Replace conflict monitor with FinRepair formula audit

.gitignore

@@ -0,0 +1,6 @@
+.venv/
+__pycache__/
+.pytest_cache/
+.env
+local_annotations/
+local_bundle/

Dockerfile

@@ -1,20 +0,0 @@
-FROM python:3.12-slim
-
-WORKDIR /app
-
-COPY requirements.txt ./requirements.txt
-RUN pip install --no-cache-dir -r requirements.txt
-
-COPY app ./app
-COPY sources.yaml ./sources.yaml
-
-ENV PYTHONPATH=/app
-ENV SOURCES_FILE=/app/sources.yaml
-ENV DATABASE_URL=sqlite:////app/data/dev.db
-ENV ALLOWED_ORIGINS=*
-ENV PORT=7860
-
-RUN mkdir -p /app/data
-
-EXPOSE 7860
-CMD ["sh", "-c", "uvicorn app.main:app --host 0.0.0.0 --port ${PORT}"]

README.md

@@ -1,13 +1,38 @@
 ---
-title: Conflict Monitor Demo
-emoji: 📡
-colorFrom: blue
-colorTo: red
-sdk: docker
-app_port: 7860
+title: FinRepair Formula Audit
+colorFrom: gray
+colorTo: blue
+sdk: gradio
+sdk_version: 6.20.0
+python_version: 3.12
+app_file: app.py
 pinned: false
+hf_oauth: true
+hf_oauth_expiration_minutes: 43200
 ---
 
-# Conflict Monitor Demo
+# FinRepair Formula Audit
 
-Public demo web app.
+Private, identity-aware human audit interface for the frozen Seed60 formula graph
+sample. The Space contains application code only. Cases, images, and per-annotator
+records live in the private `BoomQ/FinRepair-Formula-Audit` dataset repository.
+
+## Space configuration
+
+- Secret: `HF_TOKEN`, with read/write access to the private dataset repository.
+- Variable: `FINREPAIR_DATASET_REPO=BoomQ/FinRepair-Formula-Audit`.
+- Optional variable: `ANNOTATOR_ALLOWLIST=user-a,user-b`.
+- Optional variable: `FINREPAIR_DATASET_REVISION=<commit>` to pin a data revision.
+
+Every data-loading, navigation, and saving callback checks the Hugging Face OAuth
+profile independently. Annotation files are stored as
+`annotations/<annotator>.json` and never bundled into the public Space repository.
+
+## Local development
+
+```bash
+FINREPAIR_BUNDLE_DIR=/path/to/seed60_formula_audit_v1 \
+FINREPAIR_DEV_USER=local-reviewer \
+FINREPAIR_ANNOTATION_CACHE=./local_annotations \
+python app.py
+```

app.py

@@ -0,0 +1,426 @@
+"""Gradio application for the FinRepair Seed60 formula audit."""
+
+from __future__ import annotations
+
+import html
+import os
+import shutil
+from pathlib import Path
+from typing import Any
+
+os.environ.setdefault("GRADIO_TEMP_DIR", f"/tmp/finrepair-gradio-{os.getuid()}")
+
+import gradio as gr
+
+from audit_core import (
+    CONFIDENCE_LEVELS,
+    DECISIONS,
+    ISSUE_TYPES,
+    AnnotationStore,
+    AnnotationValidationError,
+    AuthorizationError,
+    CaseBundle,
+    StorageError,
+    authorize,
+    build_annotation,
+    resolve_bundle_root,
+)
+
+
+BUNDLE = CaseBundle.load(
+    resolve_bundle_root(),
+    verify_images=os.getenv("FINREPAIR_VERIFY_IMAGES", "1") != "0",
+)
+STORE = AnnotationStore.from_environment()
+CSS_PATH = Path(__file__).with_name("styles.css")
+IMAGE_EXPORT_DIR = Path(os.environ["GRADIO_TEMP_DIR"]) / "selected-images"
+IMAGE_EXPORT_DIR.mkdir(parents=True, exist_ok=True)
+
+
+def export_image(case: dict[str, Any]) -> str:
+    source = Path(BUNDLE.image_path(case))
+    destination = IMAGE_EXPORT_DIR / f"{case['image_sha256']}.png"
+    if not destination.is_file():
+        shutil.copy2(source, destination)
+    return str(destination)
+
+
+def proposal_rows(case: dict[str, Any]) -> list[list[Any]]:
+    proposal = case["proposal"]
+    rows = [["Parent", proposal["parent"]["row_id"], proposal["parent"]["label"], "-"]]
+    rows.extend(
+        ["Child", child["row_id"], child["label"], child["weight"]]
+        for child in proposal["children"]
+    )
+    return rows
+
+
+def extracted_rows(case: dict[str, Any]) -> list[list[str]]:
+    parent_id = case["proposal"]["parent"]["row_id"]
+    child_ids = {child["row_id"] for child in case["proposal"]["children"]}
+    result = []
+    for row in case["rows"]:
+        role = "Parent" if row["row_id"] == parent_id else "Child" if row["row_id"] in child_ids else ""
+        result.append([row["row_id"], row["text"], role])
+    return result
+
+
+def row_choices(case: dict[str, Any]) -> list[tuple[str, str]]:
+    return [(f"{row['row_id']} | {row['text']}", row["row_id"]) for row in case["rows"]]
+
+
+def proposed_children(case: dict[str, Any]) -> list[list[Any]]:
+    return [[child["row_id"], child["weight"]] for child in case["proposal"]["children"]]
+
+
+def progress_html(username: str | None, document: dict[str, Any] | None) -> str:
+    completed = len((document or {}).get("annotations", {}))
+    total = len(BUNDLE.cases)
+    percent = (100 * completed / total) if total else 0
+    user_text = html.escape(username) if username else "Not signed in"
+    return (
+        '<div class="progress-shell">'
+        f'<div class="progress-copy"><strong>{completed} / {total}</strong> reviewed'
+        f'<span>{percent:.1f}%</span><span class="annotator-name">{user_text}</span></div>'
+        f'<progress max="{total}" value="{completed}"></progress></div>'
+    )
+
+
+def restore_values(case: dict[str, Any], annotation: dict[str, Any] | None) -> tuple[Any, ...]:
+    if not annotation:
+        return (
+            None,
+            [],
+            None,
+            case["proposal"]["parent"]["row_id"],
+            proposed_children(case),
+            "",
+            "",
+        )
+    correction = annotation.get("correction") or {}
+    corrected_children = correction.get("children")
+    child_table = (
+        [[child["row_id"], child["weight"]] for child in corrected_children]
+        if corrected_children
+        else proposed_children(case)
+    )
+    return (
+        annotation.get("decision"),
+        annotation.get("issues") or [],
+        annotation.get("confidence"),
+        correction.get("parent_row_id") or case["proposal"]["parent"]["row_id"],
+        child_table,
+        annotation.get("equivalence_explanation") or "",
+        annotation.get("notes") or "",
+    )
+
+
+def case_view(
+    index: int,
+    username: str,
+    document: dict[str, Any],
+    status: str = "Ready",
+) -> tuple[Any, ...]:
+    index = max(0, min(int(index), len(BUNDLE.cases) - 1))
+    case = BUNDLE.cases[index]
+    annotation = document["annotations"].get(case["audit_id"])
+    decision, issues, confidence, parent, children, equivalent, notes = restore_values(
+        case, annotation
+    )
+    reviewed = "Reviewed" if annotation else "Not reviewed"
+    case_title = (
+        f"### Case {index + 1} of {len(BUNDLE.cases)}"
+        f"\n`{case['audit_id']}` | {case['relation_size']} relation terms | {reviewed}"
+    )
+    return (
+        index,
+        progress_html(username, document),
+        f"**{status}**",
+        case_title,
+        export_image(case),
+        proposal_rows(case),
+        extracted_rows(case),
+        decision,
+        issues,
+        confidence,
+        gr.update(choices=row_choices(case), value=parent),
+        children,
+        equivalent,
+        notes,
+    )
+
+
+def locked_view(message: str) -> tuple[Any, ...]:
+    return (
+        0,
+        progress_html(None, None),
+        f"**{message}**",
+        "### Sign in required",
+        None,
+        [],
+        [],
+        None,
+        [],
+        None,
+        gr.update(choices=[], value=None),
+        [["", ""]],
+        "",
+        "",
+    )
+
+
+def load_initial(profile: gr.OAuthProfile | None) -> tuple[Any, ...]:
+    try:
+        username = authorize(profile)
+        document = STORE.load(username)
+    except (AuthorizationError, StorageError) as exc:
+        return locked_view(str(exc))
+    annotations = document["annotations"]
+    index = next(
+        (i for i, case in enumerate(BUNDLE.cases) if case["audit_id"] not in annotations),
+        0,
+    )
+    return case_view(index, username, document, "Progress loaded")
+
+
+def navigate(index: int, direction: str, profile: gr.OAuthProfile | None) -> tuple[Any, ...]:
+    try:
+        username = authorize(profile)
+        document = STORE.load(username)
+    except (AuthorizationError, StorageError) as exc:
+        return locked_view(str(exc))
+    current = int(index or 0)
+    if direction == "previous":
+        target = max(0, current - 1)
+    elif direction == "next":
+        target = min(len(BUNDLE.cases) - 1, current + 1)
+    else:
+        annotations = document["annotations"]
+        scan = list(range(current + 1, len(BUNDLE.cases))) + list(range(0, current + 1))
+        target = next(
+            (i for i in scan if BUNDLE.cases[i]["audit_id"] not in annotations),
+            current,
+        )
+    return case_view(target, username, document)
+
+
+def go_previous(index: int, profile: gr.OAuthProfile | None) -> tuple[Any, ...]:
+    return navigate(index, "previous", profile)
+
+
+def go_next(index: int, profile: gr.OAuthProfile | None) -> tuple[Any, ...]:
+    return navigate(index, "next", profile)
+
+
+def go_next_unreviewed(index: int, profile: gr.OAuthProfile | None) -> tuple[Any, ...]:
+    return navigate(index, "unreviewed", profile)
+
+
+def save_and_next(
+    index: int,
+    decision: str | None,
+    issues: list[str] | None,
+    confidence: str | None,
+    corrected_parent: str | None,
+    corrected_children: Any,
+    equivalence_explanation: str,
+    notes: str,
+    profile: gr.OAuthProfile | None,
+) -> tuple[Any, ...]:
+    try:
+        username = authorize(profile)
+        document = STORE.load(username)
+        current = max(0, min(int(index or 0), len(BUNDLE.cases) - 1))
+        case = BUNDLE.cases[current]
+        annotation = build_annotation(
+            case=case,
+            username=username,
+            decision=decision,
+            issues=issues,
+            confidence=confidence,
+            corrected_parent=corrected_parent,
+            corrected_children=corrected_children,
+            equivalence_explanation=equivalence_explanation,
+            notes=notes,
+        )
+        document["annotations"][case["audit_id"]] = annotation
+        synced = STORE.save(username, document)
+    except (AuthorizationError, AnnotationValidationError, StorageError) as exc:
+        raise gr.Error(str(exc)) from exc
+
+    annotations = document["annotations"]
+    scan = list(range(current + 1, len(BUNDLE.cases))) + list(range(0, current + 1))
+    target = next(
+        (i for i in scan if BUNDLE.cases[i]["audit_id"] not in annotations),
+        current,
+    )
+    status = "Saved and synced" if synced else "Saved to local development storage"
+    return case_view(target, username, document, status)
+
+
+def build_app() -> gr.Blocks:
+    output_components: list[gr.Component] = []
+    with gr.Blocks(
+        title="FinRepair Formula Audit",
+        fill_width=True,
+    ) as demo:
+        case_index = gr.State(0)
+
+        with gr.Row(elem_id="topbar"):
+            gr.Markdown("# FinRepair Formula Audit", elem_id="product-title")
+            auth_button = gr.LoginButton(size="sm", elem_id="auth-button")
+
+        progress = gr.HTML(progress_html(None, None), elem_id="audit-progress")
+        status = gr.Markdown("**Waiting for sign-in**", elem_id="save-status")
+        case_title = gr.Markdown("### Sign in required", elem_id="case-title")
+
+        image = gr.Image(
+            label="Financial statement",
+            type="filepath",
+            interactive=False,
+            buttons=["fullscreen"],
+            height="min(58vh, 680px)",
+            elem_id="statement-image",
+        )
+
+        with gr.Row(equal_height=False, elem_id="workbench"):
+            with gr.Column(scale=6, min_width=420):
+                candidate = gr.Dataframe(
+                    headers=["Role", "Row", "Label", "Weight"],
+                    datatype=["str", "str", "str", "str"],
+                    type="array",
+                    interactive=False,
+                    wrap=True,
+                    max_height=340,
+                    column_widths=[90, 90, "60%", 90],
+                    label="Proposed relation",
+                    buttons=["fullscreen", "copy"],
+                )
+                all_rows = gr.Dataframe(
+                    headers=["Row", "Extracted label", "Candidate role"],
+                    datatype=["str", "str", "str"],
+                    type="array",
+                    interactive=False,
+                    wrap=True,
+                    max_height=420,
+                    column_widths=[90, "70%", 120],
+                    show_search="search",
+                    label="All statement rows",
+                    buttons=["fullscreen", "copy"],
+                )
+
+            with gr.Column(scale=4, min_width=340, elem_id="annotation-panel"):
+                decision = gr.Radio(
+                    choices=list(DECISIONS),
+                    label="Decision",
+                )
+                confidence = gr.Radio(
+                    choices=list(CONFIDENCE_LEVELS),
+                    label="Confidence",
+                )
+                issues = gr.CheckboxGroup(
+                    choices=list(ISSUE_TYPES),
+                    label="Issues",
+                )
+                corrected_parent = gr.Dropdown(
+                    choices=[],
+                    label="Corrected parent",
+                    filterable=True,
+                )
+                corrected_children = gr.Dataframe(
+                    headers=["Row ID", "Weight"],
+                    datatype=["str", "number"],
+                    type="array",
+                    interactive=True,
+                    row_count=2,
+                    column_widths=["65%", "35%"],
+                    label="Corrected children",
+                )
+                equivalence = gr.Textbox(
+                    label="Equivalence explanation",
+                    lines=2,
+                    max_lines=4,
+                )
+                notes = gr.Textbox(label="Notes", lines=3, max_lines=6)
+
+        with gr.Row(elem_id="navigation"):
+            previous_button = gr.Button("Previous", variant="secondary")
+            next_button = gr.Button("Next", variant="secondary")
+            unreviewed_button = gr.Button("Next unreviewed", variant="secondary")
+            save_button = gr.Button("Save and next", variant="primary")
+
+        output_components = [
+            case_index,
+            progress,
+            status,
+            case_title,
+            image,
+            candidate,
+            all_rows,
+            decision,
+            issues,
+            confidence,
+            corrected_parent,
+            corrected_children,
+            equivalence,
+            notes,
+        ]
+
+        demo.load(
+            load_initial,
+            inputs=None,
+            outputs=output_components,
+            show_progress="minimal",
+            api_visibility="private",
+        )
+        previous_button.click(
+            go_previous,
+            inputs=[case_index],
+            outputs=output_components,
+            show_progress="minimal",
+            api_visibility="private",
+        )
+        next_button.click(
+            go_next,
+            inputs=[case_index],
+            outputs=output_components,
+            show_progress="minimal",
+            api_visibility="private",
+        )
+        unreviewed_button.click(
+            go_next_unreviewed,
+            inputs=[case_index],
+            outputs=output_components,
+            show_progress="minimal",
+            api_visibility="private",
+        )
+        save_button.click(
+            save_and_next,
+            inputs=[
+                case_index,
+                decision,
+                issues,
+                confidence,
+                corrected_parent,
+                corrected_children,
+                equivalence,
+                notes,
+            ],
+            outputs=output_components,
+            show_progress="full",
+            api_visibility="private",
+        )
+    return demo
+
+
+demo = build_app()
+
+
+if __name__ == "__main__":
+    demo.launch(
+        server_name="0.0.0.0",
+        server_port=int(os.getenv("PORT", "7860")),
+        css_paths=[str(CSS_PATH)],
+        footer_links=[],
+        show_error=True,
+    )

