nexo-brain 5.3.13 → 5.3.15

package/src/db/_outcomes 2.py

@@ -0,0 +1,800 @@
+from __future__ import annotations
+"""NEXO DB — Outcome tracker v1."""
+
+import datetime
+import json
+from typing import Any
+
+from db._core import get_db
+
+VALID_METRIC_SOURCES = {
+    "manual",
+    "followup_status",
+    "decision_outcome",
+    "protocol_task_status",
+    "nexo_sqlite",
+}
+VALID_TARGET_OPERATORS = {"gte", "lte", "eq"}
+OUTCOME_PATTERN_MIN_RESOLVED = 3
+OUTCOME_PATTERN_MAX_EVIDENCE = 5
+OUTCOME_PATTERN_LEARNING_SUCCESS_BOOST = 0.9
+OUTCOME_PATTERN_LEARNING_RISK_REDUCTION = 0.6
+OUTCOME_PATTERN_LEARNING_SUCCESS_PENALTY = -1.1
+OUTCOME_PATTERN_LEARNING_RISK_PENALTY = 0.8
+OUTCOME_PATTERN_MIN_SUCCESS_RATE = 0.75
+OUTCOME_PATTERN_MAX_FAILURE_RATE = 0.25
+
+
+def _utcnow_iso() -> str:
+    return datetime.datetime.now().isoformat(timespec="seconds")
+
+
+def _normalize_deadline(deadline: str = "", *, default_days: int = 7) -> str:
+    clean = (deadline or "").strip()
+    if clean:
+        return clean
+    return (datetime.datetime.now() + datetime.timedelta(days=default_days)).isoformat(timespec="seconds")
+
+
+def _normalize_source(metric_source: str) -> str:
+    clean = (metric_source or "manual").strip().lower()
+    if clean not in VALID_METRIC_SOURCES:
+        raise ValueError(f"metric_source must be one of: {', '.join(sorted(VALID_METRIC_SOURCES))}")
+    return clean
+
+
+def _normalize_operator(target_operator: str) -> str:
+    clean = (target_operator or "gte").strip().lower()
+    if clean not in VALID_TARGET_OPERATORS:
+        raise ValueError(f"target_operator must be one of: {', '.join(sorted(VALID_TARGET_OPERATORS))}")
+    return clean
+
+
+def _as_float(value: Any) -> float | None:
+    if value is None or value == "":
+        return None
+    try:
+        return float(value)
+    except (TypeError, ValueError):
+        return None
+
+
+def _append_note(existing: str | None, extra: str) -> str:
+    extra = (extra or "").strip()
+    if not extra:
+        return (existing or "").strip()
+    existing = (existing or "").strip()
+    if not existing:
+        return extra
+    if extra in existing:
+        return existing
+    return f"{existing}\n{extra}"
+
+
+def _format_scalar(value: Any) -> str:
+    if value is None:
+        return ""
+    if isinstance(value, float):
+        return f"{value:.4f}".rstrip("0").rstrip(".")
+    return str(value)
+
+
+def _pattern_key(*, area: str, task_type: str, goal_profile_id: str, selected_choice: str) -> str:
+    return json.dumps(
+        {
+            "area": (area or "").strip(),
+            "task_type": (task_type or "").strip(),
+            "goal_profile_id": (goal_profile_id or "").strip(),
+            "selected_choice": (selected_choice or "").strip(),
+        },
+        ensure_ascii=False,
+        sort_keys=True,
+    )
+
+
+def _context_label(*, area: str, task_type: str, goal_profile_id: str) -> str:
+    bits = []
+    if (area or "").strip():
+        bits.append(f"area={area.strip()}")
+    if (task_type or "").strip():
+        bits.append(f"task_type={task_type.strip()}")
+    if (goal_profile_id or "").strip():
+        bits.append(f"profile={goal_profile_id.strip()}")
+    return ", ".join(bits) if bits else "contexto general"
+
+
+def _compare(actual_value: float, target_value: float, operator: str) -> bool:
+    if operator == "gte":
+        return actual_value >= target_value
+    if operator == "lte":
+        return actual_value <= target_value
+    return actual_value == target_value
+
+
+def _get_outcome_row(conn, outcome_id: int):
+    return conn.execute("SELECT * FROM outcomes WHERE id = ?", (int(outcome_id),)).fetchone()
+
+
+def _update_outcome(
+    outcome_id: int,
+    *,
+    status: str | None = None,
+    actual_value: float | None = None,
+    actual_value_text: str | None = None,
+    checked_at: str | None = None,
+    notes: str | None = None,
+    learning_id: int | None = None,
+) -> dict:
+    conn = get_db()
+    row = _get_outcome_row(conn, outcome_id)
+    if not row:
+        return {"error": f"Outcome {outcome_id} not found"}
+
+    updates: list[str] = []
+    params: list[Any] = []
+
+    if status is not None:
+        updates.append("status = ?")
