session-memory Phase 2: evidence-bar + bloat guard (T04)

gating.py: two-tier evidence bar (OQ5) — promote floor (frequency/sessions/
cost_impact) plus a stricter distribution-eligibility floor that sets a
promoted pattern to approved+distribution_ready vs provisional. Wired into
review() so thin approvals land provisional. bloat_warnings flags duplicate
and near-duplicate (same signal-type+locus) candidates (OQ6). [curate]/
[curate.gate] knobs in config.toml. 6 new tests; suite 64/64 green.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

session_memory/config.toml

@@ -31,6 +31,21 @@ enabled = true
 root    = "~/.grok/sessions"
 glob    = "*/*/chat_history.jsonl"
 
+# Curate phase (AGENTIC-WP-0004): catalog location + promotion evidence bar.
+[curate]
+catalog_dir = "session_memory/catalog"               # files-first Pattern Catalog (committed)
+review_log  = "session_memory/.store/reviews.jsonl"  # remembered decisions (gitignored)
+
+# Evidence bar (OQ5): floors to promote at all, and stricter floors to be
+# distribution-eligible (status=approved, distribution_ready=true).
+[curate.gate]
+min_frequency             = 2      # >= this many supporting signals to promote
+min_sessions              = 2      # >= this many distinct sessions
+min_cost_impact           = 0.0
+dist_require_cross_flavor = false  # require cross-flavor evidence to distribute
+dist_min_frequency        = 3
+dist_min_cost_impact      = 0.0
+
 # cwd basename -> domain slug. Used to tag sessions with their Custodian domain.
 [repo_domain_map]
 agentic-resources = "helix_forge"

session_memory/curate/gating.py

@@ -0,0 +1,117 @@
+"""Promotion evidence-bar + bloat guard (design OQ5/OQ6; T04).
+
+Two gates protect the catalog:
+
+* **Evidence bar (OQ5)** — a candidate must clear configurable floors
+  (frequency, distinct supporting sessions) before it may be promoted at all.
+  A separate, stricter bar decides whether the promoted pattern is
+  *distribution-eligible* (``status="approved"``, ``distribution_ready=True``)
+  vs. merely ``provisional`` — the minimum trustworthy evidence before a pattern
+  is allowed near live agent environments.
+
+* **Bloat guard (OQ6)** — flags candidates that would add little: a duplicate of
+  an already-cataloged pattern, or a near-duplicate sharing the same
+  signal-type+locus. Keeps the catalog lean so agent context budgets aren't
+  degraded by low-value instructions.
+
+Knobs live under ``[curate]`` in ``config.toml``; :func:`gate_config` reads them
+with safe defaults so the module also works config-free (tests).
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Optional
+
+from .schema import SolutionPattern
+
+
+@dataclass
+class GateConfig:
+    # promotion floor (OQ5)
+    min_frequency: int = 2
+    min_sessions: int = 2
+    min_cost_impact: float = 0.0
+    # distribution-eligibility floor (stricter; OQ5)
+    dist_require_cross_flavor: bool = False
+    dist_min_frequency: int = 3
+    dist_min_cost_impact: float = 0.0
+
+
+def gate_config(config: Optional[dict] = None) -> GateConfig:
+    c = (config or {}).get("curate", {}) if config else {}
+    g = c.get("gate", {}) if isinstance(c, dict) else {}
+    return GateConfig(
+        min_frequency=g.get("min_frequency", 2),
+        min_sessions=g.get("min_sessions", 2),
+        min_cost_impact=g.get("min_cost_impact", 0.0),
+        dist_require_cross_flavor=g.get("dist_require_cross_flavor", False),
+        dist_min_frequency=g.get("dist_min_frequency", 3),
+        dist_min_cost_impact=g.get("dist_min_cost_impact", 0.0),
+    )
+
+
+@dataclass
+class GateResult:
+    promotable: bool
+    distribution_ready: bool
+    status: str  # "approved" if distribution-ready else "provisional"
+    reasons: list = field(default_factory=list)
+
+
+def _n_sessions(candidate: dict) -> int:
+    return len(candidate.get("sessions", []) or [])
+
+
+def evaluate(candidate: dict, config: Optional[GateConfig] = None) -> GateResult:
+    """Decide whether a candidate may be promoted, and at what trust level."""
+    cfg = config or GateConfig()
+    reasons: list[str] = []
+
+    freq = candidate.get("frequency", 0)
+    sessions = _n_sessions(candidate)
+    impact = candidate.get("cost_impact", 0.0)
+
+    promotable = True
+    if freq < cfg.min_frequency:
+        promotable = False
+        reasons.append(f"frequency {freq} < min {cfg.min_frequency}")
+    if sessions < cfg.min_sessions:
+        promotable = False
+        reasons.append(f"sessions {sessions} < min {cfg.min_sessions}")
+    if impact < cfg.min_cost_impact:
+        promotable = False
+        reasons.append(f"cost_impact {impact} < min {cfg.min_cost_impact}")
+
+    dist = promotable
+    if cfg.dist_require_cross_flavor and not candidate.get("cross_flavor", False):
+        dist = False
+        reasons.append("not cross-flavor (required for distribution)")
+    if freq < cfg.dist_min_frequency:
+        dist = False
+        reasons.append(f"frequency {freq} < distribution min {cfg.dist_min_frequency}")
+    if impact < cfg.dist_min_cost_impact:
+        dist = False
+        reasons.append(f"cost_impact {impact} < distribution min {cfg.dist_min_cost_impact}")
+
+    return GateResult(
+        promotable=promotable,
+        distribution_ready=bool(dist),
+        status="approved" if dist else "provisional",
+        reasons=reasons,
+    )
+
+
+def bloat_warnings(candidate: dict, existing: list[SolutionPattern]) -> list[str]:
+    """Flag low-value adds against what is already catalogued (OQ6)."""
+    warnings: list[str] = []
+    cand_id = SolutionPattern.make_id(candidate["key"])
+    _, sig_type, locus = (candidate["key"].split(":", 2) + ["", ""])[:3]
+    for p in existing:
+        if p.id == cand_id:
+            warnings.append(f"duplicate of catalogued pattern {p.id}")
+            continue
+        p_parts = (p.provenance.source_key.split(":", 2) + ["", ""])[:3]
+        if (p_parts[1], p_parts[2]) == (sig_type, locus):
+            warnings.append(f"near-duplicate of {p.id} (same {sig_type}/{locus})")
+    return warnings

session_memory/curate/review.py

@@ -22,6 +22,7 @@ from datetime import datetime, timezone
 from typing import Callable, Optional
 
 from .catalog import Catalog
+from .gating import GateConfig, evaluate
 from .schema import Provenance, Resolution, Scope, SolutionPattern
 
 APPROVE = "approve"
@@ -46,8 +47,13 @@ def evidence_fingerprint(candidate: dict) -> str:
     return hashlib.sha1(json.dumps(payload, sort_keys=True).encode("utf-8")).hexdigest()
 
 
-def candidate_to_pattern(candidate: dict) -> SolutionPattern:
-    """Build a (provisional) Solution Pattern from a detect candidate."""
+def candidate_to_pattern(candidate: dict, *, status: str = "provisional",
+                         distribution_ready: bool = False) -> SolutionPattern:
+    """Build a Solution Pattern from a detect candidate.
+
+    ``status``/``distribution_ready`` come from the evidence gate (T04); they
+    default to a provisional, non-distribution-ready pattern when ungated.
+    """
     src = candidate["key"]
     flavors = list(candidate.get("flavors", []))
     hints = {f: {"target": _DEFAULT_TARGET.get(f, ""), "note": "TODO: refine rendering"}
@@ -62,7 +68,8 @@ def candidate_to_pattern(candidate: dict) -> SolutionPattern:
         scope=Scope(flavors=flavors, repos=list(candidate.get("repos", []))),
         provenance=Provenance(source_key=src, evidence=dict(candidate), promoted_at=_now()),
         rendering_hints=hints,
-        status="provisional",
+        status=status,
+        distribution_ready=distribution_ready,
     )
 
 
@@ -112,8 +119,14 @@ class ReviewResult:
 
 
 def review(candidates: list[dict], decide: Decider, catalog: Catalog,
-           log: ReviewLog) -> ReviewResult:
-    """Run each candidate through ``decide``; promote approvals into ``catalog``."""
+           log: ReviewLog, gate: Optional[GateConfig] = None) -> ReviewResult:
+    """Run each candidate through ``decide``; promote approvals into ``catalog``.
+
+    When a ``gate`` (T04 evidence bar) is supplied, the promoted pattern's
+    ``status``/``distribution_ready`` are set from the gate evaluation, so an
+    approved-but-thin candidate lands as ``provisional`` rather than
+    distribution-ready.
+    """
     result = ReviewResult()
     for cand in candidates:
         key = cand["key"]
@@ -125,7 +138,11 @@ def review(candidates: list[dict], decide: Decider, catalog: Catalog,
             result.deferred.append(key)
             continue  # not a final decision — leave for a later pass
         if action == APPROVE:
-            cat_action = catalog.upsert(candidate_to_pattern(cand))
+            g = evaluate(cand, gate) if gate is not None else None
+            pattern = (candidate_to_pattern(cand, status=g.status,
+                                            distribution_ready=g.distribution_ready)
+                       if g is not None else candidate_to_pattern(cand))
+            cat_action = catalog.upsert(pattern)
             result.approved.append((key, cat_action))
         elif action == REJECT:
             result.rejected.append(key)