`, favicon, and body of `Root()` to\nmatch your app's existing `index.html`, and stamp anything else your shell\nreferences directly (images, css) with `immutableFileUrl()`.\n\n1. Create `frontend/root.tsx`:\n\n```tsx\n/** @jsxImportSource npm:hono@4/jsx */\nimport { raw } from \"npm:hono@4/html\";\nimport { immutableFileUrl } from \"https://esm.town/v/std/utils/index.ts\";\n\n// The HTML shell for the app. It is never cached, so immutableFileUrl can stamp\n// the current val version into each asset URL — the browser caches those\n// immutably, and publishing the val changes the URLs, refetching exactly once.\nexport function Root() {\n return (\n <>\n {raw(\"<!DOCTYPE html>\")}\n <html lang=\"en\">\n <head>\n <meta charSet=\"UTF-8\" />\n <meta name=\"viewport\" content=\"width=device-width, initial-scale=1.0\" />\n <title>React Hono Val Town Starter

[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_416":417,"_418":419},"routes/_app.x.$handle.$name",{"_11":12,"_13":14,"_15":-7,"_18":19,"_17":-7,"_312":-5,"_20":313,"_22":314,"_315":316,"_317":318,"_26":319,"_24":320,"_321":-5,"_322":-5,"_323":54,"_324":325,"_326":45,"_327":-5,"_328":-5,"_329":330},"routes/_app.x.$handle.$name.code.$",{"_11":12,"_13":14,"_15":16,"_17":-7,"_18":19,"_20":21,"_22":23,"_24":25,"_26":27,"_28":-5,"_29":-5,"_30":-5,"_31":-5,"_32":-5},"handle","justinstevens","name","caselens-canada","branchName","main","version","path","server/analyze.ts","project",{"_33":254,"_13":14,"_46":255,"_256":-5,"_257":54,"_258":259,"_260":259,"_261":262,"_263":-5,"_264":265,"_266":76,"_267":45,"_268":-5,"_269":-5,"_270":69,"_271":272,"_273":274,"_275":276,"_277":278,"_279":-5,"_280":281,"_282":281,"_283":284,"_285":284,"_286":284,"_287":281,"_288":281,"_289":290,"_291":292,"_293":294,"_295":54},"tree",[58,59,60,61,62,63],"file",{"_33":34,"_13":35,"_36":37,"_38":39,"_40":41,"_42":43,"_44":45,"_46":47,"_17":48,"_49":50,"_51":52,"_53":54,"_55":-5,"_18":56},"contents",[],"upstream","pull","ahead","behind","envVars","id","e0b97043-3896-4c6d-9163-5ed461b71a7b","analyze.ts","fileId","72f22fbe-aa23-11f1-96b6-1607ee4eb77e","parentId","5c27b29b-f289-4386-8c91-f92f50dc4466","branchId","9da8d75d-5b67-4db9-bfe9-96a9f922df07","branchVersion",14,"latest",true,"createdAt",["D",1788720538890],3,"fileType","script","content","/**\n * The analysis pipeline.\n *\n * Stages:\n * 1. Segment text into paragraphs + extract case metadata (server/text.ts).\n * 2. Produce structured highlight *candidates* — either from a real LLM\n * (server/ai.ts) or, when no model is configured (or it fails), from an\n * honest heuristic engine (this file).\n * 3. Anchor every candidate to a verbatim, character-exact span, dedupe\n * overlaps, and prefer the most important span.\n *\n * The guarantee: the quoted text in every highlight is a character-exact\n * substring of the uploaded judgment. Anything that cannot be anchored is\n * dropped rather than kept loose.\n */\nimport { LEGEND, type Analysis, type CaseMetadata, type CategoryKey, type Highlight, type LegalType, type Paragraph } from \"../types.ts\";\nimport { emptyMetadata, extractMetadata, splitParagraphs } from \"./text.ts\";\nimport { analyzeWithAI, type RawCandidate } from \"./ai.ts\";\n\n/* ------------------------------------------------------------------ */\n/* Anchoring */\n/* ------------------------------------------------------------------ */\n\nfunction normalizeForMatch(s: string): string {\n return s.replace(/\\s+/g, \" \").replace(/\\u2018|\\u2019/g, \"'\").replace(/\\u201c|\\u201d/g, '\"').trim();\n}\n\n/** Find the original-character offsets of `quote` within paragraph `text`. Returns null if not verbatim. */\nfunction findQuoteOffsets(text: string, quote: string): { start: number; end: number } | null {\n const t = normalizeForMatch(text);\n const q = normalizeForMatch(quote);\n if (!q) return null;\n const idx = t.indexOf(q);\n if (idx < 0) return null;\n return { start: idx, end: idx + q.length };\n}\n\nfunction makeHighlight(p: Paragraph, quote: string, startChar: number, endChar: number, cand: { category: CategoryKey; legalType: LegalType; importance: number; confidence: number; what?: string; why?: string; proposition?: string; cite?: string; context?: string; whyMatters?: string; prominence?: string }): Highlight {\n const info = LEGEND.find((l) => l.key === cand.category)!;\n return {\n id: `h-${p.index}-${startChar}-${endChar}`,\n exactQuote: quote,\n paragraphIndex: p.index,\n paragraphNumber: p.num,\n startChar,\n endChar,\n category: cand.category,\n legalType: cand.legalType,\n importance: clamp01(cand.importance),\n confidence: clamp01(cand.confidence),\n whatJudgeIsSaying: cand.what ?? defaultWhat(cand.category),\n whyMatters: cand.whyMatters ?? cand.why ?? defaultWhy(cand.category),\n legalProposition: cand.proposition ?? defaultProposition(cand.category),\n citationUse: cand.cite ?? defaultCite(cand.category),\n contextSummary: cand.context ?? \"\",\n };\n}\n\nconst clamp01 = (n: number) => Math.max(0, Math.min(1, typeof n === \"number\" && isFinite(n) ? n : 0.5));\n\nconst defaultWhat = (c: CategoryKey) => `The court is making a ${LEGEND.find((l) => l.key === c)?.label.toLowerCase() ?? c} statement in this passage.`;\nconst defaultWhy = (c: CategoryKey) => `${LEGEND.find((l) => l.key === c)?.explanation ?? \"\"} This passage is a precise, quotable articulation a lawyer can rely on.`;\nconst defaultProposition = (c: CategoryKey) => `The court said, on the matter before it: see the quoted passage above.`;\nconst defaultCite = (c: CategoryKey) => `Citing matters: quote the passage and attach the neutral citation with the pinpoint paragraph number. Use it for the proposition stated in the passage.`;\n\nfunction anchorCandidates(paragraphs: Paragraph[], candidates: RawCandidate[]): Highlight[] {\n const out: Highlight[] = [];\n const collapsed: Paragraph[] = paragraphs;\n for (const c of candidates) {\n if (!c.exactQuote || !c.exactQuote.trim()) continue;\n let para: Paragraph | undefined;\n // Prefer the block the model pointed at; else search all paragraphs.\n if (typeof c.blockIndex === \"number\" && c.blockIndex >= 0 && c.blockIndex < collapsed.length) {\n para = collapsed[c.blockIndex];\n let offs = findQuoteOffsets(para.text, c.exactQuote);\n if (!offs && para.heading) offs = null;\n if (!offs) para = undefined;\n else {\n out.push(makeHighlight(para, para.text.slice(offs.start, offs.end), offs.start, offs.end, c));\n continue;\n }\n }\n if (para === undefined) {\n for (const p of collapsed) {\n if (p.heading) continue;\n const offs = findQuoteOffsets(p.text, c.exactQuote);\n if (offs) {\n out.push(makeHighlight(p, p.text.slice(offs.start, offs.end), offs.start, offs.end, c));\n para = p;\n break;\n }\n }\n }\n // Unanchored candidates are dropped — we never show a quote that isn't verbatim in the text.\n }\n // Deduplicate overlapping spans: when two highlights overlap in the same\n // paragraph, keep only the one with the higher importance (ties keep the first).\n const deduped: Highlight[] = [];\n const sorted = [...out].sort((a, b) => b.importance - a.importance || a.startChar - b.startChar);\n for (const h of sorted) {\n const blocked = deduped.some((d) =>\n d.paragraphIndex === h.paragraphIndex &&\n h.startChar < d.endChar &&\n h.endChar > d.startChar &&\n d.importance >= h.importance,\n );\n if (!blocked) deduped.push(h);\n }\n return deduped.sort((a, b) => a.paragraphIndex - b.paragraphIndex || a.startChar - b.startChar);\n}\n\n/* ------------------------------------------------------------------ */\n/* Heuristic engine (fallback — always honest, never invents a citation) */\n/* ------------------------------------------------------------------ */\n\ninterface Rule {\n category: CategoryKey;\n legalType: LegalType;\n rx: RegExp;\n what: string;\n why: string;\n cite: string;\n importance: number;\n}\n\n// Initial \"blockIndex\" is the model's guess; the heuristic always knows it.\nconst RULES: Rule[] = [\n { category: \"remedy\", legalType: \"remedy / order\", rx: /\\b(allow(?:ing)? the appeal|dismiss(?:ing)? the appeal|the appeal is (?:allowed|dismissed)|judgment (?:for|in favour of)|\\$[\\d,]+(?:\\.[\\d]{2})? in (?:damages|costs)|I would (?:allow|dismiss)|is reduced to|costs .{0,30}(?:payable|awarded|fixed in the amount)|substantial indemnity|no order as to costs)/i, what: \"The court states its operative disposition, remedy, or order.\", why: \"This is the court's final disposition — decisive for enforcing the outcome and for framing what actually was decided.\", cite: \"Cite this to establish the precise disposition (who won, and what relief was ordered), always anchored to the pinpoint paragraph.\", importance: 0.72 },\n { category: \"procedural\", legalType: \"procedural holding\", rx: /\\b(standing|jurisdiction|leave to appeal|interlocutory|motion to|stay of proceedings|admissib|new trial|quash|procedural|fresh evidence|discretion to)(.{0,120})/i, what: \"The court rules on a procedural point — standing, jurisdiction, admissibility, or process.\", why: \"Procedural holdings can determine whether a claim proceeds and are often cited on the applicable standard.\", cite: \"Cite to support the procedural rule or standard the court applied (e.g. standing, fresh evidence).\", importance: 0.62 },\n { category: \"credibility\", legalType: \"credibility finding\", rx: /\\b(credib|reliab|demeanou?r|corroborat|palpable and overriding error|weight of the evidence|evidentiary base|accept(?:ed|ing|ance of|s)? (?:the )?(?:evidence|testimony|account|version)|was (?:not )??(?:believable|credible)|on the (?:evidence|record)|the trial judge (?:observed|was entitled))/i, what: \"The court assesses the credibility or reliability of witnesses and evidence.\", why: \"Credibility findings are a core reason appellate courts defer, and are frequently cited on the standard of review.\", cite: \"Cite for the deferential standard applied to credibility findings, and pinpoint the finding itself.\", importance: 0.66 },\n { category: \"definition\", legalType: \"legal test / rule\", rx: /\\b(means|is defined as|defined as|shall include|the term .{0,40} means|in this context|referred to as|construed|interpretation)/i, what: \"The court adopts a definition or interpretation of a term or provision.\", why: \"A definition can carry weight beyond the immediate facts and be cited when the same term is at issue.\", cite: \"Cite to fix the meaning the court gave to a defined term or statutory expression.\", importance: 0.6 },\n { category: \"rule\", legalType: \"legal test / rule\", rx: /\\b(the (?:governing|correct|proper|applicable|legal) (?:test|standard)|the test for|a test (?:has )?(?:two|three|multi)|(?:two|three|four|five)-?part test|must (?:establish|prove|satisfy|meet)|the burden of proof|on (?:the )?(?:balance of probabilities|standard of proof|civil standard)|bears (?:the )?burden|requirement)/i, what: \"The court states the governing legal test, standard, or burden of proof.\", why: \"This is the legal rule a subsequent court (and counsel) apply — prime territory for citation as to 'the law'.\", cite: \"Cite as authority for the legal test or standard, with a pinpoint to the paragraph stating it.\", importance: 0.8 },\n { category: \"ratio\", legalType: \"ratio decidendi\", rx: /\\b(I (?:conclude|hold|am satisfied that|find that the (?:law|principle|test))|the (?:law|principle|rule) is (?:settled|clear|well-established|well established)|it is well established|in my view, ?the (?:law|test|principle)|the (?:principle|ration?) of|the answer to the|the essential question)/i, what: \"The court articulates the core principle that decides the case — the ratio.\", why: \"This is the binding core of the decision; identifying it precisely is the heart of case triage.\", cite: \"Cite as the ratio decidendi of the decision; be careful to state it within the factual context.\", importance: 0.88 },\n { category: \"fact\", legalType: \"key factual finding\", rx: /\\b(I find that|the evidence (?:shows|establishes|supports|reveals)|it is (?:undisputed|not disputed|clear from the record)|the record (?:shows|establishes)|the trial judge found|the parties agreed|was completed|the contract (?:provided|was fixed)|dated [A-Z][a-z]+ \\d{1,2})/i, what: \"The court makes or relies on a finding of fact.\", why: \"Key factual findings anchor the application of law and often anchor damages or disposition.\", cite: \"Cite to establish the underlying facts the court relied on — useful context, not normally the ratio.\", importance: 0.55 },\n { category: \"application\", legalType: \"application of law to facts\", rx: /\\b((?:therefore|accordingly|in this case|on the facts|applied to|as a result|here,|in the circumstances|the (?:respondent|appellant|plaintiff|defendant)'s conduct|performance would not be|quantum|mitigation credit|in the amount of))/i, what: \"The court applies the law to the specific facts before it.\", why: \"Application passages show how the rule operates on facts — persuasive analogies for similar cases.\", cite: \"Cite as an example of the rule applied; lead with the facts, then the rule.\", importance: 0.5 },\n];\n\nfunction truncate(toSentence: string, max: number): string {\n let s = toSentence.replace(/\\n+/g, \" \").replace(/\\s+/g, \" \").trim();\n const el = s.indexOf(\". \");\n if (el > 20 && el < max) s = s.slice(0, el + 1);\n return s.length > max ? s.slice(0, max - 1) + \"…\" : s;\n}\n\nfunction extractQuoteForRule(para: Paragraph, rule: Rule): { q: string; start: number; end: number } | null {\n const text = para.text.replace(/\\s+/g, \" \");\n const m = text.match(rule.rx);\n if (!m || typeof m.index !== \"number\") return null;\n const hit = m.index;\n // Expand to enclosing sentence bounds for a clean quote.\n let start = hit;\n const prevBoundary = text.lastIndexOf(\". \", start < 0 ? 0 : start);\n if (prevBoundary >= 0 && start - prevBoundary < 260) start = prevBoundary + 2;\n let end = hit + (m[0]?.length ?? 20);\n const nextBoundary = text.indexOf(\". \", end);\n if (nextBoundary >= 0 && nextBoundary - end < 160) end = nextBoundary + 1;\n if (end - start > 240) end = start + 240;\n const q = text.slice(start, end).trim();\n if (!q) return null;\n const rel = findQuoteOffsets(para.text, q);\n if (!rel) return null;\n return { q, start: rel.start, end: rel.end };\n}\n\nfunction heuristicAnalyze(paragraphs: Paragraph[]): Highlight[] {\n const out: Highlight[] = [];\n for (const p of paragraphs) {\n if (p.heading) continue;\n if (p.text.trim().length < 80) continue;\n let best: { rule: Rule; span: { q: string; start: number; end: number } } | null = null;\n // Highest-priority matching rule wins.\n for (const rule of RULES) {\n rule.rx.lastIndex = 0;\n if (rule.rx.test(p.text)) {\n const span = extractQuoteForRule(p, rule);\n if (span) {\n best = { rule, span };\n break;\n }\n }\n }\n if (!best) continue;\n const { rule, span } = best;\n const proposition = truncate(`The court stated (para. ${p.num}): ${span.q}`, 200);\n out.push(\n makeHighlight(p, span.q, span.start, span.end, {\n category: rule.category,\n legalType: rule.legalType,\n importance: rule.importance + (p.text.length > 240 ? 