+        params.append(status)
+    if actual_value is not None:
+        updates.append("actual_value = ?")
+        params.append(actual_value)
+    if actual_value_text is not None:
+        updates.append("actual_value_text = ?")
+        params.append(actual_value_text.strip())
+    if checked_at is not None:
+        updates.append("checked_at = ?")
+        params.append(checked_at)
+    if notes is not None:
+        updates.append("notes = ?")
+        params.append(notes)
+    if learning_id is not None:
+        updates.append("learning_id = ?")
+        params.append(learning_id)
+
+    updates.append("updated_at = datetime('now')")
+    params.append(int(outcome_id))
+    conn.execute(f"UPDATE outcomes SET {', '.join(updates)} WHERE id = ?", params)
+    conn.commit()
+    row = _get_outcome_row(conn, outcome_id)
+    return dict(row) if row else {"error": f"Outcome {outcome_id} not found after update"}
+
+
+def create_outcome(
+    action_type: str,
+    description: str,
+    expected_result: str,
+    *,
+    metric_source: str = "manual",
+    metric_query: str = "",
+    baseline_value: float | None = None,
+    target_value: float | None = None,
+    target_operator: str = "gte",
+    deadline: str = "",
+    action_id: str = "",
+    session_id: str = "",
+    notes: str = "",
+) -> dict:
+    conn = get_db()
+    clean_action_type = (action_type or "").strip()
+    clean_description = (description or "").strip()
+    clean_expected = (expected_result or "").strip()
+    if not clean_action_type:
+        return {"error": "action_type is required"}
+    if not clean_description:
+        return {"error": "description is required"}
+    if not clean_expected:
+        return {"error": "expected_result is required"}
+
+    try:
+        clean_source = _normalize_source(metric_source)
+        clean_operator = _normalize_operator(target_operator)
+    except ValueError as exc:
+        return {"error": str(exc)}
+
+    clean_query = (metric_query or "").strip()
+    if clean_source in {"followup_status", "decision_outcome", "protocol_task_status"} and not (action_id or "").strip():
+        return {"error": f"action_id is required for metric_source='{clean_source}'"}
+    if clean_source == "nexo_sqlite":
+        if not clean_query:
+            return {"error": "metric_query is required for metric_source='nexo_sqlite'"}
+        query_upper = clean_query.upper()
+        if ";" in clean_query.rstrip(";"):
+            return {"error": "metric_query must be a single SELECT statement"}
+        if not query_upper.startswith("SELECT "):
+            return {"error": "metric_query must start with SELECT"}
+        for forbidden in ("INSERT ", "UPDATE ", "DELETE ", "DROP ", "ALTER ", "ATTACH ", "DETACH ", "PRAGMA "):
+            if forbidden in query_upper:
+                return {"error": "metric_query must be read-only"}
+
+    cursor = conn.execute(
+        """INSERT INTO outcomes (
+               action_type, action_id, session_id, description, expected_result,
+               metric_source, metric_query, baseline_value, target_value,
+               target_operator, deadline, notes
+           ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)""",
+        (
+            clean_action_type,
+            (action_id or "").strip(),
+            (session_id or "").strip(),
+            clean_description,
+            clean_expected,
+            clean_source,
+            clean_query,
+            baseline_value,
+            target_value,
+            clean_operator,
+            _normalize_deadline(deadline),
+            (notes or "").strip(),
+        ),
+    )
+    conn.commit()
+    row = _get_outcome_row(conn, cursor.lastrowid)
+    return dict(row) if row else {"error": "Outcome insert failed"}
+
+
+def get_outcome(outcome_id: int) -> dict | None:
+    conn = get_db()
+    row = _get_outcome_row(conn, outcome_id)
+    return dict(row) if row else None
+
+
+def list_outcomes(status: str = "", action_type: str = "", limit: int = 50) -> list[dict]:
+    conn = get_db()
+    clauses = []
+    params: list[Any] = []
+    if status:
+        clauses.append("status = ?")
+        params.append((status or "").strip().lower())
+    if action_type:
+        clauses.append("action_type = ?")
+        params.append((action_type or "").strip())
+    where = f"WHERE {' AND '.join(clauses)}" if clauses else ""
+    rows = conn.execute(
+        f"""SELECT * FROM outcomes
+            {where}
+            ORDER BY
+              CASE status
+                WHEN 'pending' THEN 0
+                WHEN 'missed' THEN 1
+                WHEN 'met' THEN 2
+                ELSE 3
+              END,
+              deadline ASC,
+              created_at DESC
+            LIMIT ?""",
+        params + [max(1, int(limit))],
+    ).fetchall()
+    return [dict(row) for row in rows]
+
+
+def pending_outcomes_due(deadline_before: str | None = None, limit: int = 100) -> list[dict]:
+    conn = get_db()
+    cutoff = deadline_before or _utcnow_iso()
+    rows = conn.execute(
+        """SELECT * FROM outcomes
+           WHERE status = 'pending' AND deadline <= ?
+           ORDER BY deadline ASC, id ASC
+           LIMIT ?""",
+        (cutoff, max(1, int(limit))),
+    ).fetchall()
+    return [dict(row) for row in rows]
+
+
+def find_pending_outcomes_by_action(action_type: str, action_id: str, *, metric_source: str = "") -> list[dict]:
+    conn = get_db()
+    clauses = [
+        "status = 'pending'",
+        "action_type = ?",
+        "action_id = ?",
+    ]
+    params: list[Any] = [(action_type or "").strip(), (action_id or "").strip()]
+    if metric_source:
+        clauses.append("metric_source = ?")
+        params.append((metric_source or "").strip().lower())
+    rows = conn.execute(
+        f"SELECT * FROM outcomes WHERE {' AND '.join(clauses)} ORDER BY deadline ASC, id ASC",
+        params,
+    ).fetchall()
+    return [dict(row) for row in rows]
+
+
+def cancel_outcome(outcome_id: int, reason: str = "") -> dict:
+    conn = get_db()
+    row = _get_outcome_row(conn, outcome_id)
+    if not row:
+        return {"error": f"Outcome {outcome_id} not found"}
+    notes = _append_note(row["notes"], f"Cancelled: {(reason or '').strip() or 'no reason provided'}")
+    return _update_outcome(
+        int(outcome_id),
+        status="cancelled",
+        checked_at=_utcnow_iso(),
+        notes=notes,
+    )
+
+
+def _create_miss_learning(row: dict, actual_value: float | None, actual_value_text: str, note: str) -> int | None:
+    if row.get("learning_id"):
+        return int(row["learning_id"])
+    from db._learnings import create_learning
+
+    summary_bits = [
+        f"Outcome #{row['id']} missed.",
+        f"Action type: {row.get('action_type', '')}.",
+        f"Action id: {row.get('action_id', '') or 'N/A'}.",
+        f"Description: {row.get('description', '')}.",
+        f"Expected: {row.get('expected_result', '')}.",
+    ]
+    if actual_value is not None:
+        summary_bits.append(f"Actual numeric value: {actual_value}.")
+    if actual_value_text:
+        summary_bits.append(f"Actual evidence: {actual_value_text}.")
+    if note:
+        summary_bits.append(f"Why missed: {note}.")
+    learning = create_learning(
+        category="outcomes",
+        title=f"Outcome missed: {str(row.get('description', ''))[:80]}",
+        content=" ".join(summary_bits),
+        reasoning=f"Auto-created from missed outcome #{row['id']}.",
+        prevention="Review the action, expected result, target, or deadline before repeating the same move.",
+        applies_to=f"outcome:{row['id']}",
+    )
+    return int(learning["id"]) if learning and learning.get("id") else None
+
+
+def _status_from_protocol_task(task_row: dict | None, *, deadline_passed: bool) -> tuple[str, float | None, str, str]:
+    if not task_row:
+        if deadline_passed:
+            return "missed", None, "protocol task missing", "Linked protocol task not found."
+        return "pending", None, "", ""
+    status = str(task_row.get("status") or "").strip().lower()
+    if status == "done":
+        evidence = (task_row.get("outcome_notes") or task_row.get("close_evidence") or task_row.get("goal") or "").strip()
+        return "met", 1.0, evidence, "Linked protocol task closed as done."
+    if status in {"failed", "cancelled"}:
+        evidence = (task_row.get("outcome_notes") or task_row.get("close_evidence") or status).strip()
+        return "missed", 0.0, evidence, f"Linked protocol task closed as {status}."
+    if deadline_passed:
+        return "missed", None, status or "open", f"Deadline passed while linked protocol task remained {status or 'open'}."
+    return "pending", None, status, ""
+
+
+def evaluate_outcome(
+    outcome_id: int,
+    *,
+    actual_value: float | None = None,
+    actual_value_text: str = "",
+    create_learning_on_miss: bool = True,
+) -> dict:
+    conn = get_db()
+    row = _get_outcome_row(conn, outcome_id)
+    if not row:
+        return {"error": f"Outcome {outcome_id} not found"}
+    row_d = dict(row)
+    if row_d.get("status") != "pending":
+        row_d["evaluation"] = "skipped_non_pending"
+        return row_d
+
+    now_iso = _utcnow_iso()
+    deadline_passed = str(row_d.get("deadline") or "") <= now_iso
+    source = (row_d.get("metric_source") or "manual").strip().lower()
+    target = _as_float(row_d.get("target_value"))
+    operator = (row_d.get("target_operator") or "gte").strip().lower()
+
+    status = "pending"
+    actual_num = _as_float(actual_value)
+    actual_text = (actual_value_text or "").strip()
+    note = ""
+
+    if source == "manual":
+        if actual_num is not None:
+            actual_text = actual_text or _format_scalar(actual_num)
+            if target is None:
+                status = "met"
+                note = "Manual evidence recorded."
+            elif _compare(actual_num, target, operator):
+                status = "met"
+                note = f"Manual check met target via operator '{operator}'."
+            elif deadline_passed:
+                status = "missed"
+                note = f"Manual check below target and deadline passed (actual={actual_num}, target={target}, op={operator})."
+            else:
+                note = f"Manual check recorded current value (actual={actual_num}, target={target}, op={operator})."
+        elif actual_text:
+            status = "met"
+            note = "Manual textual evidence recorded."
+        elif deadline_passed:
+            status = "missed"
+            note = "Deadline passed with no manual evidence recorded."
+
+    elif source == "followup_status":
+        followup = conn.execute(
+            "SELECT status, description, verification FROM followups WHERE id = ?",
+            (row_d.get("action_id", ""),),
+        ).fetchone()
+        followup_status = str(followup["status"]) if followup else ""
+        if followup and followup_status.upper().startswith("COMPLETED"):
+            status = "met"
+            actual_num = 1.0
+            actual_text = (followup["verification"] or followup["description"] or followup_status or "").strip()
+            note = "Linked followup completed."
+        elif deadline_passed:
+            status = "missed"
+            actual_text = actual_text or (followup_status or "missing")
+            note = f"Deadline passed while linked followup remained {followup_status or 'missing'}."
+
+    elif source == "decision_outcome":
+        decision = conn.execute(
+            "SELECT outcome, status, decision FROM decisions WHERE id = ?",
+            (row_d.get("action_id", ""),),
+        ).fetchone()
+        decision_outcome = (decision["outcome"] or "").strip() if decision else ""
+        if decision and decision_outcome:
+            status = "met"
+            actual_num = 1.0
+            actual_text = decision_outcome
+            note = "Linked decision outcome recorded."
+        elif deadline_passed:
+            status = "missed"
+            actual_text = actual_text or (str(decision["status"]) if decision else "missing")
+            note = f"Deadline passed with no linked decision outcome for decision {row_d.get('action_id', '') or 'N/A'}."
+
+    elif source == "protocol_task_status":
+        task = conn.execute(
+            "SELECT status, goal, close_evidence, outcome_notes FROM protocol_tasks WHERE task_id = ?",
+            (row_d.get("action_id", ""),),
+        ).fetchone()
+        status, actual_num, actual_text, note = _status_from_protocol_task(dict(task) if task else None, deadline_passed=deadline_passed)
+
+    elif source == "nexo_sqlite":
+        query = (row_d.get("metric_query") or "").strip()
+        if not query:
+            return {"error": f"Outcome {outcome_id} has empty metric_query"}
+        if ";" in query.rstrip(";") or not query.upper().startswith("SELECT "):
+            return {"error": "Outcome metric_query must be a single SELECT statement"}
+        fetched = conn.execute(query).fetchone()
+        scalar = fetched[0] if fetched else None
+        actual_num = _as_float(scalar)
+        actual_text = _format_scalar(scalar)
+        if actual_num is None:
+            if deadline_passed:
+                status = "missed"
+                note = f"SQLite query did not return a numeric scalar before deadline. Got: {actual_text or 'empty'}."
+            else:
+                note = f"SQLite query returned non-numeric scalar: {actual_text or 'empty'}."
+        elif target is None:
+            if actual_num:
+                status = "met"
+                note = f"SQLite query returned truthy scalar {actual_num}."
+            elif deadline_passed:
+                status = "missed"
+                note = "SQLite query returned falsy scalar and deadline passed."
+            else:
+                note = f"SQLite query current scalar is {actual_num}."
+        elif _compare(actual_num, target, operator):
+            status = "met"
+            note = f"SQLite scalar met target (actual={actual_num}, target={target}, op={operator})."
+        elif deadline_passed:
+            status = "missed"
+            note = f"SQLite scalar missed target at deadline (actual={actual_num}, target={target}, op={operator})."
+        else:
+            note = f"SQLite scalar recorded but target not reached yet (actual={actual_num}, target={target}, op={operator})."
+
+    learning_id = row_d.get("learning_id")
+    combined_notes = _append_note(row_d.get("notes"), note)
+    if status == "missed" and create_learning_on_miss:
+        learning_id = _create_miss_learning(row_d, actual_num, actual_text, note)
+
+    updated = _update_outcome(
+        int(outcome_id),
+        status=status,
+        actual_value=actual_num,
+        actual_value_text=actual_text,
+        checked_at=now_iso,
+        notes=combined_notes,
+        learning_id=int(learning_id) if learning_id else None,
+    )
+    if "error" not in updated:
+        updated["evaluation"] = status
+    return updated
+
+
+def set_linked_outcomes_met(
+    action_type: str,
+    action_id: str,
+    *,
+    metric_source: str = "",
+    actual_value: float | None = 1.0,
+    actual_value_text: str = "",
+    note: str = "",
+) -> list[dict]:
+    rows = find_pending_outcomes_by_action(action_type, action_id, metric_source=metric_source)
+    updated: list[dict] = []
+    for row in rows:
+        updated.append(
+            _update_outcome(
+                int(row["id"]),
+                status="met",
+                actual_value=actual_value,
+                actual_value_text=actual_value_text or row.get("actual_value_text", ""),
+                checked_at=_utcnow_iso(),
+                notes=_append_note(row.get("notes"), note or "Linked action reached success state."),
+            )
+        )
+    return updated
+
+
+def list_outcome_pattern_candidates(
+    *,
+    min_resolved: int = OUTCOME_PATTERN_MIN_RESOLVED,
+    min_success_rate: float = OUTCOME_PATTERN_MIN_SUCCESS_RATE,
+    max_failure_rate: float = OUTCOME_PATTERN_MAX_FAILURE_RATE,
+    limit: int = 20,
+) -> list[dict]:
+    conn = get_db()
+    if not conn.execute("SELECT 1 FROM sqlite_master WHERE type='table' AND name='cortex_evaluations'").fetchone():
+        return []
+
+    rows = conn.execute(
+        """SELECT
+               e.area,
+               e.task_type,
+               e.goal_profile_id,
+               e.selected_choice,
+               SUM(CASE WHEN o.status = 'met' THEN 1 ELSE 0 END) AS met,
+               SUM(CASE WHEN o.status = 'missed' THEN 1 ELSE 0 END) AS missed,
+               COUNT(*) AS resolved_outcomes,
+               MAX(e.created_at) AS last_seen_at
+           FROM cortex_evaluations e
+           JOIN outcomes o ON o.id = e.linked_outcome_id
+           WHERE (e.selected_choice IS NOT NULL AND trim(e.selected_choice) != '')
+             AND o.status IN ('met', 'missed')
+           GROUP BY e.area, e.task_type, e.goal_profile_id, e.selected_choice
+           HAVING COUNT(*) >= ?
