Require golden fixture cutover gates (#105)

* Require golden fixture cutover gates

* Fix test file formatting for CI

Co-authored-by: mrjf <180956+mrjf@users.noreply.github.com>

* Fix parity coverage manifest for new Python tests

Co-authored-by: mrjf <180956+mrjf@users.noreply.github.com>

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: mrjf <180956+mrjf@users.noreply.github.com>

Author: mrjf

.crane/scripts/score.go

@@ -63,6 +63,9 @@ type CutoverGates struct {
 	FunctionalContracts     float64 `json:"functional_contracts"`
 	StateDiffContracts      float64 `json:"state_diff_contracts"`
 	PythonBehaviorContracts float64 `json:"python_behavior_contracts"`
+	GoldenFixtureCorpus     string  `json:"golden_fixture_corpus"`
+	AllGoGoldenTests        string  `json:"all_go_golden_tests"`
+	NoPythonRuntime         string  `json:"no_python_runtime_dependency"`
 	KnownExceptions         int     `json:"known_exceptions"`
 	GoTests                 string  `json:"go_tests"`
 	PythonTests             string  `json:"python_tests"`
@@ -99,6 +102,9 @@ type Score struct {
 	PythonTestsPassing     bool            `json:"python_tests_passing"`
 	GoTestsPassing         bool            `json:"go_tests_passing"`
 	BenchmarksPassing      bool            `json:"benchmarks_passing"`
+	GoldenFixtureCorpus    bool            `json:"golden_fixture_corpus"`
+	AllGoGoldenTests       bool            `json:"all_go_golden_tests"`
+	NoPythonRuntime        bool            `json:"no_python_runtime_dependency"`
 	ParityPassing          int             `json:"parity_passing"`
 	ParityTotal            int             `json:"parity_total"`
 	SourceTestsPassing     int             `json:"source_tests_passing"`
@@ -143,6 +149,9 @@ func computeScore(input scanInput, getenv getenvFunc) (Score, error) {
 	functional := RatioGate{}
 	stateDiff := RatioGate{}
 	behaviorContracts := RatioGate{}
+	goldenFixtureCorpus := BoolGate{}
+	allGoGoldenTests := BoolGate{}
+	noPythonRuntime := BoolGate{}
 
 	for scanner.Scan() {
 		line := scanner.Text()
@@ -151,7 +160,21 @@ func computeScore(input scanInput, getenv getenvFunc) (Score, error) {
 		}
 		if gate, ok := parseGateEvent(line); ok {
 			eventsSeen++
-			applyGateEvent(gate, &pythonReference, &surface, &help, &functional, &stateDiff, &behaviorContracts, &knownExceptions, &pythonTests, &benchmarks)
+			applyGateEvent(
+				gate,
+				&pythonReference,
+				&surface,
+				&help,
+				&functional,
+				&stateDiff,
+				&behaviorContracts,
+				&goldenFixtureCorpus,
+				&allGoGoldenTests,
+				&noPythonRuntime,
+				&knownExceptions,
+				&pythonTests,
+				&benchmarks,
+			)
 			continue
 		}
 
@@ -163,7 +186,21 @@ func computeScore(input scanInput, getenv getenvFunc) (Score, error) {
 
