feat(profile): /api/profile + eligibility filter + inference framework (ADR-0014 steps 4-6)

Step 4 — /api/profile read-through API:
  GET  /api/profile          → { user, prefs, consents, contexts }
  PATCH /api/profile/prefs/:scope  upsert user_preferences (source='user')
  PATCH /api/profile/consents      grant / revoke consent keys
  PATCH /api/profile/contexts      create / activate / deactivate contexts
  Legacy consentGiven bit folded in as data:core fallback.

Step 5 — registry-driven eligibility filter:
  fetchRegistry() exported from agent-registry.ts.
  profile/eligibility.ts: getEligibleAgentIds(userId) — filters by required
  consents, silenced_in_contexts, and user_preferences[enabled=false].
  fetchOrchestratorTip filters agent_outputs to eligible set before calling
  ml/serving /recommend. Fail-closed: registry unavailable → empty set.

Step 6 — shared context-inference framework (#111) + time-of-day proof (#112):
  ml/agents/inference/: UserHistory, FeedbackEvent, run_inference().
  Framework: cold-start, min_history gating, error fallback, structured logs.
  TimeOfDayAgent v1.1.0: inferred_params=[preferred_hour]; also reads
  quiet_start/quiet_end from agent_prefs. agent_prefs injected by TS caller.
  AgentInput gains agent_prefs field.
  ml/serving: POST /agents/{agent_id}/infer endpoint.
  agent-outputs.ts computeAndStore: loads prefs before compute, calls /infer
  after, persists results (source='inferred'); user overrides never touched.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

ml/agents/base.py

@@ -15,6 +15,11 @@ class AgentInput:
     profile: dict[str, float | None]           # profile feature values keyed by feature name
     feedback_history: list[dict] = field(default_factory=list)  # [{action, dwell_ms, created_at}, …]
     now: datetime = field(default_factory=lambda: datetime.now(timezone.utc))
+    # Per-agent inferred/user prefs loaded from user_preferences (ADR-0014 §3).
+    # Keys match the agent's pref_schema + inferred_params. 'user' source takes
+    # precedence over 'inferred' source; the caller resolves priority before
+    # passing this dict in.
+    agent_prefs: dict = field(default_factory=dict)
 
 
 @dataclass

ml/agents/inference/__init__.py

@@ -0,0 +1,9 @@
+"""Shared context-inference framework (ADR-0014 §3, issue #111).
+
+Each agent's manifest declares InferredParams; this package owns the
+scheduling contract, history data model, and write path to user_preferences.
+"""
+from .framework import run_inference
+from .history import FeedbackEvent, UserHistory
+
+__all__ = ["run_inference", "FeedbackEvent", "UserHistory"]

ml/agents/inference/framework.py

@@ -0,0 +1,59 @@
+"""run_inference — core of the context-inference framework (ADR-0014 §3).
+
+Contract:
+    run_inference(manifest, history) → dict[key, value]
+
+Semantics:
+  - For each InferredParam in manifest.inferred_params:
+      - If len(history.events) < param.min_history → emit cold_start_default.
+      - Otherwise → call param.infer(history) and emit the result.
+  - Returns {key: value} ready for the caller to persist to user_preferences
+    with source='inferred'.
+  - User overrides (source='user') are handled by the caller's upsert logic;
+    this function has no DB access.
+"""
+from __future__ import annotations
+
+import logging
+import time
+from typing import Any
+
+from ..manifest import AgentManifest
+from .history import UserHistory
+
+log = logging.getLogger(__name__)
+
+
+def run_inference(manifest: AgentManifest, history: UserHistory) -> dict[str, Any]:
+    """Evaluate all InferredParams for an agent and return {key: inferred_value}."""
+    result: dict[str, Any] = {}
+    n = len(history.events)
+
+    for param in manifest.inferred_params:
+        t0 = time.monotonic()
+        if param.infer is None:
+            result[param.key] = param.cold_start_default
+            continue
+        if n < param.min_history:
+            value = param.cold_start_default
+            source = "cold_start"
+        else:
+            try:
+                value = param.infer(history)
+                source = "inferred"
+            except Exception as exc:
+                log.warning(
+                    "inference_error agent=%s param=%s error=%s — using cold_start_default",
+                    manifest.id, param.key, exc,
+                )
+                value = param.cold_start_default
+                source = "error_fallback"
+
+        latency_ms = round((time.monotonic() - t0) * 1000, 1)
+        log.info(
+            "inference_param agent=%s param=%s source=%s value=%r history_len=%d latency_ms=%s",
+            manifest.id, param.key, source, value, n, latency_ms,
+        )
+        result[param.key] = value
+
+    return result

