measure: claim-24 unfloored semantic leg vs bm25lex baseline - net +0, NOT-SHOWN

gains q14,q25 (gold at global cosine rank 1, previously discarded by the 0.60
floor); losses q15,q28 (the semantic leg was EMPTY on those queries under the
floor, so filling it turns a 2-leg rotation into a 3-leg one and halves the
associative leg's share of the top 5). Guards held: exact_rare 6/6, phrase 7/7,
nonsense 2/3, superseded 2/3 outranks. recall@10 61.8->65.4pp, latency 0.99x.
Baseline reproduced from source (results-bm25base-rerun.json is byte-identical
to the committed results-bm25lex.json), candidate deterministic across 2 runs.

tools/retrieval-eval/ceiling.py

@@ -1,43 +0,0 @@
-import json,sys,pickle,numpy as np,itertools
-sys.path.insert(0,'.')
-from policy2 import legs3,outcome,G,NODES,merge
-# cache per-query leg id-lists, floored and unfloored
-cache={}
-for q in G['queries']:
-    Lf,Sf,Af=legs3(q['query'])
-    Lu,Su,Au=legs3(q['query'],unfloor=True)
-    cache[q['id']]=dict(L=Lf,Sf=Sf,A=Af,Su=Su,Au=Au)
-pickle.dump(cache,open('ceil.pkl','wb'))
-def mrg(pattern,L,S,A,lim=10):
-    out=[];p={'L':0,'S':0,'A':0};src={'L':L,'S':S,'A':A}
-    i=0
-    while len(out)<lim:
-        prog=False
-        for ch in pattern:
-            lst=src[ch]
-            if p[ch]<len(lst):
-                x=lst[p[ch]];p[ch]+=1;prog=True
-                if x not in out: out.append(x)
-                if len(out)>=lim: return out
-        if not prog: break
-    return out
-def ev(pattern,unfl):
-    res={}
-    for q in G['queries']:
-        c=cache[q['id']]
-        S=c['Su'] if unfl else c['Sf']
-        ids=[NODES[i]['id'] for i in mrg(pattern,c['L'],S,c['A'],10)]
-        res[q['id']]=outcome(q,ids)
-    return res
-base=ev('LSA',False)
-print("baseline",sum(base.values()))
-best=[]
-pats=['LSA','LAS','SLA','ALS','SAL','ASL','LSSA','LSASA','LSAA','LSSAA','LSAS','SSLA','LLSA','SALSA','LSAAS']
-for unfl in (False,True):
-    for p in pats:
-        r=ev(p,unfl)
-        g=sorted(k for k in base if r[k] and not base[k]);l=sorted(k for k in base if base[k] and not r[k])
-        best.append((len(g)-len(l),p,unfl,g,l))
-best.sort(reverse=True)
-for n,p,u,g,l in best[:10]:
-    print("net=%+d  pat=%-6s unfloor=%s gains=%s losses=%s"%(n,p,u,g,l))

tools/retrieval-eval/cmp-bm25-vs-claim24.json

@@ -1,6 +1,6 @@
 {
  "baseline": "bm25lex",
- "candidate": "wsclaim24",
+ "candidate": "claim24",
  "n_shared_queries": 38,
  "fixed_by_candidate": [
   "q14",
@@ -8,13 +8,11 @@
  ],
  "broken_by_candidate": [
   "q15",
-  "q28",
-  "q33",
-  "q34"
+  "q28"
  ],
- "discordant": 6,
- "net_queries": -2,
- "mcnemar_exact_p": 0.6875,
+ "discordant": 4,
+ "net_queries": 0,
+ "mcnemar_exact_p": 1.0,
  "min_detectable_swing_queries": 6,
  "observed_run_to_run_drift_queries": 0,
  "noise_floor_queries": 6,
@@ -82,22 +80,22 @@
   "n_scored": 35,
   "hit@5": 0.7428571428571429,
   "recall@5": 0.5768475572047,
-  "recall@10": 0.6563414759843332,
+  "recall@10": 0.6537440733869305,
   "precision@5": 0.19428571428571437,
-  "mrr@10": 0.5026530612244898,
-  "nonsense_clean": "0/3",
+  "mrr@10": 0.5021428571428572,
+  "nonsense_clean": "2/3",
   "superseded_outranks": "2/3",
-  "latency_ms_p50": 524.8,
-  "latency_ms_p95": 738.7,
-  "latency_ms_max": 755.8,
+  "latency_ms_p50": 1173.7,
+  "latency_ms_p95": 1623.0,
+  "latency_ms_max": 1647.9,
   "errors": 0,
   "by_category": {
    "associative": {
     "n": 6,
     "hit@5": 0.5,
     "recall@5": 0.07575757575757576,
-    "recall@10": 0.13636363636363635,
-    "mrr@10": 0.23214285714285712
+    "recall@10": 0.12121212121212122,
+    "mrr@10": 0.22916666666666666
    },
    "exact_rare": {
     "n": 6,
@@ -108,8 +106,8 @@
    },
    "nonsense": {
     "n": 3,
-    "clean": 0,
-    "avg_false_positives": 10.0
+    "clean": 2,
+    "avg_false_positives": 3.3333333333333335
    },
    "paraphrase": {
     "n": 13,
@@ -137,6 +135,12 @@
  },
  "repeat_variance": {
   "baseline": {
+   "runs": 3,
+   "hit@5_min": 0.7428571428571429,
+   "hit@5_max": 0.7428571428571429,
+   "spread_queries": 0
+  },
+  "candidate": {
    "runs": 2,
    "hit@5_min": 0.7428571428571429,
    "hit@5_max": 0.7428571428571429,

tools/retrieval-eval/comparison-bm25lex-vs-wordstart.json

@@ -1,146 +0,0 @@
-{
- "baseline": "bm25lex",
- "candidate": "wordstart",
- "n_shared_queries": 38,
- "fixed_by_candidate": [
-  "q35"
- ],
- "broken_by_candidate": [],
- "discordant": 1,
- "net_queries": 1,
- "mcnemar_exact_p": 1.0,
- "min_detectable_swing_queries": 6,
- "observed_run_to_run_drift_queries": 0,
- "noise_floor_queries": 6,
- "verdict": "no measurable difference",
- "baseline_aggregate": {
-  "n_queries": 38,
-  "n_scored": 35,
-  "hit@5": 0.7428571428571429,
-  "recall@5": 0.5536485340056769,
-  "recall@10": 0.6175677497106068,
-  "precision@5": 0.20000000000000007,
-  "mrr@10": 0.5021428571428571,
-  "nonsense_clean": "2/3",
-  "superseded_outranks": "2/3",
-  "latency_ms_p50": 1184.4,
-  "latency_ms_p95": 1620.0,
-  "latency_ms_max": 1655.4,
-  "errors": 0,
-  "by_category": {
-   "associative": {
-    "n": 6,
-    "hit@5": 0.6666666666666666,
-    "recall@5": 0.08857808857808858,
-    "recall@10": 0.23310023310023312,
-    "mrr@10": 0.25
-   },
-   "exact_rare": {
-    "n": 6,
-    "hit@5": 1.0,
-    "recall@5": 1.0,
-    "recall@10": 1.0,
-    "mrr@10": 1.0
-   },
-   "nonsense": {
-    "n": 3,
-    "clean": 2,
-    "avg_false_positives": 3.3333333333333335
-   },
-   "paraphrase": {
-    "n": 13,
-    "hit@5": 0.6153846153846154,
-    "recall@5": 0.6153846153846154,
-    "recall@10": 0.6153846153846154,
-    "mrr@10": 0.2846153846153846
-   },
-   "phrase": {
-    "n": 7,
-    "hit@5": 1.0,
-    "recall@5": 0.5494614512471656,
-    "recall@10": 0.6023242630385487,
-    "mrr@10": 0.8214285714285714
-   },
-   "superseded": {
-    "n": 3,
-    "hit@5": 0.3333333333333333,
-    "recall@5": 0.3333333333333333,
-    "recall@10": 0.6666666666666666,
-    "mrr@10": 0.20833333333333334,
-    "outranks": 2
-   }
-  }
- },
- "candidate_aggregate": {
-  "n_queries": 38,
-  "n_scored": 35,
-  "hit@5": 0.7428571428571429,
-  "recall@5": 0.5536485340056769,
-  "recall@10": 0.6175677497106068,
-  "precision@5": 0.20000000000000007,
-  "mrr@10": 0.5021428571428571,
-  "nonsense_clean": "3/3",
-  "superseded_outranks": "2/3",
-  "latency_ms_p50": 542.6,
-  "latency_ms_p95": 741.3,
-  "latency_ms_max": 758.8,
-  "errors": 0,
-  "by_category": {
-   "associative": {
-    "n": 6,
-    "hit@5": 0.6666666666666666,
-    "recall@5": 0.08857808857808858,
-    "recall@10": 0.23310023310023312,
-    "mrr@10": 0.25
-   },
-   "exact_rare": {
-    "n": 6,
-    "hit@5": 1.0,
-    "recall@5": 1.0,
-    "recall@10": 1.0,
-    "mrr@10": 1.0
-   },
-   "nonsense": {
-    "n": 3,
-    "clean": 3,
-    "avg_false_positives": 0.0
-   },
-   "paraphrase": {
-    "n": 13,
-    "hit@5": 0.6153846153846154,
-    "recall@5": 0.6153846153846154,
-    "recall@10": 0.6153846153846154,
-    "mrr@10": 0.2846153846153846
-   },
-   "phrase": {
-    "n": 7,
-    "hit@5": 1.0,
-    "recall@5": 0.5494614512471656,
-    "recall@10": 0.6023242630385487,
-    "mrr@10": 0.8214285714285714
-   },
-   "superseded": {
-    "n": 3,
-    "hit@5": 0.3333333333333333,
-    "recall@5": 0.3333333333333333,
-    "recall@10": 0.6666666666666666,
-    "mrr@10": 0.20833333333333334,
-    "outranks": 2
-   }
-  }
- },
- "repeat_variance": {
-  "baseline": {
-   "runs": 2,
-   "hit@5_min": 0.7428571428571429,
-   "hit@5_max": 0.7428571428571429,
-   "spread_queries": 0
-  },
-  "candidate": {
-   "runs": 2,
-   "hit@5_min": 0.7428571428571429,
-   "hit@5_max": 0.7428571428571429,
-   "spread_queries": 0
-  }
- }
-}

