diff --git a/AGENTS.md b/AGENTS.md
index cca53aa..2f4e6a6 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -1,7 +1,20 @@
 # Agent Instructions — Photo Analyzer
 
-Read `CLAUDE.md` and `INTEGRATED_PIPELINE_CONCEPT.md` before implementation. Preserve
-the established project invariants: never inspect or process content beneath any
+`INTEGRATED_PIPELINE_CONCEPT.md` is the authoritative product and architecture
+foundation for every user story. Before claiming or resuming implementation, read it
+in full, then use it to interpret the story's scope, acceptance criteria, donor
+strategy, architecture boundaries, safety invariants, fixture requirements, and test
+obligations. Read `CLAUDE.md` as operational donor context after the concept. If an
+issue, backlog Markdown file, or implementation idea conflicts with the concept, do
+not silently follow the conflicting source: preserve the concept and stop for an
+explicit decision when the conflict materially changes scope or behavior.
+
+Re-read the complete concept whenever context has been compacted/lost, the concept has
+changed since the story began, or an implementation session is resumed without a
+reliable record that the current version was read. A summary, prior familiarity, or
+reading only the story's phase is not sufficient.
+
+Preserve the established project invariants: never inspect or process content beneath any
 `_IGNORE/` directory; write verified EXIF before Immich upload; use managed
 `immich-go` uploads rather than External Libraries; keep operations resume-safe; and
 treat the existing CLI implementations as primary donors rather than rewriting them.
@@ -54,22 +67,26 @@ When asked to start, resume, or continue implementation, work from this isolated
 repository and use `work_item/scripts/work-item`. Gitea issues and their dependencies
 are the authoritative backlog.
 
-1. Run `work_item/scripts/work-item claim`.
-2. Run `work_item/scripts/work-item status` when resuming or whenever branch/state is
+1. Read the current `INTEGRATED_PIPELINE_CONCEPT.md` completely and treat it as the
+   foundation for all decisions in the story. Then read `CLAUDE.md` for operational
+   donor context.
+2. Run `work_item/scripts/work-item claim`.
+3. Run `work_item/scripts/work-item status` when resuming or whenever branch/state is
    uncertain. Work only on the claimed story and its generated feature branch.
-3. Read the entire issue, linked specification, dependencies, and acceptance criteria.
-4. Inspect the legacy CLI donors before replacing applicable behavior. Update the donor
+4. Read the entire issue, linked specification, dependencies, and acceptance criteria.
+   Reconcile them with the concept before designing or changing code.
+5. Inspect the legacy CLI donors before replacing applicable behavior. Update the donor
    ledger and characterization tests required by the story.
-5. Implement every acceptance criterion and its automated tests.
-6. Run story-specific tests and the accumulated regression suite required by the epic.
-7. Run `work_item/scripts/work-item submit --test "<story-specific command>"` without
+6. Implement every acceptance criterion and its automated tests.
+7. Run story-specific tests and the accumulated regression suite required by the epic.
+8. Run `work_item/scripts/work-item submit --test "<story-specific command>"` without
    `--yes` and review the reported diff and file list.
-8. If the changes are correct and safe, rerun with `--yes`. This commits, pushes, opens
+9. If the changes are correct and safe, rerun with `--yes`. This commits, pushes, opens
    a pull request, records test evidence, and moves the issue to review.
-9. After review and successful CI, run
+10. After review and successful CI, run
    `work_item/scripts/work-item complete --merge`. This verifies the merge, closes the
    issue, returns to updated `main`, and removes the local feature branch.
-10. Continue with the next eligible story by returning to step 1.
+11. Continue with the next eligible story by returning to step 1.
 
 If genuinely blocked, run
 `work_item/scripts/work-item block --reason "<precise reason>"` and stop. Never skip to
diff --git a/donor_ledger/README.md b/donor_ledger/README.md
new file mode 100644
index 0000000..4f84ad4
--- /dev/null
+++ b/donor_ledger/README.md
@@ -0,0 +1,23 @@
+# Donor ledger
+
+`ledger.json` is the machine-readable migration inventory for the existing Photo
+Analyzer and NSFW Tagger implementations. It is intentionally kept beside the live
+donors until the archival gates in `INTEGRATED_PIPELINE_CONCEPT.md` are satisfied.
+
+Each row records the current source symbol, the intended treatment, its destination in
+the integrated application, the reason for that decision, and the stable
+characterization-test IDs protecting current behavior. `migration_status` remains
+`inventoried` in US01-01: this story characterizes code but does not relocate it.
+
+Classification meanings:
+
+- `reuse`: preserve the implementation essentially unchanged;
+- `extract`: move the coherent behavior behind a thin service/integration adapter;
+- `refactor`: preserve the useful behavior while changing an incompatible boundary;
+- `replace`: intentionally supersede unsafe or architecturally incompatible behavior.