+           ORDER BY resolved_outcomes DESC, last_seen_at DESC
+           LIMIT ?""",
+        (max(1, int(min_resolved)), max(1, int(limit * 3))),
+    ).fetchall()
+
+    candidates: list[dict] = []
+    for row in rows:
+        resolved = int(row["resolved_outcomes"] or 0)
+        met = int(row["met"] or 0)
+        missed = int(row["missed"] or 0)
+        if resolved <= 0:
+            continue
+        success_rate = round(met / resolved, 3)
+        candidate_type = ""
+        if success_rate >= float(min_success_rate):
+            candidate_type = "reinforce_strategy"
+        elif success_rate <= float(max_failure_rate):
+            candidate_type = "avoid_strategy"
+        if not candidate_type:
+            continue
+
+        evidence_rows = conn.execute(
+            """SELECT e.id AS evaluation_id, o.id AS outcome_id, o.status, o.description, e.created_at
+               FROM cortex_evaluations e
+               JOIN outcomes o ON o.id = e.linked_outcome_id
+               WHERE e.area = ?
+                 AND e.task_type = ?
+                 AND e.goal_profile_id = ?
+                 AND e.selected_choice = ?
+                 AND o.status IN ('met', 'missed')
+               ORDER BY e.created_at DESC, e.id DESC
+               LIMIT ?""",
+            (
+                row["area"] or "",
+                row["task_type"] or "",
+                row["goal_profile_id"] or "",
+                row["selected_choice"] or "",
+                OUTCOME_PATTERN_MAX_EVIDENCE,
+            ),
+        ).fetchall()
+        evidence = [dict(item) for item in evidence_rows]
+        context = _context_label(
+            area=row["area"] or "",
+            task_type=row["task_type"] or "",
+            goal_profile_id=row["goal_profile_id"] or "",
+        )
+        selected_choice = (row["selected_choice"] or "").strip()
+        if candidate_type == "reinforce_strategy":
+            rationale = (
+                f"La estrategia '{selected_choice}' acumula {met}/{resolved} outcomes met en {context}."
+            )
+        else:
+            rationale = (
+                f"La estrategia '{selected_choice}' acumula {missed}/{resolved} outcomes missed en {context}."
+            )
+
+        pattern_key = _pattern_key(
+            area=row["area"] or "",
+            task_type=row["task_type"] or "",
+            goal_profile_id=row["goal_profile_id"] or "",
+            selected_choice=selected_choice,
+        )
+        candidates.append(
+            {
+                "pattern_key": pattern_key,
+                "candidate_type": candidate_type,
+                "area": row["area"] or "",
+                "task_type": row["task_type"] or "",
+                "goal_profile_id": row["goal_profile_id"] or "",
+                "selected_choice": selected_choice,
+                "resolved_outcomes": resolved,
+                "met": met,
+                "missed": missed,
+                "success_rate": success_rate,
+                "last_seen_at": row["last_seen_at"],
+                "context_label": context,
+                "rationale": rationale,
+                "evidence": evidence,
+                "suggested_skill_candidate": (
+                    candidate_type == "reinforce_strategy" and resolved >= max(4, int(min_resolved))
+                ),
+            }
+        )
+
+    candidates.sort(
+        key=lambda item: (
+            0 if item["candidate_type"] == "avoid_strategy" else 1,
+            -item["resolved_outcomes"],
+            item["selected_choice"],
+        )
+    )
+    return candidates[: max(1, int(limit))]
+
+
+def capture_outcome_pattern(
+    pattern_key: str,
+    *,
+    target: str = "learning",
+    category: str = "outcomes",
+) -> dict:
+    clean_target = (target or "learning").strip().lower()
+    if clean_target != "learning":
+        return {"error": f"Unsupported target: {target}"}
+
+    clean_key = (pattern_key or "").strip()
+    if not clean_key:
+        return {"error": "pattern_key is required"}
+
+    candidates = list_outcome_pattern_candidates(limit=200)
+    candidate = next((item for item in candidates if item["pattern_key"] == clean_key), None)
+    if not candidate:
+        return {"error": "Pattern candidate not found or no longer qualifies"}
+
+    selected_choice = candidate["selected_choice"]
+    context = candidate["context_label"]
+    applies_to = f"outcome-pattern:{clean_key}"
+    if candidate["candidate_type"] == "reinforce_strategy":
+        title = f"Prefer {selected_choice} in {context}"
+        prevention = (
+            f"When a comparable context appears, default to '{selected_choice}' unless fresh evidence or constraints override it."