ml/agents/inference/history.py

@@ -0,0 +1,29 @@
+"""UserHistory — normalised view of a user's feedback events for inference."""
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from datetime import datetime, timezone
+
+
+@dataclass
+class FeedbackEvent:
+    action: str              # 'done' | 'dismiss' | 'snooze' | 'helpful' | 'not_helpful'
+    dwell_ms: int | None
+    created_at: str          # ISO 8601
+
+    @property
+    def hour(self) -> int:
+        """Hour of day (0-23) when the feedback was recorded."""
+        try:
+            dt = datetime.fromisoformat(self.created_at.replace("Z", "+00:00"))
+        except ValueError:
+            return 12
+        if dt.tzinfo is None:
+            dt = dt.replace(tzinfo=timezone.utc)
+        return dt.hour
+
+
+@dataclass
+class UserHistory:
+    user_id: str
+    events: list[FeedbackEvent] = field(default_factory=list)

ml/agents/tests/test_agents.py

@@ -153,7 +153,7 @@ class TestTimeOfDayAgent:
 
     def test_snapshot_keys(self):
         out = self.agent.compute(_inp())
-        assert {"hour", "day_of_week", "preferred_hour"} == set(out.signals_snapshot)
+        assert {"hour", "day_of_week", "preferred_hour", "quiet_start", "quiet_end"} == set(out.signals_snapshot)
 