+
+Run the story suite with:
+
+```bash
+work_item/scripts/python -m unittest discover -s tests/characterization -v
+```
diff --git a/donor_ledger/ledger.json b/donor_ledger/ledger.json
new file mode 100644
index 0000000..67f296f
--- /dev/null
+++ b/donor_ledger/ledger.json
@@ -0,0 +1,180 @@
+{
+  "schema_version": 1,
+  "story": "US01-01",
+  "concept": "INTEGRATED_PIPELINE_CONCEPT.md",
+  "classifications": ["reuse", "extract", "refactor", "replace"],
+  "required_areas": [
+    "discovery", "hashing", "imaging", "nsfw", "vision", "exif",
+    "database", "ui", "configuration", "logging", "cancellation", "error_behavior"
+  ],
+  "entries": [
+    {
+      "id": "DONOR-001", "area": "discovery",
+      "sources": ["photo_analyzer.py::discover_photos", "photo_analyzer.py::SUPPORTED_EXTENSIONS"],
+      "classification": "extract", "target": "src/photo_pipeline/services/inventory.py",
+      "rationale": "Preserve sorted recursive discovery and supported formats behind the shared path policy.",
+      "intentional_changes": "All callers use stable asset IDs and one centralized boundary/exclusion policy.",
+      "test_ids": ["CHAR-001"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-002", "area": "discovery",
+      "sources": ["nsfwtag/scoring.py::discover_images", "nsfwtag/__init__.py::EXTS"],
+      "classification": "replace", "target": "src/photo_pipeline/services/inventory.py",
+      "rationale": "Its shallow/recursive switch and stable ordering are useful, but a second discovery policy and narrower formats conflict with the shared inventory.",
+      "intentional_changes": "Safety consumes canonical inventory assets rather than walking paths itself.",
+      "test_ids": ["CHAR-001"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-003", "area": "hashing",
+      "sources": ["photo_analyzer.py::_sha1_file", "photo_analyzer.py::_phash_image", "photo_analyzer.py::ensure_hashes"],
+      "classification": "refactor", "target": "src/photo_pipeline/services/inventory.py",
+      "rationale": "Retain streaming exact hashing and DCT pHash behavior while adding SHA-256, normalized pixels, versions, bounded workers, and stable IDs.",
+      "intentional_changes": "SHA-256 becomes byte identity; pHash is evidence rather than automatic fuzzy exclusion.",
+      "test_ids": ["CHAR-002"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-004", "area": "hashing",
+      "sources": ["photo_analyzer.py::cluster_duplicates", "photo_analyzer.py::mark_duplicates", "photo_analyzer.py::reconcile_moved"],
+      "classification": "refactor", "target": "src/photo_pipeline/services/duplicates.py",
+      "rationale": "Keep Hamming-distance evidence and reconciliation knowledge, but path-keyed union-find and automatic largest-file selection do not meet review and identity rules.",
+      "intentional_changes": "Exact matches may be recommended; fuzzy matches require durable human decisions and negative links.",
+      "test_ids": ["CHAR-002", "CHAR-003"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-005", "area": "imaging",
+      "sources": ["photo_analyzer.py::prepare_image"],
+      "classification": "extract", "target": "src/photo_pipeline/services/analysis.py",
+      "rationale": "The in-memory RGB conversion, long-edge resize, and JPEG normalization are proven analysis preparation behavior.",
+      "intentional_changes": "Orientation, color profile, pixel limits, and decoder version become explicit and tested.",
+      "test_ids": ["CHAR-002"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-006", "area": "imaging",
+      "sources": ["nsfwtag/server.py::serve_review", "nsfwtag/webapp.py::render_page"],
+      "classification": "refactor", "target": "src/photo_pipeline/services/thumbnails.py",
+      "rationale": "Preserve useful review presentation behavior, but arbitrary path image serving cannot be the integrated thumbnail boundary.",
+      "intentional_changes": "Managed oriented thumbnails are resolved only by asset ID and bounded size.",
+      "test_ids": ["CHAR-007"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-007", "area": "nsfw",
+      "sources": ["nsfwtag/scoring.py::score_images", "nsfwtag/__init__.py::MODEL_ID", "nsfwtag/__init__.py::DEFAULT_THRESHOLD"],
+      "classification": "extract", "target": "src/photo_pipeline/integrations/nsfw_model.py",
+      "rationale": "Retain the selected local model, batching, score interpretation, and error isolation behind an injectable integration.",
+      "intentional_changes": "SQLite is authoritative; CSV becomes optional export and only canonical assets are scored.",
+      "test_ids": ["CHAR-006"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-008", "area": "nsfw",
+      "sources": ["nsfwtag/__main__.py::main", "nsfw_tag.py::main"],
+      "classification": "refactor", "target": "src/photo_pipeline/services/safety.py",
