(#25) Divergence check -> review nudge #32

Merged
jacisjake merged 2 commits from task/25-divergence-nudge into main 2026-10-28 11:37:35 +00:00
4 changed files with 252 additions and 0 deletions

No files matched your search

+131
View File
@@ -0,0 +1,131 @@
package main
import (
"encoding/json"
"net/http"
"net/http/httptest"
"testing"
"time"
)
// fakeNoulJev answers every question in a request with a fixed noul probability.
func fakeNoulJev(t *testing.T, noul float64) *JevClient {
t.Helper()
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
var body struct {
Questions map[string]any `json:"questions"`
}
json.NewDecoder(r.Body).Decode(&body)
out := map[string]any{}
for q := range body.Questions {
out[q] = map[string]any{"type": "noul", "noul": noul}
}
json.NewEncoder(w).Encode(map[string]any{"model": "jev-1.14.0", "answers": out,
"usage": map[string]int{"input_tokens": 40, "output_tokens": 0}})
}))
t.Cleanup(srv.Close)
return &JevClient{url: srv.URL, key: "tk", model: "jev-latest", threshold: 0.7, client: srv.Client()}
}
func postDeliver(t *testing.T, ts *httptest.Server, id, author, refersTo, text string) string {
t.Helper()
code, out := post(t, ts.URL+"/api/nests/"+id+"/messages",
`{"author":"`+author+`","kind":"deliver","refers_to":"`+refersTo+`","text":"`+text+`"}`)
if code != http.StatusCreated {
t.Fatalf("deliver %s: %d %v", author, code, out)
}
return out["msg_id"].(string)
}
func findNudge(ts *httptest.Server, id, target string) map[string]any {
for _, m := range nestMessagesNoT(ts, id) {
if m["kind"] == "nudge" && m["refers_to"] == target {
return m
}
}
return nil
}
func nestMessagesNoT(ts *httptest.Server, id string) []map[string]any {
resp, err := http.Get(ts.URL + "/api/nests/" + id + "/messages")
if err != nil {
return nil
}
defer resp.Body.Close()
var msgs []map[string]any
json.NewDecoder(resp.Body).Decode(&msgs)
return msgs
}
func waitForNudge(ts *httptest.Server, id, target string) map[string]any {
deadline := time.Now().Add(2 * time.Second)
for {
if n := findNudge(ts, id, target); n != nil {
return n
}
if time.Now().After(deadline) {
return nil
}
time.Sleep(10 * time.Millisecond)
}
}
func TestRestatementGetsNudge(t *testing.T) {
ts, s := newTestNestServer(t)
s.p.Flock.SetClassifier(fakeNoulJev(t, 0.8)) // above band 0.7
id := newNest(t, ts, "Lab", `"Hugin","Munin"`)
r := postRequest(t, ts, id, "conspiracy", "is this risky")
postDeliver(t, ts, id, "Hugin", r, "ship it")
d2 := postDeliver(t, ts, id, "Munin", r, "also ship it")
n := waitForNudge(ts, id, d2)
if n == nil {
t.Fatalf("restated deliver must get a nudge, log: %+v", s.msgs.All(id))
}
if n["to"] != "Munin" || n["author"] != "harness" {
t.Fatalf("nudge must address the restating crow from the harness, got %v", n)
}
if !containsMsg(s.msgs.All(id), d2) {
t.Fatal("the restated deliver itself must stay in the log")
}
}
func TestDistinctDeliversNoNudge(t *testing.T) {
ts, s := newTestNestServer(t)
s.p.Flock.SetClassifier(fakeNoulJev(t, 0.2)) // below band
id := newNest(t, ts, "Lab", `"Hugin","Munin"`)
r := postRequest(t, ts, id, "conspiracy", "is this risky")
postDeliver(t, ts, id, "Hugin", r, "ship it")
d2 := postDeliver(t, ts, id, "Munin", r, "wait for the audit")
time.Sleep(300 * time.Millisecond)
if n := findNudge(ts, id, d2); n != nil {
t.Fatalf("a distinct deliver must not be nudged, got %v", n)
}
}
func TestNudgeSpendsNoModelTokens(t *testing.T) {
ts, s := newTestNestServer(t)
s.p.Flock.SetClassifier(fakeNoulJev(t, 0.8))
id := newNest(t, ts, "Lab", `"Hugin","Munin"`)
r := postRequest(t, ts, id, "conspiracy", "is this risky")
postDeliver(t, ts, id, "Hugin", r, "ship it")
d2, err := s.msgs.Append(Message{NestID: id, Kind: "deliver", Author: "Munin", Text: "ship", RefersTo: r, Timestamp: time.Now().UTC().Format(time.RFC3339)})
if err != nil {
t.Fatal(err)
}
b := &Budget{name: "nest", limit: 1}
s.checkDivergence(nestOf(s, id), d2, b)
if b.Spent() != 0 {
t.Fatalf("a review nudge must spend no model tokens, spent=%d", b.Spent())
}
}
func containsMsg(msgs []Message, msgID string) bool {
for _, m := range msgs {
if m.MsgID() == msgID {
return true
}
}
return false
}
+55
View File
@@ -252,6 +252,61 @@ func (j *JevClient) ClassifyKind(ctx context.Context, text string) (string, map[
return a.Choice, a.Probabilities, true
}
// divergenceQuestions asks whether a message restates an earlier point. It is
// a noul: JEV_STRATEGIES strategy 3 (crisp yes/no) and noul has no confidence,
// so the probability itself is thresholded (band = threshold).