app/static/app.js

@@ -1,551 +0,0 @@
-const eventList = document.getElementById("eventList");
-const detail = document.getElementById("detail");
-const liveEventPanel = document.getElementById("liveEventPanel");
-const liveDetailPanel = document.getElementById("liveDetailPanel");
-const backToEventsBtn = document.getElementById("backToEventsBtn");
-const criticalList = document.getElementById("criticalList");
-const airportBoardBody = document.getElementById("airportBoardBody");
-const countryFilter = document.getElementById("countryFilter");
-const languageFilter = document.getElementById("languageFilter");
-const flightCountryFilter = document.getElementById("flightCountryFilter");
-const flightShowUnknown = document.getElementById("flightShowUnknown");
-const factcheckClaim = document.getElementById("factcheckClaim");
-const factcheckOptionalLangs = document.getElementById("factcheckOptionalLangs");
-const factcheckMaxEvidence = document.getElementById("factcheckMaxEvidence");
-const factcheckRunBtn = document.getElementById("factcheckRunBtn");
-const factcheckResult = document.getElementById("factcheckResult");
-const switchButtons = document.querySelectorAll(".switch-btn");
-let activeViewId = "liveView";
-let factcheckRecord = null;
-let factcheckSelectedLang = "all";
-
-const MIDDLE_EAST_COUNTRIES = [
-  "Bahrain",
-  "Egypt",
-  "Iran",
-  "Iraq",
-  "Israel",
-  "Jordan",
-  "Kuwait",
-  "Lebanon",
-  "Oman",
-  "Palestine",
-  "Qatar",
-  "Saudi Arabia",
-  "Syria",
-  "Turkey",
-  "UAE",
-  "Yemen",
-];
-
-const DEFAULT_LANGUAGE_OPTIONS = ["en", "ar", "fa", "he", "tr", "ckb", "ku", "ru"];
-const LANGUAGE_LABELS = {
-  en: "English",
-  ar: "Arabic",
-  fa: "Persian",
-  he: "Hebrew",
-  tr: "Turkish",
-  ckb: "Kurdish (Sorani)",
-  ku: "Kurdish (Kurmanji)",
-  ru: "Russian",
-  fr: "French",
-  it: "Italian",
-  da: "Danish",
-};
-
-const CATEGORY_LABELS = {
-  airport_flights: "airport/flights",
-  us_base_activity: "US base activity",
-  bombardment: "bombardment",
-  general: "general",
-};
-
-async function api(path, options = {}) {
-  const res = await fetch(path, options);
-  if (!res.ok) {
-    const text = await res.text();
-    throw new Error(text || `Request failed: ${res.status}`);
-  }
-  return res.json();
-}
-
-function toPlainText(input) {
-  if (!input) return "";
-  const node = document.createElement("div");
-  node.innerHTML = input;
-  return (node.textContent || node.innerText || "").trim();
-}
-
-function escapeHtml(input) {
-  if (!input) return "";
-  return String(input)
-    .replaceAll("&", "&")
-    .replaceAll("<", "&lt;")
-    .replaceAll(">", "&gt;")
-    .replaceAll('"', "&quot;")
-    .replaceAll("'", "&#39;");
-}
-
-function normalizeLang(value) {
-  if (!value) return "unknown";
-  return String(value).trim().toLowerCase() || "unknown";
-}
-
-function languageLabel(code) {
-  const normalized = normalizeLang(code);
-  if (normalized === "unknown") return "Unknown";
-  return LANGUAGE_LABELS[normalized] || normalized.toUpperCase();
-}
-
-function statusTag(status) {
-  return `<span class="tag ${status}">${status}</span>`;
-}
-
-function categoryTag(category) {
-  const label = CATEGORY_LABELS[category] || category;
-  return `<span class="tag category">${label}</span>`;
-}
-
-function signalTag(signal) {
-  return `<span class="tag signal-${signal}">${signal}</span>`;
-}
-
-function flightTag(status) {
-  return `<span class="tag flight-${status}">${status}</span>`;
-}
-
-function sourceTag(sourceType) {
-  const value = sourceType || "unknown";
-  return `<span class="tag source-${value}">${value}</span>`;
-}
-
-function trustTag(level) {
-  const value = level || "low";
-  return `<span class="tag trust-${value}">${value}</span>`;
-}
-
-function showEventListPanel() {
-  liveEventPanel.classList.remove("hidden");
-  liveDetailPanel.classList.add("hidden");
-}
-
-function showEventDetailPanel() {
-  liveEventPanel.classList.add("hidden");
-  liveDetailPanel.classList.remove("hidden");
-}
-
-function eventLine(event) {
-  return `
-    <li data-id="${event.id}">
-      <div><strong>${event.title}</strong></div>
-      <small>${sourceTag(event.source_type)} ${trustTag(event.trust_level)}</small>
-    </li>
-  `;
-}
-
-function criticalCard(item) {
-  if (!item.event_id) {
-    return `
-      <div class="critical-item">
-        <h3>${item.label}</h3>
-        <div>${signalTag(item.signal)}</div>
-        <p>No recent event for current filter.</p>
-      </div>
-    `;
-  }
-
-  return `
-    <div class="critical-item">
-      <h3>${item.label}</h3>
-      <div>${signalTag(item.signal)} ${statusTag(item.status || "unverified")}</div>
-      <p><strong>${item.title}</strong></p>
-      <p>${toPlainText(item.summary)}</p>
-      <small>${item.country || "-"} | ${new Date(item.last_seen_at).toLocaleString()} | conf=${(item.confidence ?? 0).toFixed(2)}</small>
-    </div>
-  `;
-}
-
-function setCountryOptions(countries) {
-  const list = countries && countries.length ? countries : MIDDLE_EAST_COUNTRIES;
-  const merged = ["", ...list];
-  const html = merged
-    .map((country) => {
-      const label = country || "All";
-      const value = country;
-      return `<option value="${value}">${label}</option>`;
-    })
-    .join("");
-
-  countryFilter.innerHTML = html;
-  flightCountryFilter.innerHTML = html;
-}
-
-function setLanguageOptions(languages) {
-  const list = languages && languages.length ? languages : DEFAULT_LANGUAGE_OPTIONS;
-  const merged = ["", ...list];
-  const html = merged
-    .map((lang) => {
-      const label = lang ? languageLabel(lang) : "All Languages";
-      const value = lang;
-      return `<option value="${value}">${label}</option>`;
-    })
-    .join("");
-  languageFilter.innerHTML = html;
-}
-
-async function loadMeta() {
-  try {
-    const data = await api("/api/meta");
-    setCountryOptions(data.countries || []);
-    setLanguageOptions(data.languages || []);
-  } catch (err) {
-    console.error(err);
-    setCountryOptions([]);
-    setLanguageOptions([]);
-  }
-}
-
-async function loadCritical() {
-  const country = countryFilter.value;
-  const params = new URLSearchParams();
-  if (country) params.set("country", country);
-
-  const data = await api(`/api/critical?${params.toString()}`);
-  criticalList.innerHTML = data.map(criticalCard).join("");
-}
-
-function airportRow(item) {
-  const updated = item.last_seen_at ? new Date(item.last_seen_at).toLocaleString() : "-";
-  const links = Array.isArray(item.source_links) ? item.source_links : [];
-  let source = "-";
-
-  if (links.length > 0) {
-    source = links
-      .map((entry) => {
-        if (entry.url) {
-          return `<a href="${entry.url}" target="_blank" rel="noreferrer">${entry.label}</a>`;
-        }
-        return `<span>${entry.label}</span>`;
-      })
-      .join("<br/>");
-  } else if (item.source_url) {
-    source = `<a href="${item.source_url}" target="_blank" rel="noreferrer">${item.source_note || "source"}</a>`;
-  } else if (item.source_note) {
-    source = item.source_note;
-  }
-
-  return `
-    <tr>
-      <td>${item.country}</td>
-      <td>${item.city}</td>
-      <td>${item.name}</td>
-      <td>${item.iata}</td>
-      <td>${flightTag(item.status)}</td>
-      <td>${updated}</td>
-      <td>${source}</td>
-    </tr>
-  `;
-}
-
-async function loadAirportBoard() {
-  const selected = flightCountryFilter.value || countryFilter.value;
-  const showUnknown = flightShowUnknown.checked;
-  const params = new URLSearchParams();
-  if (selected) params.set("country", selected);
-  const data = await api(`/api/airports/status?${params.toString()}`);
-  const visible = showUnknown
-    ? data
-    : data.filter((row) => row.status !== "unknown");
-  airportBoardBody.innerHTML = visible.map(airportRow).join("");
-
-  if (visible.length === 0) {
-    airportBoardBody.innerHTML = `
-      <tr>
-        <td colspan="7">${showUnknown ? "No airport status rows for current filters." : "No non-unknown airport rows for current filters."}</td>
-      </tr>
-    `;
-  }
-}
-
-async function loadEvents() {
-  const status = document.getElementById("statusFilter").value;
-  const category = document.getElementById("categoryFilter").value;
-  const country = countryFilter.value;
-  const language = languageFilter.value;
-
-  const params = new URLSearchParams();
-  if (status) params.set("status", status);
-  if (category) params.set("category", category);
-  if (country) params.set("country", country);
-  if (language) params.set("language", language);
-  params.set("limit", "100");
-
-  const data = await api(`/api/events?${params.toString()}`);
-  eventList.innerHTML = data.map(eventLine).join("");
-  showEventListPanel();
-  detail.innerHTML = "<p>Select an event to inspect supporting sources.</p>";
-
-  for (const li of eventList.querySelectorAll("li")) {
-    li.addEventListener("click", () => {
-      showEventDetailPanel();
-      loadEventDetail(li.dataset.id);
-    });
-  }
-}
-
-function renderFactcheckResult(record) {
-  factcheckRecord = record;
