feat(ai-sdk): runnable iace-audit propose CLI + live LLM wiring (P2 slice 3)

Makes the offline proposer runnable end-to-end.

- BuildProposerInput (proposer_input.go): non-test engine->hazards path. The
  PatternMatch->Hazard converter is lifted out of the GT test files into
  production scope so both the tests and the CLI share one pipeline.
- iace-audit propose <narrative.json> [<ground-truth.json>]: detect candidates ->
  GT-screen survivors (when a ground truth is given) -> judge (HeuristicJudge by
  default, LLMJudge over ollama when IACE_PROPOSE_LLM=1) -> write the human-review
  queue to audit-reports/proposals.{md,json}. Propose-only.

Smoke run on a dishwasher narrative: 32 fired -> 3 candidates -> queue with a
confident duplicate, a confident distinct, and one punted to the LLM judge; GT
wall recall-safe. Live qwen is opt-in via env; the heuristic default keeps the
tool runnable (and CI deterministic) without a model. Proposal types 2-4
(foreign-framing gates, vocab->tag, coverage blind spots) remain for slice 4.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
Benjamin Admin
2026-06-25 09:10:43 +02:00
parent 0ce4794767
commit 8440ddfecb
5 changed files with 292 additions and 62 deletions
@@ -7,8 +7,6 @@ import (
"path/filepath"
"sort"
"testing"
"github.com/google/uuid"
)
// TestKistenhub_GTCoverage runs the Kistenhubgeraet ground truth (37 entries)
@@ -110,65 +108,6 @@ func TestKistenhub_GTCoverage(t *testing.T) {
// patternsToHazardsAndMitigations converts a pattern match output into the
// Hazard/Mitigation shapes that CompareBenchmark expects. Mirrors what
// iace_handler_init.go does in production but without DB writes.
func patternsToHazardsAndMitigations(out *MatchOutput) ([]Hazard, []Mitigation) {
hazards := make([]Hazard, 0, len(out.MatchedPatterns))
patternToHazard := make(map[string]uuid.UUID, len(out.MatchedPatterns))
for _, pm := range out.MatchedPatterns {
cat := ""
if len(pm.HazardCats) > 0 {
cat = pm.HazardCats[0]
}
zone := pm.ZoneDE
lifecycle := ""
if len(pm.ApplicableLifecycles) > 0 {
lifecycle = pm.ApplicableLifecycles[0]
}
h := Hazard{
ID: uuid.New(),
Name: pm.ScenarioDE,
Category: cat,
Description: pm.ScenarioDE,
Scenario: pm.ScenarioDE,
TriggerEvent: pm.TriggerDE,
PossibleHarm: pm.HarmDE,
AffectedPerson: pm.AffectedDE,
HazardousZone: zone,
LifecyclePhase: lifecycle,
}
if h.Name == "" {
h.Name = pm.PatternName
}
hazards = append(hazards, h)
patternToHazard[pm.PatternID] = h.ID
}
measureNames := make(map[string]string)
for _, m := range GetProtectiveMeasureLibrary() {
measureNames[m.ID] = m.Name
}
var mitigations []Mitigation
for _, sm := range out.SuggestedMeasures {
name := measureNames[sm.MeasureID]
if name == "" {
name = sm.MeasureID
}
for _, srcPattern := range sm.SourcePatterns {
hid, ok := patternToHazard[srcPattern]
if !ok {
continue
}
mitigations = append(mitigations, Mitigation{
ID: uuid.New(),
HazardID: hid,
Name: name,
})
}
}
return hazards, mitigations
}
func abbrev(s string, max int) string {
if len(s) <= max {
return s