 		if ev.Output != "" {
 			if gate, ok := parseGateEvent(ev.Output); ok {
-				applyGateEvent(gate, &pythonReference, &surface, &help, &functional, &stateDiff, &behaviorContracts, &knownExceptions, &pythonTests, &benchmarks)
+				applyGateEvent(
+					gate,
+					&pythonReference,
+					&surface,
+					&help,
+					&functional,
+					&stateDiff,
+					&behaviorContracts,
+					&goldenFixtureCorpus,
+					&allGoGoldenTests,
+					&noPythonRuntime,
+					&knownExceptions,
+					&pythonTests,
+					&benchmarks,
+				)
 			}
 			if n, ok := approvedExceptionCount(ev.Output); ok && n > knownExceptions {
 				knownExceptions = n
@@ -253,6 +290,9 @@ func computeScore(input scanInput, getenv getenvFunc) (Score, error) {
 		FunctionalContracts:     functional.Percent(),
 		StateDiffContracts:      stateDiff.Percent(),
 		PythonBehaviorContracts: behaviorContracts.Percent(),
+		GoldenFixtureCorpus:     passFail(goldenFixtureCorpus.OK()),
+		AllGoGoldenTests:        passFail(allGoGoldenTests.OK()),
+		NoPythonRuntime:         passFail(noPythonRuntime.OK()),
 		KnownExceptions:         knownExceptions,
 		GoTests:                 passFail(goTestsPass),
 		PythonTests:             passFail(pythonTests.OK()),
@@ -275,6 +315,9 @@ func computeScore(input scanInput, getenv getenvFunc) (Score, error) {
 		gates.FunctionalContracts == 1.0 &&
 		gates.StateDiffContracts == 1.0 &&
 		gates.PythonBehaviorContracts == 1.0 &&
+		gates.GoldenFixtureCorpus == "pass" &&
+		gates.AllGoGoldenTests == "pass" &&
+		gates.NoPythonRuntime == "pass" &&
 		gates.KnownExceptions == 0 &&
 		gates.GoTests == "pass" &&
 		gates.PythonTests == "pass" &&
@@ -315,6 +358,9 @@ func computeScore(input scanInput, getenv getenvFunc) (Score, error) {
 		PythonTestsPassing:     gates.PythonTests == "pass",
 		GoTestsPassing:         gates.GoTests == "pass",
 		BenchmarksPassing:      gates.Benchmarks == "pass",
+		GoldenFixtureCorpus:    gates.GoldenFixtureCorpus == "pass",
+		AllGoGoldenTests:       gates.AllGoGoldenTests == "pass",
+		NoPythonRuntime:        gates.NoPythonRuntime == "pass",
 		ParityPassing:          metrics.ParityPassing,
 		ParityTotal:            metrics.ParityTotal,
 		SourceTestsPassing:     metrics.SourceTestsPassing,
@@ -344,6 +390,9 @@ func applyGateEvent(
 	functional *RatioGate,
 	stateDiff *RatioGate,
 	behaviorContracts *RatioGate,
+	goldenFixtureCorpus *BoolGate,
+	allGoGoldenTests *BoolGate,
+	noPythonRuntime *BoolGate,
 	knownExceptions *int,
 	pythonTests *BoolGate,
 	benchmarks *RatioGate,
@@ -361,6 +410,12 @@ func applyGateEvent(
 		*stateDiff = RatioGate{Seen: true, Passing: gate.Passing, Total: gate.Total}
 	case "python_behavior_contracts":
 		*behaviorContracts = RatioGate{Seen: true, Passing: gate.Passing, Total: gate.Total}
+	case "golden_fixture_corpus":
+		*goldenFixtureCorpus = BoolGate{Seen: true, Passed: gate.Passed}
+	case "all_go_golden_tests":
+		*allGoGoldenTests = BoolGate{Seen: true, Passed: gate.Passed}
+	case "no_python_runtime_dependency":
+		*noPythonRuntime = BoolGate{Seen: true, Passed: gate.Passed}
 	case "known_exceptions":
 		*knownExceptions = gate.Count
 	case "python_tests":
@@ -412,6 +467,9 @@ func gateResults(gates CutoverGates) []GateResult {
 		{Name: "functional_contracts", Passing: gates.FunctionalContracts == 1.0},
 		{Name: "state_diff_contracts", Passing: gates.StateDiffContracts == 1.0},
 		{Name: "python_behavior_contracts", Passing: gates.PythonBehaviorContracts == 1.0},
+		{Name: "golden_fixture_corpus", Passing: gates.GoldenFixtureCorpus == "pass"},
+		{Name: "all_go_golden_tests", Passing: gates.AllGoGoldenTests == "pass"},
+		{Name: "no_python_runtime_dependency", Passing: gates.NoPythonRuntime == "pass"},
 		{Name: "python_tests_pass", Passing: gates.PythonTests == "pass"},
 		{Name: "benchmarks_pass", Passing: gates.Benchmarks == "pass"},
 		{Name: "no_known_exceptions", Passing: gates.KnownExceptions == 0},

cmd/apm/CUTOVER.md

@@ -13,7 +13,12 @@ via PyInstaller packaging and `pip install apm-cli`.
 The Go CLI currently implements:
 - `apm --help` / `apm --version` (full parity with Python)
 - `apm init [--yes] [PROJECT_NAME]` (functional, creates apm.yml)
-- Per-command `--help` for all 26 commands (golden-file verified)
+- Per-command `--help` for all 26 commands (initial golden-file coverage)
+
+The checked-in `cmd/apm/testdata/golden/` files are the start of the
+cutover corpus, not final completion proof. Final completion requires the
+full command matrix below to be represented as committed fixtures and replayed
+by Go without invoking the Python runtime.
 
 Most remaining commands are wired at the CLI surface. That is not enough for
 cutover. A command that prints success without writing the expected files,
@@ -84,8 +89,18 @@ are true:
 4. Python-vs-Go parity tests pass for all commands in the matrix
 5. Migration benchmarks pass real fixture-backed command workloads and emit a
    passing counted `benchmarks` gate
-6. `go build ./cmd/apm` produces a single static binary
-7. CI passes on the crane PR branch (`crane/crane-migration-python-to-go-full-apm-cli-rewrite`)
+6. The final Python-reference parity run has been frozen into a committed,
+   versioned golden fixture corpus. The corpus must include CLI inventory,
+   help and usage output, error output, exit codes, generated files, lockfiles,
+   config files, managed-file manifests, deterministic cache/config layout, and
+   audit artifacts for the full command matrix.
+7. An all-Go golden replay passes against that corpus with no live Python
+   oracle. The replay must build `cmd/apm` and compare only the Go binary
+   against checked-in fixtures.
+8. A no-Python-runtime check passes: `APM_PYTHON_BIN` is unset, the Python CLI
+   is hidden or unavailable to the replay, and the golden replay still passes.
+9. `go build ./cmd/apm` produces a single static binary
+10. CI passes on the crane PR branch (`crane/crane-migration-python-to-go-full-apm-cli-rewrite`)
 
 ## Cutover Steps
 
@@ -102,13 +117,14 @@ When conditions are met:
 
 ## Python Compatibility Shim
 
-Until all commands are implemented in Go, the Python CLI remains the
-authoritative `apm` command. The Go binary is available as `apm-go`
-for testing.
+Until all commands are implemented in Go and the golden replay gate passes, the
+Python CLI remains the authoritative `apm` command. The Go binary is available
+as `apm-go` for testing.
 
-The shim removal plan: once the command matrix passes functional tests,
-the Python entrypoint is replaced by the Go binary in the same PR that
-passes the final parity tests.
+The shim removal plan: once the command matrix passes functional tests, the
+final Python-reference behavior is frozen into golden fixtures. Only after the
+all-Go replay passes without a Python runtime can the Python entrypoint be
+replaced by the Go binary.
 
 ## Timeline