tools/retrieval-eval/diag6.py

@@ -0,0 +1,110 @@
+import numpy as np, json, urllib.request, collections, math, re, sys, time, pickle, os
+SP="/private/tmp/claude-501/-Users-timlingo/82369039-a20e-4b5a-8a5e-28234a57b996/scratchpad"
+EV="/Users/timlingo/Development/neuron-technologies/_wt-bm25lex/tools/retrieval-eval/"
+np.seterr(all='ignore')
+t0=time.time()
+M=np.load(SP+'/emb.npy'); eids=open(SP+'/ids.txt',encoding='utf-8',errors='surrogateescape').read().split('\n')
+eidx={k:i for i,k in enumerate(eids)}
+d=json.load(open('/Users/timlingo/neuron-memory-backups/snapshot-pre-repair-20260806.json',encoding='utf-8',errors='surrogateescape'))
+STRUCT={"identity","contains","superseded_by","references","embodies","demonstrated_by","canonical-self","depends_on","currently_holds","activates"}
+adj=collections.defaultdict(list)
+for e in d['edges']:
+    if e.get('relation') not in STRUCT: continue
+    w=float(e.get('weight') or 0.0)
+    adj[e['from_id']].append((e['to_id'],w)); adj[e['to_id']].append((e['from_id'],w))
+nodes=d['nodes']
+N={n['id']:n for n in nodes}
+PRINT=re.compile(r'^[\x20-\x7e]+$')
+ids=[]; hay=[]; dl=[]; sal=[]; addressable=[]
+for n in nodes:
+    i=n.get('id') or ''
+    h=((n.get('content') or '')+'\x00'+(n.get('label') or '')+'\x00'+(n.get('tags') or '')).lower()
+    ids.append(i); hay.append(h); dl.append(len(h)); sal.append(float(n.get('salience') or 0.0))
+    addressable.append(bool(PRINT.match(i)))
+del d
+NN=len(ids); avgdl=sum(dl)/NN
+print("nodes=%d avgdl=%.0f  %.1fs"%(NN,avgdl,time.time()-t0),file=sys.stderr)
+gold={q['id']:q for q in json.load(open(EV+"gold_set.json"),)['queries']}
+CACHE={}
+def emb(t):
+    if t in CACHE: return CACHE[t]
+    b=json.dumps({"model":"nomic-embed-text","prompt":t}).encode()
+    r=urllib.request.Request("http://127.0.0.1:11434/api/embeddings",data=b,headers={"Content-Type":"application/json"})
+    v=np.array(json.load(urllib.request.urlopen(r,timeout=60))["embedding"],dtype=np.float32)
+    v=v/(np.linalg.norm(v)+1e-9); CACHE[t]=v; return v
+K1,B=1.2,0.75
+def lexleg(query, lim=10):
+    toks=[]
+    for w in query.split():
+        wl=w.lower()
+        if wl not in toks: toks.append(wl)
+    nt=len(toks)
+    masks=[]; df=[0]*nt
+    for i in range(NN):
+        if not addressable[i]: continue
+        h=hay[i]; m=0; sc=0
+        for t in range(nt):
+            if toks[t] in h: m|=(1<<t); sc+=1; df[t]+=1
+        if sc: masks.append((i,m,sc))
+    idf=[math.log(1.0+(NN-df[t]+0.5)/(df[t]+0.5)) for t in range(nt)]
+    scored=[]
+    for i,m,sc in masks:
+        norm=1.0-B+B*dl[i]/avgdl
+        s=0.0
+        for t in range(nt):
+            if m&(1<<t): s+=idf[t]*(K1+1.0)/(1.0+K1*norm)
+        scored.append((s,i))
+    scored.sort(key=lambda x:(-x[0], -sal[x[1]]))
+    return [ids[i] for s,i in scored[:lim]], len(masks), sum(df)
+FIRE=0.02; DECAY=0.7; DEPTH=2; SEED_MIN=0.60; ASSOC_MAX=64
+def assoc(seeds, s):
+    act={x:1.0 for x in seeds}; seen={x:2 for x in seeds}
+    Q=[(x,0) for x in seeds]; h=0
+    while h<len(Q):
+        cur,hop=Q[h]; h+=1
+        if hop>=DEPTH: continue
+        p=act[cur]
+        for oid,w in adj.get(cur,()):
+            n=N.get(oid)
+            if not n or n.get('node_type') in ('Tag','InternalStateEvent'): continue
+            na=p*w*DECAY*float(n.get('salience') or 0.0)
+            if na<FIRE: continue
+            if oid in seen and na<=act.get(oid,0): continue
+            act[oid]=na
+            if oid not in seen: seen[oid]=1
+            Q.append((oid,hop+1))
+    out=[]
+    for k,v in seen.items():
+        if v!=1 or k not in eidx: continue
+        c=float(s[eidx[k]])
+        if c<=0: continue
+        out.append((c,k))
+    out.sort(reverse=True)
+    return [k for c,k in out[:ASSOC_MAX]]
+
+LEGS={}
+for qid,q in gold.items():
+    v=emb(q['query']); s=M@v; s[~np.isfinite(s)]=-1
+    L,nmatch,dfsum=lexleg(q['query'])
+    ordr=np.argsort(-s)
+    Sall=[eids[j] for j in ordr[:40] if PRINT.match(eids[j] or '')]
+    seeds=[x for x in L[:3] if x in N]
+    seeds=seeds+[eids[j] for j in ordr[:8] if eids[j] in N and eids[j] not in seeds and PRINT.match(eids[j] or '')]
+    A=assoc(seeds,s) if seeds else []
+    A=[x for x in A if PRINT.match(x or '')]
+    LEGS[qid]=dict(L=L,Sall=Sall,A=A,scos={x:float(s[eidx[x]]) for x in set(Sall[:20]+A[:20]+list(q.get('relevant') or [])) if x in eidx},nmatch=nmatch)
+pickle.dump(LEGS,open(SP+'/legs6.pkl','wb'))
+
+FOCUS=['q14','q17','q23','q24','q25','q30','q32','q33','q34','q35','q36','q37','q38']
+for qid in FOCUS:
+    q=gold[qid]; g=LEGS[qid]; rel=set(q.get('relevant') or [])
+    def rk(lst):
+        for i,x in enumerate(lst):
+            if x in rel: return i+1
+        return None
+    print("%s %-12s nmatch=%-6d Lrank=%s Srank=%s Arank=%s |A|=%d"%(
+        qid,q['category'],g['nmatch'],rk(g['L']),rk(g['Sall']),rk(g['A']),len(g['A'])))
+    for r in list(rel)[:2]:
+        print("     rel cos=%.3f"%(g['scos'].get(r,-9)))
+    print("     topS cos:", ["%.3f"%g['scos'].get(x,-9) for x in g['Sall'][:3]])
+print("elapsed %.1fs"%(time.time()-t0),file=sys.stderr)

tools/retrieval-eval/legs.py

@@ -1,117 +0,0 @@
-import json,pickle,os,math,urllib.request,numpy as np
-S='/private/tmp/claude-501/-Users-timlingo/82369039-a20e-4b5a-8a5e-28234a57b996/scratchpad/sim/'
-C=pickle.load(open(S+'corpus.pkl','rb'))
-NODES=C['nodes']; EDGES=C['edges']; N=len(NODES)
-E=np.load(S+'emb.npy'); HAVE=np.load(S+'have.npy')
-En=E/np.maximum(np.linalg.norm(E,axis=1,keepdims=True),1e-12)
-LAYERS={int(l['layer_id']):l for l in (C['layers'] or [])} if C['layers'] else {}
-TRANS=set(i for i,l in LAYERS.items() if l.get('transparent'))
-def addressable(s):
-    if not s: return False
-    return all(0x20<=ord(ch)<=0x7e for ch in s)
-ADDR=np.array([addressable(n['id']) for n in NODES])
-OK=np.array([ (n['layer_id'] not in TRANS) and ADDR[i] for i,n in enumerate(NODES)])
-SAL=np.array([n['salience'] for n in NODES])
-LOW=[ (n['content']+'\x00'+n['label']+'\x00'+n['tags']).lower() for n in NODES]
-DL=np.array([float(len(n['content'])+len(n['label'])+len(n['tags'])) for n in NODES])
-IDX={}
-for i,n in enumerate(NODES):
-    IDX.setdefault(n['id'],i)
-STRUCT={"identity","contains","superseded_by","references","embodies","demonstrated_by","canonical-self","depends_on","currently_holds","activates"}
-ADJ_F=[[] for _ in range(N)]; ADJ_T=[[] for _ in range(N)]
-for e in EDGES:
-    a=IDX.get(e['from']); b=IDX.get(e['to'])
-    if a is None or b is None: continue
-    ADJ_F[a].append((e,b)); ADJ_T[b].append((e,a))
-EXCL=np.array([n['node_type'] in ('Tag','InternalStateEvent') for n in NODES])
-avgdl_all=None
-def tokenize(q):
-    out=[]
-    for t in q.split():
-        if not any(t.lower()==x.lower() for x in out): out.append(t)
-    return out
-_qcache={}
-def qemb(q):
-    if q in _qcache: return _qcache[q]
-    body=json.dumps({"model":"nomic-embed-text","prompt":q}).encode()
-    r=urllib.request.urlopen(urllib.request.Request("http://127.0.0.1:11434/api/embeddings",data=body,headers={"Content-Type":"application/json"}),timeout=30)
-    v=np.array(json.loads(r.read())["embedding"],dtype=np.float32)
-    v=v/np.linalg.norm(v); _qcache[q]=v; return v
-K1,B=1.2,0.75
-SEED_MIN=0.60; SEED_K=8; ASSOC_SEEDS=3; DEPTH=2; FIRE=0.02; AMAX=64; DECAY=0.7
-def legs(query):
-    toks=tokenize(query)
-    masks=[]; 
-    hit_idx=[]; hit_mask=[]
-    df=[0]*len(toks)
-    lt=[t.lower() for t in toks]
-    for i in range(N):
-        if not OK[i]: continue
-        s=LOW[i]; m=0
-        for t,tok in enumerate(lt):
-            if tok in s: m|=(1<<t)
-        if m:
-            hit_idx.append(i); hit_mask.append(m)
-            for t in range(len(toks)):
-                if m>>t&1: df[t]+=1
-    dl_n=int(OK.sum()); avgdl=float(DL[OK].sum()/max(dl_n,1))
-    idf=[math.log(1.0+((dl_n-d+0.5)/(d+0.5))) for d in df]
-    L=[]
-    for j,i in enumerate(hit_idx):
-        norm=1.0-B+B*(DL[i]/avgdl); w=0.0
-        for t in range(len(toks)):
-            if hit_mask[j]>>t&1: w+=idf[t]*(K1+1.0)/(1.0+K1*norm)
-        L.append((i,w,SAL[i]))
-    L.sort(key=lambda x:(-x[1],-x[2]))
-    qv=qemb(query)
-    cos=En@qv
-    cos=np.where(HAVE&OK,cos,-2.0)
-    order=np.argsort(-cos)
-    semfull=[(int(i),float(cos[i])) for i in order[:400]]
-    Sleg=[(i,(c-SEED_MIN)/(1-SEED_MIN)) for i,c in semfull if c>SEED_MIN]
-    semseed=[i for i,c in semfull[:SEED_K] if c>0.0]
-    # assoc
-    act={}; seen={}; qq=[]
-    for i,_,_ in L[:ASSOC_SEEDS]:
-        act[i]=1.0; seen[i]=2; qq.append((i,0))
-    for i in semseed:
-        if i in seen: continue
-        act[i]=1.0; seen[i]=2; qq.append((i,0))
-    qh=0
-    while qh<len(qq):
-        cur,h=qq[qh]; qh+=1
-        if h>=DEPTH: continue
-        parent=act[cur]
-        for e,oi in ADJ_F[cur]+ADJ_T[cur]:
-            if e['rel'] not in STRUCT: continue
-            if EXCL[oi]: continue
-            na=parent*e['w']*DECAY*SAL[oi]
-            if na<FIRE: continue
-            if seen.get(oi) and na<=act.get(oi,0): continue
-            act[oi]=na
-            if not seen.get(oi): seen[oi]=1
-            if len(qq)<AMAX*4: qq.append((oi,h+1))
-    A=[]
-    for i,st in seen.items():
-        if st!=1: continue
-        if not OK[i] or not HAVE[i]: continue
-        c=float(cos[i])
-        if c<=0.0: continue
-        A.append((i,c))
-    A.sort(key=lambda x:-x[1]); A=A[:AMAX]
-    return L,Sleg,A,cos
-def interleave3(L,Sl,A,lim=10):
-    out=[]; li=si=ai=0
-    while len(out)<lim and (li<len(L) or si<len(Sl) or ai<len(A)):
-        if li<len(L):
-            if L[li][0] not in out: out.append(L[li][0])
-            li+=1
-        if len(out)>=lim: break
-        if si<len(Sl):
-            if Sl[si][0] not in out: out.append(Sl[si][0])
-            si+=1
-        if len(out)>=lim: break
-        if ai<len(A):
-            if A[ai][0] not in out: out.append(A[ai][0])
-            ai+=1
-    return out

tools/retrieval-eval/policy2.py

@@ -1,102 +0,0 @@
-import json,sys,pickle,numpy as np
-sys.path.insert(0,'.')
-from legs import *
-GP='/Users/timlingo/Development/neuron-technologies/_wt-bm25lex/tools/retrieval-eval/'
-G=json.load(open(GP+'gold_set.json'))
-def legs3(query, sem_sal=False, assoc_sal=False, unfloor=False, sem_cap=None):
-    toks=tokenize(query)
-    hit_idx=[];hit_mask=[];df=[0]*len(toks);lt=[t.lower() for t in toks]
-    for i in range(N):
-        if not OK[i]: continue
-        s=LOW[i];m=0
-        for t,tok in enumerate(lt):
-            if tok in s: m|=(1<<t)
-        if m:
-            hit_idx.append(i);hit_mask.append(m)
-            for t in range(len(toks)):
-                if m>>t&1: df[t]+=1
-    dl_n=int(OK.sum());avgdl=float(DL[OK].sum()/max(dl_n,1))
-    idf=[math.log(1.0+((dl_n-d+0.5)/(d+0.5))) for d in df]
-    L=[]
-    for j,i in enumerate(hit_idx):
-        norm=1.0-B+B*(DL[i]/avgdl);w=0.0
-        for t in range(len(toks)):
-            if hit_mask[j]>>t&1: w+=idf[t]*(K1+1.0)/(1.0+K1*norm)
-        L.append((i,w,SAL[i]))
-    L.sort(key=lambda x:(-x[1],-x[2]))
-    if not L: return [],[],[]
-    qv=qemb(query);cos=En@qv;cos=np.where(HAVE&OK,cos,-2.0)
-    order=np.argsort(-cos)[:600]
-    cand=[int(i) for i in order if cos[i]>(0.0 if unfloor else SEED_MIN)]
-    key=(lambda i:(SAL[i] if sem_sal else 1.0)*float(cos[i]))
-    Sl=sorted(cand,key=lambda i:-key(i))
-    if sem_cap: Sl=Sl[:sem_cap]
-    semseed=[int(i) for i in order[:SEED_K] if cos[i]>0.0]
-    act={};seen={};qq=[]
-    for i,_,_ in L[:ASSOC_SEEDS]:
-        act[i]=1.0;seen[i]=2;qq.append((i,0))
-    for i in semseed:
-        if i in seen: continue
-        act[i]=1.0;seen[i]=2;qq.append((i,0))
-    qh=0
-    while qh<len(qq):
-        cur,h=qq[qh];qh+=1
-        if h>=DEPTH: continue
-        parent=act[cur]
-        for e,oi in ADJ_F[cur]+ADJ_T[cur]:
-            if e['rel'] not in STRUCT: continue
-            if EXCL[oi]: continue
-            na=parent*e['w']*DECAY*SAL[oi]
-            if na<FIRE: continue
-            if seen.get(oi) and na<=act.get(oi,0): continue
-            act[oi]=na
-            if not seen.get(oi): seen[oi]=1
-            if len(qq)<AMAX*4: qq.append((oi,h+1))
-    A=[]
-    for i,st in seen.items():
-        if st!=1 or not OK[i] or not HAVE[i]: continue
-        c=float(cos[i])
-        if c<=0.0: continue
-        A.append((i,(SAL[i] if assoc_sal else 1.0)*c))
-    A.sort(key=lambda x:-x[1]);A=[i for i,_ in A[:AMAX]]
-    return [i for i,_,_ in L],Sl,A
-def merge(L,S,A,lim=10):
-    out=[];li=si=ai=0
-    while len(out)<lim and (li<len(L) or si<len(S) or ai<len(A)):
-        if li<len(L):
-            if L[li] not in out: out.append(L[li])
-            li+=1
-        if len(out)>=lim: break
-        if si<len(S):
-            if S[si] not in out: out.append(S[si])
-            si+=1
-        if len(out)>=lim: break
-        if ai<len(A):
-            if A[ai] not in out: out.append(A[ai])
-            ai+=1
-    return out
-def outcome(q,ids):
-    c=q['category']
-    if c=='nonsense': return len(ids)==0
-    if c=='superseded':
-        a,b=q['must_outrank']
-        if a not in ids: return False
-        if b not in ids: return True
-        return ids.index(a)<ids.index(b)
-    return any(x in ids[:5] for x in q['relevant'])
-def run(**kw):
-    return {q['id']:outcome(q,[NODES[i]['id'] for i in merge(*legs3(q['query'],**kw),10)]) for q in G['queries']}
-base=run()
-print("baseline",sum(base.values()),"/38  misses:",[k for k,v in base.items() if not v])
-import itertools
-for name,kw in [
-  ('sem_sal(floored)',dict(sem_sal=True)),
-  ('unfloor',dict(unfloor=True)),
-  ('unfloor+sem_sal',dict(unfloor=True,sem_sal=True)),
-  ('assoc_sal',dict(assoc_sal=True)),
-  ('unfloor+sem_sal+assoc_sal',dict(unfloor=True,sem_sal=True,assoc_sal=True)),
-  ('sem_sal+assoc_sal(floored)',dict(sem_sal=True,assoc_sal=True)),
-]:
-    r=run(**kw)
-    g=sorted(k for k in base if r[k] and not base[k]); l=sorted(k for k in base if base[k] and not r[k])
-    print("%-28s net=%+d gains=%s losses=%s"%(name,len(g)-len(l),g,l))