-  const claimStruct = record.claim_struct || {};
-  const result = record.result || {};
-  const signals = result.signals || {};
-  const languages = Array.isArray(record.languages) ? record.languages : [];
-  const evidence = Array.isArray(record.evidence) ? record.evidence : [];
-  const langCounts = {};
-  for (const row of evidence) {
-    const lang = normalizeLang(row.lang);
-    langCounts[lang] = (langCounts[lang] || 0) + 1;
-  }
-
-  const availableLangs = Object.keys(langCounts).sort();
-  if (factcheckSelectedLang !== "all" && !availableLangs.includes(factcheckSelectedLang)) {
-    factcheckSelectedLang = "all";
-  }
-
-  const filteredEvidence =
-    factcheckSelectedLang === "all"
-      ? evidence
-      : evidence.filter((row) => normalizeLang(row.lang) === factcheckSelectedLang);
-
-  const languageTags = languages
-    .map((lang) => `<span class="tag category">${escapeHtml(languageLabel(lang))}</span>`)
-    .join("");
-  const languageCards = [
-    `<button type="button" class="lang-card ${factcheckSelectedLang === "all" ? "active" : ""}" data-lang="all">All (${evidence.length})</button>`,
-    ...availableLangs.map((lang) => {
-      const count = langCounts[lang];
-      const active = factcheckSelectedLang === lang ? "active" : "";
-      return `<button type="button" class="lang-card ${active}" data-lang="${escapeHtml(lang)}">${escapeHtml(languageLabel(lang))} (${count})</button>`;
-    }),
-  ].join("");
-
-  const claimStructHtml = `
-    <div class="factcheck-grid">
-      <div><strong>actor</strong><div>${escapeHtml(claimStruct.actor || "-")}</div></div>
-      <div><strong>location</strong><div>${escapeHtml(claimStruct.location || "-")}</div></div>
-      <div><strong>time</strong><div>${escapeHtml(claimStruct.time || "-")}</div></div>
-      <div><strong>action</strong><div>${escapeHtml(claimStruct.action || "-")}</div></div>
-      <div><strong>object</strong><div>${escapeHtml(claimStruct.object || "-")}</div></div>
-      <div><strong>casualty</strong><div>${escapeHtml(claimStruct.casualty || "-")}</div></div>
-    </div>
-  `;
-
-  const evidenceRows = filteredEvidence
-    .map((row) => {
-      const when = row.published_at ? new Date(row.published_at).toLocaleString() : "-";
-      const title = escapeHtml(row.title || "-");
-      const snippet = escapeHtml(row.snippet || "-");
-      const lang = escapeHtml(languageLabel(row.lang || "unknown"));
-      const sourceType = escapeHtml(row.source_type || "news");
-      const stance = escapeHtml(row.stance || "unclear");
-      const url = escapeHtml(row.url || "#");
-      return `
-      <tr>
-        <td>${lang}</td>
-        <td>${sourceType}</td>
-        <td>${stance}</td>
-        <td>${title}<div class="muted">${snippet}</div></td>
-        <td>${when}</td>
-        <td><a href="${url}" target="_blank" rel="noreferrer">source</a></td>
-      </tr>
-    `;
-    })
-    .join("");
-
-  const evidenceTable = evidenceRows
-    ? `
-      <div class="table-wrap">
-        <table>
-          <thead>
-            <tr>
-              <th>Lang</th>
-              <th>Source</th>
-              <th>Stance</th>
-              <th>Evidence</th>
-              <th>Published</th>
-              <th>Link</th>
-            </tr>
-          </thead>
-          <tbody>${evidenceRows}</tbody>
-        </table>
-      </div>
-    `
-    : "<p>No matching evidence found for the selected language in current local dataset.</p>";
-
-  factcheckResult.innerHTML = `
-    <div class="factcheck-summary">
-      <p><strong>Claim ID:</strong> ${escapeHtml(record.claim_id || "-")}</p>
-      <p><strong>Claim:</strong> ${escapeHtml(record.claim_text || "-")}</p>
-      <p><strong>Languages:</strong> ${languageTags || "-"}</p>
-      ${claimStructHtml}
-      <p>
-        <strong>Credibility:</strong> ${(result.credibility_score ?? 0).toFixed(3)}
-        ${statusTag(result.verification_level || "unverified")}
-      </p>
-      <p><strong>Signals:</strong>
-        source_diversity=${signals.source_diversity ?? 0},
-        language_diversity=${signals.language_diversity ?? 0},
-        time_geo_consistency=${signals.time_geo_consistency ?? 0},
-        contradiction_count=${signals.contradiction_count ?? 0}
-      </p>
-      <p>${escapeHtml(result.notes || "")}</p>
-    </div>
-    <h3>Evidence by Original Language</h3>
-    <div class="lang-cards">${languageCards}</div>
-    <h3>Evidence</h3>
-    ${evidenceTable}
-  `;
-
-  for (const btn of factcheckResult.querySelectorAll(".lang-card")) {
-    btn.addEventListener("click", () => {
-      factcheckSelectedLang = btn.dataset.lang || "all";
-      renderFactcheckResult(factcheckRecord);
-    });
-  }
-}
-
-async function runFactcheck() {
-  const claimText = (factcheckClaim.value || "").trim();
-  if (claimText.length < 5) {
-    factcheckResult.innerHTML = "<p>Please enter a claim with at least 5 characters.</p>";
-    return;
-  }
-
-  const payload = {
-    claim_text: claimText,
-    include_optional_languages: factcheckOptionalLangs.checked,
-    max_evidence: Number(factcheckMaxEvidence.value || 8),
-  };
-
-  factcheckRunBtn.disabled = true;
-  factcheckRunBtn.textContent = "Running...";
-  try {
-    factcheckSelectedLang = "all";
-    const data = await api("/api/factcheck", {
-      method: "POST",
-      headers: { "Content-Type": "application/json" },
-      body: JSON.stringify(payload),
-    });
-    renderFactcheckResult(data);
-  } catch (err) {
-    factcheckResult.innerHTML = `<p>Error: ${escapeHtml(err.message)}</p>`;
-  } finally {
-    factcheckRunBtn.disabled = false;
-    factcheckRunBtn.textContent = "Run Fact Check";
-  }
-}
-
-async function loadEventDetail(id) {
-  const data = await api(`/api/events/${id}`);
-  const evidenceHtml = data.evidence
-    .map(
-      (e) => `
-      <div class="evidence-item">
-        <div>${statusTag(e.stance === "supports" ? "confirmed" : "conflicting")} <strong>${e.source_name}</strong> (${e.source_priority})</div>
-        <div>${new Date(e.published_at).toLocaleString()} | score=${(e.final_score ?? 0).toFixed(2)}</div>
-        <div>${e.title}</div>
-        <a href="${e.source_url}" target="_blank" rel="noreferrer">source link</a>
-      </div>
-    `
-    )
-    .join("");
-
-  detail.innerHTML = `
-    <h3>${data.title}</h3>
-    <p>${statusTag(data.status)} ${categoryTag(data.category)} ${toPlainText(data.summary)}</p>
-    <p><small>${data.country || "-"} | ${data.actor || "-"} | confidence=${data.confidence.toFixed(2)}</small></p>
-    ${evidenceHtml || "<p>No evidence yet.</p>"}
-  `;
-}
-
-async function bindActions() {
-  document.getElementById("refreshBtn").addEventListener("click", refreshActiveView);
-  countryFilter.addEventListener("change", refreshActiveView);
-  document.getElementById("statusFilter").addEventListener("change", () => {
-    if (activeViewId === "liveView") loadEvents();
-  });
-  document.getElementById("categoryFilter").addEventListener("change", () => {
-    if (activeViewId === "liveView") loadEvents();
-  });
-  languageFilter.addEventListener("change", () => {
-    if (activeViewId === "liveView") loadEvents();
-  });
-  flightCountryFilter.addEventListener("change", () => {
-    if (activeViewId === "flightView") loadAirportBoard();
-  });
-  flightShowUnknown.addEventListener("change", () => {
-    if (activeViewId === "flightView") loadAirportBoard();
-  });
-  backToEventsBtn.addEventListener("click", showEventListPanel);
-  factcheckRunBtn.addEventListener("click", runFactcheck);
-
-  document.getElementById("bootstrapBtn").addEventListener("click", async () => {
-    await api("/api/bootstrap", { method: "POST" });
-    await refreshActiveView();
-  });
-
-  document.getElementById("ingestBtn").addEventListener("click", async () => {
-    await api("/api/ingest/run", { method: "POST" });
-    await refreshActiveView();
-  });
-
-  for (const btn of switchButtons) {
-    btn.addEventListener("click", () => activateView(btn.dataset.view));
-  }
-
-  await loadMeta();
-  flightShowUnknown.checked = false;
-  flightCountryFilter.value = countryFilter.value;
-  await activateView(activeViewId);
-}
-
-function activateButton(viewId) {
-  for (const button of switchButtons) {
-    button.classList.toggle("active", button.dataset.view === viewId);
-  }
-}
-
-function activateSection(viewId) {
-  for (const view of document.querySelectorAll(".view")) {
-    view.classList.toggle("active", view.id === viewId);
-  }
-}
-
-async function activateView(viewId) {
-  activeViewId = viewId;
-  activateButton(viewId);
-  activateSection(viewId);
-  await refreshActiveView();
-}
-
-async function refreshActiveView() {
-  if (activeViewId === "flightView") {
-    await loadAirportBoard();
-    return;
-  }
-
-  if (activeViewId === "signalView") {
-    await loadCritical();
-    return;
-  }
-
-  if (activeViewId === "factcheckView") {
-    return;
-  }
-
-  await loadEvents();
-}
-
-bindActions().catch((err) => {
-  console.error(err);
-  detail.innerHTML = `<p>Error: ${err.message}</p>`;
-});

