpus now adds custom_exception_pytest_receipt as a post-promotion stability eval where a broad pytest.raises(ExpiredCouponError) observer strengthens to exact pytest.raises(..., match=...) evidence for a custom exception type, verifies with a focused pytest command, exports a bounded test-only packet, and closes the canonical Python exception gap through ripr outcome.
• The corpus now adds unittest_exception_path_receipt as a post-promotion stability eval where a broad self.assertRaises(ValueError) observer strengthens to exact self.assertRaisesRegex(...) evidence, verifies with python -m unittest, exports a bounded test-only packet, and closes the canonical Python exception gap through ripr outcome.
• Route response-constructor assignments such as response = Response(status_code=422, detail="coupon expired") now route through the field/object repair path for statically recognized Python route owners. The corpus records api_exception_response_pytest_receipt, where a route catches an application exception, constructs a response object, recommends assert response.status_code == 422, verifies with focused pytest, exports a bounded test-only packet, and closes the canonical Python gap through ripr outcome.
• The corpus now records decorator_indirection_no_packet_eval as a post-promotion fail-closed stability eval where RIPR sees a related pytest exact oracle but refuses to emit a repair card, agent packet, verify success, or receipt movement because the changed owner is wrapped by a runtime decorator.
• The corpus now records missing_import_graph_no_packet_eval as a post-promotion fail-closed stability eval where RIPR sees a related pytest exact oracle but refuses to emit a repair card, agent packet, verify success, or receipt movement because the changed behavior depends on an imported implementation outside the static preview import graph.
• The corpus now records metaprogramming_no_packet_eval as a post-promotion fail-closed stability eval where RIPR sees a related pytest exact oracle but refuses to emit a repair card, agent packet, verify success, or receipt movement because the changed behavior depends on runtime-created class semantics.
• The corpus now records mocked_module_no_packet_eval as a post-promotion fail-closed stability eval where RIPR sees a related pytest exact oracle but refuses to emit a repair card, agent packet, verify success, or receipt movement because the related test depends on unittest.mock.patch runtime substitution.
• The corpus now records opaque_custom_helper_no_packet_eval as a post-promotion fail-closed stability eval where RIPR sees a related pytest method and custom assert_* helper oracle but refuses to emit a repair card, agent packet, verify success, or receipt movement because the helper body is opaque to the preview adapter.
• The corpus now records property_based_no_packet_eval as a post-promotion fail-closed stability eval where RIPR sees a related Hypothesis-style pytest method and weak relational oracle but refuses to emit a repair card, agent packet, verify success, or receipt movement because generated inputs do not prove that the changed discriminator is covered.
• The corpus now records unsupported_syntax_no_packet_eval as a post-promotion fail-closed stability eval where RIPR sees a related pytest exact oracle but refuses to emit a repair card, agent packet, verify success, or receipt movement because the changed lambda-return expression is outside the preview adapter's precise syntax model.
• The corpus now records generated_file_no_packet_eval as a post-promotion fail-closed stability eval where a detectable generated Python diff such as src/schema_pb2.py is excluded before probe generation, so RIPR records no repair card, agent packet, verify success, or receipt movement even when a related pytest import exists.
• The corpus now records no_related_test_no_packet_eval under no_action_cases as an ordinary fail-closed no-action eval: RIPR reports no_static_path for a changed Python owner with only unrelated pytest text mention evidence, and the dogfood report records no repair card, no agent packet, not-applicable verify/receipt results, and no receipt movement.
• The corpus now records already_observed_no_packet_eval under no_action_cases as an ordinary fail-closed no-action eval: RIPR reports exposed for a changed Python owner with a related exact-value pytest oracle, and the dogfood report records no repair card, no agent packet, not-applicable verify/receipt results, and no receipt movement because there is no missing proof to route.
• The corpus now records heuristic_only_no_packet_eval under no_action_cases as an ordinary fail-closed no-action eval: RIPR reports weakly_exposed for a changed Python owner with only heuristic pytest test-name similarity, and the dogfood report records no repair card, no agent packet, not-applicable verify/receipt results, and no receipt movement because uncertain related-test proximity is not safe repair-routing evidence.
• Dogfood quality metrics now include agent-packet boundary validity so a future eval that lacks packet scope, stop conditions, or forbidden-file protection fails the checked quality gate instead of counting as usable.
• cargo xtask dogfood now requires the receipt-backed no-config pyproject boundary, async return-value, pytest exception, custom exception, unittest exception, API JSON detail, log output, argparse CLI output, Click CLI output, Typer CLI output, CLI exit-code, Flask route JSON detail, FastAPI route JSON detail, API exception-response, src/ package-layout boundary, unittest return-value, and unittest dict-field eval rows in addition to the original boundary/API/CLI/mixed cases, so those Python closure proofs cannot disappear from the corpus without failing the checked dogfood gate.
• The corpus now adds log_output_pytest_receipt as a post-promotion stability eval where a changed logger.warning(...) side effect routes to a log-output repair card, exports a bounded test-only packet, verifies with a focused pytest/caplog command, and closes the canonical Python call/output-effect gap through ripr outcome.
• The corpus now adds cli_exit_code_pytest_receipt as a post-promotion stability eval where a changed literal sys.exit(2) side effect routes to a CLI exit-code repair card, exports a bounded test-only packet, verifies with focused pytest, and closes the canonical Python call/output-effect gap through ripr outcome.
• The corpus now adds argparse_cli_output_pytest_receipt as a post-promotion stability eval where a changed static print(...) side effect inside an argparse-shaped command routes to a CLI output repair card, exports a bounded test-only packet, verifies with focused pytest, and closes the canonical Python call/output-effect gap through ripr outcome without RIPR importing argparse or executing parser setup.
• The corpus now adds click_cli_output_pytest_receipt as a post-promotion stability eval where a changed click.echo("shipment queued") side effect routes to a CLI output repair card, exports a bounded test-only packet, verifies with focused pytest against a local Click-shaped stub, and closes the canonical Python call/output-effect gap through ripr outcome.
• The corpus now adds typer_cli_output_pytest_receipt as a post-promotion stability eval where a changed typer.echo("shipment queued") side effect under a static Typer @app.command() owner routes to a CLI output repair card, exports a bounded test-only packet, verifies with focused pytest against a local Typer-shaped stub, and closes the canonical Python call/output-effect gap through ripr outcome.