tools/retrieval-eval/results-bm25base-rerun.json

@@ -1,15 +1,15 @@
 {
- "label": "wordstart",
- "soul_binary": "/private/tmp/claude-501/-Users-timlingo/82369039-a20e-4b5a-8a5e-28234a57b996/scratchpad/soul-wordstart",
- "soul_md5": "32d4cf77672658a5f49dc7c9213e3ba2",
+ "label": "bm25base-rerun",
+ "soul_binary": "/private/tmp/claude-501/-Users-timlingo/82369039-a20e-4b5a-8a5e-28234a57b996/scratchpad/soul-bm25base",
+ "soul_md5": "36c8dfa09c073b85fe7e00b02904d0ed",
  "corpus": "/Users/timlingo/neuron-eval-corpora/snapshot-pre-repair-20260806-embedded.json",
  "corpus_nodes": 78768,
  "corpus_edges": 14214,
- "gold_set": "/Users/timlingo/Development/neuron-technologies/_wt-eval/tools/retrieval-eval/gold_set.json",
+ "gold_set": "/Users/timlingo/Development/neuron-technologies/_wt-bm25lex/tools/retrieval-eval/gold_set.json",
  "limit": 10,
  "port": 7894,
- "wall_clock_s": 28.9,
- "child_pid": 1420,
+ "wall_clock_s": 50.6,
+ "child_pid": 93451,
  "child_confirmed_dead": true,
  "aggregate": {
   "n_queries": 38,
@@ -19,11 +19,11 @@
   "recall@10": 0.6175677497106068,
   "precision@5": 0.20000000000000007,
   "mrr@10": 0.5021428571428571,
-  "nonsense_clean": "3/3",
+  "nonsense_clean": "2/3",
   "superseded_outranks": "2/3",
-  "latency_ms_p50": 542.6,
-  "latency_ms_p95": 741.3,
-  "latency_ms_max": 758.8,
+  "latency_ms_p50": 1186.1,
+  "latency_ms_p95": 1632.5,
+  "latency_ms_max": 1669.8,
   "errors": 0,
   "by_category": {
    "associative": {
@@ -42,8 +42,8 @@
    },
    "nonsense": {
     "n": 3,
-    "clean": 3,
-    "avg_false_positives": 0.0
+    "clean": 2,
+    "avg_false_positives": 3.3333333333333335
    },
    "paraphrase": {
     "n": 13,
@@ -78,7 +78,7 @@
     "mem-7f61beb4-271c-4feb-9f6e-1c9c837a6226"
    ],
    "n_returned": 1,
-   "latency_ms": 170.4,
+   "latency_ms": 302.9,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -103,7 +103,7 @@
     "ctx-74ed"
    ],
    "n_returned": 10,
-   "latency_ms": 210.4,
+   "latency_ms": 336.7,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -119,7 +119,7 @@
     "mem-1ba7c67d-85b9-4c2e-9fe2-39f8b0477091"
    ],
    "n_returned": 1,
-   "latency_ms": 167.1,
+   "latency_ms": 290.2,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -135,7 +135,7 @@
     "mem-89c02aae-d3ca-43f9-9e5d-eb369896276c"
    ],
    "n_returned": 1,
-   "latency_ms": 166.9,
+   "latency_ms": 307.3,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -153,7 +153,7 @@
     "project-Add_inference_url_config_to_Neuron_MCP__Route_summarization_gen_tasks_to_Pantheon__keep_frontier_for_complex_reasoning_"
    ],
    "n_returned": 3,
-   "latency_ms": 162.3,
+   "latency_ms": 331.2,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -169,7 +169,7 @@
     "knw-c72597c5-c23d-4c08-8e9e-996dadf26a99"
    ],
    "n_returned": 1,
-   "latency_ms": 162.2,
+   "latency_ms": 302.2,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -194,7 +194,7 @@
     "bl-b8af6601-a8cb-41b5-aef5-ab8a57432dd5"
    ],
    "n_returned": 10,
-   "latency_ms": 306.7,
+   "latency_ms": 595.3,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -219,7 +219,7 @@
     "knw-e94982a2-358d-4f2f-af31-8ee0fcec07c6"
    ],
    "n_returned": 10,
-   "latency_ms": 261.2,
+   "latency_ms": 526.1,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.1875,
@@ -244,7 +244,7 @@
     "kn-f230b362-b201-4402-9833-4160c89ab3d4"
    ],
    "n_returned": 10,
-   "latency_ms": 272.9,
+   "latency_ms": 516.8,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.3333333333333333,
@@ -269,7 +269,7 @@
     "bl-18a9d1e4-1484-474c-bf6b-c6173212181b"
    ],
    "n_returned": 10,
-   "latency_ms": 265.5,
+   "latency_ms": 558.8,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.1111111111111111,
@@ -294,7 +294,7 @@
     "kn-6061318f-046b-4935-907d-8eafdce14930"
    ],
    "n_returned": 10,
-   "latency_ms": 271.1,
+   "latency_ms": 513.4,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -315,11 +315,11 @@
     "bl-5b17bd3b-0c41-46cb-a710-6fa4429692ff",
     "bl-967536a0-d49d-44fb-8cfb-b31b40bcbfae",
     "bl-8b58d9bc-352b-4842-a7f8-a6254b5d1e25",
-    "?Q??m?;?u?'",
+    "2c56a7a9-5323-4ce4-ba09-35836ba15d54",
     "bl-39cec462-c80c-4970-a3aa-91fe83053bde"
    ],
    "n_returned": 10,
-   "latency_ms": 412.6,
+   "latency_ms": 867.9,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.21428571428571427,
@@ -338,13 +338,13 @@
     "?",
     "bl-ec84b63d-b278-4944-8d7f-4aa7a51c0315",
     "?",
-    "mem-fb44a2fc-7405-41ff-87b3-84643ac07313",
+    "830ca37a-d334-4e41-ba89-64893dc8d628",
     "?",
-    "mem-a3c97012-5fa3-4915-a839-2c75c72005e0",
+    "ce9636dc-85a5-4dae-9e07-74ea2fcc6307",
     "?"
    ],
    "n_returned": 10,
-   "latency_ms": 377.3,
+   "latency_ms": 752.9,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -361,15 +361,15 @@
     "kn-58874a74-b96f-4883-9e08-45707f4bd3ee",
     "knw-d788a210-613b-4c49-9486-88bbc9d4716f",
     "kn-0bb4f021-56de-4947-a35b-a37209e7ba21",
-    "kn-13f60407-7b70-4db1-964f-ea1f8196efbd",
+    "66c63082-b4da-4aa1-8fee-848db8a83210",
     "kn-6061318f-046b-4935-907d-8eafdce14930",
     "mem-a535f205-bc4c-4058-9171-6263c496044a",
     "kn-eb1b9e18-3dc6-4b9b-9cc6-86e0ae6b6be8",
-    "mem-0228da71-d7f7-4f3b-b7b3-c5eede42b62a",
+    "kn-13f60407-7b70-4db1-964f-ea1f8196efbd",
     "ctx-4a41"
    ],
    "n_returned": 10,
-   "latency_ms": 758.8,
+   "latency_ms": 1591.4,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -387,14 +387,14 @@
     "kn-db9f141b-dbe3-4037-92e0-4bb9be0e5e6e",
     "mem-16efddd1-c43d-4a42-9d78-f54fb82bd277",
     "kn-eb1b9e18-3dc6-4b9b-9cc6-86e0ae6b6be8",
-    "7452eb8f-be01-4b55-aec1-ff0c29e790f6",
+    "f0eb6b13-909c-4674-91ef-23301d3abc8b",
     "kn-a99cefe3-5e83-4050-98d8-6c69f57c7c71",
-    "532277bf-2959-4beb-ae0d-b018c97678ee",
+    "30a44d10-2487-420e-bf61-3892e4343c92",
     "54608b69-78b6-4239-b60f-b8206cfecacc",
-    "d4015bd7-c592-4ed8-8574-1f15ad37af75"
+    "bfb5809e-d19a-4d3f-8c1a-796db622ad9d"
    ],
    "n_returned": 10,
-   "latency_ms": 741.3,
+   "latency_ms": 1660.1,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -407,19 +407,19 @@
    "category": "paraphrase",
    "query": "choosing an uncomfortable fact over a pleasant fiction",
    "returned": [
-    "tag-fiction",
+    "mem-ef878e30-5851-4e82-8588-745415108941",
     "kn-13f60407-7b70-4db1-964f-ea1f8196efbd",
     "knw-4aebd815-4eaf-49d7-954b-03595f3d48be",
-    "mem-8d690e9d-a7e9-4062-b2f8-e2064294e463",
+    "tag-fiction",
     "knw-8fd9836c-cc39-49df-8d61-babda626cc88",
-    "mem-ce793303-c5a5-4586-a232-a3426edd9ec7",
+    "mem-8d690e9d-a7e9-4062-b2f8-e2064294e463",
     "knw-e94982a2-358d-4f2f-af31-8ee0fcec07c6",
-    "mem-443bd012-fc9a-4088-b236-de5157a1ef92",
+    "mem-ce793303-c5a5-4586-a232-a3426edd9ec7",
     "knw-d357b6bb-ad8a-4791-b516-426aea45fa5b",
-    "mem-ca4d6a34-d354-413f-bc86-126cc17ca81c"
+    "mem-443bd012-fc9a-4088-b236-de5157a1ef92"
    ],
    "n_returned": 10,
-   "latency_ms": 642.2,
+   "latency_ms": 1345.5,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -432,19 +432,19 @@
    "category": "paraphrase",
    "query": "a tight payload beats a bloated one",
    "returned": [
-    "bl-680b24a9-edc3-4a9d-847a-bff0b46b568c",
+    "bl-8de20bcf-7149-4f48-b67c-e7f9758fd6e5",
     "bl-798d135f-3987-4ccd-8de6-70ca2f358337",
-    "08f0d1e2-8d0e-42e3-9f0a-8186ae31ec7e",
+    "bl-680b24a9-edc3-4a9d-847a-bff0b46b568c",
     "knw-08559f5c-2306-4220-a146-398c74f1643c",
-    "4da5dbaf-46e5-4f3e-b474-f60d9f8241d3",
+    "bl-164b520b-c503-49db-89f9-bd2fdf4215f5",
     "knw-f6ed7d00-bf7d-42ce-9e40-77cf3406e918",
-    "mem-434be7c8-88cb-4039-b79a-1da4ac4de783",
+    "1219277c-1b95-45ec-95a2-07b4a47a4d92",
     "knw-920c891f-bb8c-48c4-9afc-018ef12dcdc4",
-    "a708dd6e-fe73-4f2f-a21e-89daa0985487",
+    "08f0d1e2-8d0e-42e3-9f0a-8186ae31ec7e",
     "bl-79ce4464-5dd6-49bd-9b0c-9803549d0665"
    ],
    "n_returned": 10,
-   "latency_ms": 537.2,
+   "latency_ms": 1070.7,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -459,17 +459,17 @@
    "returned": [
     "bl-4476e856-c567-4b49-8ff7-d7dca3e5715e",
     "kn-eb1b9e18-3dc6-4b9b-9cc6-86e0ae6b6be8",
-    "mem-e5cc63c0-8701-49d6-855a-e387fe087771",
-    "kn-a99cefe3-5e83-4050-98d8-6c69f57c7c71",
-    "mem-75e490d1-f0a9-4b73-8cfc-8daecfaf6f38",
-    "kn-58874a74-b96f-4883-9e08-45707f4bd3ee",
     "a1000001-0000-0000-0000-000000000010",
-    "015644f5-8194-4af0-800d-dd4a0cd71396",
-    "bfad516b-c306-4c4c-874a-a347c46c05c2",
-    "kn-e0423482-cfa5-4796-8689-8495c93b66bc"
+    "a1000001-0000-0000-0000-000000000009",
+    "bl-448bc514-c2f1-4520-a9b1-1f3a73678d26",
+    "a1000001-0000-0000-0000-000000000012",
+    "43098881-e044-482b-8e92-471728a8ba8b",
+    "kn-a99cefe3-5e83-4050-98d8-6c69f57c7c71",
+    "mem-e5cc63c0-8701-49d6-855a-e387fe087771",
+    "a1000001-0000-0000-0000-000000000001"
    ],
    "n_returned": 10,
-   "latency_ms": 725.1,
+   "latency_ms": 1395.8,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -486,15 +486,15 @@
     "kn-0bb4f021-56de-4947-a35b-a37209e7ba21",
     "knw-7902acca-604e-409b-8faf-ad85424211d0",
     "kn-f230b362-b201-4402-9833-4160c89ab3d4",
-    "345b6420-e004-4d2e-b55c-6a729393fa99",
+    "d6b12ecf-702b-4101-b1bb-09ed9b220b29",
     "kn-5de5a9ac-fd15-45ab-bf18-77566781cf40",
-    "mem-92a7fdc5-9dd0-48cf-a691-506058de3838",
+    "c608a095-c98b-4bfa-bfe1-1611c1320290",
     "knw-ed33e669-0790-44cb-a036-958d605c6fea",
-    "a1000001-0000-0000-0000-000000000010",
+    "451ae007-4219-4096-89fe-fa2e045fbeb1",
     "kn-db9f141b-dbe3-4037-92e0-4bb9be0e5e6e"
    ],
    "n_returned": 10,
-   "latency_ms": 562.8,
+   "latency_ms": 1243.4,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -519,7 +519,7 @@
