Rename output/ -> outputs/; add docs/CHANGELOG.md

- Generated artifacts now write to outputs/ (data/raw = inputs stay under data/;
the old data/out is legacy and gitignored).
- Update all path refs: src/trust_demo.py, streamlit_app.py, tests/run_eval.py,
.gitignore, README.md, CLAUDE.md.
- Add docs/CHANGELOG.md documenting big changes per PR since the fork's initial commit.

Verified: ruff clean; 23 tests pass; run_eval writes outputs/results.json.

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

.gitignore

@@ -1,8 +1,8 @@
 # NHS PHI safety: raw data, generated output, and the re-identification vault NEVER leave the repo
 data/*
 !data/.gitkeep
-output/*
-!output/.gitkeep
+outputs/*
+!outputs/.gitkeep
 *.vault.json
 
 # Python / packaging

CLAUDE.md

@@ -9,8 +9,8 @@ data leaves a Trust. Encode Club "Trusted Data & AI Infrastructure" hackathon; f
 python -m venv .venv; .\.venv\Scripts\Activate.ps1
 pip install -e ".[app,dev]"; python -m spacy download en_core_web_lg
 
-python tests/run_eval.py --compare --limit 300   # VERIFIABLE SIGNAL: rules vs presidio+rules -> output/results.json
-python -m src.trust_demo                          # two NHS Trusts share only de-identified data -> output/
+python tests/run_eval.py --compare --limit 300   # VERIFIABLE SIGNAL: rules vs presidio+rules -> outputs/results.json
+python -m src.trust_demo                          # two NHS Trusts share only de-identified data -> outputs/
 streamlit run streamlit_app.py                    # demo (Try-it / Metrics / Governance / Two-Trust)
 python -m pytest tests/ -v
 
@@ -29,7 +29,7 @@ python -m pytest tests/ -v
   WITHOUT spaCy/Presidio (the fallback path). snake_case / PascalCase.
 
 ## Data rules (treat the synthetic notes as if real NHS PHI)
-- `data/raw/`, `output/`, and any vault export are gitignored — never commit. Never paste note text
+- `data/raw/`, `outputs/`, and any vault export are gitignored — never commit. Never paste note text
   into prompts; point at file paths.
 - The note→patient join (`src/data.py` ground truth) is the EVAL-ONLY oracle. It must NEVER feed
   detection/transform — that is data leakage and invalidates the metric.

README.md

@@ -88,7 +88,7 @@ src/                 the package
 tests/               unit tests + run_eval.py (the evaluation CLI)
 docs/                tool card + project docs
 data/                input CSVs (gitignored)
-output/              generated artifacts: results.json, manifests (gitignored)
+outputs/              generated artifacts: results.json, manifests (gitignored)
 streamlit_app.py     the demo UI            pyproject.toml   packaging + lint/test config
 ```
 
@@ -113,8 +113,8 @@ pip install -e ".[app,dev]"
 python -m spacy download en_core_web_lg   # or en_core_web_sm for a faster, lighter run
 
 # 3) run
-python tests/run_eval.py --compare --limit 300   # reproduce the table -> output/results.json
-python -m src.trust_demo                          # two NHS Trusts share only de-identified data -> output/
+python tests/run_eval.py --compare --limit 300   # reproduce the table -> outputs/results.json
+python -m src.trust_demo                          # two NHS Trusts share only de-identified data -> outputs/
 streamlit run streamlit_app.py                    # demo: Try-it · Metrics · Governance · Two-Trust
 pytest -q                                          # unit tests
 ```

