Add de-identified download, remove FLock.io tech refs, add hackathon footer

- Try-it tab: download the de-identified output as JSON or CSV — a single sanitised
note, or the full sanitised dataset (only de-identified text is exported).
- Remove FLock.io as a technical reference across the UI and docs; keep the
hackathon credit only.
- Footer on every tab: "Live demo for the FLock Sovereign AI Challenge at the
Encode Vibe Coding Hackathon, hosted by Encode Hub."
- Align the event name to "Encode Vibe Coding Hackathon — FLock Sovereign AI
Challenge (Encode Hub)" across README / CLAUDE / tool_card / report / changelog.
- CHANGELOG: log this live-demo polish.

CLAUDE.md

@@ -1,7 +1,7 @@
 # NoteGuard — NHS Clinical-Note PII Sanitisation
 
 Sanitise-at-source: detect + de-identify PII in free-text NHS clinical notes so only de-identified
-data leaves a Trust. Encode Club "Trusted Data & AI Infrastructure" hackathon; fork of `NoteGuard/`.
+data leaves a Trust. Encode Vibe Coding Hackathon — FLock Sovereign AI Challenge (Encode Hub); fork of `NoteGuard/`.
 
 ## Commands
 ```bash

README.md

@@ -17,9 +17,9 @@ collaborative / federated training.
 
 > Federated learning lets institutions train without moving data. NoteGuard is the **privacy-preserving
 > on-ramp** that makes the data safe to train on in the first place — the missing layer in front of an
-> NHS Secure Data Environment / the Federated Data Platform / FLock.io.
+> NHS Secure Data Environment / the Federated Data Platform.
 
-Encode Club hackathon — *Trusted Data & AI Infrastructure*. Built on **Microsoft Presidio** + **spaCy**,
+Encode Vibe Coding Hackathon — *FLock Sovereign AI Challenge* (hosted by Encode Hub). Built on **Microsoft Presidio** + **spaCy**,
 evaluated on [NHSEDataScience/synthetic_clinical_notes](https://huggingface.co/datasets/NHSEDataScience/synthetic_clinical_notes).
 
 ## What makes this more than "just Presidio"
@@ -66,7 +66,7 @@ The rules→engine drop is the headline: it shows, with numbers, exactly what th
                  │                     patient-consistent + date-shift, vault│     (no PHI leaves)
                  └─────────────────────────────────────────────────────────┘
         same gate runs inside Trust B ──►  ┌────────────────────────────┐
-                                           │  shared de-identified pool  │ ──► federated AI / FLock.io
+                                           │  shared de-identified pool  │ ──► federated AI training
                                            └────────────────────────────┘
 ```
 

docs/CHANGELOG.md

@@ -1,7 +1,18 @@
 # 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.
+grouped by pull request / milestone (newest first). Encode Vibe Coding Hackathon — FLock Sovereign AI Challenge (hosted by Encode Hub).
+
+## Live-demo polish — 2026-06-21
+Direct UI / docs changes on `main` for the public demo.
+- **Download de-identified output** from the Try-it tab: a single sanitised note *or* the full
+  sanitised dataset, as JSON or CSV — only the de-identified text is exported (raw PHI never leaves).
+- Removed FLock.io as a technical reference throughout (UI + docs); kept only the hackathon credit.
+- Added a hackathon footer to every tab.
+- Simplified the UI copy: dropped model / algorithm names and file / command references; renamed the
+  "Metrics & Leakage" tab; 1-based note index; 50 sample notes in the browser.
+- Fixed the Streamlit `outputs/` path after the app moved to the repo root (the Metrics and Two-Trust
+  "Run" buttons read / wrote the wrong location).
 