var divergenceQuestions = map[string]any{
"restates_earlier": map[string]any{
"type": "noul",
"instructions": "Does `this` make a point already made in `earlier`, adding no new angle?",
"criteria": map[string]string{
"true": "It repeats an earlier point without a genuinely new angle.",
"false": "It adds something different from what `earlier` already said.",
},
},
}
// ClassifyDivergence returns the probability that a message restates earlier
// ones, and ok when a value came back (noul always carries one; ok is false
// only on transport error or a missing answer).
func (j *JevClient) ClassifyDivergence(ctx context.Context, state any) (float64, bool) {
if j == nil {
return 0, false
}
answers, _, _, err := j.SystemOne(ctx, state, divergenceQuestions)
if err != nil {
return 0, false
}
a := answers["restates_earlier"]
if a.Noul == nil {
return 0, false
}
return *a.Noul, true
}
// ClassifyDivergence runs the restatement check under the flock guardrails
// (cap, timeout, flight log with the probability recorded).
func (f *Flock) ClassifyDivergence(ctx context.Context, state any, budgets ...*Budget) (float64, bool) {
if f.jev == nil {
return 0, false
}
for _, b := range budgets {
if err := b.Check(); err != nil {
return 0, false
}
}
f.sem <- struct{}{}
defer func() { <-f.sem }()
cctx, cancel := context.WithTimeout(ctx, f.g.CrowTimeout)
defer cancel()
start := time.Now()
p, ok := f.jev.ClassifyDivergence(cctx, state)
if f.log != nil {
f.log.Write(Event{Kind: "divergence", Crow: "Sentry", Model: f.jev.model, Ms: time.Since(start).Milliseconds(), Prob: map[string]float64{"restates_earlier": p}, Note: fmt.Sprint(ok && p >= f.jev.threshold)})
}
return p, ok
}
// ClassifyKind tags a nest message under the same guardrails as any call:
// flock cap, timeout and flight log, with the probabilities recorded.
func (f *Flock) ClassifyKind(ctx context.Context, text string, budgets ...*Budget) (string, map[string]float64, bool) {
+40
View File
@@ -184,6 +184,10 @@ func (s *Server) postNestMessage(w http.ResponseWriter, r *http.Request) {
}
}
if kind == "deliver" {
go s.checkDivergence(nest, record)
}
writeJSON(w, http.StatusCreated, nestRecord(record))
}
@@ -298,6 +302,42 @@ func (s *Server) respondOne(nest Nest, req Message, crow string, lens bool) {
}
if record, err := s.msgs.Append(reply); err == nil {
s.emitNestMessage(record)
if record.Kind == "deliver" {
s.checkDivergence(nest, record)
}
}
}
// checkDivergence asks Jev whether this deliver restates earlier delivers that
// answer the same request. Above the band the crow gets a review nudge; the
// deliver itself stays in the log (nudges never silence).
func (s *Server) checkDivergence(nest Nest, m Message, budgets ...*Budget) {
if m.Kind != "deliver" {
return
}
earlier := s.msgs.EarlierDelivers(nest.ID, m.RefersTo, m.ID)
if len(earlier) == 0 {
return
}
if _, nudged := s.msgs.NudgeFor(nest.ID, m.MsgID()); nudged {
return // one nudge per deliver; no loop
}
prob, ok := s.p.Flock.ClassifyDivergence(s.ctx, map[string]any{"this": m.Text, "earlier": earlier}, budgets...)
if !ok || prob < s.p.Flock.jev.threshold {
return
}
nudge := Message{
NestID: nest.ID,
Kind: "nudge",
Author: "harness",
Text: "This reads like a point already made in the thread; review the earlier delivers and add your distinct angle or let it rest.",
To: m.Author,
RefersTo: m.MsgID(),
TaggedBy: "harness",
Timestamp: time.Now().UTC().Format(time.RFC3339),
}
if record, err := s.msgs.Append(nudge); err == nil {
s.emitNestMessage(record)
}
}
+26
View File
@@ -132,6 +132,32 @@ func (s *MessageStore) FirstClaim(nestID, referentID string) (string, bool) {
return "", false
}
// EarlierDelivers returns the texts of deliver messages answering the same
// referent, oldest first, excluding the one with id excludeID.
func (s *MessageStore) EarlierDelivers(nestID, referentID string, excludeID int) []string {
s.mu.Lock()
defer s.mu.Unlock()
var out []string
for _, m := range s.log[nestID] {
if m.Kind == "deliver" && m.RefersTo == referentID && m.ID != excludeID {
out = append(out, m.Text)
}
}
return out
}
// NudgeFor reports whether a nudge already references a given message.
func (s *MessageStore) NudgeFor(nestID, targetMsgID string) (Message, bool) {
s.mu.Lock()
defer s.mu.Unlock()
for _, m := range s.log[nestID] {
if m.Kind == "nudge" && m.RefersTo == targetMsgID {
return m, true
}
}
return Message{}, false
}
func (s *MessageStore) save(nestID string) error {
if err := os.MkdirAll(s.dir, 0o755); err != nil {
return err