| 1 | package main |
| 2 | |
| 3 | import ( |
| 4 | "reflect" |
| 5 | "testing" |
| 6 | |
| 7 | "reasonix/internal/ablation" |
| 8 | ) |
| 9 | |
| 10 | func TestBuildRunTaskArgsEnablesUnattendedWorkspaceWrites(t *testing.T) { |
| 11 | cfg := suiteConfig{model: "e2e"} |
| 12 | got := buildRunTaskArgs(cfg, "metrics.json", "run.trajectory.jsonl", 12, "fix it") |
| 13 | want := []string{ |
| 14 | "run", "--auto", "--metrics", "metrics.json", |
| 15 | "--trajectory", "run.trajectory.jsonl", |
| 16 | "--model", "e2e", "--max-steps", "12", "fix it", |
| 17 | } |
| 18 | if !reflect.DeepEqual(got, want) { |
| 19 | t.Fatalf("run task args = %v, want %v", got, want) |
| 20 | } |
| 21 | } |
| 22 | |
| 23 | func TestBuildRunTaskArgsPassesTheAblationArmThrough(t *testing.T) { |
| 24 | cfg := suiteConfig{arm: ablation.New(ablation.Evidence, ablation.Planner)} |
| 25 | got := buildRunTaskArgs(cfg, "m.json", "", 0, "fix it") |
| 26 | want := []string{"run", "--auto", "--metrics", "m.json", "--ablate", "evidence,planner", "fix it"} |
| 27 | if !reflect.DeepEqual(got, want) { |
| 28 | t.Fatalf("ablated args = %v, want %v", got, want) |
| 29 | } |
| 30 | } |
| 31 | |
| 32 | func TestBuildRunTaskArgsPassesEffortThroughWithoutLegacyMode(t *testing.T) { |
| 33 | cfg := suiteConfig{effort: "low"} |
| 34 | got := buildRunTaskArgs(cfg, "m.json", "", 0, "fix it") |
| 35 | want := []string{"run", "--auto", "--metrics", "m.json", "--effort", "low", "fix it"} |
| 36 | if !reflect.DeepEqual(got, want) { |
| 37 | t.Fatalf("effort args = %v, want %v", got, want) |
| 38 | } |
| 39 | } |
| 40 | |
| 41 | func TestResolveExperimentAxesHasNoExecutionModeArm(t *testing.T) { |
| 42 | got, err := resolveExperimentAxes("evidence", "warm", "wrong") |
| 43 | if err != nil { |
| 44 | t.Fatal(err) |
| 45 | } |
| 46 | if got.arm.String() != "evidence" || got.cache != "warm" || got.anchor != "wrong" { |
| 47 | t.Fatalf("axes = %+v", got) |
| 48 | } |
| 49 | } |
| 50 | |
| 51 | func TestDefaultSuiteBudgetCoversCurrentFiveTaskBaseline(t *testing.T) { |
| 52 | // The real-provider baseline exceeded 400k after only three successful |
| 53 | // tasks. Keep enough headroom to grade all five instead of silently skipping |
| 54 | // the final scenarios as normal model and cache usage varies. |
| 55 | if defaultSuiteTokenBudget < 800_000 { |
| 56 | t.Fatalf("default suite token budget = %d, want at least 800000", defaultSuiteTokenBudget) |
| 57 | } |
| 58 | } |
| 59 | |
| 60 | func TestNormalizeCacheArm(t *testing.T) { |
| 61 | for input, want := range map[string]string{"": "cold", "cold": "cold", " WARM ": "warm"} { |
| 62 | if got, err := normalizeCacheArm(input); err != nil || got != want { |
| 63 | t.Fatalf("normalizeCacheArm(%q) = %q, %v", input, got, err) |
| 64 | } |
| 65 | } |
| 66 | if _, err := normalizeCacheArm("hot"); err == nil { |
| 67 | t.Fatal("unknown cache arm should fail") |
| 68 | } |
| 69 | } |
| 70 |