 ## [0.0.1] — 2026-06-20 — first release (Gold-RAP restructure & Hugging Face deploy)
 Branches `dev/refactor-cleancode` + `feat/hf-spaces-demo`, merged to `main`.

docs/report.md

@@ -20,7 +20,7 @@
 ## Tier 2
 
 ### 1. Owner and responsibility
-- **1.1 Organisation:** Encode Club "Trusted Data & AI Infrastructure" hackathon team (fork of `NoteGuard/`).
+- **1.1 Organisation:** Encode Vibe Coding Hackathon team — FLock Sovereign AI Challenge (fork of `NoteGuard/`).
 - **1.2 Team:** Project contributors (see repository history / `docs/CHANGELOG.md`).
 - **1.3 Senior responsible owner:** None — prototype, not in service. An SRO would be required before deployment.
 - **1.4 External supplier involvement:** No commercial supplier. Built on open-source components
@@ -110,4 +110,4 @@ false positive). Recall and leakage are the sound headline metrics.
 
 ---
 
-*NoteGuard · Encode Club "Trusted Data & AI Infrastructure" hackathon · prototype · v0.0.1*
+*NoteGuard · Encode Vibe Coding Hackathon — FLock Sovereign AI Challenge · prototype · v0.0.1*

docs/tool_card.md

@@ -12,7 +12,7 @@
 |---|---|
 | Description | De-identification gate that detects + removes PII from free-text NHS clinical notes |
 | Type | Hybrid pipeline — pure-Python rules + Microsoft Presidio (spaCy `en_core_web_lg` NER). **No model is trained**; pre-trained components are composed. |
-| Developer | Encode Club hackathon team (fork of `NoteGuard/`) |
+| Developer | Encode Vibe Coding Hackathon team (fork of `NoteGuard/`) |
 | Status / version | Prototype · v0.0.1 |
 | Repository | github.com/chaeyoonyunakim/automatic-pii-preprocessing-tool |
 
@@ -36,7 +36,7 @@ NoteGuard is a **de-identification gate** for free-text NHS clinical notes. It d
 |---|---|---|
 | Data Wrangler / IG Analyst | Before releasing notes to research or AI teams | Cannot share raw free-text; must prove zero identifier leakage |
 | SDE Operator | At the Trust boundary ingestion point | Gate between Trust raw data and the shared pool |
-| Federated AI Platform (e.g. FLock.io) | Before each training round | Needs de-identified text; cannot inspect raw Trust data |
+| Federated AI platform | Before each training round | Needs de-identified text; cannot inspect raw Trust data |
 
