V127 audit raw transcript schema and paths

.github/workflows/trace-ace-v127-transcript-schema-audit.yml

@@ -0,0 +1,41 @@
+name: Trace Ace V127 Transcript Schema Audit
+
+on:
+  pull_request:
+    branches: [agent/v121-cache-batch2]
+    paths:
+      - '.github/workflows/trace-ace-v127-transcript-schema-audit.yml'
+      - 'competitions/trace_the_ace/v127_transcript_schema_audit.py'
+  workflow_dispatch:
+
+jobs:
+  audit:
+    runs-on: ubuntu-24.04
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.12'
+      - name: Restore frozen transcript archive from GitHub cache only
+        id: transcript-cache
+        uses: actions/cache@v4
+        with:
+          path: transcripts.zip
+          key: trace-ace-transcripts-v1-603547640
+      - name: Require cache hit and frozen archive
+        run: |
+          test '${{ steps.transcript-cache.outputs.cache-hit }}' = 'true'
+          test -f transcripts.zip
+          test "$(stat -c%s transcripts.zip)" = '603547640'
+          echo 'e685b85b04694e130c25b17d09cdd1892fbda5e9fa685e98b2300114b915aa2d  transcripts.zip' | sha256sum -c -
+          unzip -tq transcripts.zip >/dev/null
+      - name: Run V127 discovery-only audit
+        run: python competitions/trace_the_ace/v127_transcript_schema_audit.py --archive transcripts.zip --sample 256 --rows-per-file 20 --out v127_transcript_schema_audit.json
+      - name: Show decision
+        run: cat v127_transcript_schema_audit.json
+      - uses: actions/upload-artifact@v4
+        with:
+          name: trace-ace-v127-transcript-schema-audit
+          path: v127_transcript_schema_audit.json
+          retention-days: 14

competitions/trace_the_ace/v127_transcript_schema_audit.py

@@ -0,0 +1,89 @@
+#!/usr/bin/env python3
+"""V127 cache-only transcript archive schema/path audit.
+
+Discovery-only. No labels and no model fitting. Inspects every CSV member name and
+header in the frozen transcript ZIP to determine whether process/provider/schema
+metadata exists below the loader abstraction. Also inspects a deterministic small
+sample for role and non-content field vocabularies. No transcript corpus mutation.
+"""
+from __future__ import annotations
+import argparse, csv, io, json, re, zipfile, hashlib
+from collections import Counter, defaultdict
+from pathlib import PurePosixPath, Path
+
+REQUIRED=("session_id","utterance_id","role","content","timestamp")
+
+def stable(s): return hashlib.sha256(s.encode()).hexdigest()
+
+def shape(v):
+    s=str(v or '')
+    s=re.sub(r'[0-9]','#',s); s=re.sub(r'[A-Fa-f]','a',s)
+    return s[:80]
+
+def main(a):
+    z=zipfile.ZipFile(a.archive)
+    members=[n for n in z.namelist() if n.lower().endswith('.csv') and not n.endswith('/')]
+    headers=Counter(); path_depth=Counter(); topdirs=Counter(); second=Counter(); extras=Counter(); bad=[]
+    member_records=[]
+    for n in members:
+        p=PurePosixPath(n); path_depth[len(p.parts)]+=1
+        if p.parts: topdirs[p.parts[0]]+=1
+        if len(p.parts)>1: second['/'.join(p.parts[:2])]+=1
+        try:
+            with z.open(n) as f:
+                first=io.TextIOWrapper(f,encoding='utf-8-sig',newline='').readline()
+            h=tuple(next(csv.reader([first]))) if first else tuple()
+        except Exception as e:
+            bad.append({'member':n,'error':repr(e)}); continue
+        headers[h]+=1
+        for c in h:
+            if c not in REQUIRED: extras[c]+=1
+        member_records.append((stable(n),n,h))
+    member_records.sort()
+    sample=member_records[:min(a.sample,len(member_records))]
+    role_values=Counter(); noncontent_values=defaultdict(Counter); timestamp_shapes=Counter(); utterance_shapes=Counter(); sid_shapes=Counter()
+    sampled_rows=0
+    for _,n,h in sample:
+        try:
+            with z.open(n) as f:
+                t=io.TextIOWrapper(f,encoding='utf-8-sig',newline='')
+                r=csv.DictReader(t)
+                for j,row in enumerate(r):
+                    sampled_rows+=1
+                    role_values[str(row.get('role',''))]+=1
+                    timestamp_shapes[shape(row.get('timestamp',''))]+=1
+                    utterance_shapes[shape(row.get('utterance_id',''))]+=1
+                    sid_shapes[shape(row.get('session_id',''))]+=1
+                    for c in h:
+                        if c not in ('content',):
+                            v=str(row.get(c,''))
+                            if len(noncontent_values[c])<200 or v in noncontent_values[c]: noncontent_values[c][v]+=1
+                    if j+1>=a.rows_per_file: break
+        except Exception as e:
+            bad.append({'member':n,'sample_error':repr(e)})
+    out={
+      'protocol':'V127_TRANSCRIPT_SCHEMA_PATH_AUDIT','archive':str(a.archive),
+      'csv_members':len(members),'bad_members':bad[:20],
+      'header_signatures':[{'columns':list(k),'count':v,'missing_required':sorted(set(REQUIRED)-set(k)),'extra_columns':sorted(set(k)-set(REQUIRED))} for k,v in headers.most_common()],
+      'extra_column_presence':dict(extras.most_common()),
+      'path_depths':dict(path_depth),'top_level_paths':topdirs.most_common(30),'second_level_paths':second.most_common(30),
+      'sample':{'files':len(sample),'rows':sampled_rows,'roles':role_values.most_common(30),
+                'timestamp_shapes':timestamp_shapes.most_common(20),'utterance_id_shapes':utterance_shapes.most_common(20),'session_id_shapes':sid_shapes.most_common(20),
+                'noncontent_unique_counts':{c:len(v) for c,v in noncontent_values.items()},
+                'noncontent_top_values':{c:v.most_common(20) for c,v in noncontent_values.items() if c not in ('content',)}},
+      'decision':{}
+    }
+    any_extra=bool(extras); schema_variants=len(headers)>1; path_variants=len(second)>1 or len(topdirs)>1
+    if any_extra:
+        verdict='EXPLICIT_EXTRA_METADATA_PRESENT'
+    elif schema_variants:
+        verdict='SCHEMA_VARIANTS_PRESENT'
+    elif path_variants:
+        verdict='PATH_FAMILY_PRESENT'
+    else:
+        verdict='NO_EXPLICIT_PROVIDER_SCHEMA_CHANNEL'
+    out['decision']={'verdict':verdict,'any_extra_columns':any_extra,'header_signature_count':len(headers),'top_level_path_count':len(topdirs),'second_level_path_count':len(second)}
+    Path(a.out).write_text(json.dumps(out,indent=2)); print(json.dumps(out,indent=2),flush=True)
+
+if __name__=='__main__':
+    p=argparse.ArgumentParser(); p.add_argument('--archive',required=True); p.add_argument('--sample',type=int,default=256); p.add_argument('--rows-per-file',type=int,default=20); p.add_argument('--out',required=True); main(p.parse_args())