feat: finish the enrollment protocol's reporting checks

Three items from DECISION §8's enrollment protocol were unimplemented and none
of them needed Tier-2 data:

- Eval reports now record provenance: catalog hash, label-file hash, image hash
  and the code revision, so a number can be traced to the inputs that produced
  it. The catalog hash covers the CSV bytes and the view names, not image bytes:
  a re-encoded view of the same product is the same catalog.
- rebuild_catalog flags cross-SKU near-duplicates as text-discriminated
  siblings. These are the near-unknown case no threshold rejects - the geometry
  is real because the sibling shares the brand's artwork - so they are reported
  rather than refused.
- --shelf-tags turns the completeness rule into a live-ready verdict. Without
  the shelf's SKU list the verdict is 'unknown', not a guess.

Adds test_rebuild_catalog.py (11 tests; the tool had no test file). Suite 261
-> 272. Also corrects score_crop_lanes' docstring, which still described the
live path as averaging local scores when the local scale decides on the newest
crop.

recognization/code/engine.py

@@ -527,9 +527,12 @@ class RecognitionEngine:
         """Both lanes' evidence for one crop: local scores and text ranking.
 
         The live path accumulates evidence across looks, so it needs the lanes
-        separately rather than one fused decision: it averages the local scores
+        separately rather than one fused decision. How it combines them depends
-        over the track's window and keeps the newest text reading, then decides
+        on the scale: the local lane decides on the newest crop (its value is a
-        once. The still path calls `decide_crop`, which fuses a single crop.
+        margin over the runner-up, and averaging in an empty observation halves
+        it below the accept line), while the cosine and inlier scales average
+        the scores over the track's window. The text reading is always the
+        newest. The still path calls `decide_crop`, which fuses a single crop.
         """
         scores = self.score_crop(crop, method)
         text_lane = self.identity_text_lane()

recognization/code/eval_identity.py

@@ -35,8 +35,10 @@ from __future__ import annotations
 
 import argparse
 import csv
+import hashlib
 import json
 import os
+import subprocess
 import sys
 import time
 from collections import Counter
@@ -86,6 +88,55 @@ def load_labels(path: str) -> Dict[str, Any]:
     return payload
 
 
+def file_hash(path: str) -> Optional[str]:
+    """SHA-256 of a file's bytes, or None when it cannot be read."""
+    try:
+        with open(path, "rb") as handle:
+            return hashlib.sha256(handle.read()).hexdigest()
+    except OSError:
+        return None
+
+
+def catalog_hash(stock_dir: str) -> Optional[str]:
+    """SHA-256 over the catalog's CSV bytes and its image names.
+
+    Image bytes are not hashed: a re-encoded view of the same product is the
+    same catalog, and hashing every packshot would make the number move for
+    reasons that do not change a single decision. A changed row or a different
+    set of views does change it.
+    """
+    digest = hashlib.sha256()
+    csv_path = os.path.join(stock_dir, "stock.csv")
+    try:
+        with open(csv_path, "rb") as handle:
+            digest.update(handle.read())
+    except OSError:
+        digest.update(b"missing-csv")
+    try:
+        names = sorted(
+            name for name in os.listdir(stock_dir)
+            if os.path.splitext(name)[1].lower() in (".jpg", ".jpeg", ".png")
+        )
+    except OSError:
+        names = []
+    digest.update(b"\n")
+    digest.update("\n".join(names).encode("utf-8"))
+    return digest.hexdigest()
+
+
+def code_revision() -> Optional[str]:
+    """The checked-out revision, or None outside a git worktree."""
+    try:
+        result = subprocess.run(
+            ["git", "rev-parse", "HEAD"],
+            cwd=os.path.dirname(os.path.abspath(__file__)),
+            capture_output=True, text=True, timeout=10,
+        )
+    except (OSError, subprocess.SubprocessError):
+        return None
+    return result.stdout.strip() or None if result.returncode == 0 else None
+
+
 def crop_of(image: Any, box: Sequence[int]) -> Any:
     x1, y1, x2, y2 = [int(v) for v in box]
     return image[max(0, y1):y2, max(0, x1):x2]
@@ -539,6 +590,15 @@ def main() -> int:
         "image": image_path,
         "stock_dir": args.stock_dir,
         "labels": args.labels,
+        # The enrollment protocol's versioning rule: every report records the
+        # catalog hash, the label-file hash and the code revision, so a number
+        # can be traced to the exact inputs that produced it.
+        "provenance": {
+            "catalog_hash": catalog_hash(args.stock_dir),
+            "labels_hash": file_hash(args.labels),
+            "image_hash": file_hash(image_path),
+            "code_revision": code_revision(),
+        },
         "facings": len(facings),
         "leakage": {"excluded": len(refused), "labels": refused},
         "hub_share_limit": HUB_SHARE,

recognization/code/rebuild_catalog.py

@@ -21,6 +21,12 @@ images by default), whose image entries use ``file`` plus
 ``capture_session_id`` or ``session_id``. Pixel equality is not a leak test:
 re-cut crops pass it, so leakage is refused by capture session instead.
 
+Two checks report rather than refuse, because both are properties of the
+catalog as a whole: cross-SKU near-duplicates are flagged as
+text-discriminated siblings (the near-unknown case only package text
+separates), and ``--shelf-tags`` turns the completeness rule into a live-ready
+verdict. Without that list the verdict is ``unknown``, not a guess.
+
     python code/rebuild_catalog.py --source /tmp/lopia-clean \
         --output stock-lopia-clean --csv stock-lopia/stock.csv
     python code/rebuild_catalog.py --source /tmp/lopia-clean \
@@ -278,6 +284,95 @@ def audit(
     return kept, refused
 
 
+def text_discriminated_siblings(
+    kept: Dict[str, List[str]], source: str, names: Optional[Dict[str, str]] = None
+) -> List[Dict[str, Any]]:
+    """Cross-SKU near-duplicates: the same artwork on two different products.
+
+    These are the near-unknown case no threshold rejects. The geometry is real
+    because the sibling shares the brand's artwork, so only the package text
+    separates them. They are kept - they are legitimately different products -
+    but flagged, because a catalog that enrolls one and not the other will
+    accept the missing one as the enrolled one.
+    """
+    views: List[Tuple[str, str, np.ndarray]] = []
+    for sku, filenames in sorted(kept.items()):
+        for filename in filenames:
+            image = cv2.imread(os.path.join(source, filename))
+            if image is not None:
+                views.append((sku, filename, image))
+    flagged: List[Dict[str, Any]] = []
+    for index, (sku, filename, image) in enumerate(views):
+        for other_sku, other_name, other in views[index + 1:]:
+            if other_sku == sku or image.shape != other.shape:
+                continue
+            similarity = cosine(image, other)
+            if similarity < DUPLICATE_COSINE:
+                continue
+            flagged.append({
+                "sku": sku,
+                "view": filename,
+                "sibling_sku": other_sku,
+                "sibling_view": other_name,
+                "cosine": round(similarity, 4),
+                "name_similarity": round(
+                    name_ocr_similarity(
+                        (names or {}).get(sku, ""), (names or {}).get(other_sku, "")
+                    ), 3,
+                ),
+            })
+    return flagged
+
+
+def live_ready_verdict(
+    kept: Dict[str, List[str]], names: Optional[Dict[str, str]], shelf_skus: Optional[Set[str]]
+) -> Dict[str, Any]:
+    """The enrollment protocol's completeness rule, as a verdict.
+
+    A bay is live-ready when every SKU on its shelf tags is enrolled or
+    explicitly listed as text-only. Without a shelf-tag list the rule cannot be
+    evaluated, and the verdict says so rather than guessing.
+    """
+    catalog = set(names or {})
+    enrolled = set(kept)
+    text_only = catalog - enrolled
+    if shelf_skus is None:
+        return {
+            "live_ready": None,
+            "reason": "no --shelf-tags given; the completeness rule needs the shelf's SKU list",
+            "enrolled": len(enrolled),
+            "text_only": sorted(text_only),
+        }
+    missing = sorted(shelf_skus - enrolled - text_only)
+    return {
+        "live_ready": not missing,
+        "reason": (
+            "every shelf SKU is enrolled or text-only"
+            if not missing
+            else "on the shelf but absent from the catalog: " + ", ".join(missing)
+        ),
+        "enrolled": len(enrolled),
+        "text_only": sorted(text_only),
+        "shelf_skus": len(shelf_skus),
+        "missing_from_catalog": missing,
+    }
+
+
+def load_shelf_skus(path: Optional[str]) -> Optional[Set[str]]:
+    """SKUs on the shelf's tags: one per line, or a JSON list."""
+    if not path:
+        return None
+    with open(path, encoding="utf-8") as handle:
+        text = handle.read()
+    try:
+        payload = json.loads(text)
+    except json.JSONDecodeError:
+        return {line.strip() for line in text.splitlines() if line.strip()}
+    if isinstance(payload, list):
+        return {str(item).strip() for item in payload if str(item).strip()}
+    raise RebuildError(f"{path}: expected a JSON list or one SKU per line")
+
+
 def write_catalog(kept: Dict[str, List[str]], source: str, output: str, csv_path: Optional[str]) -> None:
     """Copy the kept views and the whole CSV into a staging dir, then swap it in."""
     staging = tempfile.mkdtemp(prefix=".catalog-", dir=os.path.dirname(os.path.abspath(output)) or ".")
@@ -329,6 +424,9 @@ def main(argv: Optional[Sequence[str]] = None) -> int:
                         help="image capture metadata (default: SOURCE/manifest.json when present)")
     parser.add_argument("--eval-manifest",
                         help="manifest of evaluation sessions; images captured in them are refused")
+    parser.add_argument("--shelf-tags",
+                        help="the shelf's SKU list (JSON list or one per line) for the "
+                             "completeness rule; without it the verdict is 'unknown'")
     parser.add_argument("--json", help="write the audit here")
     args = parser.parse_args(argv)
 
@@ -378,6 +476,8 @@ def main(argv: Optional[Sequence[str]] = None) -> int:
     # A row with no kept image is a text-only entry: the text lane still names
     # and vetoes it from the CSV alone, so it is a coverage gap, not a refusal.
     text_only = sorted(set(names) - set(kept))
+    siblings = text_discriminated_siblings(kept, args.source, names)
+    verdict = live_ready_verdict(kept, names, load_shelf_skus(args.shelf_tags))
 
     report = {
         "source": args.source,
@@ -389,6 +489,8 @@ def main(argv: Optional[Sequence[str]] = None) -> int:
         "min_name_ocr": MIN_NAME_OCR,
         "ocr": engine.name if ocr_engine is not None else "skipped",
         "text_only_skus": text_only,
+        "text_discriminated_siblings": siblings,
+        "completeness": verdict,
         "skus": {sku: views for sku, views in sorted(kept.items())},
         "refused": refused,
     }
@@ -401,6 +503,17 @@ def main(argv: Optional[Sequence[str]] = None) -> int:
         print(f"  {sku:<10} {len(views)} view(s): {', '.join(views)}")
     if text_only:
         print(f"\ntext-only CSV row(s), no enrolled image ({len(text_only)}): {', '.join(text_only)}")
+    if siblings:
+        print(f"\ntext-discriminated sibling(s) ({len(siblings)}):")
+        for row in siblings:
+            print(f"  {row['sku']:<10} {row['view']:<20} ~ {row['sibling_sku']} "
+                  f"({row['sibling_view']}) cosine {row['cosine']}")
+    if verdict["live_ready"] is None:
+        print(f"\ncompleteness: unknown - {verdict['reason']}")
+    elif verdict["live_ready"]:
+        print(f"\ncompleteness: live-ready ({verdict['reason']})")
+    else:
+        print(f"\ncompleteness: NOT live-ready - {verdict['reason']}")
     if refused:
         print(f"\nrefused {len(refused)} view(s):")
         for row in refused:

recognization/code/test_rebuild_catalog.py

@@ -0,0 +1,122 @@
+"""
+Contract tests for the enrollment tool's reporting checks.
+
+The refusal rules (size floors, session leakage, name/OCR, duplicates) are what
+the tool exists for; these tests cover the two checks that report rather than
+refuse, because both are properties of the catalog as a whole and neither is
+visible from a single view: cross-SKU near-duplicates, and the completeness
+rule that decides whether a bay is live-ready.
+"""
+
+from __future__ import annotations
+
+import os
+import sys
+import tempfile
+import unittest
+
+current_dir = os.path.dirname(os.path.abspath(__file__))
+if current_dir not in sys.path:
+    sys.path.insert(0, current_dir)
+
+import cv2  # noqa: E402
+import numpy as np  # noqa: E402
+
+from rebuild_catalog import (  # noqa: E402
+    RebuildError,
+    live_ready_verdict,
+    load_shelf_skus,
+    text_discriminated_siblings,
+)
+
+
+def artwork(mark=None):
+    """A package face; `mark` adds the small text that separates siblings."""
+    image = np.full((400, 300, 3), 200, dtype="uint8")
+    cv2.rectangle(image, (20, 20), (280, 380), (40, 40, 180), -1)
+    cv2.putText(image, "MAYO", (60, 200), cv2.FONT_HERSHEY_SIMPLEX, 2.0, (255, 255, 255), 4)
+    if mark:
+        cv2.putText(image, mark, (60, 300), cv2.FONT_HERSHEY_SIMPLEX, 1.2, (255, 255, 255), 3)
+    return image
+
+
+class SiblingFlagTests(unittest.TestCase):
+    """The near-unknown case: same artwork, different product."""
+
+    def make_source(self, images):
+        root = tempfile.mkdtemp(prefix="siblings-")
+        for name, image in images.items():
+            cv2.imwrite(os.path.join(root, name), image)
+        return root
+
+    def test_shared_artwork_across_skus_is_flagged(self):
+        root = self.make_source({"sku_x1.jpg": artwork(), "sku_x2.jpg": artwork("HALF")})
+        flagged = text_discriminated_siblings({"sku_x1": ["sku_x1.jpg"], "sku_x2": ["sku_x2.jpg"]}, root)
+        self.assertEqual(len(flagged), 1, flagged)
+        self.assertEqual(flagged[0]["sku"], "sku_x1")
+        self.assertEqual(flagged[0]["sibling_sku"], "sku_x2")
+        self.assertGreater(flagged[0]["cosine"], 0.98)
+
+    def test_distinct_artwork_is_not_flagged(self):
+        other = np.full((400, 300, 3), 30, dtype="uint8")
+        cv2.circle(other, (150, 200), 90, (200, 200, 40), -1)
+        root = self.make_source({"sku_x1.jpg": artwork(), "sku_x2.jpg": other})
+        self.assertEqual(
+            text_discriminated_siblings({"sku_x1": ["sku_x1.jpg"], "sku_x2": ["sku_x2.jpg"]}, root), []
+        )
+
+    def test_two_views_of_one_sku_are_not_siblings(self):
+        """The same-SKU duplicate rule already covers these; this flag is cross-SKU."""
+        root = self.make_source({"sku_x1.jpg": artwork(), "sku_x1_2.jpg": artwork()})
+        self.assertEqual(
+            text_discriminated_siblings({"sku_x1": ["sku_x1.jpg", "sku_x1_2.jpg"]}, root), []
+        )
+
+
+class CompletenessVerdictTests(unittest.TestCase):
+    """The enrollment protocol's completeness rule, as a verdict."""
+
+    def test_without_a_shelf_list_the_verdict_is_unknown(self):
+        """Not a guess: the rule needs the shelf's SKU list to mean anything."""
+        verdict = live_ready_verdict({"sku_a": ["a.jpg"]}, {"sku_a": "A"}, None)
+        self.assertIsNone(verdict["live_ready"])
+        self.assertIn("shelf", verdict["reason"])
+
+    def test_every_shelf_sku_enrolled_is_live_ready(self):
+        verdict = live_ready_verdict({"sku_a": ["a.jpg"]}, {"sku_a": "A"}, {"sku_a"})
+        self.assertTrue(verdict["live_ready"])
+
+    def test_a_text_only_row_satisfies_the_rule(self):
+        """A CSV row with no image still names and vetoes from its text."""
+        verdict = live_ready_verdict({"sku_a": ["a.jpg"]}, {"sku_a": "A", "sku_b": "B"}, {"sku_a", "sku_b"})
+        self.assertTrue(verdict["live_ready"])
+
+    def test_a_shelf_sku_absent_from_the_catalog_fails(self):
+        verdict = live_ready_verdict({"sku_a": ["a.jpg"]}, {"sku_a": "A"}, {"sku_a", "sku_z"})
+        self.assertFalse(verdict["live_ready"])
+        self.assertEqual(verdict["missing_from_catalog"], ["sku_z"])
+
+
+class ShelfTagListTests(unittest.TestCase):
+    def write(self, text):
+        path = os.path.join(tempfile.mkdtemp(prefix="tags-"), "tags.json")
+        with open(path, "w", encoding="utf-8") as handle:
+            handle.write(text)
+        return path
+
+    def test_a_json_list_is_read(self):
+        self.assertEqual(load_shelf_skus(self.write('["sku_a", "sku_b"]')), {"sku_a", "sku_b"})
+
+    def test_one_sku_per_line_is_read(self):
+        self.assertEqual(load_shelf_skus(self.write("sku_a\nsku_b\n")), {"sku_a", "sku_b"})
+
+    def test_no_path_means_no_list(self):
+        self.assertIsNone(load_shelf_skus(None))
+
+    def test_a_json_object_is_refused(self):
+        with self.assertRaises(RebuildError):
+            load_shelf_skus(self.write('{"sku_a": 1}'))
+
+
+if __name__ == "__main__":
+    unittest.main()