     "bl-ef2bac68-e119-4139-b529-c7a1404ae3ac"
    ],
    "n_returned": 10,
-   "latency_ms": 686.5,
+   "latency_ms": 1598.7,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -544,7 +544,7 @@
     "bl-286b562a-5299-40e0-a32a-afa9cbdfe995"
    ],
    "n_returned": 10,
-   "latency_ms": 584.2,
+   "latency_ms": 1379.6,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -557,19 +557,19 @@
    "category": "paraphrase",
    "query": "what shifts tells you where to cut a system apart",
    "returned": [
-    "knw-723551f5-1950-42a3-8b89-b6a06913cef0",
+    "5a2c118a-87bd-4239-97a7-9e02c5991983",
     "kn-78db5396-3dbc-4481-bfc7-e4e1422feb1c",
-    "bl-4c5b385e-135a-4663-8521-96af0b491121",
+    "mem-ef878e30-5851-4e82-8588-745415108941",
     "knw-12b4b913-7a25-4b0d-844c-504c01d6725e",
-    "bl-2b00aeb0-c0fa-4a9f-8f30-4207e98b3d52",
+    "knw-723551f5-1950-42a3-8b89-b6a06913cef0",
     "knw-9707256e-ed44-4042-bd88-f90fa514e1cf",
-    "34356a36-0df5-4020-8dcc-5e7a423f8d4c",
+    "kn-22d77abe-b3c5-42fd-afcd-dcb87d924929",
     "knw-0087493b-25cd-45b0-bf46-c078c5b49718",
-    "mem-22f5f665-3ad2-4063-88b0-915849a795f5",
+    "bl-4c5b385e-135a-4663-8521-96af0b491121",
     "kn-0bb4f021-56de-4947-a35b-a37209e7ba21"
    ],
    "n_returned": 10,
-   "latency_ms": 719.6,
+   "latency_ms": 1632.5,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -584,17 +584,17 @@
    "returned": [
     "bl-8dd70cac-866d-4ff2-b9fe-b4b3c5f094bb",
     "mem-b43f6ef4-2f5a-418d-b5ce-3f21520cf6b8",
-    "knw-7902acca-604e-409b-8faf-ad85424211d0",
-    "knw-ed33e669-0790-44cb-a036-958d605c6fea",
-    "a1000001-0000-0000-0000-000000000001",
     "kn-58874a74-b96f-4883-9e08-45707f4bd3ee",
-    "bl-1b58b05c-9305-4f06-a586-a08c96008027",
+    "a0edad47-5f77-4fc3-a546-1e85f8c68e77",
+    "a1000001-0000-0000-0000-000000000001",
+    "8cbb60c5-4999-4ec1-8682-2592aedc4249",
+    "96497334-b18f-495c-9228-eeb8182bdc38",
     "mem-024598a9-ed2e-4eeb-b1e1-5410856ff132",
-    "ctx-4a41",
-    "mem-5708f4c9-3d61-4182-8543-2843698931e6"
+    "kn-5de5a9ac-fd15-45ab-bf18-77566781cf40",
+    "knw-ed33e669-0790-44cb-a036-958d605c6fea"
    ],
    "n_returned": 10,
-   "latency_ms": 634.6,
+   "latency_ms": 1491.9,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -607,19 +607,19 @@
    "category": "paraphrase",
    "query": "loved for the unedited self and not the polished exterior",
    "returned": [
-    "077d064f-3489-4c05-9aca-3782f96b51db",
-    "knw-f9ce17a7-17fc-431f-8f23-695b670ec4fa",
-    "kn-6061318f-046b-4935-907d-8eafdce14930",
-    "knw-2c46cfb4-6d4e-4822-8a1a-7d743c1e4329",
     "27e1b1a4-ad0b-49d9-812f-fedf43b8aabe",
-    "kn-5de5a9ac-fd15-45ab-bf18-77566781cf40",
+    "knw-f9ce17a7-17fc-431f-8f23-695b670ec4fa",
     "bl-87c93185-b2bf-40af-ae23-3c830c007abf",
+    "knw-2c46cfb4-6d4e-4822-8a1a-7d743c1e4329",
+    "077d064f-3489-4c05-9aca-3782f96b51db",
+    "kn-5de5a9ac-fd15-45ab-bf18-77566781cf40",
+    "kn-6061318f-046b-4935-907d-8eafdce14930",
     "knw-920c891f-bb8c-48c4-9afc-018ef12dcdc4",
-    "fce2792a-53fc-4d4a-be3b-42bd6ceb1ba7",
+    "75e036d3-c170-4e3f-acc2-e456a6850ee2",
     "a1000001-0000-0000-0000-000000000001"
    ],
    "n_returned": 10,
-   "latency_ms": 674.2,
+   "latency_ms": 1535.5,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -638,13 +638,13 @@
     "kn-5de5a9ac-fd15-45ab-bf18-77566781cf40",
     "mem-5624ec9d-62ba-4aba-8a3d-6afec6c09dd4",
     "kn-db9f141b-dbe3-4037-92e0-4bb9be0e5e6e",
-    "mem-a16deccb-16a7-419c-a013-ff824a4daa15",
-    "a1000001-0000-0000-0000-000000000009",
     "mem-833dbbcd-2400-4594-bb35-93b023049ac0",
+    "a1000001-0000-0000-0000-000000000009",
+    "mem-759e78ca-5394-4244-aa39-1c1468bc5f3e",
     "kn-13f60407-7b70-4db1-964f-ea1f8196efbd"
    ],
    "n_returned": 10,
-   "latency_ms": 571.6,
+   "latency_ms": 1249.8,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -657,19 +657,19 @@
    "category": "paraphrase",
    "query": "a childhood offering no solid foundation to inherit",
    "returned": [
-    "mem-b99efff0-00e6-40c8-9c5b-730330eef33b",
+    "bl-3f57bc69-7285-4f4a-a861-2de52efca058",
     "kn-a99cefe3-5e83-4050-98d8-6c69f57c7c71",
-    "knw-2c46cfb4-6d4e-4822-8a1a-7d743c1e4329",
-    "knw-f6ed7d00-bf7d-42ce-9e40-77cf3406e918",
-    "kn-6061318f-046b-4935-907d-8eafdce14930",
-    "knw-23c27d3b-e0d2-43a8-a80c-0a44477ae18a",
-    "kn-eb1b9e18-3dc6-4b9b-9cc6-86e0ae6b6be8",
     "kn-f230b362-b201-4402-9833-4160c89ab3d4",
-    "tag-childhood",
-    "kn-0bb4f021-56de-4947-a35b-a37209e7ba21"
+    "bl-0d8c5dfa-e163-4fef-a58b-56b0d076c5a8",
+    "kn-6061318f-046b-4935-907d-8eafdce14930",
+    "kn-db9f141b-dbe3-4037-92e0-4bb9be0e5e6e",
+    "mem-b99efff0-00e6-40c8-9c5b-730330eef33b",
+    "kn-eb1b9e18-3dc6-4b9b-9cc6-86e0ae6b6be8",
+    "kn-5b606390-a52d-4ca2-8e0e-eba141d13440",
+    "tag-childhood"
    ],
    "n_returned": 10,
-   "latency_ms": 603.4,
+   "latency_ms": 1351.0,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -690,11 +690,11 @@
     "kn-0bb4f021-56de-4947-a35b-a37209e7ba21",
     "766de879-f9d0-4a07-b6df-b43ee13763d8",
     "kn-e0423482-cfa5-4796-8689-8495c93b66bc",
-    "efe53612-6914-4936-8e3b-1e694eb174e5",
+    "3499d5da-0e9c-4de4-9bc4-8941b14e0b1f",
     "kn-78db5396-3dbc-4481-bfc7-e4e1422feb1c"
    ],
    "n_returned": 10,
-   "latency_ms": 656.2,
+   "latency_ms": 1438.4,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.18181818181818182,
@@ -719,7 +719,7 @@
     "kn-a99cefe3-5e83-4050-98d8-6c69f57c7c71"
    ],
    "n_returned": 10,
-   "latency_ms": 536.9,
+   "latency_ms": 1110.1,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.07692307692307693,
@@ -744,7 +744,7 @@
     "kn-78db5396-3dbc-4481-bfc7-e4e1422feb1c"
    ],
    "n_returned": 10,
-   "latency_ms": 562.4,
+   "latency_ms": 1186.1,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.18181818181818182,
@@ -769,7 +769,7 @@
     "mem-ab34c2f7-3243-424b-affa-25555f6cf9cc"
    ],
    "n_returned": 10,
-   "latency_ms": 556.6,
+   "latency_ms": 1193.1,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -794,7 +794,7 @@
     "kn-db9f141b-dbe3-4037-92e0-4bb9be0e5e6e"
    ],
    "n_returned": 10,
-   "latency_ms": 542.6,
+   "latency_ms": 1201.5,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.09090909090909091,
@@ -808,18 +808,18 @@
    "query": "man of the house six years old expectation",
    "returned": [
     "kn-eb1b9e18-3dc6-4b9b-9cc6-86e0ae6b6be8",
-    "knw-528dbc37-eabc-4b75-a7a5-65bf38d6018a",
+    "knw-729fc901-8335-44c4-9f3a-b150b4aa0915",
     "knw-d357b6bb-ad8a-4791-b516-426aea45fa5b",
-    "knw-473f3f24-20f6-4f39-8589-3709538eb6ac",
-    "mem-a0b7cfda-bc9e-4f40-b9a9-1722cf3f8263",
-    "knw-7902acca-604e-409b-8faf-ad85424211d0",
+    "knw-528dbc37-eabc-4b75-a7a5-65bf38d6018a",
     "knw-35940684-abc4-42f0-b942-818f66b1f69a",
-    "4f698ae6-c40e-464e-9798-50350991a188",
-    "719aa819-00a9-4f4b-a857-4f9fe5ad44d7",
+    "knw-473f3f24-20f6-4f39-8589-3709538eb6ac",
+    "?Z?<S???K ?",
+    "knw-7902acca-604e-409b-8faf-ad85424211d0",
+    "'?T?a\"B~-?8",
     "kn-6061318f-046b-4935-907d-8eafdce14930"
    ],
    "n_returned": 10,
-   "latency_ms": 683.5,
+   "latency_ms": 1424.8,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -833,7 +833,7 @@
    "query": "zqxjvw plimforth grebulon",
    "returned": [],
    "n_returned": 0,
-   "latency_ms": 350.4,
+   "latency_ms": 705.7,
    "error": null,
    "clean": true,
    "false_positives": 0
@@ -844,7 +844,7 @@
    "query": "flarnbistle quommetry",
    "returned": [],
    "n_returned": 0,
-   "latency_ms": 258.4,
+   "latency_ms": 484.4,
    "error": null,
    "clean": true,
    "false_positives": 0
@@ -853,12 +853,23 @@
    "id": "q35",
    "category": "nonsense",
    "query": "xxqzzt vurblenacht throom",
-   "returned": [],
-   "n_returned": 0,
-   "latency_ms": 348.4,
+   "returned": [
+    "bl-4476e856-c567-4b49-8ff7-d7dca3e5715e",
+    "?V?",
+    "knw-6b48dce2-f21c-452a-9db5-4e6aa61c87ca",
+    "knw-920c891f-bb8c-48c4-9afc-018ef12dcdc4",
+    "?m?\\}Q??6??",
+    "8cbb60c5-4999-4ec1-8682-2592aedc4249",
+    "kn-66a21179-2adc-4b19-a109-880cf4674d7d",
+    "?m?\\}Q??6??",
+    "knw-723551f5-1950-42a3-8b89-b6a06913cef0",
+    "kn-333542cb-6dab-4662-9725-bf7440d28bf7"
+   ],
+   "n_returned": 10,
+   "latency_ms": 740.2,
    "error": null,
-   "clean": true,
-   "false_positives": 0
+   "clean": false,
+   "false_positives": 10
   },
   {
    "id": "q36",
@@ -871,13 +882,13 @@
     "kn-b7e98d63-8b83-4911-b4d0-990602a7f575",
     "tag-__darma____cgi____patents____self-improvement____character-preservation____autonomous____kotlin____architecture__",
     "knw-e047bb42-dc5b-4383-9e88-e508dc03abe3",
-    "mem-c17aefb1-38b5-4ced-af50-fe524127e1a4",
+    "mem-e6327f52-2bda-4ce7-9471-2fffd1e172de",
     "bl-5b17bd3b-0c41-46cb-a710-6fa4429692ff",
-    "bl-145a0985-2382-400f-a7c5-c335c5e30a72",
-    "mem-82b93b21-a865-410f-9ec1-fc54121d9bb5"
+    "mem-82b93b21-a865-410f-9ec1-fc54121d9bb5",
+    "bl-145a0985-2382-400f-a7c5-c335c5e30a72"
    ],
    "n_returned": 10,
-   "latency_ms": 664.5,
+   "latency_ms": 1307.7,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -902,10 +913,10 @@
     "a1000001-0000-0000-0000-000000000002",
     "3cf706a1-3825-45d8-b0a9-06cae6cdf5b8",
     "kn-eb1b9e18-3dc6-4b9b-9cc6-86e0ae6b6be8",
-    "de3b6428-b76c-4c44-90e0-bf1dd6998027"
+    "7ac62daa-2eac-4c7a-a97e-e4203fc1b57b"
    ],
    "n_returned": 10,
-   "latency_ms": 747.3,
+   "latency_ms": 1669.8,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -933,7 +944,7 @@
     "mem-3a2cf162-d93b-4f29-86f2-5066fb7fe1f5"
    ],
    "n_returned": 10,
-   "latency_ms": 504.2,
+   "latency_ms": 1018.2,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,