0.04 : 0),\n confidence: 0.93,\n what: rule.what,\n whyMatters: rule.why,\n proposition,\n cite: rule.cite,\n context: \"\",\n }),\n );\n }\n return out;\n}\n\n/* ------------------------------------------------------------------ */\n/* Public API */\n/* ------------------------------------------------------------------ */\n\n/** Given AI candidates already typed (offsets unknown), treat blockIndex as paragraph index. */\nexport function anchorCandidatesPublic(paragraphs: Paragraph[], candidates: RawCandidate[]): Highlight[] {\n return anchorCandidates(paragraphs, candidates);\n}\n\nexport async function analyzeJudgment(text: string): Promise {\n const paragraphs = splitParagraphs(text);\n const metadata: CaseMetadata = extractMetadata(text, paragraphs);\n\n let highlights: Highlight[] = [];\n let source: \"ai\" | \"heuristic\" = \"heuristic\";\n let model: string | undefined;\n\n const heuristicResult = heuristicAnalyze(paragraphs);\n\n try {\n const ai = await analyzeWithAI(paragraphs, metadata);\n if (ai) {\n highlights = anchorCandidates(paragraphs, ai.candidates);\n source = \"ai\";\n model = ai.model;\n }\n } catch (e) {\n console.log(\"AI analysis failed, falling back to heuristic engine:\", String(e?.message ?? e));\n }\n\n if (highlights.length === 0) {\n highlights = heuristicResult;\n source = \"heuristic\";\n }\n\n return {\n source,\n model,\n engineLabel: source === \"ai\" ? `AI (${model ?? \"model\"})` : \"Built-in heuristic engine\",\n metadata,\n paragraphs,\n highlights,\n };\n}\n","deleted",false,"label",[57,35],"server",{"_33":137,"_13":138,"_36":139,"_38":-5,"_40":41,"_42":88,"_44":45,"_46":140,"_17":69,"_49":102,"_51":-5,"_53":54,"_55":-5,"_71":141,"_18":142},{"_33":99,"_13":57,"_36":39,"_38":-5,"_40":41,"_42":100,"_44":45,"_46":101,"_17":69,"_49":102,"_51":-5,"_53":54,"_55":-5,"_71":103,"_18":104},{"_33":92,"_13":93,"_36":94,"_38":-5,"_40":41,"_42":88,"_44":45,"_46":95,"_17":69,"_49":24,"_51":96,"_53":54,"_55":-5,"_71":97,"_18":98},{"_33":83,"_13":84,"_36":85,"_38":-5,"_40":41,"_42":86,"_44":45,"_46":87,"_17":88,"_49":24,"_51":89,"_53":54,"_55":-5,"_71":90,"_18":91},{"_33":74,"_13":75,"_36":76,"_38":-5,"_40":41,"_42":77,"_44":45,"_46":78,"_17":67,"_49":79,"_51":80,"_53":54,"_55":-5,"_71":81,"_18":82},{"_33":64,"_13":65,"_36":66,"_38":-5,"_40":41,"_42":67,"_44":45,"_46":68,"_17":69,"_49":50,"_51":70,"_53":54,"_55":-5,"_71":72,"_18":73},"9b30d97f-8b98-4c48-9a4d-868ad6d9c5e0","types.ts","49c0a90e-aa23-11f1-8352-1607ee4eb77e",2,["D",1788720396299],0,"/**\n * Shared types for CaseLens Canada — used by both the server pipeline\n * (server/*.ts) and the client SPA (frontend/**).\n */\n\nexport type CategoryKey =\n | \"ratio\"\n | \"rule\"\n | \"application\"\n | \"fact\"\n | \"credibility\"\n | \"procedural\"\n | \"remedy\"\n | \"obiter\"\n | \"definition\";\n\nexport interface LegendEntry {\n key: CategoryKey;\n label: string;\n short: string;\n /** what a lawyer typically calls this kind of passage */\n explanation?: string;\n bg: string; // highlight background (hex)\n text: string; // readable foreground / accent (hex)\n solid: string; // strong accent (hex)\n}\n\n/** The category legend — a single source of truth for filters, colors, labels. */\nexport const LEGEND: LegendEntry[] = [\n { key: \"ratio\", label: \"Ratio / core holding\", short: \"Ratio\", bg: \"#dbeafe\", text: \"#1e3a8a\", solid: \"#2563eb\", explanation: \"Ratio decidendi — the binding principle on which the decision turns.\" },\n { key: \"rule\", label: \"Legal test / governing rule\", short: \"Rule / test\", bg: \"#e0e7ff\", text: \"#3730a3\", solid: \"#4f46e5\", explanation: \"A statement of a legal test, standard, or rule the court applies.\" },\n { key: \"application\", label: \"Application / reasoning\", short: \"Application\", bg: \"#ccfbf1\", text: \"#115e59\", solid: \"#0d9488\", explanation: \"Reasoning applying the law to the facts of the case.\" },\n { key: \"fact\", label: \"Key factual finding\", short: \"Finding of fact\", bg: \"#fef3c7\", text: \"#92400e\", solid: \"#d97706\", explanation: \"An established or asserted fact the court relies on.\" },\n { key: \"credibility\", label: \"Credibility / evidentiary finding\", short: \"Credibility\", bg: \"#ffedd5\", text: \"#9a3412\", solid: \"#ea580c\", explanation: \"A determination about reliability, credibility, or the weight of evidence.\" },\n { key: \"procedural\", label: \"Procedural ruling\", short: \"Procedural\", bg: \"#ede9fe\", text: \"#4c1d95\", solid: \"#7c3aed\", explanation: \"A ruling about procedure, jurisdiction, standing, or admissibility.\" },\n { key: \"remedy\", label: \"Remedy / disposition / order\", short: \"Remedy / order\", bg: \"#d1fae5\", text: \"#065f46\", solid: \"#059669\", explanation: \"The remedy granted, disposition, or operative order.\" },\n { key: \"obiter\", label: \"Notable obiter\", short: \"Obiter\", bg: \"#ffe4e6\", text: \"#881337\", solid: \"#e11d48\", explanation: \"A comment made in passing that is not necessary to the decision — persuasive, not binding.\" },\n { key: \"definition\", label: \"Important definition / interpretation\", short: \"Definition\", bg: \"#cffafe\", text: \"#164e63\", solid: \"#0891b2\", explanation: \"A definition or statutory/language interpretation the court adopts.\" },\n];\n\nexport const categoryInfo = (key: CategoryKey): LegendEntry =>\n LEGEND.find((l) => l.key === key) ?? LEGEND[0];\n\nexport interface Paragraph {\n /** 0-based index, stable anchor within the analysis */\n index: number;\n /** Paragraph number shown in the margin (from the document when detected, else sequential) */\n num: string;\n /** True for heading/part labels — never highlighted */\n heading: boolean;\n /** Exact text of the paragraph */\n text: string;\n}\n\nexport interface MetadataField {\n value: string | null;\n confidence: \"detected\" | \"not-detected\";\n}\n\nexport interface CaseMetadata {\n styleOfCause: MetadataField;\n neutralCitation: MetadataField;\n court: MetadataField;\n decisionDate: MetadataField;\n judges: MetadataField;\n docket: MetadataField;\n fileNumber: MetadataField;\n}\n\nexport type LegalType =\n | \"ratio decidendi\"\n | \"legal test / rule\"\n | \"application of law to facts\"\n | \"key factual finding\"\n | \"credibility finding\"\n | \"procedural holding\"\n | \"remedy / order\"\n | \"notable obiter\";\n\nexport interface Highlight {\n id: string;\n /** exact, verbatim quote from the uploaded text (normalized whitespace) */\n exactQuote: string;\n paragraphIndex: number;\n paragraphNumber: string;\n /** character offsets into the ORIGINAL paragraph text for precise highlight rendering */\n startChar: number;\n endChar: number;\n category: CategoryKey;\n legalType: LegalType;\n importance: number; // 0–1\n confidence: number; // 0–1\n whatJudgeIsSaying: string;\n whyMatters: string;\n legalProposition: string; // \"Suggested proposition\"\n citationUse: string; // how to use/cite\n contextSummary: string;\n // user state (client-side)\n status?: \"saved\" | \"not-useful\";\n userNote?: string;\n}\n\nexport interface Analysis {\n source: \"ai\" | \"heuristic\";\n model?: string;\n engineLabel: string;\n metadata: CaseMetadata;\n paragraphs: Paragraph[];\n highlights: Highlight[];\n}\n\nexport type ProgressStage =\n | \"idle\"\n | \"Uploading\"\n | \"Extracting text\"\n | \"Identifying key passages\"\n | \"Building explanations\"\n | \"Ready\";\n","children",[],[65],"f0e0cafd-7acb-4310-8d30-6a8ed2a7d081","index.ts","01a0780a-5903-77b6-aac8-7b10e8bcd475",15,["D",1788720538950],"http","import { parseVal, serveImmutableFile } from \"https://esm.town/v/std/utils/index.ts\";\nimport { Hono } from \"npm:hono\";\nimport { Root } from \"./frontend/root.tsx\";\nimport { analyzeJudgment } from \"./server/analyze.ts\";\n\nconst app = new Hono();\n\n// Serve the (never-cached) HTML shell at the root /\napp.get(\"/\", (c) => c.html(Root()));\n\n// Serve frontend files at the version-addressed URLs Root() stamps, cached immutably\napp.get(\"/__immutable/*\", (c) => serveImmutableFile(c.req.path));\n\n// View source redirect\napp.get(\"/source\", (c) => c.redirect(parseVal().links.self.val));\n\n// The analysis endpoint. The client parses the document in the browser and\n// sends plain text here; this endpoint returns structured highlights.\napp.post(\"/api/analyze\", async (c) => {\n let body: unknown;\n try {\n body = await c.req.json();\n } catch {\n return c.json({ ok: false, error: \"Could not read request body.\" }, 400);\n }\n const text = (body as any)?.text;\n if (typeof text !== \"string\" || text.trim().length < 40) {\n return c.json(\n { ok: false, error: \"No readable judgment text was provided (minimum ~40 characters).\" },\n 400,\n );\n }\n if (text.length > 1_500_000) {\n return c.json({ ok: false, error: \"Document is too large to analyze in one pass.\" }, 413);\n }\n try {\n const analysis = await analyzeJudgment(text);\n return c.json({ ok: true, analysis });\n } catch (e: any) {\n console.log(\"analyze failed:\", e);\n return c.json({ ok: false, error: `Analysis failed: ${e?.message ?? \"unknown error\"}` }, 500);\n }\n});\n\n// Unwrap and rethrow Hono errors as the original error\napp.onError((err) => Promise.reject(err));\n\n// HTTP vals expect an exported \"fetch handler\"\nexport default app.fetch;\n",[],[75],"7ec3c0ac-c897-4a54-b3fe-46a0f1bed21d","README.md","01a0780a-5903-77b6-aac8-7592d4e98454",28,["D",1788720715968],1,"# CaseLens Canada\n\nA polished, Canadian-legal reader that helps users analyse a court or tribunal\ndecision. Upload a judgment (PDF, DOCX, or plain text), and CaseLens:\n\n- Parses it into numbered paragraphs and reads the case header — neutral citation, court, decision date, justice(s), style of cause, and docket number.\n- Identifies **citation-worthy passages** and highlights them **verbatim, in place**, against the categories in the legend below.\n- Explains each highlight in a right-side panel: what the judge is saying, why it matters, the legal type (ratio, test, application, finding, credibility, procedural, remedy, obiter, definition), a suggested proposition, and how to cite it.\n- Lets the user mark passages **Saved / Not useful**, add notes, and **export** the saved set (copy, `.txt`, `.md`).\n\nThe full uploaded judgment stays central — CaseLens never replaces it with a summary, and it never invents citations, parallel citations, history, or treatment. Anything it cannot confidently extract is shown as **\"Not detected\"**.\n\n[Try the app](url:index.ts)\n\n## Analytics engine\n\n- **Default (no key):** a built-in, transparent heuristic engine (regex + rules in `server/analyze.ts`) that works out of the box and is labelled in the UI.\n- **AI (recommended):** set an `ANTHROPIC_API_KEY` and the pipeline uses Claude for deep, structured extraction (`server/ai.ts`). `CLAUDE_MODEL` optionally overrides the default model.\n\n👉 Add ANTHROPIC_API_KEY here: https://www.val.town/x/justinstevens/caselens-canada/environment-variables?key=ANTHROPIC_API_KEY\n\nWhen AI is enabled, every highlight is still **anchored to a verbatim substring** of the uploaded text; candidates that can't be anchored are dropped. If the AI call fails, the app transparently falls back to the heuristic engine.\n\n### How anchoring works\n\n1. `server/text.ts` segments the document into paragraphs (preserving numbers/headings) and extracts case metadata (best-effort, honest).\n2. The engine produces highlight candidates with an `exactQuote` — AI or heuristic.\n3. `server/analyze.ts` finds each quote as a character-exact substring, records `startChar`/`endChar` offsets, and **dedupes overlapping spans** (keeping the more important one).\n\n## Architecture\n\n```mermaid\ngraph LR\n B[Browser] -->|\"PDF / DOCX / .txt parsed client-side\"| T[parse.ts]\n T -->|text| API[\"POST /api/analyze\"]\n API --> PIPE[\"server/analyze.ts (anchor + dedupe)\"]\n PIPE --> AI[\"server/ai.ts (Claude)\"]\n PIPE --> HE[\"server/heuristic.ts\"]\n API -->|Analysis JSON| R[\"Reader (React SPA)\"]\n R --> E[\"Export view\"]\n```\n\n- `index.ts` — Hono HTTP entrypoint; serves the SPA and `/api/analyze`.\n- `server/text.ts`, `server/analyze.ts`, `server/ai.ts` — the analysis pipeline.\n- `frontend/` — React SPA (landing, progress, two-pane reader, export).\n- `frontend/demo.ts` — a clearly-labelled **fictional** sample judgment used by the \"Try the sample judgment\" button.\n- `types.ts` — shared types + the highlight legend (single source of truth for categories and colors).\n\n## Highlight legend\n\n| Category | Meaning |\n| --- | --- |\n| Ratio / core holding | the binding principle the decision turns on |\n| Legal test / rule | a test, standard, or governing rule |\n| Application / reasoning | applying the law to the facts |\n| Key factual finding | a fact the court relies on |\n| Credibility finding | reliability / weight of evidence |\n| Procedural ruling | jurisdiction, standing, admissibility, process |\n| Remedy / disposition / order | the operative order |\n| Notable obiter | persuasive, non-binding comment |\n| Definition | definition or interpretation adopted |\n\n## Privacy\n\nDocuments are parsed **in the browser**; only the extracted plain text is sent to\nCaseLens for analysis. No document is retained server-side.\n\n## Important limitation\n\nCaseLens Canada analyses the text you upload only. It does **not** determine whether\na case is still good law — that requires a citator (e.g. keyCite/CanLII treatment\nchecking) which is not connected. Use this tool to accelerate reading, not to\nreplace your own research.\n\n## View source\n\nEvery page links to this val's source at `/source`.\n\nRun the pipeline standalone with `server/selftest.ts`.\n",[],[84],"3ba147c7-9d07-4444-9b21-71b45cae8c00","CHANGELOG.md","01a0780a-5903-77b6-aac8-70166b5b6141",["D",1788720339103],"# Changelog\n\n## 2026-07-30 — Immutable asset caching\n\nFrontend files are now served at version-addressed URLs\n(`/__immutable/45/frontend/index.tsx`) with\n`Cache-Control: public, max-age=31536000, immutable`, so repeat visits load the\nwhole client graph from the browser cache — roughly 3× faster (~665ms → ~157ms\nmeasured). Publishing your val bumps its version, which changes every URL, so\ncaches invalidate automatically. Old-version URLs return 404 after a publish\n(like Next.js build assets); a reload picks up the new version.\n\nWhat changed in the template:\n\n- `frontend/index.html` → [`frontend/root.tsx`](/frontend/root.tsx): the HTML\n shell is now JSX and stamps asset URLs with `immutableFileUrl()`\n- [`index.ts`](/index.ts): serves `Root()` at `/`, and one\n `app.get(\"/__immutable/*\", (c) => serveImmutableFile(c.req.path))` route\n serves all frontend files\n\n### Upgrading an existing remix\n\nYour copy of this template doesn't update automatically. Two options:\n\n**Minimal (most of the benefit):** point both routes at `serveImmutableFile` —\nyour existing bare URLs redirect into versioned space, the new route serves\nthem, and the whole module graph gets cached (costs one redirect per page view):\n\n```ts\nimport { serveImmutableFile } from \"https://esm.town/v/std/utils/index.ts\";\n\napp.get(\"/__immutable/*\", (c) => serveImmutableFile(c.req.path));\napp.get(\"/frontend/**/*\", (c) => serveImmutableFile(c.req.path)); // was serveFile\n```\n\n**Full upgrade (also removes the redirect):** three self-contained steps —\nno other files needed. Adjust the ``, favicon, and body of `Root()` to\nmatch your app's existing `index.html`, and stamp anything else your shell\nreferences directly (images, css) with `immutableFileUrl()`.\n\n1. Create `frontend/root.tsx`:\n\n```tsx\n/** @jsxImportSource npm:hono@4/jsx */\nimport { raw } from \"npm:hono@4/html\";\nimport { immutableFileUrl } from \"https://esm.town/v/std/utils/index.ts\";\n\n// The HTML shell for the app. It is never cached, so immutableFileUrl can stamp\n// the current val version into each asset URL — the browser caches those\n// immutably, and publishing the val changes the URLs, refetching exactly once.\nexport function Root() {\n return (\n <>\n {raw(\"<!DOCTYPE html>\")}\n <html lang=\"en\">\n <head>\n <meta charSet=\"UTF-8\" />\n <meta name=\"viewport\" content=\"width=device-width, initial-scale=1.0\" />\n <title>React Hono Val Town Starter\n