+        )
+    else:
+        title = f"Avoid {selected_choice} in {context}"
+        prevention = (
+            f"When a comparable context appears, do not default to '{selected_choice}' until the evidence base changes."
+        )
+
+    conn = get_db()
+    existing = conn.execute(
+        """SELECT * FROM learnings
+           WHERE status = 'active' AND applies_to = ?
+           ORDER BY updated_at DESC, id DESC
+           LIMIT 1""",
+        (applies_to,),
+    ).fetchone()
+    if existing:
+        return {
+            "ok": True,
+            "created": False,
+            "target": clean_target,
+            "candidate": candidate,
+            "learning": dict(existing),
+        }
+
+    evidence_refs = ", ".join(
+        f"eval#{item['evaluation_id']}/outcome#{item['outcome_id']}:{item['status']}"
+        for item in candidate["evidence"][:OUTCOME_PATTERN_MAX_EVIDENCE]
+    )
+    content = (
+        f"{candidate['rationale']} "
+        f"Success rate: {candidate['success_rate']:.3f}. "
+        f"Resolved outcomes: {candidate['resolved_outcomes']} "
+        f"(met={candidate['met']}, missed={candidate['missed']}). "
+        f"Evidence: {evidence_refs or 'none'}."
+    )
+    reasoning = (
+        "Structured outcome pattern captured from repeated resolved cortex-linked outcomes. "
+        f"Pattern key: {clean_key}."
+    )
+    from db._learnings import create_learning
+
+    learning = create_learning(
+        category=(category or "outcomes").strip(),
+        title=title,
+        content=content,
+        reasoning=reasoning,
+        prevention=prevention,
+        applies_to=applies_to,
+    )
+    return {
+        "ok": True,
+        "created": True,
+        "target": clean_target,
+        "candidate": candidate,
+        "learning": learning,
+    }
+
+
+def get_outcome_pattern_learning_signal(
+    *,
+    area: str = "",
+    task_type: str = "",
+    goal_profile_id: str = "",
+    selected_choice: str = "",
+) -> dict:
+    clean_choice = (selected_choice or "").strip()
+    if not clean_choice:
+        return {
+            "active": False,
+            "pattern_key": "",
+            "learning_id": 0,
+            "mode": "",
+            "title": "",
+            "success_adjustment": 0.0,
+            "risk_adjustment": 0.0,
+        }
+
+    pattern_key = _pattern_key(
+        area=area or "",
+        task_type=task_type or "",
+        goal_profile_id=goal_profile_id or "",
+        selected_choice=clean_choice,
+    )
+    applies_to = f"outcome-pattern:{pattern_key}"
+    conn = get_db()
+    row = conn.execute(
+        """SELECT id, title
+           FROM learnings
+           WHERE status = 'active' AND applies_to = ?
+           ORDER BY updated_at DESC, id DESC
+           LIMIT 1""",
+        (applies_to,),
+    ).fetchone()
+    if not row:
+        return {
+            "active": False,
+            "pattern_key": pattern_key,
+            "learning_id": 0,
+            "mode": "",
+            "title": "",
+            "success_adjustment": 0.0,
+            "risk_adjustment": 0.0,
+        }
+
+    title = str(row["title"] or "").strip()
+    if title.startswith("Prefer "):
+        mode = "prefer"
+        success_adjustment = OUTCOME_PATTERN_LEARNING_SUCCESS_BOOST
+        risk_adjustment = -OUTCOME_PATTERN_LEARNING_RISK_REDUCTION
+    elif title.startswith("Avoid "):
+        mode = "avoid"
+        success_adjustment = OUTCOME_PATTERN_LEARNING_SUCCESS_PENALTY
+        risk_adjustment = OUTCOME_PATTERN_LEARNING_RISK_PENALTY
+    else:
+        mode = "observe"
+        success_adjustment = 0.0
+        risk_adjustment = 0.0
+
+    return {
+        "active": mode in {"prefer", "avoid"},
+        "pattern_key": pattern_key,
+        "learning_id": int(row["id"] or 0),
+        "mode": mode,
+        "title": title,
+        "success_adjustment": round(success_adjustment, 2),
+        "risk_adjustment": round(risk_adjustment, 2),
+    }