app/static/styles.css

@@ -1,521 +0,0 @@
-:root {
-  --bg: #fcf8ef;
-  --ink: #1f2937;
-  --primary: #0f766e;
-  --accent: #b45309;
-  --panel: rgba(255, 255, 255, 0.82);
-  --border: rgba(15, 118, 110, 0.25);
-}
-
-* {
-  box-sizing: border-box;
-}
-
-body {
-  margin: 0;
-  font-family: "IBM Plex Sans", "Helvetica Neue", sans-serif;
-  color: var(--ink);
-  background: radial-gradient(circle at 10% 10%, #fef3c7 0%, var(--bg) 55%);
-  min-height: 100vh;
-  position: relative;
-}
-
-.bg-shape {
-  position: fixed;
-  z-index: -1;
-  border-radius: 999px;
-  filter: blur(40px);
-  opacity: 0.5;
-}
-
-.bg-a {
-  width: 260px;
-  height: 260px;
-  background: #14b8a6;
-  top: 40px;
-  right: 10%;
-}
-
-.bg-b {
-  width: 320px;
-  height: 320px;
-  background: #f59e0b;
-  bottom: 40px;
-  left: 5%;
-}
-
-.hero,
-main {
-  width: min(1100px, 92vw);
-  margin: 0 auto;
-}
-
-.hero {
-  padding: 32px 0 18px;
-}
-
-.hero h1 {
-  margin: 0;
-  font-size: clamp(1.8rem, 4vw, 2.8rem);
-}
-
-.actions {
-  display: flex;
-  gap: 10px;
-  margin-top: 14px;
-}
-
-.top-actions {
-  align-items: end;
-  flex-wrap: wrap;
-}
-
-.top-actions label {
-  display: flex;
-  flex-direction: column;
-  gap: 6px;
-  font-size: 0.88rem;
-}
-
-.top-actions select {
-  border: 1px solid #d1d5db;
-  border-radius: 8px;
-  padding: 7px;
-  min-width: 160px;
-}
-
-.view-switch {
-  margin-top: 14px;
-  display: flex;
-  gap: 8px;
-  flex-wrap: wrap;
-}
-
-.switch-btn {
-  border: 1px solid var(--border);
-  border-radius: 999px;
-  padding: 7px 12px;
-  background: rgba(255, 255, 255, 0.85);
-}
-
-.switch-btn.active {
-  background: var(--primary);
-  color: #fff;
-  border-color: var(--primary);
-}
-
-button {
-  border: 1px solid var(--border);
-  background: white;
-  color: var(--ink);
-  border-radius: 10px;
-  padding: 8px 12px;
-  cursor: pointer;
-}
-
-button:hover {
-  border-color: var(--primary);
-}
-
-.filters {
-  display: grid;
-  grid-template-columns: repeat(auto-fit, minmax(180px, 1fr));
-  gap: 10px;
-  background: var(--panel);
-  border: 1px solid var(--border);
-  border-radius: 12px;
-  padding: 12px;
-}
-
-.view {
-  display: none;
-}
-
-.view.active {
-  display: block;
-}
-
-.critical {
-  background: var(--panel);
-  border: 1px solid var(--border);
-  border-radius: 12px;
-  padding: 12px;
-  margin-bottom: 12px;
-}
-
-.flight-board {
-  background: var(--panel);
-  border: 1px solid var(--border);
-  border-radius: 12px;
-  padding: 12px;
-  margin-bottom: 12px;
-}
-
-.factcheck {
-  background: var(--panel);
-  border: 1px solid var(--border);
-  border-radius: 12px;
-  padding: 12px;
-  margin-bottom: 12px;
-}
-
-.critical .note {
-  margin: 6px 0 10px;
-  font-size: 0.9rem;
-}
-
-.flight-board .note {
-  margin: 6px 0 10px;
-  font-size: 0.9rem;
-}
-
-.factcheck .note {
-  margin: 6px 0 10px;
-  font-size: 0.9rem;
-}
-
-.flight-controls {
-  display: flex;
-  gap: 12px;
-  align-items: end;
-  flex-wrap: wrap;
-  margin-bottom: 10px;
-}
-
-.flight-controls label {
-  display: flex;
-  flex-direction: column;
-  gap: 6px;
-  font-size: 0.88rem;
-}
-
-.flight-controls select {
-  border: 1px solid #d1d5db;
-  border-radius: 8px;
-  padding: 7px;
-  min-width: 180px;
-}
-
-.flight-controls .check {
-  flex-direction: row;
-  align-items: center;
-  gap: 6px;
-  margin-bottom: 2px;
-}
-
-.factcheck-form label {
-  display: flex;
-  flex-direction: column;
-  gap: 6px;
-  font-size: 0.88rem;
-}
-
-.factcheck-form textarea,
-.factcheck-controls select {
-  border: 1px solid #d1d5db;
-  border-radius: 8px;
-  padding: 9px;
-  width: 100%;
-}
-
-.factcheck-controls {
-  margin-top: 10px;
-  display: flex;
-  gap: 12px;
-  flex-wrap: wrap;
-  align-items: end;
-}
-
-.factcheck-controls .check {
-  flex-direction: row;
-  align-items: center;
-  gap: 6px;
-  margin-bottom: 4px;
-}
-
-.factcheck-result {
-  margin-top: 14px;
-  border-top: 1px solid #e5e7eb;
-  padding-top: 12px;
-}
-
-.factcheck-grid {
-  margin: 10px 0;
-  display: grid;
-  grid-template-columns: repeat(auto-fit, minmax(180px, 1fr));
-  gap: 8px;
-}
-
-.factcheck-grid > div {
-  border: 1px solid #e5e7eb;
-  border-radius: 8px;
-  background: rgba(255, 255, 255, 0.7);
-  padding: 8px;
-}
-
-.lang-cards {
-  display: flex;
-  flex-wrap: wrap;
-  gap: 8px;
-  margin: 8px 0 10px;
-}
-
-.lang-card {
-  border: 1px solid #cbd5e1;
-  background: #ffffff;
-  color: #1f2937;
-  border-radius: 999px;
-  padding: 6px 10px;
-  font-size: 0.82rem;
-  cursor: pointer;
-}
-
-.lang-card:hover {
-  border-color: var(--primary);
-}
-
-.lang-card.active {
-  background: var(--primary);
-  color: #fff;
-  border-color: var(--primary);
-}
-
-.muted {
-  color: #64748b;
-  font-size: 0.84rem;
-  margin-top: 4px;
-}
-
-.table-wrap {
-  overflow: auto;
-}
-
-table {
-  width: 100%;
-  border-collapse: collapse;
-}
-
-th,
-td {
-  text-align: left;
-  border-bottom: 1px solid #e5e7eb;
-  padding: 8px;
-  vertical-align: top;
-  font-size: 0.9rem;
-}
-
-th {
-  font-size: 0.78rem;
-  letter-spacing: 0.03em;
-  text-transform: uppercase;
-  color: #334155;
-}
-
-.critical-list {
-  display: grid;
-  grid-template-columns: repeat(auto-fit, minmax(220px, 1fr));
-  gap: 10px;
-}
-
-.critical-item {
-  border: 1px solid #e5e7eb;
-  border-radius: 10px;
-  padding: 10px;
-  background: rgba(255, 255, 255, 0.75);
-}
-
-.critical-item h3 {
-  margin: 0 0 6px;
-  font-size: 1rem;
-}
-
-.critical-item p {
-  margin: 6px 0;
-}
-
-.filters label {
-  display: flex;
-  flex-direction: column;
-  gap: 6px;
-  font-size: 0.88rem;
-}
-
-.filters input,
-.filters select {
-  border: 1px solid #d1d5db;
-  border-radius: 8px;
-  padding: 7px;
-}
-
-.live-stage {
-  margin: 16px 0 30px;
-}
-
-.live-stage article {
-  background: var(--panel);
-  border: 1px solid var(--border);
-  border-radius: 12px;
-  padding: 12px;
-  backdrop-filter: blur(2px);
-}
-
-.detail-head {
-  margin-bottom: 10px;
-}
-
-.hidden {
-  display: none;
-}
-
-.grid {
-  display: grid;
-  grid-template-columns: 1fr;
-  gap: 16px;
-  margin: 16px 0 30px;
-}
-
-@media (min-width: 980px) {
-  .grid {
-    grid-template-columns: 1fr 1fr;
-  }
-}
-
-article {
-  background: var(--panel);
-  border: 1px solid var(--border);
-  border-radius: 12px;
-  padding: 12px;
-  backdrop-filter: blur(2px);
-}
-
-.event-list {
-  list-style: none;
-  margin: 0;
-  padding: 0;
-  display: flex;
-  flex-direction: column;
-  gap: 9px;
-}
-
-.event-list li {
-  border: 1px solid #e5e7eb;
-  border-radius: 10px;
-  padding: 10px;
-  cursor: pointer;
-}
-
-.event-list li:hover {
-  border-color: var(--primary);
-}
-
-.tag {
-  display: inline-block;
-  font-size: 0.75rem;
-  padding: 2px 8px;
-  border-radius: 999px;
-  margin-right: 4px;
-}
-
-.tag.confirmed {
-  background: #d1fae5;
-}
-
-.tag.unverified {
-  background: #fee2e2;
-}
-
-.tag.conflicting {
-  background: #fef3c7;
-}
-
-.tag.official {
-  background: #d1fae5;
-}
-
-.tag.cross_verified {
-  background: #dbeafe;
-}
-
-.tag.partial {
-  background: #fef3c7;
-}
-
-.tag.category {
-  background: #dbeafe;
-}
-
-.tag.signal-open {
-  background: #d1fae5;
-}
-
-.tag.signal-disrupted,
-.tag.signal-under_attack,
-.tag.signal-active {
-  background: #fee2e2;
-}
-
-.tag.signal-monitoring,
-.tag.signal-heightened,
-.tag.signal-de_escalating {
-  background: #fef3c7;
-}
-
-.tag.flight-open {
-  background: #d1fae5;
-}
-
-.tag.flight-limited {
-  background: #fef3c7;
-}
-
-.tag.flight-closed {
-  background: #fee2e2;
-}
-
-.tag.flight-unknown {
-  background: #e5e7eb;
-}
-
-.tag.source-official {
-  background: #dcfce7;
-}
-
-.tag.source-news {
-  background: #dbeafe;
-}
-
-.tag.source-social {
-  background: #fef3c7;
-}
-
-.tag.source-notam {
-  background: #fae8ff;
-}
-
-.tag.source-unknown {
-  background: #e5e7eb;
-}
-
-.tag.trust-high {
-  background: #22c55e;
-  color: #fff;
-}
-
-.tag.trust-medium {
-  background: #f59e0b;
-  color: #fff;
-}
-
-.tag.trust-low {
-  background: #ef4444;
-  color: #fff;
-}
-
-.detail .evidence-item {
-  border-top: 1px solid #e5e7eb;
-  padding: 8px 0;
-}
-
-.detail a {
-  color: #0f766e;
-}

