@pentatonic-ai/ai-agent-sdk 0.10.32 → 0.10.33

package/dist/index.cjs

@@ -878,7 +878,7 @@ function fireAndForgetEmit(clientConfig, sessionOpts, messages, result, model) {
 }
 
 // src/telemetry.js
-var VERSION = "0.10.32";
+var VERSION = "0.10.33";
 var TELEMETRY_URL = "https://sdk-telemetry.philip-134.workers.dev";
 function machineId() {
   const raw = typeof process !== "undefined" ? `${process.env?.USER || process.env?.USERNAME || "u"}:${process.platform || "x"}:${process.arch || "x"}` : "browser";

package/dist/index.js

@@ -847,7 +847,7 @@ function fireAndForgetEmit(clientConfig, sessionOpts, messages, result, model) {
 }
 
 // src/telemetry.js
-var VERSION = "0.10.32";
+var VERSION = "0.10.33";
 var TELEMETRY_URL = "https://sdk-telemetry.philip-134.workers.dev";
 function machineId() {
   const raw = typeof process !== "undefined" ? `${process.env?.USER || process.env?.USERNAME || "u"}:${process.platform || "x"}:${process.arch || "x"}` : "browser";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pentatonic-ai/ai-agent-sdk",
-  "version": "0.10.32",
+  "version": "0.10.33",
   "description": "TES SDK — LLM observability and lifecycle tracking via Pentatonic Thing Event System. Track token usage, tool calls, and conversations. Manage things through event-sourced lifecycle stages with AI enrichment and vector search.",
   "type": "module",
   "main": "./dist/index.cjs",

package/packages/memory-engine-v2/extractor-async/worker.py

@@ -193,13 +193,24 @@ _PROV_REL = _cap_prov("relationships.provenance_event_ids", "EXCLUDED.provenance
 # lever, this is the monitoring + active-learning sample (drift detection and
 # fresh teacher-gold on live traffic). Default conservative-ish 5%.
 STUDENT_SAMPLE_RATE = float(os.environ.get("STUDENT_SAMPLE_RATE", "0.05"))
-# Fact categories that always go to the teacher regardless of the student's
-# output (high-value, cheap to over-escalate). Comma-separated, lowercased.
+# Fact categories that go to the teacher regardless of the student's output
+# (high-value, cheap to over-escalate). Comma-separated, lowercased.
 HIGH_VALUE_CATEGORIES = {
     c.strip().lower()
     for c in os.environ.get("HIGH_VALUE_CATEGORIES", "decision,commitment").split(",")
     if c.strip()
 }
+# Fraction of high-value-class events that ESCALATE to the teacher. 1.0 (default)
+# = the original always-teacher gate. The gate was set at cascade launch when the
+# student was unproven; since then the clean student has zero quality-fail
+# escalations in prod and shadow-eval parity with the teacher — while the HV gate
+# routed ~57% of a decision/commitment-heavy backlog to the teacher (90% of all
+# escalations, 2026-07-09), pinning drain to teacher capacity with the student
+# idle. Lowering this (e.g. 0.25) makes HV student-primary with an elevated
+# teacher sample as the safety net/monitoring corpus on exactly the high-value
+# class. Instant rollback: set back to 1.0. Escalated-anyway HV events keep the
+# 'high_value_class' ledger reason, so the mix stays queryable.
+HIGH_VALUE_SAMPLE_RATE = float(os.environ.get("HIGH_VALUE_SAMPLE_RATE", "1.0"))
 
 
 # KV-text output format constants. We dropped JSON output (and the
@@ -1078,10 +1089,14 @@ def escalation_decision(
     for em in {e.lower() for e in _EMAIL_RE.findall(blob)}:
         if em not in known:
             return True, "grounding_violation"
-    # 2. High-value fact class — always teacher.
+    # 2. High-value fact class — teacher at HIGH_VALUE_SAMPLE_RATE (1.0 = the
+    # original always-teacher gate; lower = student-primary with an elevated
+    # teacher sample on exactly this class — see the config note).
     for f in student_result.get("facts", []):
         if (f.get("category") or "").lower() in HIGH_VALUE_CATEGORIES:
-            return True, "high_value_class"
+            if HIGH_VALUE_SAMPLE_RATE >= 1.0 or random.random() < HIGH_VALUE_SAMPLE_RATE:
+                return True, "high_value_class"
+            break  # HV but sampled-out → student writes; still eligible for gate 3
     # 3. Random monitoring/active-learning sample (not a quality lever).
     if STUDENT_SAMPLE_RATE > 0 and random.random() < STUDENT_SAMPLE_RATE:
         return True, "random_sample"
@@ -2964,7 +2979,8 @@ async def amain():
             f"cascade ENABLED — student-primary "
             f"(model={STUDENT_MODEL}, endpoint={STUDENT_ENDPOINT or '(unset!)'}, "
             f"sample_rate={STUDENT_SAMPLE_RATE}, "
-            f"high_value={sorted(HIGH_VALUE_CATEGORIES)})"
+            f"high_value={sorted(HIGH_VALUE_CATEGORIES)}, "
+            f"hv_sample_rate={HIGH_VALUE_SAMPLE_RATE})"
         )
         if not STUDENT_ENDPOINT:
             log.warning(