@a9i5k4/dsh-auto-memory 3.1.5 → 3.1.6

package/python/worker_semantic_v1.py

@@ -6,7 +6,7 @@ Extends the tested M7-0/M7-1 fake worker (python/worker_v1.py) WITHOUT
 touching its protocol semantics: same JSONL framing, same validators, same
 index_sync rejection matrix, same atomic derived-corpus persistence. Adds:
 
-  - after a successful index_sync commit: chunk (m7_chunk_v1) + embed
+  - after a successful index_sync commit: chunk (m7_chunk_pre_v1) + embed
     (frozen provider) every record and persist versioned vectors with an
     identity block under <dsh-home>/memory/semantic/ (atomic replace)
   - on startup: reuse persisted vectors only when the identity block
@@ -19,7 +19,7 @@ index_sync rejection matrix, same atomic derived-corpus persistence. Adds:
 
 Embedding backend is selected by an optional JSON config file passed via
 the DSH_M7_EMBEDDING_CONFIG environment variable (no CLI change, no JS
-change): {"provider":"bge-m3-v1"|"hash-v1", "modelDir":...,
+change): {"provider":"bge-m3-pre-v1"|"hash-pre-v1", "modelDir":...,
 "modelRevision":..., "dimension":1024, "torchThreads":16}.
 Without the env var the worker degrades to fake-worker behavior (protocol
 alive, embedding not ready) - never crashes, never changes ack semantics.
@@ -51,24 +51,24 @@ SHADOW_TOP_K = 8
 
 
 def canonical_workspace_key(key):
-    """Byte-twin of lib/evidence-store.js canonicalWorkspaceKey:
+    """Byte-twin of lib/evidence-store-pre.js canonicalWorkspaceKey:
     path.resolve + backslash->slash + lowercase."""
     return os.path.abspath(str(key == None and '' or key)).replace('\\', '/').lower()
 
 
 def wsref_of(workspace_key):
-    """Byte-twin of evidence-store.js workspaceRefOf. JS owns identity;
+    """Byte-twin of evidence-store-pre.js workspaceRefOf. JS owns identity;
     this is a deterministic reproduction of its published pure function so
     the worker can apply the workspace/scope/miv triple filter required by
     the M7-7.5 hardening audit (P1: isolation must be explicit, never an
     artifact of differing miv values)."""
     canon = canonical_workspace_key(workspace_key)
     return 'wsr_' + hashlib.sha256(
-        ('evidence-wsref-v1\u0000' + canon).encode('utf-8')).hexdigest()[:32]
+        ('evidence-wsref-pre-v1\u0000' + canon).encode('utf-8')).hexdigest()[:32]
 
 
 def _tokenize(text, stopwords=frozenset()):
-    """lexical_v2 parity tokenizer: NFKC + CJK 2-gram + ascii tokens."""
+    """lexical_pre_v2 parity tokenizer: NFKC + CJK 2-gram + ascii tokens."""
     t = unicodedata.normalize('NFKC', str(text)).lower()
     out = []
     for run in __import__('re').findall(r'[\u4e00-\u9fff]+|[a-z0-9_./-]+', t):
@@ -111,7 +111,7 @@ class LexicalBM25:
         return s
 