app/templates/index.html

@@ -1,163 +0,0 @@
-<!DOCTYPE html>
-<html lang="en">
-  <head>
-    <meta charset="UTF-8" />
-    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
-    <title>Conflict Monitor v0.1</title>
-    <link rel="stylesheet" href="/static/styles.css" />
-  </head>
-  <body>
-    <div class="bg-shape bg-a"></div>
-    <div class="bg-shape bg-b"></div>
-
-    <header class="hero">
-      <h1>Conflict Monitor v0.1</h1>
-      <p>Live conflict intelligence with focused operational views.</p>
-      <div class="actions top-actions">
-        <label>
-          Country
-          <select id="countryFilter"></select>
-        </label>
-        <button id="refreshBtn">Refresh</button>
-        <button id="bootstrapBtn">Load Sources</button>
-        <button id="ingestBtn">Run Ingestion</button>
-      </div>
-      <div class="view-switch" role="tablist" aria-label="Main views">
-        <button class="switch-btn active" data-view="liveView" type="button">Live Feed</button>
-        <button class="switch-btn" data-view="flightView" type="button">Flight Board</button>
-        <button class="switch-btn" data-view="signalView" type="button">Safety Signals</button>
-        <button class="switch-btn" data-view="factcheckView" type="button">Fact Check</button>
-      </div>
-    </header>
-
-    <main>
-      <section class="view active" id="liveView">
-        <section class="filters">
-          <label>
-            Status
-            <select id="statusFilter">
-              <option value="">All</option>
-              <option value="confirmed">confirmed</option>
-              <option value="unverified">unverified</option>
-              <option value="conflicting">conflicting</option>
-            </select>
-          </label>
-          <label>
-            Focus
-            <select id="categoryFilter">
-              <option value="">All</option>
-              <option value="airport_flights">airport / flights</option>
-              <option value="us_base_activity">US base activity</option>
-              <option value="bombardment">bombardment</option>
-              <option value="general">general</option>
-            </select>
-          </label>
-          <label>
-            Language
-            <select id="languageFilter">
-              <option value="">All</option>
-            </select>
-          </label>
-        </section>
-
-        <section class="live-stage">
-          <article id="liveEventPanel">
-            <h2>Events</h2>
-            <ul id="eventList" class="event-list"></ul>
-          </article>
-
-          <article id="liveDetailPanel" class="hidden">
-            <div class="detail-head">
-              <button id="backToEventsBtn" type="button">Back to Events</button>
-            </div>
-            <h2>Evidence Chain</h2>
-            <div id="detail" class="detail">
-              <p>Select an event to inspect supporting sources.</p>
-            </div>
-          </article>
-        </section>
-      </section>
-
-      <section class="view" id="flightView">
-        <section class="flight-board" id="flightBoard">
-          <h2>Middle East Airport Flight Status</h2>
-          <p class="note">Single board view for airport operations: open / limited / closed / unknown.</p>
-          <div class="flight-controls">
-            <label>
-              Flight Country
-              <select id="flightCountryFilter"></select>
-            </label>
-            <label class="check">
-              <input id="flightShowUnknown" type="checkbox" />
-              Show Unknown
-            </label>
-          </div>
-          <div class="table-wrap">
-            <table>
-              <thead>
-                <tr>
-                  <th>Country</th>
-                  <th>City</th>
-                  <th>Airport</th>
-                  <th>IATA</th>
-                  <th>Status</th>
-                  <th>Last Update</th>
-                  <th>Source</th>
-                </tr>
-              </thead>
-              <tbody id="airportBoardBody"></tbody>
-            </table>
-          </div>
-        </section>
-      </section>
-
-      <section class="view" id="signalView">
-        <section class="critical" id="criticalPanel">
-          <h2>Critical Safety Signals</h2>
-          <p class="note">
-            Use this as a monitoring aid. For immediate safety decisions, confirm with official
-            airline, embassy, and civil-defense channels.
-          </p>
-          <div id="criticalList" class="critical-list"></div>
-        </section>
-      </section>
-
-      <section class="view" id="factcheckView">
-        <section class="factcheck" id="factcheckPanel">
-          <h2>Cross-Lingual Fact Check</h2>
-          <div class="factcheck-form">
-            <label>
-              Claim
-              <textarea
-                id="factcheckClaim"
-                rows="4"
-                placeholder="Example: Flights at Beirut airport were suspended after overnight strikes."
-              ></textarea>
-            </label>
-            <div class="factcheck-controls">
-              <label class="check">
-                <input id="factcheckOptionalLangs" type="checkbox" />
-                Include optional languages (ckb, ku, ru)
-              </label>
-              <label>
-                Max Evidence
-                <select id="factcheckMaxEvidence">
-                  <option value="5">5</option>
-                  <option value="8" selected>8</option>
-                  <option value="12">12</option>
-                  <option value="20">20</option>
-                </select>
-              </label>
-              <button id="factcheckRunBtn" type="button">Run Fact Check</button>
-            </div>
-          </div>
-          <div id="factcheckResult" class="factcheck-result">
-            <p>Enter a claim and run the pipeline to generate a structured fact-check record.</p>
-          </div>
-        </section>
-      </section>
-    </main>
-
-    <script src="/static/app.js"></script>
-  </body>
-</html>

audit_core.py

@@ -0,0 +1,384 @@
+"""Core data, validation, authorization, and persistence for the audit Space."""
+
+from __future__ import annotations
+
+import hashlib
+import json
+import math
+import os
+import re
+import threading
+from dataclasses import dataclass
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any, Iterable
+
+from huggingface_hub import HfApi, hf_hub_download, snapshot_download
+from huggingface_hub.errors import EntryNotFoundError, HfHubHTTPError
+
+
+ANNOTATION_SCHEMA_VERSION = "1.0"
+DECISIONS = ("Exact", "Equivalent", "Incorrect", "Unclear")
+CONFIDENCE_LEVELS = ("High", "Medium", "Low")
+ISSUE_TYPES = (
+    "Wrong parent",
+    "Missing child",
+    "Extra child",
+    "Wrong sign or weight",
+    "Ambiguous decomposition",
+    "Table crop or row text issue",
+    "Other",
+)
+
+
+class AuthorizationError(ValueError):
+    """Raised when an annotation callback has no authorized identity."""
+
+
+class AnnotationValidationError(ValueError):
+    """Raised when a submitted annotation is incomplete or malformed."""
+
+
+class StorageError(RuntimeError):
+    """Raised when durable annotation persistence fails."""
+
+
+def utc_now() -> str:
+    return datetime.now(timezone.utc).isoformat()
+
+
+def sha256_file(path: Path) -> str:
+    digest = hashlib.sha256()
+    with path.open("rb") as handle:
+        for chunk in iter(lambda: handle.read(1024 * 1024), b""):
+            digest.update(chunk)
+    return digest.hexdigest()
+
+
+def _read_jsonl(path: Path) -> list[dict[str, Any]]:
+    records: list[dict[str, Any]] = []
+    with path.open("r", encoding="utf-8") as handle:
+        for line_number, line in enumerate(handle, 1):
+            if not line.strip():
+                continue
+            try:
+                records.append(json.loads(line))
+            except json.JSONDecodeError as exc:
+                raise ValueError(f"Invalid JSON at {path}:{line_number}") from exc
+    return records
+
+
+@dataclass(frozen=True)
+class CaseBundle:
+    root: Path
+    cases: tuple[dict[str, Any], ...]
+    by_id: dict[str, dict[str, Any]]
+
+    @classmethod
+    def load(cls, root: str | Path, verify_images: bool = True) -> "CaseBundle":
+        root_path = Path(root).expanduser().resolve()
+        case_path = root_path / "cases.jsonl"
+        lock_path = root_path / "LOCK.json"
+        if not case_path.is_file() or not lock_path.is_file():
+            raise ValueError(f"Incomplete audit bundle: {root_path}")
+
+        lock = json.loads(lock_path.read_text(encoding="utf-8"))
+        expected_cases_hash = lock.get("artifacts", {}).get("cases.jsonl")
+        if not expected_cases_hash or sha256_file(case_path) != expected_cases_hash:
+            raise ValueError("The frozen cases.jsonl hash does not match LOCK.json")
+
+        cases = sorted(_read_jsonl(case_path), key=lambda row: int(row["order"]))
+        expected_count = int(lock.get("image_count", 0))
+        if not cases or len(cases) != expected_count:
+            raise ValueError(
+                f"Bundle case count mismatch: found {len(cases)}, expected {expected_count}"
+            )
+
+        by_id: dict[str, dict[str, Any]] = {}
+        orders: set[int] = set()
+        for case in cases:
+            audit_id = str(case.get("audit_id", ""))
+            order = int(case.get("order", 0))
+            if not audit_id or audit_id in by_id or order in orders:
+                raise ValueError(f"Duplicate or empty case identity: {audit_id!r}/{order}")
+            rows = {str(row["row_id"]) for row in case.get("rows", [])}
+            parent = case.get("proposal", {}).get("parent", {}).get("row_id")
+            children = case.get("proposal", {}).get("children", [])
+            if parent not in rows or not children:
+                raise ValueError(f"Malformed proposal in {audit_id}")
+            if any(child.get("row_id") not in rows for child in children):
+                raise ValueError(f"Proposal references an absent row in {audit_id}")
+            image_path = root_path / str(case.get("image_file", ""))
+            if not image_path.is_file():
+                raise ValueError(f"Missing image for {audit_id}: {image_path}")
+            if verify_images and sha256_file(image_path) != case.get("image_sha256"):
+                raise ValueError(f"Image hash mismatch for {audit_id}")
+            by_id[audit_id] = case
+            orders.add(order)
+
+        if orders != set(range(1, len(cases) + 1)):
+            raise ValueError("Case order must be contiguous and one-indexed")
+        return cls(root=root_path, cases=tuple(cases), by_id=by_id)
+
+    def image_path(self, case: dict[str, Any]) -> str:
+        return str((self.root / str(case["image_file"])).resolve())
+
+
+def resolve_bundle_root() -> Path:
+    local_root = os.getenv("FINREPAIR_BUNDLE_DIR", "").strip()
+    if local_root:
+        return Path(local_root).expanduser().resolve()
+
+    repo_id = os.getenv("FINREPAIR_DATASET_REPO", "").strip()
+    token = os.getenv("HF_TOKEN", "").strip()
+    if not repo_id or not token:
+        raise RuntimeError(
+            "Set FINREPAIR_BUNDLE_DIR locally, or configure both "
+            "FINREPAIR_DATASET_REPO and HF_TOKEN in the Space."
+        )
+    revision = os.getenv("FINREPAIR_DATASET_REVISION", "").strip() or None
+    path = snapshot_download(
+        repo_id=repo_id,
+        repo_type="dataset",
+        revision=revision,
+        token=token,
+        allow_patterns=("cases.jsonl", "LOCK.json", "summary.json", "images/*.png"),
+    )
+    return Path(path).resolve()
+
+
+def profile_username(profile: Any | None) -> str | None:
+    if profile is not None:
+        for field in ("preferred_username", "username", "name"):
+            value = profile.get(field) if isinstance(profile, dict) else getattr(profile, field, None)
+            if isinstance(value, str) and value.strip():
+                return value.strip()
+    dev_user = os.getenv("FINREPAIR_DEV_USER", "").strip()
+    return dev_user or None
+
+
+def authorize(profile: Any | None) -> str:
+    username = profile_username(profile)
+    if not username:
+        raise AuthorizationError("Sign in with Hugging Face to access audit cases.")
+    allowlist = {
+        item.strip().casefold()
+        for item in os.getenv("ANNOTATOR_ALLOWLIST", "").split(",")
+        if item.strip()
+    }
+    if allowlist and username.casefold() not in allowlist:
+        raise AuthorizationError(f"Hugging Face user {username!r} is not an authorized annotator.")
+    return username
+
+
+def safe_user_key(username: str) -> str:
+    slug = re.sub(r"[^a-z0-9._-]+", "-", username.casefold()).strip("-._") or "annotator"
+    suffix = hashlib.sha256(username.casefold().encode("utf-8")).hexdigest()[:10]
+    return f"{slug[:60]}-{suffix}"
+
+
+def _table_rows(value: Any) -> Iterable[list[Any]]:
+    if value is None:
+        return []
+    if hasattr(value, "values") and hasattr(value.values, "tolist"):
+        return value.values.tolist()
+    if isinstance(value, dict):
+        columns = list(value.values())
+        return [list(row) for row in zip(*columns)] if columns else []
+    if isinstance(value, (list, tuple)):
+        return value
+    raise AnnotationValidationError("Corrected children must be a two-column table.")
+
+
+def normalize_children(value: Any, valid_row_ids: set[str]) -> list[dict[str, int | float | str]]:
+    children: list[dict[str, int | float | str]] = []
+    seen: set[str] = set()
+    for raw in _table_rows(value):
+        row = list(raw) if isinstance(raw, (list, tuple)) else [raw]
+        row_id = str(row[0]).strip() if row and row[0] is not None else ""
+        raw_weight = row[1] if len(row) > 1 else None
+        if not row_id and (raw_weight is None or str(raw_weight).strip() == ""):
+            continue
+        if row_id not in valid_row_ids:
+            raise AnnotationValidationError(f"Unknown corrected child row: {row_id or '<blank>'}")
+        if row_id in seen:
+            raise AnnotationValidationError(f"Corrected child row is duplicated: {row_id}")
+        try:
+            weight = float(raw_weight)
+        except (TypeError, ValueError) as exc:
+            raise AnnotationValidationError(f"Weight for {row_id} must be numeric.") from exc
+        if not math.isfinite(weight) or weight == 0:
+            raise AnnotationValidationError(f"Weight for {row_id} must be finite and non-zero.")
+        normalized_weight: int | float = int(weight) if weight.is_integer() else weight
+        children.append({"row_id": row_id, "weight": normalized_weight})
+        seen.add(row_id)
+    return children
+
+
+def build_annotation(
+    case: dict[str, Any],
+    username: str,
+    decision: str | None,
+    issues: list[str] | None,
+    confidence: str | None,
+    corrected_parent: str | None,
+    corrected_children: Any,
+    equivalence_explanation: str | None,
+    notes: str | None,
+) -> dict[str, Any]:
+    if decision not in DECISIONS:
+        raise AnnotationValidationError("Select a decision before saving.")
+    if confidence not in CONFIDENCE_LEVELS:
+        raise AnnotationValidationError("Select a confidence level before saving.")
+    unknown_issues = set(issues or []) - set(ISSUE_TYPES)
+    if unknown_issues:
+        raise AnnotationValidationError(f"Unknown issue types: {sorted(unknown_issues)}")
+
+    row_ids = {str(row["row_id"]) for row in case["rows"]}
+    parent = str(corrected_parent or "").strip()
+    children = normalize_children(corrected_children, row_ids)
+    explanation = str(equivalence_explanation or "").strip()
+    note_text = str(notes or "").strip()
+
+    if decision == "Equivalent" and not explanation:
+        raise AnnotationValidationError("Explain why the proposed relation is equivalent.")
+    if decision == "Incorrect":
+        if parent not in row_ids:
+            raise AnnotationValidationError("Select a valid corrected parent row.")
+        if len(children) < 2:
+            raise AnnotationValidationError("An incorrect relation needs at least two corrected children.")
+        if parent in {str(child["row_id"]) for child in children}:
+            raise AnnotationValidationError("The corrected parent cannot also be a child.")
+    if decision == "Unclear" and not note_text:
+        raise AnnotationValidationError("Describe why this case is unclear.")
+
+    correction = None
+    if decision == "Incorrect":
+        correction = {"parent_row_id": parent, "children": children}
+    return {
+        "annotation_schema_version": ANNOTATION_SCHEMA_VERSION,
+        "audit_id": case["audit_id"],
+        "proposal_hash": case["proposal_hash"],
+        "annotator": username,
+        "annotated_at_utc": utc_now(),
+        "decision": decision,
+        "confidence": confidence,
+        "issues": list(issues or []),
+        "correction": correction,
+        "equivalence_explanation": explanation or None,
+        "notes": note_text or None,
+    }
+
+
+class AnnotationStore:
+    """Per-user JSON persistence with an atomic local cache and optional HF sync."""
+
+    def __init__(
+        self,
+        local_root: str | Path,
+        repo_id: str | None = None,
+        token: str | None = None,
+    ) -> None:
+        self.local_root = Path(local_root).expanduser().resolve()
+        self.local_root.mkdir(parents=True, exist_ok=True)
+        self.repo_id = (repo_id or "").strip() or None
+        self.token = (token or "").strip() or None
+        self._locks: dict[str, threading.Lock] = {}
+        self._locks_guard = threading.Lock()
+
+    @classmethod
+    def from_environment(cls) -> "AnnotationStore":
+        return cls(
+            local_root=os.getenv("FINREPAIR_ANNOTATION_CACHE", "/tmp/finrepair_annotations"),
+            repo_id=os.getenv("FINREPAIR_ANNOTATION_REPO")
+            or os.getenv("FINREPAIR_DATASET_REPO"),
+            token=os.getenv("HF_TOKEN"),
+        )
+
+    def _lock_for(self, username: str) -> threading.Lock:
+        key = safe_user_key(username)
+        with self._locks_guard:
+            return self._locks.setdefault(key, threading.Lock())
+
+    def _relative_path(self, username: str) -> str:
+        return f"annotations/{safe_user_key(username)}.json"
+
+    def _local_path(self, username: str) -> Path:
+        return self.local_root / f"{safe_user_key(username)}.json"
+
+    @staticmethod
+    def empty_document(username: str) -> dict[str, Any]:
+        return {
+            "annotation_schema_version": ANNOTATION_SCHEMA_VERSION,
+            "annotator": username,
+            "updated_at_utc": None,
+            "annotations": {},
+        }
+
+    def load(self, username: str) -> dict[str, Any]:
+        with self._lock_for(username):
+            local_path = self._local_path(username)
+            if self.repo_id and self.token:
+                try:
+                    remote_path = hf_hub_download(
+                        repo_id=self.repo_id,
+                        repo_type="dataset",
+                        filename=self._relative_path(username),
+                        token=self.token,
+                        force_download=True,
+                    )
+                    document = json.loads(Path(remote_path).read_text(encoding="utf-8"))
+                    self._write_local(local_path, document)
+                    return self._validate_document(document, username)
+                except EntryNotFoundError:
+                    pass
+                except (HfHubHTTPError, OSError, json.JSONDecodeError) as exc:
+                    raise StorageError(f"Could not load the remote annotation record: {exc}") from exc
+            if local_path.is_file():
+                try:
+                    document = json.loads(local_path.read_text(encoding="utf-8"))
+                except (OSError, json.JSONDecodeError) as exc:
+                    raise StorageError(f"Could not load the local annotation record: {exc}") from exc
+                return self._validate_document(document, username)
+            return self.empty_document(username)
+
+    def save(self, username: str, document: dict[str, Any]) -> bool:
+        with self._lock_for(username):
+            document = self._validate_document(document, username)
+            document["updated_at_utc"] = utc_now()
+            local_path = self._local_path(username)
+            self._write_local(local_path, document)
+            if not self.repo_id or not self.token:
+                return False
+            try:
+                HfApi(token=self.token).upload_file(
+                    path_or_fileobj=str(local_path),
+                    path_in_repo=self._relative_path(username),
+                    repo_id=self.repo_id,
+                    repo_type="dataset",
+                    commit_message=f"Save audit progress for {safe_user_key(username)}",
+                )
+            except (HfHubHTTPError, OSError) as exc:
+                raise StorageError(
+                    "The annotation was cached locally but could not be synced to the private dataset."
+                ) from exc
+            return True
+
+    @staticmethod
+    def _validate_document(document: dict[str, Any], username: str) -> dict[str, Any]:
+        if not isinstance(document, dict) or not isinstance(document.get("annotations"), dict):
+            raise StorageError("Annotation document has an invalid structure.")
+        owner = document.get("annotator")
+        if owner != username:
+            raise StorageError(f"Annotation owner mismatch: expected {username!r}, found {owner!r}")
+        return document
+
+    @staticmethod
+    def _write_local(path: Path, document: dict[str, Any]) -> None:
+        path.parent.mkdir(parents=True, exist_ok=True)
+        temporary = path.with_suffix(path.suffix + ".tmp")
+        temporary.write_text(
+            json.dumps(document, indent=2, sort_keys=True, ensure_ascii=True) + "\n",
+            encoding="utf-8",
+        )
+        temporary.replace(path)