 
 # ── RecentPatternsAgent ───────────────────────────────────────────────────────

ml/agents/tests/test_inference.py

@@ -0,0 +1,120 @@
+"""Tests for the inference framework and time-of-day #112 proof."""
+from __future__ import annotations
+
+import sys, os
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "..", ".."))
+
+import pytest
+from datetime import datetime, timezone
+
+from ml.agents.inference.history import FeedbackEvent, UserHistory
+from ml.agents.inference.framework import run_inference
+from ml.agents.time_of_day import TimeOfDayAgent, MANIFEST as TOD_MANIFEST, MANIFEST
+from ml.agents.base import AgentInput
+
+
+_NOW = datetime(2026, 5, 1, 14, 0, 0, tzinfo=timezone.utc)  # Thursday 14:00
+
+
+def _inp(**kwargs) -> AgentInput:
+    defaults = dict(user_id="u1", tasks=[], profile={}, now=_NOW, agent_prefs={})
+    defaults.update(kwargs)
+    return AgentInput(**defaults)
+
+
+def _event(action: str, hour: int) -> FeedbackEvent:
+    ts = f"2026-05-01T{hour:02d}:00:00+00:00"
+    return FeedbackEvent(action=action, dwell_ms=60_000 if action == "done" else 500, created_at=ts)
+
+
+class TestRunInference:
+    def test_cold_start_when_below_min_history(self):
+        history = UserHistory(user_id="u1", events=[_event("done", 9)] * 5)  # only 5 < 10
+        result = run_inference(TOD_MANIFEST, history)
+        assert result["preferred_hour"] is None  # cold_start_default
+
+    def test_infers_preferred_hour_as_mode(self):
+        # 7 events at 09:00, 3 at 17:00 → preferred_hour should be 9
+        events = [_event("done", 9)] * 7 + [_event("done", 17)] * 3
+        history = UserHistory(user_id="u1", events=events)
+        result = run_inference(TOD_MANIFEST, history)
+        assert result["preferred_hour"] == 9
+
+    def test_infers_preferred_hour_from_majority_hour(self):
+        events = [_event("done", 20)] * 6 + [_event("done", 8)] * 4
+        history = UserHistory(user_id="u1", events=events)
+        result = run_inference(TOD_MANIFEST, history)
+        assert result["preferred_hour"] == 20
+
+    def test_no_inferred_params_returns_empty(self):
+        from ml.agents.manifest import AgentManifest
+        bare = AgentManifest(
+            id="bare", version="1.0.0", description="", pref_schema={},
+            context_schema=[], required_consents=[], output_contract={}, ttl_sec=300,
+        )
+        history = UserHistory(user_id="u1", events=[_event("done", 9)] * 20)
+        result = run_inference(bare, history)
+        assert result == {}
+
+    def test_cold_start_fallback_on_infer_error(self):
+        """infer() raising should fall back to cold_start_default, not crash."""
+        from ml.agents.manifest import InferredParam, AgentManifest
+
+        def _bad_infer(h):
+            raise RuntimeError("oops")
+
+        m = AgentManifest(
+            id="boom", version="1.0.0", description="", pref_schema={},
+            context_schema=[], required_consents=[], output_contract={}, ttl_sec=300,
+            inferred_params=[InferredParam(key="x", ttl_sec=60, cold_start_default=42, min_history=1, infer=_bad_infer)],
+        )
+        history = UserHistory(user_id="u1", events=[_event("done", 9)] * 5)
+        result = run_inference(m, history)
+        assert result["x"] == 42
+
+
+class TestTimeOfDayAgentWithInference:
+    agent = TimeOfDayAgent()
+
+    def test_uses_preferred_hour_from_agent_prefs(self):
+        inp = _inp(agent_prefs={"preferred_hour": 9}, now=datetime(2026, 5, 1, 9, 0, 0, tzinfo=timezone.utc))
+        out = self.agent.compute(inp)
+        assert "peak productivity hour" in out.prompt_text.lower() or "peak" in out.prompt_text
+
+    def test_quiet_window_noon_suppressed(self):
+        inp = _inp(
+            agent_prefs={"quiet_start": "22:00", "quiet_end": "07:00"},
+            now=datetime(2026, 5, 1, 23, 0, 0, tzinfo=timezone.utc),
+        )
+        out = self.agent.compute(inp)
+        assert "quiet window" in out.prompt_text
+
+    def test_quiet_window_not_in_window(self):
+        inp = _inp(
+            agent_prefs={"quiet_start": "22:00", "quiet_end": "07:00"},
+            now=datetime(2026, 5, 1, 14, 0, 0, tzinfo=timezone.utc),
+        )
+        out = self.agent.compute(inp)
+        assert "quiet window" not in out.prompt_text
+
+    def test_agent_prefs_override_profile(self):
+        # agent_prefs.preferred_hour wins over profile.preferred_hour
+        inp = _inp(
+            profile={"preferred_hour": 8},
+            agent_prefs={"preferred_hour": 14},
+            now=datetime(2026, 5, 1, 14, 0, 0, tzinfo=timezone.utc),
+        )
+        out = self.agent.compute(inp)
+        assert "peak productivity hour (14:00)" in out.prompt_text
+
+    def test_no_prefs_falls_back_to_profile(self):
+        inp = _inp(profile={"preferred_hour": 10}, now=datetime(2026, 5, 1, 10, 0, 0, tzinfo=timezone.utc))
+        out = self.agent.compute(inp)
+        assert "peak" in out.prompt_text
+
+    def test_version_bumped(self):
+        assert MANIFEST.version == "1.1.0"
+
+    def test_manifest_has_preferred_hour_param(self):
+        keys = {p.key for p in MANIFEST.inferred_params}
+        assert "preferred_hour" in keys

ml/agents/time_of_day.py

@@ -1,14 +1,26 @@
 from __future__ import annotations
+
+from collections import Counter
 from typing import ClassVar
+
 from .base import BaseAgent, AgentInput, AgentOutput
-from .manifest import AgentManifest
+from .inference.history import UserHistory
+from .manifest import AgentManifest, InferredParam
 
 _DOW_NAMES = ["Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday", "Sunday"]
 