+      "rationale": "Keep CLI compatibility temporarily while moving orchestration to durable jobs and explicit review decisions.",
+      "intentional_changes": "The API never invokes the legacy CLI and cloud analysis requires confirmed SFW state.",
+      "test_ids": ["CHAR-008"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-009", "area": "vision",
+      "sources": ["photo_analyzer.py::ANALYSIS_PROMPT", "photo_analyzer.py::analyze_image"],
+      "classification": "extract", "target": "src/photo_pipeline/integrations/vision.py",
+      "rationale": "Preserve provider-compatible image requests, prompt context, fenced-JSON handling, token accounting, and retry categories.",
+      "intentional_changes": "Typed validation, persisted prompt/model versions, bounded jittered retries, and stale-result fencing are added.",
+      "test_ids": ["CHAR-005"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-010", "area": "exif",
+      "sources": ["photo_analyzer.py::build_exif_caption_from_result", "photo_analyzer.py::write_exif", "photo_analyzer.py::read_existing_exif"],
+      "classification": "refactor", "target": "src/photo_pipeline/integrations/exiftool.py",
+      "rationale": "Retain additive caption/tag projection and exiftool knowledge while making ownership and read-back verification explicit.",
+      "intentional_changes": "Each metadata stage snapshots non-owned fields, verifies its projection, and refreshes post-write hashes.",
+      "test_ids": ["CHAR-004", "CHAR-009"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-011", "area": "exif",
+      "sources": ["nsfwtag/exif.py::write_keyword", "nsfwtag/exif.py::remove_keyword", "nsfwtag/exif.py::read_marks"],
+      "classification": "extract", "target": "src/photo_pipeline/integrations/exiftool.py",
+      "rationale": "The idempotent keyword commands and mutually exclusive review semantics are reusable in a shared adapter.",
+      "intentional_changes": "Both SFW and NSFW decisions are persisted and verified without removing unrelated metadata.",
+      "test_ids": ["CHAR-009"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-012", "area": "database",
+      "sources": ["photo_analyzer.py::SCHEMA", "photo_analyzer.py::get_db", "photo_analyzer.py::_migrate_schema"],
+      "classification": "refactor", "target": "src/photo_pipeline/models/",
+      "rationale": "Preserve historical schema and FTS migration knowledge while introducing SQLAlchemy repositories, Alembic, WAL, foreign keys, and stable asset identity.",
+      "intentional_changes": "Paths become occurrences, not primary identity; migrations are versioned and transactional where possible.",
+      "test_ids": ["CHAR-003"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-013", "area": "database",
+      "sources": ["webapp/query.py::search", "webapp/query.py::facets", "webapp/query.py::stats"],
+      "classification": "extract", "target": "src/photo_pipeline/repositories/assets.py",
+      "rationale": "FTS query sanitization, paging, facets, album labels, and stats are useful read-model behavior.",
+      "intentional_changes": "Repositories own SQL and return typed stable-ID projections.",
+      "test_ids": ["CHAR-003"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-014", "area": "ui",
+      "sources": ["nsfwtag/webapp.py::folder_tree_html", "nsfwtag/webapp.py::render_page", "nsfwtag/review.html"],
+      "classification": "refactor", "target": "frontend/js/views/safety.js",
+      "rationale": "Preserve folder navigation, score cards, review controls, lightbox interaction, and visual language.",
+      "intentional_changes": "Semantic static HTML/CSS/JS fetch JSON from /api/v1; Python no longer renders operational cards.",
+      "test_ids": ["CHAR-007"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-015", "area": "ui",
+      "sources": ["webapp/page.py::render_page", "webapp/server.py::serve"],
+      "classification": "replace", "target": "frontend/index.html",
+      "rationale": "The design is a donor, but server-generated application state and ad-hoc HTTP routing conflict with the application-shell/API architecture.",
+      "intentional_changes": "FastAPI serves a data-free shell and versioned JSON/SSE endpoints.",
+      "test_ids": ["CHAR-007"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-016", "area": "configuration",
+      "sources": ["photo_analyzer.py::load_env_file", "photo_analyzer.py::_env_int", "photo_analyzer.py::_env_bool"],
+      "classification": "replace", "target": "src/photo_pipeline/config.py",
+      "rationale": "Environment precedence is useful, but a manual untyped parser cannot validate the integrated app's paths, secrets, and worker settings.",
+      "intentional_changes": "Pydantic settings provide typed validation and secret references.",
+      "test_ids": ["CHAR-010"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-017", "area": "logging",
+      "sources": ["photo_analyzer.py::log_history", "photo_analyzer.py::_record_ratelimit"],
+      "classification": "refactor", "target": "src/photo_pipeline/jobs/events.py",
+      "rationale": "Per-item history, token evidence, and throttle diagnostics remain valuable operational events.",
+      "intentional_changes": "Structured durable events carry job, asset, attempt, and operation IDs with retention controls.",
+      "test_ids": ["CHAR-011"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-018", "area": "cancellation",
+      "sources": ["photo_analyzer.py::_handle_sigint", "webapp/runner.py::Runner"],
+      "classification": "replace", "target": "src/photo_pipeline/jobs/coordinator.py",
+      "rationale": "The cooperative stop intent is correct, but in-memory events and subprocess ownership do not survive restart or provide fencing.",
+      "intentional_changes": "Cancellation is a durable job transition checked between items and recovered by heartbeat/lease state.",
+      "test_ids": ["CHAR-012"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-019", "area": "error_behavior",
+      "sources": ["photo_analyzer.py::mark_error", "photo_analyzer.py::analyze_image", "photo_analyzer.py::write_exif"],
+      "classification": "refactor", "target": "src/photo_pipeline/services/analysis.py",
+      "rationale": "Keep item-level failures, retry categories, raw-response diagnostics, and explicit EXIF failure behavior.",
+      "intentional_changes": "Errors gain stable codes, attempts, retryability, and unknown/divergent states rather than broad status strings.",
+      "test_ids": ["CHAR-005", "CHAR-010"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-020", "area": "error_behavior",
+      "sources": ["nsfwtag/scoring.py::score_images", "nsfwtag/exif.py::_read_keywords", "webapp/runner.py::Runner"],
+      "classification": "refactor", "target": "src/photo_pipeline/services/safety.py",
+      "rationale": "Per-file inference isolation and tolerant metadata reads are useful, while process errors need durable domain outcomes.",
+      "intentional_changes": "Failures are persisted per attempt; unavailable integrations block precisely and never imply a safety decision.",
+      "test_ids": ["CHAR-006", "CHAR-010"], "migration_status": "inventoried"
+    },
+    {
+      "id": "DONOR-021", "area": "configuration",
+      "sources": ["photo_analyzer.py::main", "nsfwtag/__main__.py::main"],
+      "classification": "refactor", "target": "src/photo_pipeline/__main__.py",
+      "rationale": "The current flags document supported workflows and compatibility needs, but long operations must become service-backed durable jobs.",
+      "intentional_changes": "Compatibility CLIs call shared services; API and worker management use typed commands.",
+      "test_ids": ["CHAR-008"], "migration_status": "inventoried"
+    }
+  ]
+}
diff --git a/tests/characterization/test_donors.py b/tests/characterization/test_donors.py
new file mode 100644
index 0000000..a403489
--- /dev/null
+++ b/tests/characterization/test_donors.py
@@ -0,0 +1,295 @@
+from __future__ import annotations
+
+import base64
+import importlib
+import json
+import logging
+import os
+import subprocess
+import sys
+import tempfile
+import unittest
+from pathlib import Path
+from types import SimpleNamespace
+from unittest import mock
+
+from PIL import Image
+
+
+REPO = Path(__file__).resolve().parents[2]
+FIXTURES = REPO / "tests" / "fixtures" / "characterization"
+
+
+class DonorCharacterizationTests(unittest.TestCase):
+    """Golden contracts for donor behavior; these are not target-architecture tests."""
+
+    @classmethod
+    def setUpClass(cls):
+        cls.import_dir = tempfile.TemporaryDirectory(prefix="donor-import-")
+        old_cwd = Path.cwd()
+        sys.path.insert(0, str(REPO))
+        try:
+            os.chdir(cls.import_dir.name)
+            cls.analyzer = importlib.import_module("photo_analyzer")
+            cls.nsfw_scoring = importlib.import_module("nsfwtag.scoring")
+            cls.nsfw_exif = importlib.import_module("nsfwtag.exif")
+            cls.nsfw_webapp = importlib.import_module("nsfwtag.webapp")
+            cls.web_query = importlib.import_module("webapp.query")
+        finally:
+            os.chdir(old_cwd)
+        cls.manifest = json.loads((FIXTURES / "manifest.json").read_text())["fixtures"]
+        cls.expected = json.loads((FIXTURES / "expected_outputs.json").read_text())["outputs"]
+
+    @classmethod
+    def tearDownClass(cls):
+        for handler in list(logging.getLogger().handlers) + list(
+            logging.getLogger("history").handlers
+        ):
+            handler.close()
+        cls.import_dir.cleanup()
+
+    def setUp(self):
+        self.tmp = tempfile.TemporaryDirectory(prefix="donor-fixtures-")
+        self.root = Path(self.tmp.name)
+
+    def tearDown(self):
+        self.analyzer._stop.clear()
+        self.tmp.cleanup()
+
+    def _make_discovery_library(self) -> None:
+        recipes = {item["id"]: item for item in self.manifest}
+        colors = {
+            "DISC-ROOT-JPEG": (10, 20, 30),
+            "DISC-UPPER-JPEG": (40, 50, 60),
+            "DISC-NESTED-WEBP": (70, 80, 90),
+            "DISC-TIFF": (100, 110, 120),
+        }
+        sizes = {
+            "DISC-ROOT-JPEG": (12, 8),
+            "DISC-UPPER-JPEG": (8, 12),
+            "DISC-NESTED-WEBP": (9, 9),
+            "DISC-TIFF": (16, 4),
+        }
+        for fixture_id, color in colors.items():
+            path = self.root / recipes[fixture_id]["relative_path"]
+            path.parent.mkdir(parents=True, exist_ok=True)
+            Image.new("RGB", sizes[fixture_id], color).save(path)
+        unsupported = self.root / recipes["DISC-UNSUPPORTED"]["relative_path"]
+        unsupported.write_text("not-an-image", encoding="utf-8")
+
+    def _relative(self, paths) -> list[str]:
+        return [str(Path(path).relative_to(self.root)) for path in paths]
+
+    def test_discovery_outputs(self):
+        self._make_discovery_library()
+
+        analyzer_paths = self.analyzer.discover_photos(self.root)
+        nsfw_recursive = self.nsfw_scoring.discover_images(self.root, recursive=True)
+        nsfw_shallow = self.nsfw_scoring.discover_images(self.root, recursive=False)
+
+        self.assertEqual(
+            self._relative(analyzer_paths), self.expected["analyzer_recursive_discovery"]
+        )
+        self.assertEqual(self._relative(nsfw_recursive), self.expected["nsfw_recursive_discovery"])
+        self.assertEqual(self._relative(nsfw_shallow), self.expected["nsfw_shallow_discovery"])
+
+    def test_image_preparation_and_hash_outputs(self):
+        raw_path = self.root / "hash.bin"
+        raw_path.write_bytes(b"photo-pipeline-donor")
+        self.assertEqual(self.analyzer._sha1_file(raw_path), self.expected["hash_bytes_sha1"])
+
+        gradient = Image.new("RGB", (40, 20))
+        gradient.putdata(
+            [(x * 5 % 256, y * 11 % 256, (x + y) * 7 % 256) for y in range(20) for x in range(40)]
+        )
+        image_path = self.root / "gradient.png"
+        gradient.save(image_path)
+        old_edge = self.analyzer.MAX_LONG_EDGE
+        try:
+            self.analyzer.MAX_LONG_EDGE = 16
+            encoded, mime = self.analyzer.prepare_image(image_path)
+        finally:
+            self.analyzer.MAX_LONG_EDGE = old_edge
+        prepared_path = self.root / "prepared.jpg"
+        prepared_path.write_bytes(base64.b64decode(encoded))
+        with Image.open(prepared_path) as prepared:
+            self.assertEqual(prepared.size, (16, 8))
+            self.assertEqual(prepared.mode, "RGB")
+        self.assertEqual(mime, "image/jpeg")
+        phash = self.analyzer._phash_image(image_path)
+        self.assertRegex(phash or "", r"^[0-9a-f]{16}$")
+        self.assertEqual(phash, self.analyzer._phash_image(image_path))
+
+    def test_database_status_and_fts_outputs(self):
+        db_path = self.root / "characterization.sqlite"
+        conn = self.analyzer.get_db(str(db_path))
+        self.addCleanup(conn.close)
+        photo_path = str(self.root / "lake.jpg")
+        self.analyzer.upsert_pending(conn, photo_path)
+        result = {
+            "description": "Three adults walk beside a lake.",
+            "tags": ["people", "lake", "summer"],
+            "people_count": 3,
+            "setting": "outdoor",
+            "time_of_day": "afternoon",
+            "season": "summer",
+            "mood": "relaxed",
+            "location_hint": "Como, Italy",
+            "approx_year": 2021,
+        }
+        self.analyzer.mark_analyzed(conn, photo_path, result, json.dumps(result))
+        row = conn.execute("SELECT * FROM photos WHERE path = ?", (photo_path,)).fetchone()
+        self.assertEqual(row["status"], "analyzed")
+        self.assertEqual(json.loads(row["tags"]), result["tags"])
+        found = self.web_query.search(conn, q="lake")
+        self.assertEqual(found["total"], 1)
+        self.assertEqual(found["rows"][0]["path"], photo_path)
+        self.analyzer.mark_exif_written(conn, photo_path)
+        self.assertEqual(conn.execute("SELECT status FROM photos").fetchone()[0], "exif_written")
+
+    def test_caption_and_variant_outputs(self):
+        result = {
+            "description": "Three adults walk beside a lake.",
+            "tags": ["people", "lake", "summer"],
+            "mood": "relaxed",
+            "location_hint": "Como, Italy",
+            "approx_year": 2021,
+        }
+        self.assertEqual(
+            self.analyzer.build_exif_caption_from_result(result), self.expected["caption"]
+        )
+        self.assertEqual(
+            self.analyzer._strip_variant_markers("IMG_0001-bearbeitet (1920x1080)"),
+            self.expected["variant_base"],
+        )
+
+    def test_vision_dry_run_output(self):
+        result, raw, tokens = self.analyzer.analyze_image(
+            None, self.root / "fixture.jpg", dry_run=True
+        )
+        self.assertEqual(result["tags"], self.expected["dry_run_tags"])
+        self.assertEqual(json.loads(raw), result)
+        self.assertEqual(tokens, {"prompt": 0, "completion": 0, "total": 0})
+
+    def test_nsfw_cache_output(self):
+        cache_path = self.root / "scores.csv"
+        values = self.expected["nsfw_cache"]
+        self.nsfw_scoring._save_cache(cache_path, values)
+        self.assertEqual(self.nsfw_scoring.load_cache(cache_path), values)
+        self.assertEqual(
+            cache_path.read_text(encoding="utf-8").splitlines(),
+            ["path,nsfw_score", "a.jpg,0.1250", "b.jpg,0.9876"],
+        )
+
+    def test_review_html_output(self):
+        logical = self.root / "album & one" / "<portrait>.jpg"
+        page = self.nsfw_webapp.render_page([(logical, 0.8123), (logical, 0.7)], 0.6)
+        self.assertIn("&lt;portrait&gt;.jpg", page)
+        self.assertIn("album &amp; one", page)
+        self.assertIn(f'data-score="{0.8123:.4f}"', page)
+        self.assertIn(self.expected["ui_total"], page)
+        self.assertIn(self.expected["ui_threshold"], page)
+        self.assertEqual(page.count('class="card"'), 1)
+
+    def test_cli_entry_point_help(self):
+        env = os.environ.copy()
+        env["PYTHONPATH"] = str(REPO)
+        commands = [
+            ([sys.executable, str(REPO / "photo_analyzer.py"), "--help"], "--group-variants"),
+            ([sys.executable, str(REPO / "nsfw_tag.py"), "--help"], "--review-min"),
+            ([sys.executable, "-m", "nsfwtag", "--help"], "--threshold"),
+        ]
+        for command, marker in commands:
+            with self.subTest(command=command):
+                result = subprocess.run(
+                    command,
+                    cwd=self.root,
+                    env=env,
+                    capture_output=True,
+                    text=True,
+                    timeout=30,
+                )
+                self.assertEqual(result.returncode, 0, result.stderr)
+                self.assertIn(marker, result.stdout)
+
+    def test_exif_command_contracts(self):
+        calls = []
+
+        def fake_run(command, **kwargs):
+            calls.append(command)
+            if "-json" in command:
+                return subprocess.CompletedProcess(
+                    command,
+                    0,
+                    stdout=json.dumps(
+                        [{"ImageDescription": "User caption", "Keywords": ["family"]}]
+                    ),
+                    stderr="",
+                )
+            return subprocess.CompletedProcess(command, 0, stdout="", stderr="")
+
+        with mock.patch.object(self.analyzer.subprocess, "run", side_effect=fake_run):
+            self.assertTrue(
+                self.analyzer.write_exif("fixture.jpg", "AI caption", ["family", "lake"])
+            )
+        write_command = calls[-1]
+        self.assertIn("-overwrite_original", write_command)
+        self.assertIn("-ImageDescription=AI caption | User caption", write_command)
+        self.assertEqual(write_command.count("-Keywords=family"), 1)
+        self.assertEqual(write_command.count("-Keywords=lake"), 1)
+
+        with mock.patch.object(
+            self.nsfw_exif.subprocess,
+            "run",
+            return_value=subprocess.CompletedProcess([], 0, stdout="", stderr=""),
+        ) as run:
+            self.assertTrue(self.nsfw_exif.write_keyword("fixture.jpg"))
+            command = run.call_args.args[0]
+        self.assertIn("-Keywords-=nsfw", command)
+        self.assertIn("-Keywords+=nsfw", command)
+        self.assertIn("-Subject-=nsfw", command)
+        self.assertIn("-Subject+=nsfw", command)
+
+    def test_error_and_configuration_fallbacks(self):
+        with mock.patch.dict(os.environ, {"MAX_WORKERS": "not-an-int"}):
+            self.assertEqual(self.analyzer._env_int("MAX_WORKERS", 3), 3)
+        invalid = subprocess.CompletedProcess([], 0, stdout="not-json", stderr="")
+        with mock.patch.object(self.nsfw_exif.subprocess, "run", return_value=invalid):
+            self.assertEqual(self.nsfw_exif._read_keywords(["fixture.jpg"]), {})
+
+    def test_history_log_shape(self):
+        sink = mock.Mock()
+        result = {
+            "description": "Fixture description",
+            "tags": ["fixture"],
+            "mood": "calm",
+            "setting": "indoor",
+            "people_count": 0,
+            "location_hint": None,
+            "approx_year": None,
+        }
+        with mock.patch.object(self.analyzer.history_log, "info", sink):
+            self.analyzer.log_history(
+                "fixture.jpg", "analyzed", result=result, tokens={"total": 12}
+            )
+        entry = json.loads(sink.call_args.args[0])
+        self.assertEqual(entry["path"], "fixture.jpg")
+        self.assertEqual(entry["status"], "analyzed")
+        self.assertEqual(entry["tokens_total"], 12)
+        self.assertEqual(entry["tags"], ["fixture"])
+        self.assertRegex(entry["ts"], r"^\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}$")
+
+    def test_cooperative_cancellation_contract(self):
+        self.analyzer._stop.clear()
+        fake_stderr = SimpleNamespace(write=mock.Mock(), flush=mock.Mock())
+        with (
+            mock.patch.object(self.analyzer.sys, "__stderr__", fake_stderr),
+            mock.patch.object(self.analyzer.log, "warning"),
+        ):
+            self.analyzer._handle_sigint(None, None)
+        self.assertTrue(self.analyzer._stop.is_set())
+        fake_stderr.write.assert_called_once_with("\a")
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/tests/characterization/test_ledger.py b/tests/characterization/test_ledger.py
new file mode 100644
index 0000000..f016cce
--- /dev/null
+++ b/tests/characterization/test_ledger.py
@@ -0,0 +1,62 @@
+from __future__ import annotations
+
+import ast
+import json
+import unittest
+from pathlib import Path
+
+REPO = Path(__file__).resolve().parents[2]
+LEDGER = REPO / "donor_ledger" / "ledger.json"
+TRACEABILITY = REPO / "tests" / "characterization" / "test_traceability.json"
+DONOR_TESTS = REPO / "tests" / "characterization" / "test_donors.py"
+
+
+def source_symbols(path: Path) -> set[str]:
+    tree = ast.parse(path.read_text(encoding="utf-8"), filename=str(path))
+    names = set()
+    for node in ast.walk(tree):
+        if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef, ast.ClassDef)):
+            names.add(node.name)
+        elif isinstance(node, (ast.Assign, ast.AnnAssign)):
+            targets = node.targets if isinstance(node, ast.Assign) else [node.target]
+            names.update(t.id for t in targets if isinstance(t, ast.Name))
+        elif isinstance(node, ast.ImportFrom):
+            names.update(alias.asname or alias.name for alias in node.names)
+    return names
+
+
+class DonorLedgerLintTests(unittest.TestCase):
+    def test_ledger_is_complete_and_resolvable(self):
+        ledger = json.loads(LEDGER.read_text(encoding="utf-8"))
+        traceability = json.loads(TRACEABILITY.read_text(encoding="utf-8"))
+        entries = ledger["entries"]
+        donor_test_symbols = source_symbols(DONOR_TESTS)
+
+        self.assertEqual(ledger["story"], "US01-01")
+        self.assertEqual(len({entry["id"] for entry in entries}), len(entries))
+        self.assertEqual(set(ledger["required_areas"]), {entry["area"] for entry in entries})
+        self.assertFalse((REPO / "legacy_cli_archive").exists())
+
+        allowed = set(ledger["classifications"])
+        for entry in entries:
+            with self.subTest(entry=entry["id"]):
+                self.assertIn(entry["classification"], allowed)
+                self.assertTrue(entry["rationale"].strip())
+                self.assertTrue(entry["target"].strip())
+                self.assertTrue(entry["test_ids"])
+                self.assertEqual(entry["migration_status"], "inventoried")
+                for test_id in entry["test_ids"]:
+                    self.assertIn(test_id, traceability)
+                    class_name, method_name = traceability[test_id].split(".", 1)
+                    self.assertEqual(class_name, "DonorCharacterizationTests")
+                    self.assertIn(method_name, donor_test_symbols)
+                for source in entry["sources"]:
+                    file_name, separator, symbol = source.partition("::")
+                    source_path = REPO / file_name
+                    self.assertTrue(source_path.is_file(), source)
+                    if separator:
+                        self.assertIn(symbol, source_symbols(source_path), source)
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/tests/characterization/test_traceability.json b/tests/characterization/test_traceability.json
new file mode 100644
index 0000000..00fe0e6
--- /dev/null
+++ b/tests/characterization/test_traceability.json
@@ -0,0 +1,15 @@
+{
+  "CHAR-001": "DonorCharacterizationTests.test_discovery_outputs",
+  "CHAR-002": "DonorCharacterizationTests.test_image_preparation_and_hash_outputs",
+  "CHAR-003": "DonorCharacterizationTests.test_database_status_and_fts_outputs",
+  "CHAR-004": "DonorCharacterizationTests.test_caption_and_variant_outputs",
+  "CHAR-005": "DonorCharacterizationTests.test_vision_dry_run_output",