tools/retrieval-eval/results-claim24.json

@@ -1,37 +1,37 @@
 {
- "label": "wsclaim24",
- "soul_binary": "/private/tmp/claude-501/-Users-timlingo/82369039-a20e-4b5a-8a5e-28234a57b996/scratchpad/soul-wsclaim24",
- "soul_md5": "a377d9c0e5282e842ed98384c8b50d2c",
+ "label": "claim24",
+ "soul_binary": "/private/tmp/claude-501/-Users-timlingo/82369039-a20e-4b5a-8a5e-28234a57b996/scratchpad/soul-claim24",
+ "soul_md5": "676ebffb91f00046dcea895f5aa7efba",
  "corpus": "/Users/timlingo/neuron-eval-corpora/snapshot-pre-repair-20260806-embedded.json",
  "corpus_nodes": 78768,
  "corpus_edges": 14214,
- "gold_set": "/Users/timlingo/Development/neuron-technologies/_wt-eval/tools/retrieval-eval/gold_set.json",
+ "gold_set": "/Users/timlingo/Development/neuron-technologies/_wt-bm25lex/tools/retrieval-eval/gold_set.json",
  "limit": 10,
  "port": 7893,
- "wall_clock_s": 29.9,
- "child_pid": 99726,
+ "wall_clock_s": 52.3,
+ "child_pid": 92441,
  "child_confirmed_dead": true,
  "aggregate": {
   "n_queries": 38,
   "n_scored": 35,
   "hit@5": 0.7428571428571429,
   "recall@5": 0.5768475572047,
-  "recall@10": 0.6563414759843332,
+  "recall@10": 0.6537440733869305,
   "precision@5": 0.19428571428571437,
-  "mrr@10": 0.5026530612244898,
-  "nonsense_clean": "0/3",
+  "mrr@10": 0.5021428571428572,
+  "nonsense_clean": "2/3",
   "superseded_outranks": "2/3",
-  "latency_ms_p50": 524.8,
-  "latency_ms_p95": 738.7,
-  "latency_ms_max": 755.8,
+  "latency_ms_p50": 1173.7,
+  "latency_ms_p95": 1623.0,
+  "latency_ms_max": 1647.9,
   "errors": 0,
   "by_category": {
    "associative": {
     "n": 6,
     "hit@5": 0.5,
     "recall@5": 0.07575757575757576,
-    "recall@10": 0.13636363636363635,
-    "mrr@10": 0.23214285714285712
+    "recall@10": 0.12121212121212122,
+    "mrr@10": 0.22916666666666666
    },
    "exact_rare": {
     "n": 6,
@@ -42,8 +42,8 @@
    },
    "nonsense": {
     "n": 3,
-    "clean": 0,
-    "avg_false_positives": 10.0
+    "clean": 2,
+    "avg_false_positives": 3.3333333333333335
    },
    "paraphrase": {
     "n": 13,
@@ -87,7 +87,7 @@
     "696e609c-da7a-4394-8a0c-106ba07dc6c3"
    ],
    "n_returned": 10,
-   "latency_ms": 173.2,
+   "latency_ms": 284.2,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -112,7 +112,7 @@
     "ctx-fae1"
    ],
    "n_returned": 10,
-   "latency_ms": 158.1,
+   "latency_ms": 338.1,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -137,7 +137,7 @@
     "x??I?cB?]p?"
    ],
    "n_returned": 10,
-   "latency_ms": 160.8,
+   "latency_ms": 300.4,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -162,7 +162,7 @@
     "mem-2d1ea831-cccd-4f0f-86b9-2cbbc89dc3e0"
    ],
    "n_returned": 10,
-   "latency_ms": 158.0,
+   "latency_ms": 285.2,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -187,7 +187,7 @@
     "bl-2632242e-80b1-4d88-8368-7065b5de5b34"
    ],
    "n_returned": 10,
-   "latency_ms": 170.3,
+   "latency_ms": 318.5,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -212,7 +212,7 @@
     "??"
    ],
    "n_returned": 10,
-   "latency_ms": 175.8,
+   "latency_ms": 331.3,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -237,7 +237,7 @@
     "bl-39dad13d-7105-4049-8224-dc3c34fdb1f3"
    ],
    "n_returned": 10,
-   "latency_ms": 293.8,
+   "latency_ms": 582.9,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -257,12 +257,12 @@
     "mem-c9bec303-a638-4a11-a490-f38410d448cf",
     "knw-ed33e669-0790-44cb-a036-958d605c6fea",
     "knw-f671966c-3387-4848-abca-b5deec122e00",
-    "7cf3f825-d7b6-4864-b0ae-52939cf1ae84",
+    "2795fdbb-ee2a-4009-aa30-3f1c25b6e77c",
     "kn-db9f141b-dbe3-4037-92e0-4bb9be0e5e6e",
     "knw-4aebd815-4eaf-49d7-954b-03595f3d48be"
    ],
    "n_returned": 10,
-   "latency_ms": 251.2,
+   "latency_ms": 520.6,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.1875,
@@ -287,7 +287,7 @@
     "kn-f230b362-b201-4402-9833-4160c89ab3d4"
    ],
    "n_returned": 10,
-   "latency_ms": 264.4,
+   "latency_ms": 531.5,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.2222222222222222,
@@ -312,7 +312,7 @@
     "tag-harmonic-framework"
    ],
    "n_returned": 10,
-   "latency_ms": 262.5,
+   "latency_ms": 502.8,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.1111111111111111,
@@ -337,7 +337,7 @@
     "??_/Pr?????"
    ],
    "n_returned": 10,
-   "latency_ms": 257.7,
+   "latency_ms": 506.1,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -358,11 +358,11 @@
     "bl-2515d870-e35e-443b-ba20-5150bbc73fed",
     "bl-5b17bd3b-0c41-46cb-a710-6fa4429692ff",
     "bl-a0982e7c-e165-4da3-a11d-619fa0b535b0",
-    "?Q??m?;?u?'",
+    "2c56a7a9-5323-4ce4-ba09-35836ba15d54",
     "knw-cf13b883-d947-4cf8-b86b-cd9c6f0748d6"
    ],
    "n_returned": 10,
-   "latency_ms": 393.5,
+   "latency_ms": 861.8,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.21428571428571427,
@@ -381,13 +381,13 @@
     "7774a16c-1027-4e3b-a21e-67f1f95a4acd",
     "?",
     "bl-3c719d9a-cba1-47f4-b097-52cdeccc7c0d",
-    "mem-fb44a2fc-7405-41ff-87b3-84643ac07313",
+    "830ca37a-d334-4e41-ba89-64893dc8d628",
     "?",
-    "mem-ce88adf9-3f3c-47ac-a7d3-83af7b290e68",
+    "ce9636dc-85a5-4dae-9e07-74ea2fcc6307",
     "?"
    ],
    "n_returned": 10,
-   "latency_ms": 333.4,
+   "latency_ms": 758.5,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -406,13 +406,13 @@
     "knw-d788a210-613b-4c49-9486-88bbc9d4716f",
     "art-4a99aa1a-489b-4b43-958b-25217adb1aad",
     "kn-6061318f-046b-4935-907d-8eafdce14930",
-    "kn-13f60407-7b70-4db1-964f-ea1f8196efbd",
+    "66c63082-b4da-4aa1-8fee-848db8a83210",
     "kn-db9f141b-dbe3-4037-92e0-4bb9be0e5e6e",
     "kn-eb1b9e18-3dc6-4b9b-9cc6-86e0ae6b6be8",
     "mem-a535f205-bc4c-4058-9171-6263c496044a"
    ],
    "n_returned": 10,
-   "latency_ms": 755.8,
+   "latency_ms": 1590.9,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -429,15 +429,15 @@
     "art-79042b8b-6192-440f-90b0-60708f7e6325",
     "mem-16efddd1-c43d-4a42-9d78-f54fb82bd277",
     "daf9d558-2265-49b0-b4fb-be0b80f3cdb1",
-    "bfb5809e-d19a-4d3f-8c1a-796db622ad9d",
+    "863bfd72-50d2-487f-85ae-ecd48e2a501f",
     "art-ddfcd045-2c3b-4a1e-9966-fec5ce44e1dd",
-    "f3364a57-956a-4e78-a500-b3b66a4e3066",
+    "5c1c3404-ad4c-49f1-8c0c-81262812731e",
     "15a90710-5372-4d3d-aee5-2d7287d2a492",
-    "341afa66-6a49-43fe-9d5b-3f1032bb6904",
+    "465fa29f-1ce3-4769-962a-1c32dfed8347",
     "mem-da21c52c-04a5-4f92-8fba-f10aac47e027"
    ],
    "n_returned": 10,
-   "latency_ms": 720.3,
+   "latency_ms": 1647.9,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -459,10 +459,10 @@
     "mem-d1cfde0a-37f1-4bff-9a06-8eddbbf259f6",
     "kn-db9f141b-dbe3-4037-92e0-4bb9be0e5e6e",
     "knw-f671966c-3387-4848-abca-b5deec122e00",
-    "tag-factions"
+    "mem-16efddd1-c43d-4a42-9d78-f54fb82bd277"
    ],
    "n_returned": 10,
-   "latency_ms": 616.0,
+   "latency_ms": 1338.6,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -475,19 +475,19 @@
    "category": "paraphrase",
    "query": "a tight payload beats a bloated one",
    "returned": [
-    "bl-680b24a9-edc3-4a9d-847a-bff0b46b568c",
+    "bl-8de20bcf-7149-4f48-b67c-e7f9758fd6e5",
     "mem-628437a6-47b0-4d81-8112-7e78499723d5",
     "bl-798d135f-3987-4ccd-8de6-70ca2f358337",
-    "08f0d1e2-8d0e-42e3-9f0a-8186ae31ec7e",
+    "bl-680b24a9-edc3-4a9d-847a-bff0b46b568c",
     "knw-12b4b913-7a25-4b0d-844c-504c01d6725e",
     "knw-08559f5c-2306-4220-a146-398c74f1643c",
-    "4da5dbaf-46e5-4f3e-b474-f60d9f8241d3",
+    "bl-164b520b-c503-49db-89f9-bd2fdf4215f5",
     "o#?CW????y:",
     "knw-f6ed7d00-bf7d-42ce-9e40-77cf3406e918",
-    "mem-434be7c8-88cb-4039-b79a-1da4ac4de783"
+    "1219277c-1b95-45ec-95a2-07b4a47a4d92"
    ],
    "n_returned": 10,
-   "latency_ms": 518.5,
+   "latency_ms": 1060.5,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -506,13 +506,13 @@
     "a1000001-0000-0000-0000-000000000010",
     "bl-9ce4128a-9436-4b06-82bc-8a6faafa81e0",
     "a1000001-0000-0000-0000-000000000009",
-    "mem-32203649-3213-4d6d-86fd-3d657ac70d77",
+    "43098881-e044-482b-8e92-471728a8ba8b",
     "? t?'?B?+??",
     "a1000001-0000-0000-0000-000000000012",
-    "mem-da21c52c-04a5-4f92-8fba-f10aac47e027"
+    "bl-448bc514-c2f1-4520-a9b1-1f3a73678d26"
    ],
    "n_returned": 10,
-   "latency_ms": 742.9,
+   "latency_ms": 1388.8,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -534,10 +534,10 @@
     "345b6420-e004-4d2e-b55c-6a729393fa99",
     "kn-a99cefe3-5e83-4050-98d8-6c69f57c7c71",
     "kn-5de5a9ac-fd15-45ab-bf18-77566781cf40",
-    "mem-92a7fdc5-9dd0-48cf-a691-506058de3838"
+    "74f4776a-d0ea-44e4-b94f-7c87d0179ef3"
    ],
    "n_returned": 10,
-   "latency_ms": 549.8,
+   "latency_ms": 1231.7,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -562,7 +562,7 @@
     "bl-ef2bac68-e119-4139-b529-c7a1404ae3ac"
    ],
    "n_returned": 10,
-   "latency_ms": 682.4,
+   "latency_ms": 1590.7,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -587,7 +587,7 @@
     "bl-56a50e97-9a85-4e81-b6c9-3e3d26482f1d"
    ],
    "n_returned": 10,
-   "latency_ms": 575.0,
+   "latency_ms": 1380.0,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -600,19 +600,19 @@
    "category": "paraphrase",
    "query": "what shifts tells you where to cut a system apart",
    "returned": [
-    "knw-723551f5-1950-42a3-8b89-b6a06913cef0",
+    "5a2c118a-87bd-4239-97a7-9e02c5991983",
     "bl-06c13965-082b-417d-9561-93d6e958ae5d",
     "kn-78db5396-3dbc-4481-bfc7-e4e1422feb1c",
-    "bl-4c5b385e-135a-4663-8521-96af0b491121",
+    "knw-723551f5-1950-42a3-8b89-b6a06913cef0",
     "kn-f230b362-b201-4402-9833-4160c89ab3d4",
     "knw-12b4b913-7a25-4b0d-844c-504c01d6725e",
-    "bl-2b00aeb0-c0fa-4a9f-8f30-4207e98b3d52",
+    "kn-22d77abe-b3c5-42fd-afcd-dcb87d924929",
     "mem-7b74cac0-905f-4c35-9688-fbcce105a177",
     "knw-9707256e-ed44-4042-bd88-f90fa514e1cf",
-    "34356a36-0df5-4020-8dcc-5e7a423f8d4c"
+    "bl-4c5b385e-135a-4663-8521-96af0b491121"
    ],
    "n_returned": 10,
-   "latency_ms": 711.3,
+   "latency_ms": 1623.0,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -627,17 +627,17 @@
    "returned": [
     "bl-8dd70cac-866d-4ff2-b9fe-b4b3c5f094bb",
     "mem-b43f6ef4-2f5a-418d-b5ce-3f21520cf6b8",
-    "knw-7902acca-604e-409b-8faf-ad85424211d0",
-    "knw-ed33e669-0790-44cb-a036-958d605c6fea",
+    "8cbb60c5-4999-4ec1-8682-2592aedc4249",
+    "a0edad47-5f77-4fc3-a546-1e85f8c68e77",
     "a1000001-0000-0000-0000-000000000001",
     "kn-5de5a9ac-fd15-45ab-bf18-77566781cf40",
-    "bl-1b58b05c-9305-4f06-a586-a08c96008027",
+    "96497334-b18f-495c-9228-eeb8182bdc38",
     "mem-024598a9-ed2e-4eeb-b1e1-5410856ff132",
     "kn-78db5396-3dbc-4481-bfc7-e4e1422feb1c",
-    "mem-5708f4c9-3d61-4182-8543-2843698931e6"
+    "knw-ed33e669-0790-44cb-a036-958d605c6fea"
    ],
    "n_returned": 10,
-   "latency_ms": 616.7,
+   "latency_ms": 1471.0,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -650,19 +650,19 @@
    "category": "paraphrase",
    "query": "loved for the unedited self and not the polished exterior",
    "returned": [
-    "077d064f-3489-4c05-9aca-3782f96b51db",
+    "27e1b1a4-ad0b-49d9-812f-fedf43b8aabe",
     "kn-13f60407-7b70-4db1-964f-ea1f8196efbd",
     "knw-2c46cfb4-6d4e-4822-8a1a-7d743c1e4329",
-    "kn-6061318f-046b-4935-907d-8eafdce14930",
+    "bl-87c93185-b2bf-40af-ae23-3c830c007abf",
     "knw-c72597c5-c23d-4c08-8e9e-996dadf26a99",
     "kn-5de5a9ac-fd15-45ab-bf18-77566781cf40",
-    "27e1b1a4-ad0b-49d9-812f-fedf43b8aabe",
+    "077d064f-3489-4c05-9aca-3782f96b51db",
     "kn-db9f141b-dbe3-4037-92e0-4bb9be0e5e6e",
     "knw-920c891f-bb8c-48c4-9afc-018ef12dcdc4",
-    "fce2792a-53fc-4d4a-be3b-42bd6ceb1ba7"
+    "kn-6061318f-046b-4935-907d-8eafdce14930"
    ],
    "n_returned": 10,
-   "latency_ms": 658.4,
+   "latency_ms": 1516.6,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -687,7 +687,7 @@
     "mem-a16deccb-16a7-419c-a013-ff824a4daa15"
    ],
    "n_returned": 10,
-   "latency_ms": 566.3,
+   "latency_ms": 1224.2,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -700,19 +700,19 @@
    "category": "paraphrase",
    "query": "a childhood offering no solid foundation to inherit",
    "returned": [
-    "mem-b99efff0-00e6-40c8-9c5b-730330eef33b",
+    "bl-3f57bc69-7285-4f4a-a861-2de52efca058",
     "kn-a99cefe3-5e83-4050-98d8-6c69f57c7c71",
-    "knw-2c46cfb4-6d4e-4822-8a1a-7d743c1e4329",
-    "knw-f6ed7d00-bf7d-42ce-9e40-77cf3406e918",
-    "kn-6061318f-046b-4935-907d-8eafdce14930",
-    "knw-23c27d3b-e0d2-43a8-a80c-0a44477ae18a",
-    "kn-eb1b9e18-3dc6-4b9b-9cc6-86e0ae6b6be8",
     "kn-f230b362-b201-4402-9833-4160c89ab3d4",
-    "tag-childhood",
+    "bl-0d8c5dfa-e163-4fef-a58b-56b0d076c5a8",
+    "kn-6061318f-046b-4935-907d-8eafdce14930",
+    "kn-db9f141b-dbe3-4037-92e0-4bb9be0e5e6e",
+    "mem-b99efff0-00e6-40c8-9c5b-730330eef33b",
+    "kn-eb1b9e18-3dc6-4b9b-9cc6-86e0ae6b6be8",
+    "kn-5b606390-a52d-4ca2-8e0e-eba141d13440",
     "kn-0bb4f021-56de-4947-a35b-a37209e7ba21"
    ],
    "n_returned": 10,
-   "latency_ms": 589.6,
+   "latency_ms": 1335.9,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -733,11 +733,11 @@
     "766de879-f9d0-4a07-b6df-b43ee13763d8",
     "mem-5f76880b-bafb-4716-8e15-90f8ef59bebc",
     "kn-e0423482-cfa5-4796-8689-8495c93b66bc",
-    "28ae74a1-9d47-4874-9279-43c1f90c0f64",
+    "3499d5da-0e9c-4de4-9bc4-8941b14e0b1f",
     "mem-bce80169-2d46-4b3e-9ebe-8498e26f0a89"
    ],
    "n_returned": 10,
-   "latency_ms": 627.8,
+   "latency_ms": 1421.9,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.18181818181818182,
@@ -762,7 +762,7 @@
     "5eb24168-c2d3-4842-85db-2070e0e14923"
    ],
    "n_returned": 10,
-   "latency_ms": 524.3,
+   "latency_ms": 1102.3,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -787,7 +787,7 @@
     "?V?"
    ],
    "n_returned": 10,
-   "latency_ms": 526.0,
+   "latency_ms": 1173.7,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.18181818181818182,
@@ -812,7 +812,7 @@
     "???I?cB?Zx?"
    ],
    "n_returned": 10,
-   "latency_ms": 526.0,
+   "latency_ms": 1187.7,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -837,7 +837,7 @@
     "kn-13f60407-7b70-4db1-964f-ea1f8196efbd"
    ],
    "n_returned": 10,
-   "latency_ms": 524.8,
+   "latency_ms": 1192.0,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 0.09090909090909091,
@@ -853,85 +853,63 @@
     "kn-eb1b9e18-3dc6-4b9b-9cc6-86e0ae6b6be8",
     "mem-5305665c-6b5b-45b7-89ae-5d2fb0b896ac",
     "4f698ae6-c40e-464e-9798-50350991a188",
-    "mem-a0b7cfda-bc9e-4f40-b9a9-1722cf3f8263",
-    "kn-6061318f-046b-4935-907d-8eafdce14930",
     "knw-35940684-abc4-42f0-b942-818f66b1f69a",
+    "kn-6061318f-046b-4935-907d-8eafdce14930",
+    "?Z?<S???K ?",
+    "mem-a0b7cfda-bc9e-4f40-b9a9-1722cf3f8263",
     "kn-db9f141b-dbe3-4037-92e0-4bb9be0e5e6e",
-    "719aa819-00a9-4f4b-a857-4f9fe5ad44d7",
-    "mem-d396d789-0f7f-4366-a008-5d8801c8f2eb",
-    "kn-0bb4f021-56de-4947-a35b-a37209e7ba21"
+    "'?T?a\"B~-?8",
+    "mem-d396d789-0f7f-4366-a008-5d8801c8f2eb"
    ],
    "n_returned": 10,
-   "latency_ms": 668.9,
+   "latency_ms": 1425.3,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
-   "recall@10": 0.18181818181818182,
+   "recall@10": 0.09090909090909091,
    "precision@5": 0.0,
-   "mrr@10": 0.14285714285714285
+   "mrr@10": 0.125
   },
   {
    "id": "q33",
    "category": "nonsense",
    "query": "zqxjvw plimforth grebulon",
-   "returned": [
-    "$\\?l????T?",
-    "project-Deploy_Ollama_on_Legion_k8s__Traefik_route_at_ollama_neuralplatform_ai__8B_model_seeded_",
-    "c??Z??I?E??",
-    "?^?l????K8?",
-    "${????X?6#E",
-    "???Z??I?b??",
-    "??m???|Y`0?",
-    "??m???|Y`0?",
-    "??m???|Y`0?",
-    "=?m???|YH??"
-   ],
-   "n_returned": 10,
-   "latency_ms": 338.4,
+   "returned": [],
+   "n_returned": 0,
+   "latency_ms": 714.9,
    "error": null,
-   "clean": false,
-   "false_positives": 10
+   "clean": true,
+   "false_positives": 0
   },
   {
    "id": "q34",
    "category": "nonsense",
    "query": "flarnbistle quommetry",
-   "returned": [
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??"
-   ],
-   "n_returned": 10,
-   "latency_ms": 248.9,
+   "returned": [],
+   "n_returned": 0,
+   "latency_ms": 496.0,
    "error": null,
-   "clean": false,
-   "false_positives": 10
+   "clean": true,
+   "false_positives": 0
   },
   {
    "id": "q35",
    "category": "nonsense",
    "query": "xxqzzt vurblenacht throom",
    "returned": [
+    "bl-4476e856-c567-4b49-8ff7-d7dca3e5715e",
     "?V?",
+    "knw-6b48dce2-f21c-452a-9db5-4e6aa61c87ca",
+    "knw-920c891f-bb8c-48c4-9afc-018ef12dcdc4",
     "?m?\\}Q??6??",
+    "8cbb60c5-4999-4ec1-8682-2592aedc4249",
+    "kn-66a21179-2adc-4b19-a109-880cf4674d7d",
     "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??",
-    "?m?\\}Q??6??"
+    "knw-723551f5-1950-42a3-8b89-b6a06913cef0",
+    "kn-333542cb-6dab-4662-9725-bf7440d28bf7"
    ],
    "n_returned": 10,
-   "latency_ms": 338.6,
+   "latency_ms": 729.4,
    "error": null,
    "clean": false,
    "false_positives": 10
@@ -947,13 +925,13 @@
     "kn-b7e98d63-8b83-4911-b4d0-990602a7f575",
     "tag-__darma____cgi____patents____self-improvement____character-preservation____autonomous____kotlin____architecture__",
     "knw-e047bb42-dc5b-4383-9e88-e508dc03abe3",
-    "mem-c17aefb1-38b5-4ced-af50-fe524127e1a4",
+    "mem-e6327f52-2bda-4ce7-9471-2fffd1e172de",
     "bl-145a0985-2382-400f-a7c5-c335c5e30a72",
-    "bl-5b17bd3b-0c41-46cb-a710-6fa4429692ff",
-    "mem-82b93b21-a865-410f-9ec1-fc54121d9bb5"
+    "mem-82b93b21-a865-410f-9ec1-fc54121d9bb5",
+    "bl-5b17bd3b-0c41-46cb-a710-6fa4429692ff"
    ],
    "n_returned": 10,
-   "latency_ms": 644.8,
+   "latency_ms": 1306.9,
    "error": null,
    "hit@5": 1.0,
    "recall@5": 1.0,
@@ -962,7 +940,7 @@
    "mrr@10": 0.5,
    "outranks": true,
    "rank_correct": 2,
-   "rank_stale": 9
+   "rank_stale": 10
   },
   {
    "id": "q37",
@@ -978,10 +956,10 @@
     "13705072-4515-4124-963d-083af490494f",
     "3cf706a1-3825-45d8-b0a9-06cae6cdf5b8",
     "kn-5de5a9ac-fd15-45ab-bf18-77566781cf40",
-    "4f225001-3a51-4a68-8d38-c8ecac3412de"
+    "96eb59a0-5603-4c9e-835c-1de53f2319bc"
    ],
    "n_returned": 10,
-   "latency_ms": 738.7,
+   "latency_ms": 1647.4,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,
@@ -1009,7 +987,7 @@
     "mem-3a2cf162-d93b-4f29-86f2-5066fb7fe1f5"
    ],
    "n_returned": 10,
-   "latency_ms": 498.6,
+   "latency_ms": 1003.4,
    "error": null,
    "hit@5": 0.0,
    "recall@5": 0.0,