 
+def _infer_preferred_hour(history: UserHistory) -> int:
+    """Mode hour of day across all 'done' feedback events; falls back to 9."""
+    done_hours = [e.hour for e in history.events if e.action == "done"]
+    if not done_hours:
+        return 9
+    return Counter(done_hours).most_common(1)[0][0]
+
+
 MANIFEST = AgentManifest(
     id="time-of-day",
-    version="1.0.0",
+    version="1.1.0",  # bumped: inferred_params added (ADR-0014 §3, #112)
     description="Frames the current moment relative to the user's productive peak and quiet hours.",
     pref_schema={
         "type": "object",
@@ -30,6 +42,15 @@ MANIFEST = AgentManifest(
     required_consents=["data:core", "agent:time-of-day"],
     output_contract={"type": "snippet", "format": "free_text"},
     ttl_sec=900,
+    inferred_params=[
+        InferredParam(
+            key="preferred_hour",
+            ttl_sec=3_600,       # recompute hourly
+            cold_start_default=None,
+            min_history=10,      # need at least 10 feedback events to be meaningful
+            infer=_infer_preferred_hour,
+        ),
+    ],
 )
 
 
@@ -42,31 +63,63 @@ class TimeOfDayAgent(BaseAgent):
     def compute(self, inp: AgentInput) -> AgentOutput:
         hour = inp.now.hour
         dow = inp.now.weekday()  # 0=Monday … 6=Sunday
-        preferred = inp.profile.get("preferred_hour")
         is_weekend = dow >= 5
 
+        # agent_prefs (inferred or user-set) take precedence over ML profile features.
+        preferred_raw = inp.agent_prefs.get("preferred_hour", inp.profile.get("preferred_hour"))
+        preferred = int(preferred_raw) if preferred_raw is not None else None
+
+        quiet_start: str | None = inp.agent_prefs.get("quiet_start")
+        quiet_end: str | None = inp.agent_prefs.get("quiet_end")
+        in_quiet = self._in_quiet_window(hour, quiet_start, quiet_end)
+
         parts = [f"It is {hour:02d}:00 on {_DOW_NAMES[dow]} ({self._label(hour)})."]
 
         if is_weekend:
             parts.append("Weekend context — prefer personal or reflective tips over work tasks.")
 
+        if in_quiet:
+            parts.append(
+                f"User is in their quiet window ({quiet_start}–{quiet_end}) — "
+                "avoid urgent or demanding tips."
+            )
+
         if preferred is not None:
-            ph = int(preferred)
-            delta = min(abs(hour - ph), 24 - abs(hour - ph))  # circular distance
+            delta = min(abs(hour - preferred), 24 - abs(hour - preferred))
             if delta == 0:
                 parts.append(
-                    f"This is the user's peak productivity hour ({ph:02d}:00) — "
-                    f"a high-impact tip is appropriate."
+                    f"This is the user's peak productivity hour ({preferred:02d}:00) — "
+                    "a high-impact tip is appropriate."
                 )
             elif delta <= 2:
-                parts.append(f"Approaching the user's peak productivity window ({ph:02d}:00).")
+                parts.append(f"Approaching the user's peak productivity window ({preferred:02d}:00).")
         else:
             parts.append("No preferred-hour data yet.")
 
         prompt = " ".join(parts)
-        snapshot = {"hour": hour, "day_of_week": dow, "preferred_hour": preferred}
+        snapshot = {
+            "hour": hour,
+            "day_of_week": dow,
+            "preferred_hour": preferred,
+            "quiet_start": quiet_start,
+            "quiet_end": quiet_end,
+        }
         return self._make_output(inp, prompt, snapshot)
 
+    @staticmethod
+    def _in_quiet_window(hour: int, start: str | None, end: str | None) -> bool:
+        if not start or not end:
+            return False
+        try:
+            sh = int(start.split(":")[0])
+            eh = int(end.split(":")[0])
+        except (ValueError, IndexError):
+            return False
+        if sh <= eh:
+            return sh <= hour < eh
+        # wraps midnight e.g. 22:00–07:00
+        return hour >= sh or hour < eh
+
     @staticmethod
     def _label(hour: int) -> str:
         if 5 <= hour < 12: