| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "testing" |
| 5 | |
| 6 | "reasonix/internal/ablation" |
| 7 | "reasonix/internal/evidence" |
| 8 | ) |
| 9 | |
| 10 | func TestEvidenceAblationStandsDownTheReadinessGate(t *testing.T) { |
| 11 | writer := evidence.Receipt{ToolName: "write_file", Success: true, Write: true, Paths: []string{"a.go"}} |
| 12 | todo := evidence.Receipt{ToolName: "todo_write", Success: true, Todos: []evidence.TodoItem{{Content: "edit", Status: "in_progress"}}} |
| 13 | |
| 14 | gated := &Agent{evidence: readinessLedger(writer, todo)} |
| 15 | if gated.finalReadinessCheckFor().reason == "" { |
| 16 | t.Fatal("control arm must still gate an incomplete todo after a write") |
| 17 | } |
| 18 | |
| 19 | off := &Agent{evidence: readinessLedger(writer, todo), ablation: ablation.New(ablation.Evidence)} |
| 20 | if got := off.finalReadinessCheckFor().reason; got != "" { |
| 21 | t.Fatalf("evidence ablation still gated the final answer: %q", got) |
| 22 | } |
| 23 | |
| 24 | unrelated := &Agent{evidence: readinessLedger(writer, todo), ablation: ablation.New(ablation.Planner)} |
| 25 | if unrelated.finalReadinessCheckFor().reason == "" { |
| 26 | t.Fatal("an unrelated ablation must not disable the readiness gate") |
| 27 | } |
| 28 | } |
| 29 | |
| 30 | func TestCompactionAblationCollapsesTheCachePreservingDeferral(t *testing.T) { |
| 31 | full := &Agent{contextWindow: 100_000, softCompactRatio: 0.5, toolResultSnipRatio: 0.7, compactRatio: 0.8} |
| 32 | soft, snip, high := full.compactThresholds() |
| 33 | if soft != 50_000 || snip != 70_000 || high != 80_000 { |
| 34 | t.Fatalf("control thresholds = %d/%d/%d, want 50000/70000/80000", soft, snip, high) |
| 35 | } |
| 36 | |
| 37 | off := &Agent{contextWindow: 100_000, softCompactRatio: 0.5, toolResultSnipRatio: 0.7, compactRatio: 0.8, |
| 38 | ablation: ablation.New(ablation.Compaction)} |
| 39 | soft, snip, high = off.compactThresholds() |
| 40 | if soft != 50_000 || snip != soft || high != soft { |
| 41 | t.Fatalf("ablated thresholds = %d/%d/%d, want all three at 50000", soft, snip, high) |
| 42 | } |
| 43 | } |
| 44 |