 ---
 
@@ -151,11 +151,11 @@ NHS Trust (raw notes)
     │
     ▼  NHS SDE / FDP shared pool
     │
-    ▼  Federated AI  (e.g. FLock.io)
+    ▼  Federated AI
 ```
 
 Same privacy model as OpenSAFELY: *code comes to the data, data never leaves*.
 
 ---
 
-*NoteGuard · Encode Club "Trusted Data & AI Infrastructure" hackathon · internal use only*
+*NoteGuard · Encode Vibe Coding Hackathon — FLock Sovereign AI Challenge · internal use only*

src/trust_demo.py

@@ -4,7 +4,7 @@ Each Trust holds its own patients and its own re-identification vault. It
 sanitises its notes LOCALLY and contributes only the
 de-identified text + a content-free audit manifest to a shared pool. Raw notes and
 vaults never leave the Trust. This is the sanitise-at-source gate that sits in
-front of a federated SDE / FLock.io training round.
+front of a federated SDE training round.
 
     python -m src.trust_demo                 # pseudonymise, 300 notes
     python -m src.trust_demo redaction 600

streamlit_app.py

@@ -13,6 +13,7 @@ import sys
 from collections import Counter
 from pathlib import Path
 
+import pandas as pd
 import streamlit as st
 
 REPO = Path(__file__).resolve().parent
@@ -105,13 +106,13 @@ with tab_try:
         if source == "Sample note" and NOTES:
             idx = st.number_input("Note index", 1, len(NOTES), 1, step=1)
             rec = NOTES[int(idx) - 1]
-            text, person_id = rec.text, rec.person_id
+            text, person_id, note_id = rec.text, rec.person_id, rec.note_id
         else:
             text = st.text_area("Clinical note (messy free-text)", height=200,
                                 value="Pt John Smith, NHS no 943 476 5919, DOB 02/03/1981, lives SW1A 1AA. "
                                       "Admitted Manchester Royal Infirmary Ward 9. "
                                       "Reviewed by Dr Lee, GMC 1234567.")
-            person_id = "demo"
+            person_id, note_id = "demo", "pasted"
 
     if text.strip():
         result = Pipeline(detector, PseudonymVault()).sanitise(text, method, person_id)
@@ -148,6 +149,40 @@ with tab_try:
         else:
             st.success("All detections auto-confirmed (score ≥ threshold). No human review needed.", icon="✅")
 
+        st.markdown("##### 4) Download this de-identified note")
+        one = [{"note_id": note_id, "method": method, "sanitised_text": result.sanitised}]
+        d1, d2 = st.columns(2)
+        d1.download_button("⬇ Download JSON", json.dumps(one, ensure_ascii=False, indent=2),
+                           file_name="noteguard_note.json", mime="application/json",
+                           use_container_width=True)
+        d2.download_button("⬇ Download CSV", pd.DataFrame(one).to_csv(index=False),
+                           file_name="noteguard_note.csv", mime="text/csv",
+                           use_container_width=True)
+
+    st.divider()
+    with st.expander("⬇ Download the full de-identified dataset"):
+        st.caption("De-identify a batch of notes and export **only the sanitised text** — "
+                   "the original PHI never leaves the gate.")
+        ca, cb = st.columns([3, 1])
+        n_all = ca.slider("Notes to de-identify", 50, 1600, 200, step=50, key="dataset_n")
+        if cb.button("Prepare", use_container_width=True):
+            with st.spinner(f"De-identifying {n_all} notes…"):
+                pipe = Pipeline(detector, PseudonymVault())  # one vault → patient-consistent
+                rows = [{"note_id": r.note_id, "method": method,
+                         "sanitised_text": pipe.sanitise(r.text, method, r.person_id).sanitised}
+                        for r in load_notes(limit=n_all) if r.text]
+                st.session_state["dataset_rows"] = rows
+        rows = st.session_state.get("dataset_rows")
+        if rows:
+            st.success(f"{len(rows)} notes de-identified — ready to download.")
+            e1, e2 = st.columns(2)
+            e1.download_button("⬇ Download JSON", json.dumps(rows, ensure_ascii=False, indent=2),
+                               file_name="noteguard_dataset.json", mime="application/json",
+                               use_container_width=True)
+            e2.download_button("⬇ Download CSV", pd.DataFrame(rows).to_csv(index=False),
+                               file_name="noteguard_dataset.csv", mime="text/csv",
+                               use_container_width=True)
+
 # ---------------------------------------------------------------- Metrics
 with tab_metrics:
     st.markdown(
@@ -248,7 +283,7 @@ NHS Trust (raw notes)
     ▼  NHS Secure Data Environment / Federated Data Platform pool
     │   (same model as OpenSAFELY: code comes to data, data never leaves)
     │
-    ▼  Federated AI training  (e.g. FLock.io round)
+    ▼  Federated AI training
         each Trust trains locally; only model gradients are shared
 ```
     """)
@@ -283,6 +318,13 @@ with tab_trust:
             st.metric("De-identified notes", summary["shared_pool_size"])
             st.metric("Raw records shared", summary["raw_records_shared"])
             st.metric("Total residual leaks", summary["total_residual_leaks"])
-            st.caption("→ ready for federated AI / FLock.io")
+            st.caption("→ ready for federated AI training")
     else:
         st.info("Click **Run two-Trust demo** above.")
+
+# ---------------------------------------------------------------- Footer (all tabs)
+st.divider()
+st.caption(
+    "Live demo for the **FLock Sovereign AI Challenge** at the Encode Vibe Coding Hackathon, "
+    "hosted by Encode Hub."
+)