 # ---- M7-6 activation policy (default: shadow calibration only) ----
-ACTIVATION_POLICY_VERSION = 'm7_semantic_threshold_v1'
+ACTIVATION_POLICY_VERSION = 'm7_semantic_threshold_pre_v1'
 DEFAULT_ACTIVATION_POLICY = {
     'mode': 'shadow',            # 'shadow' = calibrate/log only; 'active' = emit frames
     'tOn': 0.62, 'tOff': 0.52,   # dual threshold, T_on > T_off (hysteresis)
@@ -166,7 +166,7 @@ class SemanticWorker(base.Worker):
         # 'shadow'(默认)=只记 shadow 行零发射;'canary-explicit'=仅 explicit 车道
         # 的 emit 决策发 activation_request 帧;'active' 预留。非法值回退 shadow
         # (fail closed)。此开关属 JS/用户运营面,不进策略工件——阈值权威仍在
-        # activation_policy_v2.json(append-only),发射节流依赖 M6 收件箱的
+        # activation_policy_pre_v2.json(append-only),发射节流依赖 M6 收件箱的
         # 硬校验+cooldown+TTL+latest-wins,worker 侧不重复限速。
         _em = str(self.embedding_config.get('activationEmitMode') or 'shadow')
         self.activation_emit_mode = _em if _em in (
@@ -184,9 +184,9 @@ class SemanticWorker(base.Worker):
                              'policies')
             try:
                 self._fv2 = featv2.load_and_verify_policy(
-                    os.path.join(pol_dir, 'recall_intent_lr_v1.json'),
+                    os.path.join(pol_dir, 'recall_intent_lr_pre_v1.json'),
                     os.path.join(pol_dir,
-                                 'activation_policy_v2.json'))
+                                 'activation_policy_pre_v2.json'))
             except Exception as exc:  # fail closed, retrieval unaffected
                 self._fv2_invalid = str(exc)[:200]
                 base.diag('featuresV2-policy-invalid: ' + self._fv2_invalid)
@@ -361,7 +361,7 @@ class SemanticWorker(base.Worker):
         payload = {
             'schemaVersion': 1,
             'namespace': base.NAMESPACE,
-            'policyVersion': 'semantic_vectors_v1',
+            'policyVersion': 'semantic_vectors_pre_v1',
             'identity': identity,
             'workspaceRef': ws_ref,
             'scope': scope,
@@ -627,7 +627,7 @@ class SemanticWorker(base.Worker):
             if len(excerpt.encode('utf-8')) > 480:
                 excerpt = excerpt[:150]
             cands.append({
-                'candidateId': 'cand_' + base.first32(
+                'candidateId': 'cand_pre_' + base.first32(
                     base.sha_str('m7-semantic-cand\u0000' + obs + '\u0000' +
                                  c['memoryId'] + '\u0000' + str(i))),
                 'memoryId': c['memoryId'], 'anchorId': c['anchorId'],
@@ -639,8 +639,8 @@ class SemanticWorker(base.Worker):
             })
         if not cands:
             return None
-        activation_id = 'act_' + base.first32(
-            base.sha_str('m7-semantic-activation-v1\u0000' + obs))
+        activation_id = 'act_pre_' + base.first32(
+            base.sha_str('m7-semantic-activation-pre-v1\u0000' + obs))
         created = req.get('sentAt', 0)
         ttl = int(pol['ttlSteps'])
         return {
@@ -736,7 +736,7 @@ class SemanticWorker(base.Worker):
 
     # ---------- M7-7 judgement shadow (audit only, never writes) ----------
 
-    JUDGEMENT_POLICY = 'judgement_shadow_v1'
+    JUDGEMENT_POLICY = 'judgement_shadow_pre_v1'
 
     _J_MARKERS = ('CORRECTION', 'UPDATED', 'REVISED', 'FREEZE',
                   'HARD RULE', 'DECISION reversing', '纠正', '更新:',
@@ -911,7 +911,7 @@ class SemanticWorker(base.Worker):
             self._append_shadow({
                 'schemaVersion': 1,
                 'namespace': base.NAMESPACE,
-                'policyVersion': 'semantic_shadow_v1',
+                'policyVersion': 'semantic_shadow_pre_v1',
                 'observationId': str(p.get('observationId', '')),
                 'workerEpoch': str(req.get('workerEpoch', '')),
                 'memoryIndexVersion': miv,
@@ -1063,7 +1063,7 @@ class SemanticWorker(base.Worker):
 
     def _intent_config_hash(self):
         ip_path = os.path.join(os.path.dirname(os.path.abspath(__file__)),
-                               'policies', 'recall_intent_lr_v1.json')
+                               'policies', 'recall_intent_lr_pre_v1.json')
         try:
             with open(ip_path, encoding='utf-8') as f:
                 ip = json.load(f)
@@ -1367,7 +1367,7 @@ def main():
     args, _unknown = ap.parse_known_args()
     if args.selftest:
         base.run_selftest()
-        w = SemanticWorker('ep', '', {'provider': 'hash-v1',
+        w = SemanticWorker('ep', '', {'provider': 'hash-pre-v1',
                                       'dimension': 64})
         view = w.embedding_view()
         assert view['enabled'] is True and view['ready'] is False

package/python/worker_v1.py

@@ -24,9 +24,9 @@ import sys
 import tempfile
 import time
 
-PROTOCOL = 'm7_wire_v1'
-NAMESPACE = 'dsh-auto-memory'
-INDEX_POLICY = 'index_sync_v1'
+PROTOCOL = 'm7_wire_pre_v1'
+NAMESPACE = 'dsh-auto-memory-pre'
+INDEX_POLICY = 'index_sync_pre_v1'
 MAX_LINE_BYTES = 256 * 1024
 MAX_RECORDS_PER_PAGE = 64
 # ★2026-09-17 修「索引永久不就绪 → 语义唤回全程降级」：JS 侧契约是「新 miv latest-wins；
@@ -47,9 +47,9 @@ PY_TYPES = frozenset(['health_result', 'context_ack', 'index_ack', 'activation_r
 
 RE_MEMORY_ID = re.compile(r'^mem_[0-9a-f]{32}$')
 RE_HEX64 = re.compile(r'^[0-9a-f]{64}$')
-RE_IDX = re.compile(r'^idx_[0-9a-f]{32}$')
+RE_IDX = re.compile(r'^idx_pre_[0-9a-f]{32}$')
 RE_WSR = re.compile(r'^wsr_[0-9a-f]{32}$')
-RE_OBS = re.compile(r'^obs_')
+RE_OBS = re.compile(r'^obs_pre_')
 RE_SOURCE_REF = re.compile(r'^(user|workspace|workspace-log):[A-Za-z0-9._\u4e00-\u9fff-]+$')
 
 _stderr_used = {'n': 0}
@@ -84,7 +84,7 @@ def dumps(v):
 
 
 def canonical(v):
-    """Byte-identical twin of lib/m7-wire.js canonicalJson()."""
+    """Byte-identical twin of lib/m7-wire-pre.js canonicalJson()."""
     if v is None:
         return 'null'
     if isinstance(v, bool):
@@ -105,7 +105,7 @@ class ProtocolError(Exception):
         self.detail = detail
 
 
-# ---------- payload validators (mirror of lib/m7-wire.js) ----------
+# ---------- payload validators (mirror of lib/m7-wire-pre.js) ----------
 
 
 def validate_semantic_record(rec):
@@ -126,7 +126,7 @@ def validate_semantic_record(rec):
     need(RE_HEX64.match(str(rec.get('recordDigest', ''))), 'recordDigest')
     need(rec.get('heading') is None or isinstance(rec.get('heading'), str), 'heading')
     need(isinstance(rec.get('text'), str), 'text')
-    need(isinstance(rec.get('chunkId'), str) and str(rec.get('chunkId')).startswith('chk_'), 'chunkId')
+    need(isinstance(rec.get('chunkId'), str) and str(rec.get('chunkId')).startswith('chk_pre_'), 'chunkId')
     ordinal = rec.get('chunkOrdinal')
     count = rec.get('chunkCount')
     need(isinstance(ordinal, int) and not isinstance(ordinal, bool) and ordinal >= 0, 'chunkOrdinal')
@@ -219,7 +219,7 @@ class Worker:
 
     def maybe_activation(self, req, p):
         """Deterministic fake ActivationRequestPre copied from JS-owned provenance.
-        Emits only when the frame carries valid identity + idx_ index version;
+        Emits only when the frame carries valid identity + idx_pre_ index version;
         otherwise stays silent (fail closed). Field-compatible with M6 validator."""
         miv = str((p.get('index') or {}).get('memoryIndexVersion', ''))
         if not RE_IDX.match(miv):
@@ -257,9 +257,9 @@ class Worker:
                 continue
             if not RE_HEX64.match(fdig) or not RE_HEX64.match(rdig):
                 continue
-            activation_id = 'act_' + first32(sha_str('m7-fake-activation-v1\u0000' + obs))
+            activation_id = 'act_pre_' + first32(sha_str('m7-fake-activation-pre-v1\u0000' + obs))
             cand = {
-                'candidateId': 'cand_' + first32(sha_str(activation_id + '\u0000' + mid + '\u0000' + str(i))),
+                'candidateId': 'cand_pre_' + first32(sha_str(activation_id + '\u0000' + mid + '\u0000' + str(i))),
                 'memoryId': mid,
                 'anchorId': aid,
                 'scope': rscope,
@@ -273,7 +273,7 @@ class Worker:
             candidates.append(cand)
         if not candidates:
             return None
-        activation_id = 'act_' + first32(sha_str('m7-fake-activation-v1\u0000' + obs))
+        activation_id = 'act_pre_' + first32(sha_str('m7-fake-activation-pre-v1\u0000' + obs))
         created = req.get('sentAt', 0)
         ttl_steps = 2
         activation = {
@@ -290,7 +290,7 @@ class Worker:
             'contextVersion': cv,
             'memoryIndexVersion': miv,
             'threshold': {
-                'policyVersion': 'm7_fake_threshold_v1',
+                'policyVersion': 'm7_fake_threshold_pre_v1',
                 'score': 0.92,
                 'threshold': 0.8,
                 'reason': 'deterministic fake activation (python worker)',
@@ -326,7 +326,7 @@ class Worker:
         require(isinstance(p, dict), 'invalid-payload', 'not-object')
         require(p.get('schemaVersion') == 1, 'invalid-payload', 'schemaVersion')
         sync_id = str(p.get('syncId', ''))
-        require(sync_id.startswith('syn_'), 'invalid-payload', 'syncId')
+        require(sync_id.startswith('syn_pre_'), 'invalid-payload', 'syncId')
         ws_ref = str(p.get('workspaceRef', ''))
         require(RE_WSR.match(ws_ref), 'invalid-payload', 'workspaceRef')
         scope = p.get('scope')
@@ -469,7 +469,7 @@ class Worker:
         for i in range(st['pageCount']):
             flat.extend(st['pages'][i])
         expected_final = sha_hex(canonical({
-            'kind': 'index_sync_final_v1',
+            'kind': 'index_sync_final_pre_v1',
             'syncId': st['syncId'],
             'memoryIndexVersion': st['memoryIndexVersion'],
             'workspaceRef': st['workspaceRef'],
@@ -511,7 +511,7 @@ class Worker:
                 'records': e['records'],
             })
         payload = {'schemaVersion': 1, 'namespace': NAMESPACE,
-                   'policyVersion': 'semantic_derived_v1', 'entries': entries}
+                   'policyVersion': 'semantic_derived_pre_v1', 'entries': entries}
         data = (dumps(payload) + '\n').encode('utf-8')
         try:
             os.makedirs(dir_path, exist_ok=True)
@@ -584,10 +584,10 @@ def run_selftest():
     assert frames[0]['frameId'] == 'res_' + first32(sha_str('r1:health_result'))
     checks += 1
     push = {'requestId': 'r2', 'workerEpoch': 'ep', 'sentAt': 20, 'frameId': 'f2', 'type': 'context_push',
-            'payload': {'kind': 'context_push', 'observationId': 'obs_' + '0' * 32,
+            'payload': {'kind': 'context_push', 'observationId': 'obs_pre_' + '0' * 32,
                         'session': {'sessionId': 's', 'agentId': 'a', 'workspaceKey': 'w', 'scope': 'Workspace'},
                         'cursor': {'eventSeq': 1, 'contextVersion': 3},
-                        'index': {'memoryIndexVersion': 'idx_' + 'ab' * 16}, 'memoryRefs': []}}
+                        'index': {'memoryIndexVersion': 'idx_pre_' + 'ab' * 16}, 'memoryRefs': []}}
     f2 = w.handle_frame(push)
     assert len(f2) == 1 and f2[0]['payload']['accepted'] is True
     again = w.handle_frame(push)