tools/retrieval-eval/sim6.py

@@ -0,0 +1,135 @@
+import numpy as np, json, urllib.request, collections, math, re, sys, time
+SP="/private/tmp/claude-501/-Users-timlingo/82369039-a20e-4b5a-8a5e-28234a57b996/scratchpad"
+EV="/Users/timlingo/Development/neuron-technologies/_wt-bm25lex/tools/retrieval-eval/"
+np.seterr(all='ignore'); t0=time.time()
+M=np.load(SP+'/emb.npy'); eids=open(SP+'/ids.txt',encoding='utf-8',errors='surrogateescape').read().split('\n')
+eidx={k:i for i,k in enumerate(eids)}
+d=json.load(open('/Users/timlingo/neuron-memory-backups/snapshot-pre-repair-20260806.json',encoding='utf-8',errors='surrogateescape'))
+STRUCT={"identity","contains","superseded_by","references","embodies","demonstrated_by","canonical-self","depends_on","currently_holds","activates"}
+adj=collections.defaultdict(list)
+for e in d['edges']:
+    if e.get('relation') not in STRUCT: continue
+    w=float(e.get('weight') or 0.0)
+    adj[e['from_id']].append((e['to_id'],w)); adj[e['to_id']].append((e['from_id'],w))
+nodes=d['nodes']; N={n['id']:n for n in nodes}
+PRINT=re.compile(r'^[\x20-\x7e]+$')
+ids=[];hay=[];dl=[];sal=[];addr=[]
+for n in nodes:
+    i=n.get('id') or ''
+    h=((n.get('content') or '')+'\x00'+(n.get('label') or '')+'\x00'+(n.get('tags') or '')).lower()
+    ids.append(i);hay.append(h);dl.append(len(h));sal.append(float(n.get('salience') or 0.0));addr.append(bool(PRINT.match(i)))
+del d
+NN=len(ids); avgdl=sum(dl)/NN
+gold={q['id']:q for q in json.load(open(EV+"gold_set.json"))['queries']}
+CACHE={}
+def emb(t):
+    if t in CACHE: return CACHE[t]
+    b=json.dumps({"model":"nomic-embed-text","prompt":t}).encode()
+    r=urllib.request.Request("http://127.0.0.1:11434/api/embeddings",data=b,headers={"Content-Type":"application/json"})
+    v=np.array(json.load(urllib.request.urlopen(r,timeout=60))["embedding"],dtype=np.float32)
+    v=v/(np.linalg.norm(v)+1e-9); CACHE[t]=v; return v
+K1,B=1.2,0.75
+LEXC={}
+def lexleg(qid,query,lim=10):
+    if qid in LEXC: return LEXC[qid]
+    toks=[]
+    for w in query.split():
+        wl=w.lower()
+        if wl not in toks: toks.append(wl)
+    nt=len(toks); masks=[]; df=[0]*nt
+    for i in range(NN):
+        if not addr[i]: continue
+        h=hay[i]; m=0; sc=0
+        for t in range(nt):
+            if toks[t] in h: m|=(1<<t); sc+=1; df[t]+=1
+        if sc: masks.append((i,m))
+    idf=[math.log(1.0+(NN-df[t]+0.5)/(df[t]+0.5)) for t in range(nt)]
+    scored=[]
+    for i,m in masks:
+        norm=1.0-B+B*dl[i]/avgdl; s=0.0
+        for t in range(nt):
+            if m&(1<<t): s+=idf[t]*(K1+1.0)/(1.0+K1*norm)
+        scored.append((s,i))
+    scored.sort(key=lambda x:(-x[0],-sal[x[1]]))
+    LEXC[qid]=([ids[i] for s,i in scored[:lim]], len(masks))
+    return LEXC[qid]
+FIRE=0.02; DECAY=0.7; DEPTH=2; SEED_MIN=0.60; ASSOC_MAX=64
+def assoc(seeds, s, use_cos, order):
+    act={x:1.0 for x in seeds}; seen={x:2 for x in seeds}
+    Q=[(x,0) for x in seeds]; h=0
+    while h<len(Q):
+        cur,hop=Q[h]; h+=1
+        if hop>=DEPTH: continue
+        p=act[cur]
+        for oid,w in adj.get(cur,()):
+            n=N.get(oid)
+            if not n or n.get('node_type') in ('Tag','InternalStateEvent'): continue
+            c=1.0
+            if use_cos:
+                j=eidx.get(oid)
+                c=max(0.0,float(s[j])) if j is not None else 0.0
+            na=p*w*DECAY*float(n.get('salience') or 0.0)*c
+            if na<FIRE: continue
+            if oid in seen and na<=act.get(oid,0): continue
+            act[oid]=na
+            if oid not in seen: seen[oid]=1
+            Q.append((oid,hop+1))
+    out=[]
+    for k,v in seen.items():
+        if v!=1 or k not in eidx: continue
+        c=float(s[eidx[k]])
+        if c<=0: continue
+        out.append((act[k] if order=='act' else c,k))
+    out.sort(reverse=True)
+    return [k for c,k in out[:ASSOC_MAX] if PRINT.match(k or '')]
+def inter(legs,lim=10):
+    out=[];idx=[0]*len(legs)
+    while len(out)<lim and any(idx[i]<len(legs[i]) for i in range(len(legs))):
+        for i in range(len(legs)):
+            if idx[i]<len(legs[i]):
+                if legs[i][idx[i]] not in out: out.append(legs[i][idx[i]])
+                idx[i]+=1
+            if len(out)>=lim: break
+    return out
+def run(floor, vocabgate, use_cos, order):
+    res={}; legs={}
+    for qid,q in gold.items():
+        v=emb(q['query']); s=M@v; s[~np.isfinite(s)]=-1
+        L,nmatch=lexleg(qid,q['query'])
+        if vocabgate and nmatch==0:
+            res[qid]=[]; legs[qid]=([],[],[]); continue
+        ordr=np.argsort(-s)
+        S=[eids[j] for j in ordr[:10] if PRINT.match(eids[j] or '') and (not floor or s[j]>SEED_MIN)]
+        seeds=[x for x in L[:3] if x in N]
+        seeds=seeds+[eids[j] for j in ordr[:8] if eids[j] in N and eids[j] not in seeds and PRINT.match(eids[j] or '')]
+        A=assoc(seeds,s,use_cos,order) if seeds else []
+        res[qid]=inter([L,S,A]); legs[qid]=(L,S,A)
+    return res,legs
+def score(res,label,base=None):
+    det={}
+    for qid,q in gold.items():
+        out=res[qid][:5]
+        if q['category']=='nonsense': ok=(len(res[qid])==0)
+        elif q['category']=='superseded':
+            must=q.get('must_outrank') or {}; ok=False
+            for good,bad in (must.items() if isinstance(must,dict) else []):
+                ok = good in res[qid] and (bad not in res[qid] or res[qid].index(good)<res[qid].index(bad))
+            if not must: ok=any(r in out for r in q['relevant'])
+        else: ok=any(r in out for r in q['relevant'])
+        det[qid]=ok
+    line="%-34s true=%d/38"%(label,sum(det.values()))
+    if base is not None:
+        dd=[q for q in sorted(gold) if det[q]!=base[q]]
+        line+="  moved=%d gains=%s losses=%s"%(len(dd),[q for q in dd if det[q]],[q for q in dd if not det[q]])
+    print(line, flush=True)
+    return det
+if __name__=="__main__":
+    b,_=run(True,False,False,'cos'); base=score(b,'BASE bm25lex replica')
+    for lab,args in [
+        ("A floor-off+vocabgate",      (False,True,False,'cos')),
+        ("B A+cos-in-traversal",       (False,True,True ,'cos')),
+        ("C A+cos-trav+act-order",     (False,True,True ,'act')),
+        ("D floor-off NO gate",        (False,False,False,'cos')),
+    ]:
+        r,_=run(*args); score(r,lab,base)
+    print("elapsed %.1fs"%(time.time()-t0),file=sys.stderr)

tools/retrieval-eval/sim6b.py

@@ -0,0 +1,32 @@
+exec(open('sim6.py').read().split('if __name__')[0])
+HASSTRUCT=set(adj.keys())
+print("nodes with >=1 structural edge:",len(HASSTRUCT),file=sys.stderr)
+def run2(sfilter, seedout, lim=10):
+    res={}
+    for qid,q in gold.items():
+        v=emb(q['query']); s=M@v; s[~np.isfinite(s)]=-1
+        L,nmatch=lexleg(qid,q['query'])
+        if nmatch==0: res[qid]=[]; continue
+        ordr=np.argsort(-s)
+        cand=[eids[j] for j in ordr[:200] if PRINT.match(eids[j] or '')]
+        S=[x for x in cand if (not sfilter or x in HASSTRUCT)][:10]
+        seeds=[x for x in L[:3] if x in N]
+        semseeds=[eids[j] for j in ordr[:8] if eids[j] in N and eids[j] not in seeds and PRINT.match(eids[j] or '')]
+        seeds=seeds+semseeds
+        A=assoc(seeds,s,False,'cos') if seeds else []
+        if seedout:
+            extra=[(float(s[eidx[x]]),x) for x in semseeds if x in HASSTRUCT and x in eidx]
+            merged=[(float(s[eidx[x]]),x) for x in A if x in eidx]+extra
+            merged.sort(reverse=True)
+            seen=set(); A=[]
+            for c,x in merged:
+                if x in seen: continue
+                seen.add(x); A.append(x)
+            A=A[:ASSOC_MAX]
+        res[qid]=inter([L,S,A])
+    return res
+b,_=run(True,False,False,'cos'); base=score(b,'BASE bm25lex replica')
+a,_=run(False,True,False,'cos'); score(a,'A floor-off+vocabgate',base)
+score(run2(False,True),'E A+struct-seeds-in-graphleg',base)
+score(run2(True,False),'F A+S-restricted-to-graph',base)
+score(run2(True,True),'G  E+F',base)

tools/retrieval-eval/sim6c.py

@@ -0,0 +1,30 @@
+exec(open('sim6.py').read().split('if __name__')[0])
+HASSTRUCT=set(adj.keys())
+import json as _j
+d2=_j.load(open('/Users/timlingo/neuron-memory-backups/snapshot-pre-repair-20260806.json',encoding='utf-8',errors='surrogateescape'))
+ANYEDGE=set()
+for e in d2['edges']: ANYEDGE.add(e['from_id']); ANYEDGE.add(e['to_id'])
+del d2
+print("struct=%d anyedge=%d"%(len(HASSTRUCT),len(ANYEDGE)),file=sys.stderr)
+def run4(pool, nlegs, lim=10):
+    P = HASSTRUCT if pool=='struct' else ANYEDGE
+    res={}
+    for qid,q in gold.items():
+        v=emb(q['query']); s=M@v; s[~np.isfinite(s)]=-1
+        L,nmatch=lexleg(qid,q['query'])
+        if nmatch==0: res[qid]=[]; continue
+        ordr=np.argsort(-s)
+        cand=[eids[j] for j in ordr[:3000] if PRINT.match(eids[j] or '')]
+        S=cand[:10]
+        G=[x for x in cand if x in P][:10]
+        seeds=[x for x in L[:3] if x in N]
+        seeds=seeds+[eids[j] for j in ordr[:8] if eids[j] in N and eids[j] not in seeds and PRINT.match(eids[j] or '')]
+        A=assoc(seeds,s,False,'cos') if seeds else []
+        legs=[L,S,G,A] if nlegs==4 else [L,G,A]
+        res[qid]=inter(legs)
+    return res
+b,_=run(True,False,False,'cos'); base=score(b,'BASE bm25lex replica')
+score(run4('struct',4),'I 4leg L,S,G(struct),A',base)
+score(run4('any',4),   'J 4leg L,S,G(anyedge),A',base)
+score(run4('struct',3),'K 3leg L,G(struct),A',base)
+score(run4('any',3),   'L 3leg L,G(anyedge),A',base)