docs/CHANGELOG.md

@@ -0,0 +1,52 @@
+# Changelog
+
+Big changes since the fork of [`NoteGuard/Automatic-PII-preprocessing-tool`](https://github.com/NoteGuard/Automatic-PII-preprocessing-tool),
+grouped by pull request / milestone (newest first). Encode Club "Trusted Data & AI Infrastructure" hackathon.
+
+## Gold-RAP restructure & Hugging Face deploy — 2026-06-20
+Branches `dev/refactor-cleancode` + `feat/hf-spaces-demo`, merged to `main`.
+- Reorganised to a Gold-RAP "analysis as a product" layout: `src/` package, `tests/`, `docs/`,
+  `data/` (inputs), `outputs/` (generated artifacts).
+- Renamed package `noteguard/` → `src/`; British spelling throughout (`recognisers`).
+- Packaged via `pyproject.toml` as the single dependency source (removed `requirements.txt`).
+- CI (`.github/workflows/ci.yml`) runs `ruff` + `pytest` on every push/PR; added logging.
+- Streamlit demo moved to repo root (`streamlit_app.py`); `run_eval.py` moved under `tests/`.
+- Hugging Face Spaces (Docker SDK) deploy: `Dockerfile`, README front-matter, `DEPLOY_HF_SPACES.md`.
+- Decluttered Claude tooling (eval skill, post-edit hook, local settings) and the root changelog.
+
+## Post-PR #3 hardening — 2026-06-20
+Direct fixes on `main` after the team-briefing work.
+- Only **date-of-birth** is masked; visit / admission dates are kept (clinically useful). Human-readable
+  redaction labels (e.g. `[NHS number]`).
+- Fixed pseudonym mis-mapping caused by spaCy ORGANIZATION over-tagging.
+- Removed the roster / gazetteer feature (kept the metric honest — no circularity).
+- Consolidated the frontend on Streamlit; removed the Gradio demo.
+
+## PR #3 — Action items 1–4 from the team briefing — 2026-06-20
+- Default spaCy model `en_core_web_lg` (best UK-name recall; clinical transformer tested and rejected).
+- **Human-in-the-loop review queue**: low-confidence spans are still redacted *and* flagged for IG
+  analyst confirmation before SDE-pool admission.
+- ORGANIZATION excluded from Presidio's kept entities (it over-tagged labels like "NHS"/"GMC"); NHS
+  site names caught by a LOCATION rule instead.
+- Overlap-safe, priority-ranked span merge (precise rule entities beat broad NER spans).
+
+## PR #2 — UK entity backport — 2026-06-20
+- Added `UK_NINO`, `UK_PASSPORT`, `UK_VEHICLE_REGISTRATION` recognisers and comma-separated
+  NHS-number detection, with format-correct fake surrogates for each.
+
+## PR #1 — NoteGuard de-identification gate — 2026-06-20
+Consolidated two parallel implementations into one package.
+- One pluggable `Detector` interface: `RuleDetector` / `PresidioDetector`, degrading gracefully to
+  pure-Python rules when spaCy/Presidio are unavailable.
+- Patient-consistent pseudonymisation via a Faker(en_GB) vault + per-patient date-shift; redaction mode.
+- Evaluation harness: per-entity precision / recall / F1 **and** a measured residual-leakage rate,
+  using the dataset's structured patient/admission tables as free ground truth.
+- Two-Trust "sanitise-at-source" demo; Streamlit demo UI.
+
+## Initial NoteGuard implementation
+- Pure-Python `noteguard` package: rule recognisers (NHS Modulus-11 + context anchor, postcode, date,
+  phone, email), detection layer, redaction/pseudonymisation transforms, evaluation, pipeline.
+- Loads the NHSE synthetic clinical notes (3 CSVs joined on `person_id`/`admission_id`).
+
+## Initial commit
+- Fork of `NoteGuard/Automatic-PII-preprocessing-tool`.

src/trust_demo.py

@@ -22,7 +22,7 @@ from .evaluate import _find_all, ground_truth_spans, value_variants
 from .pipeline import Pipeline
 from .transform import PSEUDONYM, PseudonymVault
 
-OUT_DIR = Path(__file__).resolve().parent.parent / "output"
+OUT_DIR = Path(__file__).resolve().parent.parent / "outputs"
 TRUST_NAMES = {0: "Trust A (Northgate NHS Foundation Trust)", 1: "Trust B (Riverside NHS Trust)"}
 
 

streamlit_app.py

@@ -24,8 +24,8 @@ from src.evaluate import evaluate  # noqa: E402
 from src.pipeline import Pipeline  # noqa: E402
 from src.transform import PSEUDONYM, REDACTION, PseudonymVault  # noqa: E402
 
-OUT_DIR = REPO / "output"
-RESULTS = REPO / "output" / "results.json"
+OUT_DIR = REPO / "outputs"
+RESULTS = REPO / "outputs" / "results.json"
 
 ENTITY_COLORS = {
     "PERSON": "#ffd6e0", "UK_NHS": "#ffe9b3", "DATE_TIME": "#d4f4dd", "UK_POSTCODE": "#cfe8ff",

tests/run_eval.py

@@ -4,7 +4,7 @@
     python tests/run_eval.py --method pseudonym     # leakage under pseudonymisation
     python tests/run_eval.py --compare              # rules-only vs presidio+rules
 
-Writes output/results.json (consumed by the Streamlit metrics panel) and prints a summary.
+Writes outputs/results.json (consumed by the Streamlit metrics panel) and prints a summary.
 This is the pipeline's evaluation entry point; it lives under tests/ alongside the unit tests.
 """
 from __future__ import annotations
@@ -23,7 +23,7 @@ from src.detect import RuleDetector, build_detector  # noqa: E402
 from src.evaluate import EvalResult, evaluate  # noqa: E402
 from src.transform import REDACTION  # noqa: E402
 
-OUTPUT_DIR = REPO / "output"
+OUTPUT_DIR = REPO / "outputs"
 logger = logging.getLogger("noteguard.eval")
 
 
@@ -49,7 +49,7 @@ def main() -> None:
     ap.add_argument("--method", default=REDACTION, choices=["redaction", "pseudonym"])
     ap.add_argument("--no-presidio", action="store_true", help="rules only")
     ap.add_argument("--compare", action="store_true", help="rules vs presidio+rules")
-    ap.add_argument("--out", default=None, help="output JSON path (default: output/results.json)")
+    ap.add_argument("--out", default=None, help="output JSON path (default: outputs/results.json)")
     args = ap.parse_args()
 
     logger.info("loading notes (limit=%s) ...", args.limit)