+  "CHAR-006": "DonorCharacterizationTests.test_nsfw_cache_output",
+  "CHAR-007": "DonorCharacterizationTests.test_review_html_output",
+  "CHAR-008": "DonorCharacterizationTests.test_cli_entry_point_help",
+  "CHAR-009": "DonorCharacterizationTests.test_exif_command_contracts",
+  "CHAR-010": "DonorCharacterizationTests.test_error_and_configuration_fallbacks",
+  "CHAR-011": "DonorCharacterizationTests.test_history_log_shape",
+  "CHAR-012": "DonorCharacterizationTests.test_cooperative_cancellation_contract",
+  "LEDGER-001": "DonorLedgerLintTests.test_ledger_is_complete_and_resolvable"
+}
diff --git a/tests/fixtures/characterization/expected_outputs.json b/tests/fixtures/characterization/expected_outputs.json
new file mode 100644
index 0000000..67908b5
--- /dev/null
+++ b/tests/fixtures/characterization/expected_outputs.json
@@ -0,0 +1,15 @@
+{
+  "schema_version": 1,
+  "outputs": {
+    "analyzer_recursive_discovery": ["album/UPPER.JPG", "album/nested.webp", "root.jpg", "wide.tiff"],
+    "nsfw_recursive_discovery": ["album/UPPER.JPG", "album/nested.webp", "root.jpg"],
+    "nsfw_shallow_discovery": ["root.jpg"],
+    "hash_bytes_sha1": "6a961f2bf4e8cdcd04838175b8cc98d40d729fe4",
+    "variant_base": "IMG_0001",
+    "caption": "Three adults walk beside a lake. | Tags: people, lake, summer | Mood: relaxed | Location: Como, Italy | ~2021",
+    "dry_run_tags": ["dry-run", "test"],
+    "nsfw_cache": {"a.jpg": 0.125, "b.jpg": 0.9876},
+    "ui_total": "1",
+    "ui_threshold": "0.60"
+  }
+}
diff --git a/tests/fixtures/characterization/manifest.json b/tests/fixtures/characterization/manifest.json
new file mode 100644
index 0000000..b0f5f76
--- /dev/null
+++ b/tests/fixtures/characterization/manifest.json
@@ -0,0 +1,13 @@
+{
+  "schema_version": 1,
+  "fixtures": [
+    {"id": "DISC-ROOT-JPEG", "relative_path": "root.jpg", "recipe": "rgb:12x8:10,20,30"},
+    {"id": "DISC-UPPER-JPEG", "relative_path": "album/UPPER.JPG", "recipe": "rgb:8x12:40,50,60"},
+    {"id": "DISC-NESTED-WEBP", "relative_path": "album/nested.webp", "recipe": "rgb:9x9:70,80,90"},
+    {"id": "DISC-TIFF", "relative_path": "wide.tiff", "recipe": "rgb:16x4:100,110,120"},
+    {"id": "DISC-UNSUPPORTED", "relative_path": "notes.txt", "recipe": "text:not-an-image"},
+    {"id": "HASH-BYTES", "relative_path": "hash.bin", "recipe": "bytes:photo-pipeline-donor"},
+    {"id": "IMAGE-GRADIENT", "relative_path": "gradient.png", "recipe": "gradient:40x20"},
+    {"id": "UI-ESCAPED", "relative_path": "album & one/<portrait>.jpg", "recipe": "logical-only"}
+  ]
+}
diff --git a/work_item/.work-item.yml b/work_item/.work-item.yml
index a9bc586..81cbff6 100644
--- a/work_item/.work-item.yml
+++ b/work_item/.work-item.yml
@@ -10,6 +10,7 @@ workflow:
   require_ci: false
   required_tests:
     - work_item/scripts/python -m unittest discover -s work_item/tests -v
+    - work_item/scripts/python -m unittest discover -s tests/characterization -v
 
 safety:
   max_file_bytes: 5000000
diff --git a/work_item/src/work_item/core.py b/work_item/src/work_item/core.py
index ed8f0ad..4c755e2 100644
--- a/work_item/src/work_item/core.py
+++ b/work_item/src/work_item/core.py
@@ -317,7 +317,13 @@ class GitRepo:
         return branch
 
     def changed_paths(self) -> list[Path]:
-        output = self.git("status", "--porcelain=v1", "--untracked-files=all", "-z")
+        # Porcelain's leading status column is significant (an unstaged modification
+        # starts with a space). ``git()`` strips surrounding whitespace for ordinary
+        # scalar commands, so read this machine format directly and preserve it.
+        output = self.runner.run(
+            ("git", "status", "--porcelain=v1", "--untracked-files=all", "-z"),
+            cwd=self.root,
+        ).stdout
         if not output:
             return []
         entries = output.split("\0")
diff --git a/work_item/tests/test_core.py b/work_item/tests/test_core.py
index 5dd2839..aed2476 100644
--- a/work_item/tests/test_core.py
+++ b/work_item/tests/test_core.py
@@ -354,6 +354,12 @@ class GitSafetyTests(RepositoryFixture):
         run("git", "mv", "README.md", "RENAMED.md", cwd=self.repo)
         self.assertEqual(git.changed_paths(), [Path("RENAMED.md")])
 
+    def test_changed_paths_preserves_modified_and_untracked_filenames(self) -> None:
+        git = GitRepo(self.repo, self.config, Runner())
+        (self.repo / "README.md").write_text("changed\n", encoding="utf-8")
+        (self.repo / "Another file.md").write_text("new\n", encoding="utf-8")
+        self.assertEqual(set(git.changed_paths()), {Path("README.md"), Path("Another file.md")})
+
     def test_diff_check_rejects_whitespace_errors(self) -> None:
         git = GitRepo(self.repo, self.config, Runner())
         (self.repo / "README.md").write_text("seed\ntrailing whitespace   \n", encoding="utf-8")