tools/retrieval-eval/wb2.py

@@ -1,92 +0,0 @@
-import json,sys,pickle,numpy as np
-sys.path.insert(0,'.')
-from legs import *
-GP='/Users/timlingo/Development/neuron-technologies/_wt-bm25lex/tools/retrieval-eval/'
-G=json.load(open(GP+'gold_set.json'))
-def wstart(s,tok):
-    i=s.find(tok)
-    while i!=-1:
-        if i==0 or not s[i-1].isalnum(): return True
-        i=s.find(tok,i+1)
-    return False
-def legs4(query, wordstart=False, unfloor=False):
-    toks=tokenize(query); lt=[t.lower() for t in toks]
-    hit_idx=[];hit_mask=[];df=[0]*len(toks)
-    for i in range(N):
-        if not OK[i]: continue
-        s=LOW[i];m=0
-        for t,tok in enumerate(lt):
-            if tok in s and (not wordstart or wstart(s,tok)): m|=(1<<t)
-        if m:
-            hit_idx.append(i);hit_mask.append(m)
-            for t in range(len(toks)):
-                if m>>t&1: df[t]+=1
-    dl_n=int(OK.sum());avgdl=float(DL[OK].sum()/max(dl_n,1))
-    idf=[math.log(1.0+((dl_n-d+0.5)/(d+0.5))) for d in df]
-    L=[]
-    for j,i in enumerate(hit_idx):
-        norm=1.0-B+B*(DL[i]/avgdl);w=0.0
-        for t in range(len(toks)):
-            if hit_mask[j]>>t&1: w+=idf[t]*(K1+1.0)/(1.0+K1*norm)
-        L.append((i,w,SAL[i]))
-    L.sort(key=lambda x:(-x[1],-x[2]))
-    if not L: return [],[],[]
-    qv=qemb(query);cos=En@qv;cos=np.where(HAVE&OK,cos,-2.0)
-    order=np.argsort(-cos)[:600]
-    Sl=[int(i) for i in order if cos[i]>(0.0 if unfloor else SEED_MIN)]
-    semseed=[int(i) for i in order[:SEED_K] if cos[i]>0.0]
-    act={};seen={};qq=[]
-    for i,_,_ in L[:ASSOC_SEEDS]:
-        act[i]=1.0;seen[i]=2;qq.append((i,0))
-    for i in semseed:
-        if i in seen: continue
-        act[i]=1.0;seen[i]=2;qq.append((i,0))
-    qh=0
-    while qh<len(qq):
-        cur,h=qq[qh];qh+=1
-        if h>=DEPTH: continue
-        parent=act[cur]
-        for e,oi in ADJ_F[cur]+ADJ_T[cur]:
-            if e['rel'] not in STRUCT or EXCL[oi]: continue
-            na=parent*e['w']*DECAY*SAL[oi]
-            if na<FIRE: continue
-            if seen.get(oi) and na<=act.get(oi,0): continue
-            act[oi]=na
-            if not seen.get(oi): seen[oi]=1
-            if len(qq)<AMAX*4: qq.append((oi,h+1))
-    A=sorted([(i,float(cos[i])) for i,st in seen.items() if st==1 and OK[i] and HAVE[i] and cos[i]>0.0],key=lambda x:-x[1])[:AMAX]
-    return [i for i,_,_ in L],Sl,[i for i,_ in A]
-def merge(L,S,A,lim=10):
-    out=[];li=si=ai=0
-    while len(out)<lim and (li<len(L) or si<len(S) or ai<len(A)):
-        if li<len(L):
-            if L[li] not in out: out.append(L[li])
-            li+=1
-        if len(out)>=lim: break
-        if si<len(S):
-            if S[si] not in out: out.append(S[si])
-            si+=1
-        if len(out)>=lim: break
-        if ai<len(A):
-            if A[ai] not in out: out.append(A[ai])
-            ai+=1
-    return out
-def outcome(q,ids):
-    c=q['category']
-    if c=='nonsense': return len(ids)==0
-    if c=='superseded':
-        a,b=q['must_outrank']
-        if a not in ids: return False
-        if b not in ids: return True
-        return ids.index(a)<ids.index(b)
-    return any(x in ids[:5] for x in q['relevant'])
-def run(**kw):
-    return {q['id']:outcome(q,[NODES[i]['id'] for i in merge(*legs4(q['query'],**kw),10)]) for q in G['queries']}
-base=run()
-print("baseline",sum(base.values()),"misses",[k for k,v in base.items() if not v])
-for name,kw in [('wordstart',dict(wordstart=True)),
-                ('unfloor',dict(unfloor=True)),
-                ('wordstart+unfloor',dict(wordstart=True,unfloor=True))]:
-    r=run(**kw)
-    g=sorted(k for k in base if r[k] and not base[k]);l=sorted(k for k in base if base[k] and not r[k])
-    print("%-20s net=%+d gains=%s losses=%s"%(name,len(g)-len(l),g,l))

vendor/el-runtime/v1.0.0-20260501/el_runtime.c

@@ -7327,39 +7327,6 @@ static int istr_contains(const char* hay, const char* needle) {
     return 0;
 }
 
-/* Word-START-anchored variant of istr_contains.
- *
- * WHY. The retrieval match primitive is a raw substring test, so a query token
- * matches ANYWHERE inside a corpus word: "throom" matches "bathroom", "cat"
- * matches "concatenate". Measured on this corpus over the 38-query gold set,
- * that is not a rare accident — it is the bulk of some queries' candidate
- * sets. q28's lexical leg is 36,954 records of which only 13 contain a query
- * token at a word start (99.96% mid-word noise); six other queries carry
- * ~20,500 mid-word-only records each; and the nonsense control q35
- * ("xxqzzt vurblenacht throom") returns 7 records ALL of which match only
- * mid-word, which is the entire reason that control has been dirty since main.
- *
- * WHAT CHANGES. A token must begin at a word boundary — the preceding
- * character is not alphanumeric. Suffixes are still matched ("value" still
- * hits "values", "unjailbreakable" still hits "unjailbreakables"), so this is
- * strictly a prefix anchor, not whole-word equality; whole-word equality would
- * break the morphological matching the phrase category depends on.
- *
- * PROVENANCE, stated honestly: this restores no engram claim. Will's design
- * has no lexical leg at all (05-detailed-description l.64 takes "one or more
- * seed node UUIDs representing the current active context" as its input), so
- * the lexical leg is the seed-finding step that feeds the designed mechanism.
- * Cleaner seeds serve that mechanism; they do not replace it. */
-static int istr_contains_wordstart(const char* hay, const char* needle) {
-    if (!hay || !needle || !*needle) return 0;
-    size_t nl = strlen(needle);
-    for (const char* p = hay; *p; p++) {
-        if (p != hay && isalnum((unsigned char)p[-1])) continue;
-        if (strncasecmp(p, needle, nl) == 0) return 1;
-    }
-    return 0;
-}
-
 /* ── Tokenized query matching ───────────────────────────────────────────
  * The engram query surface (search / activate / goal-bias) historically
  * matched the ENTIRE raw query string as a single case-insensitive
@@ -7412,9 +7379,9 @@ static int engram_node_match_score(const EngramNode* n,
                                    char toks[][ENGRAM_QTOK_LEN], int ntok) {
     int score = 0;
     for (int t = 0; t < ntok; t++) {
-        if (istr_contains_wordstart(n->content, toks[t]) ||
-            istr_contains_wordstart(n->label,   toks[t]) ||
-            istr_contains_wordstart(n->tags,    toks[t]))
+        if (istr_contains(n->content, toks[t]) ||
+            istr_contains(n->label,   toks[t]) ||
+            istr_contains(n->tags,    toks[t]))
             score++;
     }
     return score;
@@ -7429,9 +7396,9 @@ static uint32_t engram_node_match_mask(const EngramNode* n,
                                        char toks[][ENGRAM_QTOK_LEN], int ntok) {
     uint32_t m = 0;
     for (int t = 0; t < ntok && t < 32; t++) {
-        if (istr_contains_wordstart(n->content, toks[t]) ||
-            istr_contains_wordstart(n->label,   toks[t]) ||
-            istr_contains_wordstart(n->tags,    toks[t]))
+        if (istr_contains(n->content, toks[t]) ||
+            istr_contains(n->label,   toks[t]) ||
+            istr_contains(n->tags,    toks[t]))
             m |= (uint32_t)1u << t;
     }
     return m;
@@ -9687,12 +9654,25 @@ el_val_t engram_search_json(el_val_t query, el_val_t limit) {
                     }
                     if (sem && n->emb && n->emb_dim == qdim) {
                         double c = eg_cosine(n->emb, qv, qdim);
-                        /* Semantic leg: identical to eg_sem_term(), which is
-                         * left in place and still used by engram_search().
-                         * Inlined here only so one cosine serves both uses. */
-                        if (c > ENGRAM_EMBED_SEED_MIN) {
-                            double sv = (c - ENGRAM_EMBED_SEED_MIN) / (1.0 - ENGRAM_EMBED_SEED_MIN);
-                            if (sv > 1.0) sv = 1.0;
+                        /* Semantic leg, claim 24 verbatim: "returning the node
+                         * records whose embedding vectors have the HIGHEST
+                         * COSINE SIMILARITY to a query vector" — a ranking, with
+                         * no threshold anywhere in the claim. The leg used to be
+                         * gated at ENGRAM_EMBED_SEED_MIN and rescaled onto
+                         * [SEED_MIN,1]; that constant is defined (l.6083) as the
+                         * SEED-JOIN threshold for the HippoRAG pass, and reusing
+                         * it as a result filter is not authorised by claim 24.
+                         * Measured on this corpus, it is also not a quality
+                         * gate: true paraphrase targets score 0.459-0.657 while
+                         * the nonsense controls' own nearest neighbours score
+                         * 0.553-0.622 — the distributions overlap, so no value
+                         * of the constant separates them. What actually holds
+                         * the nonsense control is corpus vocabulary (see the
+                         * nhits==0 gate below), not cosine magnitude.
+                         * Claim 32: clamp the cosine to [0,1] rather than let a
+                         * negative value invert the signal. */
+                        if (c > 0.0) {
+                            double sv = c > 1.0 ? 1.0 : c;
                             sem[nsem].idx = i; sem[nsem].sem = sv; nsem++;
                         }
                         /* Graph seeds: top-K by RAW cosine, insertion-ordered. */
@@ -9710,6 +9690,21 @@ el_val_t engram_search_json(el_val_t query, el_val_t limit) {
                         }
                     }
                 }
+                /* CORPUS-VOCABULARY GATE — the thing that actually keeps an
+                 * unfloored semantic leg from answering gibberish.
+                 * nhits == 0 means NO stored record contains ANY query token
+                 * anywhere in its content, label or tags: the query is outside
+                 * the graph's vocabulary entirely. A vector index always has a
+                 * nearest neighbour, so without this gate the semantic leg
+                 * answers "zqxjvw plimforth grebulon" with its 0.55-cosine
+                 * garbage. It is also the honest reading of Will's retrieval
+                 * contract: 05-detailed-description l.64 has the caller supply
+                 * "one or more seed node UUIDs representing the current active
+                 * context", and every leg here is downstream of finding those
+                 * seeds. No seeds, no retrieval — the graph declines rather
+                 * than confabulates. Suppressing the graph seeds too keeps the
+                 * associative leg from running off the semantic top-K alone. */
+                if (nhits == 0) { nsem = 0; nsemseed = 0; }
                 /* BM25-shaped lexical score. Binary term frequency (the match
                  * primitive is a substring test, not a count), Lucene-form IDF,
                  * and length normalisation over the corpus mean. A token that
@@ -9745,19 +9740,7 @@ el_val_t engram_search_json(el_val_t query, el_val_t limit) {
                     ? engram_assoc_leg(g, hits, nhits, semseed, nsemseed,
                                        qv, qdim, assoc, ENGRAM_ASSOC_MAX)
                     : 0;
-                /* Corpus-vocabulary gate. If no stored record contains ANY
-                 * query token in its content, label or tags, the query is
-                 * outside this graph's vocabulary: there are no seeds, and
-                 * 05-detailed-description l.64 makes retrieval downstream of
-                 * seeds ("the caller provides one or more seed node UUIDs
-                 * representing the current active context"). No seeds, no
-                 * retrieval — the graph declines rather than confabulating a
-                 * nearest neighbour for gibberish. The mechanism is iteration
-                 * 6's (feat/claim24-unfloored-semantic); it is required here
-                 * because word-start matching empties the lexical leg for
-                 * q35-style queries whose only "hits" were mid-word, and the
-                 * semantic leg would otherwise answer them anyway. */
-                int64_t* order = (nhits > 0) ? malloc((size_t)lim * sizeof(int64_t)) : NULL;
+                int64_t* order = malloc((size_t)lim * sizeof(int64_t));
                 if (order) {
                     int64_t no = engram_interleave3(hits, nhits, sem, nsem,
                                                     assoc, nassoc, lim, order);