requirements.txt

@@ -1,13 +1,2 @@
-fastapi==0.116.1
-uvicorn[standard]==0.35.0
-sqlalchemy==2.0.43
-psycopg2-binary==2.9.10
-pydantic==2.11.7
-pydantic-settings==2.10.1
-feedparser==6.0.11
-httpx==0.28.1
-beautifulsoup4==4.13.4
-python-dateutil==2.9.0.post0
-langdetect==1.0.9
-PyYAML==6.0.2
-jinja2==3.1.6
+gradio[oauth]==6.20.0
+huggingface_hub==1.23.0

sources.yaml

@@ -1,307 +0,0 @@
-sources:
-  - name: US State Department Press
-    kind: rss
-    base_url: https://www.state.gov/
-    feed_url: https://www.state.gov/press-releases/feed/
-    country: United States
-    actor: US State Department
-    priority: official
-    active: true
-
-  - name: The White House Briefing Room
-    kind: rss
-    base_url: https://www.whitehouse.gov/
-    feed_url: https://www.whitehouse.gov/briefing-room/feed/
-    country: United States
-    actor: White House
-    priority: official
-    active: true
-
-  - name: US DoD Releases
-    kind: rss
-    base_url: https://www.defense.gov/
-    feed_url: https://www.defense.gov/DesktopModules/ArticleCS/RSS.ashx?ContentType=1&Site=945&max=20
-    country: United States
-    actor: US Department of Defense
-    priority: official
-    active: true
-
-  - name: UN News Middle East
-    kind: rss
-    base_url: https://news.un.org/
-    feed_url: https://news.un.org/feed/subscribe/en/news/region/middle-east/feed/rss.xml
-    country: International
-    actor: United Nations
-    priority: official
-    active: true
-
-  - name: UN Security Council Meetings
-    kind: rss
-    base_url: https://press.un.org/
-    feed_url: https://press.un.org/en/content/security-council/rss.xml
-    country: International
-    actor: UN Security Council
-    priority: official
-    active: true
-
-  - name: UK Foreign Office
-    kind: rss
-    base_url: https://www.gov.uk/government/organisations/foreign-commonwealth-development-office
-    feed_url: https://www.gov.uk/government/organisations/foreign-commonwealth-development-office.atom
-    country: United Kingdom
-    actor: FCDO
-    priority: official
-    active: true
-
-  - name: French Foreign Ministry
-    kind: rss
-    base_url: https://www.diplomatie.gouv.fr/
-    feed_url: https://www.diplomatie.gouv.fr/en/spip.php?page=backend
-    country: France
-    actor: French Foreign Ministry
-    priority: official
-    active: true
-
-  - name: Reuters World
-    kind: rss
-    base_url: https://www.reuters.com/
-    feed_url: https://feeds.reuters.com/Reuters/worldNews
-    country: International
-    actor: Reuters
-    priority: media
-    active: true
-
-  - name: Reuters Middle East
-    kind: rss
-    base_url: https://www.reuters.com/world/middle-east/
-    feed_url: https://feeds.reuters.com/reuters/world/middleeast
-    country: International
-    actor: Reuters
-    priority: media
-    active: true
-
-  - name: AP News World
-    kind: rss
-    base_url: https://apnews.com/
-    feed_url: https://apnews.com/hub/apf-world-news?output=1
-    country: International
-    actor: Associated Press
-    priority: media
-    active: true
-
-  - name: Al Jazeera Middle East
-    kind: rss
-    base_url: https://www.aljazeera.com/
-    feed_url: https://www.aljazeera.com/xml/rss/all.xml
-    country: Qatar
-    actor: Al Jazeera
-    priority: media
-    active: true
-
-  - name: BBC World
-    kind: rss
-    base_url: https://www.bbc.com/news
-    feed_url: http://feeds.bbci.co.uk/news/world/rss.xml
-    country: United Kingdom
-    actor: BBC
-    priority: media
-    active: true
-
-  - name: BBC Arabic
-    kind: rss
-    base_url: https://www.bbc.com/arabic
-    feed_url: https://feeds.bbci.co.uk/arabic/rss.xml
-    country: International
-    actor: BBC Arabic
-    priority: media
-    active: true
-
-  - name: BBC Persian
-    kind: rss
-    base_url: https://www.bbc.com/persian
-    feed_url: https://feeds.bbci.co.uk/persian/rss.xml
-    country: Iran
-    actor: BBC Persian
-    priority: media
-    active: true
-
-  - name: BBC Turkish
-    kind: rss
-    base_url: https://www.bbc.com/turkce
-    feed_url: https://feeds.bbci.co.uk/turkce/rss.xml
-    country: Turkey
-    actor: BBC Turkish
-    priority: media
-    active: true
-
-  - name: Al Jazeera Arabic
-    kind: rss
-    base_url: https://www.aljazeera.net/
-    feed_url: https://www.aljazeera.net/aljazeerarss/ar
-    country: Qatar
-    actor: Al Jazeera Arabic
-    priority: media
-    active: true
-
-  - name: Anadolu Turkish
-    kind: rss
-    base_url: https://www.aa.com.tr/tr
-    feed_url: https://www.aa.com.tr/tr/rss/default?cat=guncel
-    country: Turkey
-    actor: Anadolu Agency
-    priority: media
-    active: true
-
-  - name: Anadolu Arabic
-    kind: rss
-    base_url: https://www.aa.com.tr/ar
-    feed_url: https://www.aa.com.tr/ar/rss/default?cat=featured
-    country: International
-    actor: Anadolu Agency Arabic
-    priority: media
-    active: true
-
-  - name: Ynet Hebrew
-    kind: rss
-    base_url: https://www.ynet.co.il/
-    feed_url: https://www.ynet.co.il/Integration/StoryRss2.xml
-    country: Israel
-    actor: Ynet
-    priority: media
-    active: true
-
-  - name: CNN World
-    kind: rss
-    base_url: https://edition.cnn.com/world
-    feed_url: http://rss.cnn.com/rss/edition_world.rss
-    country: United States
-    actor: CNN
-    priority: media
-    active: true
-
-  - name: NYTimes World
-    kind: rss
-    base_url: https://www.nytimes.com/section/world
-    feed_url: https://rss.nytimes.com/services/xml/rss/nyt/World.xml
-    country: United States
-    actor: New York Times
-    priority: media
-    active: true
-
-  - name: Guardian World
-    kind: rss
-    base_url: https://www.theguardian.com/world
-    feed_url: https://www.theguardian.com/world/rss
-    country: United Kingdom
-    actor: The Guardian
-    priority: media
-    active: true
-
-  - name: Times of Israel
-    kind: rss
-    base_url: https://www.timesofisrael.com/
-    feed_url: https://www.timesofisrael.com/feed/
-    country: Israel
-    actor: Times of Israel
-    priority: media
-    active: true
-
-  - name: Haaretz
-    kind: rss
-    base_url: https://www.haaretz.com/
-    feed_url: https://www.haaretz.com/cmlink/1.628752
-    country: Israel
-    actor: Haaretz
-    priority: media
-    active: true
-
-  - name: Jerusalem Post
-    kind: rss
-    base_url: https://www.jpost.com/
-    feed_url: https://www.jpost.com/rss/rssfeedsheadlines.aspx
-    country: Israel
-    actor: Jerusalem Post
-    priority: media
-    active: true
-
-  - name: Arab News
-    kind: rss
-    base_url: https://www.arabnews.com/
-    feed_url: https://www.arabnews.com/rss.xml
-    country: Saudi Arabia
-    actor: Arab News
-    priority: media
-    active: true
-
-  - name: Gulf News
-    kind: rss
-    base_url: https://gulfnews.com/
-    feed_url: https://gulfnews.com/rss
-    country: UAE
-    actor: Gulf News
-    priority: media
-    active: true
-
-  - name: Israeli PMO Announcements
-    kind: webpage
-    base_url: https://www.gov.il/en/departments/prime_ministers_office/news
-    country: Israel
-    actor: Prime Minister's Office
-    priority: official
-    active: true
-
-  - name: UAE MoFA Announcements
-    kind: webpage
-    base_url: https://www.mofa.gov.ae/en/mediahub/news
-    country: UAE
-    actor: Ministry of Foreign Affairs
-    priority: official
-    active: true
-
-  - name: Saudi Foreign Ministry Announcements
-    kind: webpage
-    base_url: https://www.mofa.gov.sa/en/mediacenter/news
-    country: Saudi Arabia
-    actor: Ministry of Foreign Affairs
-    priority: official
-    active: true
-
-  - name: Iranian Foreign Ministry News
-    kind: webpage
-    base_url: https://en.mfa.ir/portal/news
-    country: Iran
-    actor: Ministry of Foreign Affairs
-    priority: official
-    active: true
-
-  - name: Dubai Airports Alerts
-    kind: webpage
-    base_url: https://www.dubaiairports.ae/alert
-    country: UAE
-    actor: Dubai Airports
-    priority: official
-    active: true
-
-  - name: Hamad International Airport
-    kind: webpage
-    base_url: https://dohahamadairport.com/
-    country: Qatar
-    actor: Hamad International Airport
-    priority: official
-    active: true
-
-  - name: Israel Airports Authority
-    kind: webpage
-    base_url: https://www.iaa.gov.il/en/
-    country: Israel
-    actor: Israel Airports Authority
-    priority: official
-    active: true
-
-  - name: Kuwait International Airport
-    kind: webpage
-    base_url: https://www.kuwaitairport.gov.kw/en/
-    country: Kuwait
-    actor: Directorate General of Civil Aviation
-    priority: official
-    active: true

styles.css

@@ -0,0 +1,204 @@
+:root {
+  --ink: #17202a;
+  --muted: #5f6b76;
+  --line: #d7dde3;
+  --surface: #ffffff;
+  --surface-soft: #f5f7f8;
+  --primary: #185f8d;
+  --primary-dark: #11496d;
+  --success: #27825b;
+  --warning: #a25e15;
+}
+
+.gradio-container {
+  background: var(--surface-soft) !important;
+  color: var(--ink) !important;
+  max-width: min(1680px, 100vw) !important;
+  margin: 0 auto !important;
+  padding: 0 !important;
+  width: 100vw !important;
+}
+
+#topbar {
+  align-items: center;
+  border-bottom: 1px solid var(--line);
+  margin-bottom: 12px;
+  padding-bottom: 10px;
+}
+
+#product-title {
+  flex: 1 1 auto;
+}
+
+#product-title h1 {
+  color: var(--ink);
+  font-size: 24px;
+  line-height: 1.25;
+  margin: 0;
+}
+
+#auth-button {
+  flex: 0 0 auto;
+  min-width: 150px;
+}
+
+.progress-shell {
+  background: var(--surface);
+  border: 1px solid var(--line);
+  border-radius: 6px;
+  padding: 10px 12px;
+}
+
+.progress-copy {
+  align-items: baseline;
+  color: var(--muted);
+  display: flex;
+  font-size: 13px;
+  gap: 10px;
+  justify-content: flex-start;
+  margin-bottom: 7px;
+}
+
+.progress-copy strong {
+  color: var(--ink);
+}
+
+.annotator-name {
+  margin-left: auto;
+}
+
+progress {
+  accent-color: var(--success);
+  display: block;
+  height: 8px;
+  width: 100%;
+}
+
+#save-status {
+  color: var(--success);
+  font-size: 13px;
+  min-height: 22px;
+  padding: 5px 2px 0;
+}
+
+#case-title h3 {
+  font-size: 17px;
+  margin: 3px 0 8px;
+}
+
+#case-title p {
+  color: var(--muted);
+  font-size: 12px;
+}
+
+#statement-image {
+  background: var(--surface);
+  border-color: var(--line);
+  border-radius: 6px;
+}
+
+#statement-image img {
+  object-fit: contain !important;
+}
+
+#workbench {
+  align-items: flex-start;
+  gap: 16px;
+  margin-top: 14px;
+}
+
+#annotation-panel {
+  border-left: 1px solid var(--line);
+  padding-left: 16px;
+}
+
+#navigation {
+  background: var(--surface-soft);
+  border-top: 1px solid var(--line);
+  gap: 8px;
+  margin-top: 16px;
+  padding-top: 12px;
+}
+
+#navigation button {
+  border-radius: 5px !important;
+  min-height: 42px;
+}
+
+#navigation button.primary {
+  background: var(--primary) !important;
+  border-color: var(--primary) !important;
+}
+
+#navigation button.primary:hover {
+  background: var(--primary-dark) !important;
+}
+
+@media (max-width: 760px) {
+  #topbar,
+  #workbench,
+  #workbench > .column,
+  #navigation {
+    max-width: 100% !important;
+    min-width: 0 !important;
+    width: 100% !important;
+  }
+
+  #workbench .block {
+    max-width: 100% !important;
+    min-width: 0 !important;
+  }
+
+  #workbench {
+    flex-direction: column !important;
+  }
+
+  #workbench > .column {
+    flex: 1 1 auto !important;
+  }
+
+  #product-title h1 {
+    font-size: 20px;
+  }
+
+  #topbar {
+    align-items: stretch;
+    gap: 8px;
+  }
+
+  #auth-button {
+    min-width: 100%;
+  }
+
+  .progress-copy {
+    flex-wrap: wrap;
+  }
+
+  .annotator-name {
+    flex-basis: 100%;
+    margin-left: 0;
+    overflow-wrap: anywhere;
+  }
+
+  #statement-image {
+    height: 220px !important;
+    min-height: 220px;
+  }
+
+  #annotation-panel {
+    border-left: 0;
+    border-top: 1px solid var(--line);
+    padding-left: 0;
+    padding-top: 14px;
+  }
+
+  #navigation {
+    display: grid !important;
+    grid-template-columns: 1fr 1fr;
+  }
+
+  #navigation > div {
+    min-width: 0 !important;
+    width: 100% !important;
+  }
+}

tests/test_audit_core.py

@@ -0,0 +1,126 @@
+from __future__ import annotations
+
+import hashlib
+import json
+from pathlib import Path
+
+import pytest
+
+from audit_core import (
+    AnnotationStore,
+    AnnotationValidationError,
+    AuthorizationError,
+    CaseBundle,
+    authorize,
+    build_annotation,
+    normalize_children,
+    safe_user_key,
+)
+
+
+def make_case() -> dict:
+    return {
+        "audit_id": "audit_test",
+        "image_file": "images/audit_test.png",
+        "image_sha256": hashlib.sha256(b"image").hexdigest(),
+        "order": 1,
+        "proposal": {
+            "parent": {"row_id": "r3", "label": "Profit"},
+            "children": [
+                {"row_id": "r1", "label": "Revenue", "weight": 1},
+                {"row_id": "r2", "label": "Cost", "weight": -1},
+            ],
+        },
+        "proposal_hash": "proposal-hash",
+        "relation_size": 3,
+        "rows": [
+            {"row_id": "r1", "text": "Revenue"},
+            {"row_id": "r2", "text": "Cost"},
+            {"row_id": "r3", "text": "Profit"},
+        ],
+        "schema_version": "1.0",
+    }
+
+
+def make_bundle(tmp_path: Path) -> Path:
+    root = tmp_path / "bundle"
+    (root / "images").mkdir(parents=True)
+    case = make_case()
+    (root / case["image_file"]).write_bytes(b"image")
+    case_bytes = (json.dumps(case) + "\n").encode()
+    (root / "cases.jsonl").write_bytes(case_bytes)
+    lock = {
+        "artifacts": {"cases.jsonl": hashlib.sha256(case_bytes).hexdigest()},
+        "image_count": 1,
+    }
+    (root / "LOCK.json").write_text(json.dumps(lock), encoding="utf-8")
+    return root
+
+
+def test_case_bundle_validates_locked_cases_and_images(tmp_path: Path) -> None:
+    bundle = CaseBundle.load(make_bundle(tmp_path))
+    assert len(bundle.cases) == 1
+    assert bundle.cases[0]["audit_id"] == "audit_test"
+    assert bundle.image_path(bundle.cases[0]).endswith("images/audit_test.png")
+
+
+def test_normalize_children_rejects_duplicates_and_zero_weight() -> None:
+    assert normalize_children([["r1", "1"], ["r2", -1]], {"r1", "r2"}) == [
+        {"row_id": "r1", "weight": 1},
+        {"row_id": "r2", "weight": -1},
+    ]
+    with pytest.raises(AnnotationValidationError, match="duplicated"):
+        normalize_children([["r1", 1], ["r1", -1]], {"r1"})
+    with pytest.raises(AnnotationValidationError, match="non-zero"):
+        normalize_children([["r1", 0]], {"r1"})
+
+
+def test_decision_specific_validation() -> None:
+    case = make_case()
+    exact = build_annotation(
+        case, "alice", "Exact", [], "High", "r3", [["r1", 1], ["r2", -1]], "", ""
+    )
+    assert exact["correction"] is None
+
+    with pytest.raises(AnnotationValidationError, match="equivalent"):
+        build_annotation(
+            case, "alice", "Equivalent", [], "Medium", "r3", [], "", ""
+        )
+    with pytest.raises(AnnotationValidationError, match="at least two"):
+        build_annotation(
+            case, "alice", "Incorrect", ["Missing child"], "Low", "r3", [["r1", 1]], "", ""
+        )
+    corrected = build_annotation(
+        case,
+        "alice",
+        "Incorrect",
+        ["Wrong sign or weight"],
+        "High",
+        "r3",
+        [["r1", 1], ["r2", -1]],
+        "",
+        "checked",
+    )
+    assert corrected["correction"]["children"][1]["weight"] == -1
+
+
+def test_authorization_requires_identity_and_honors_allowlist(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.delenv("FINREPAIR_DEV_USER", raising=False)
+    monkeypatch.delenv("ANNOTATOR_ALLOWLIST", raising=False)
+    with pytest.raises(AuthorizationError, match="Sign in"):
+        authorize(None)
+    assert authorize({"preferred_username": "Alice"}) == "Alice"
+    monkeypatch.setenv("ANNOTATOR_ALLOWLIST", "bob, carol")
+    with pytest.raises(AuthorizationError, match="not an authorized"):
+        authorize({"preferred_username": "Alice"})
+    assert authorize({"preferred_username": "BOB"}) == "BOB"
+
+
+def test_local_store_round_trip_and_user_isolation(tmp_path: Path) -> None:
+    store = AnnotationStore(tmp_path)
+    alice = store.empty_document("alice")
+    alice["annotations"]["audit_test"] = {"decision": "Exact"}
+    assert store.save("alice", alice) is False
+    assert store.load("alice")["annotations"]["audit_test"]["decision"] == "Exact"
+    assert store.load("bob")["annotations"] == {}
+    assert safe_user_key("Alice") != safe_user_key("Bob")