@postalsys/email-content 0.0.4 → 0.1.0

package/dist/canonical/html.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"html.d.ts","sourceRoot":"","sources":["../../src/canonical/html.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAgDH,sFAAsF;AACtF,wBAAgB,cAAc,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAoBpD;AAQD;;;;;;GAMG;AACH,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAgBpD;AAED,yFAAyF;AACzF,MAAM,WAAW,SAAS;IACtB,GAAG,EAAE,MAAM,EAAE,CAAC;IACd,aAAa,EAAE,OAAO,CAAC;IACvB,eAAe,EAAE,MAAM,CAAC;IACxB,QAAQ,EAAE,OAAO,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,OAAO,CAAC;IAClB,0FAA0F;IAC1F,gBAAgB,EAAE,OAAO,CAAC;CAC7B;AAyBD,4DAA4D;AAC5D,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,GAAG,SAAS,CAqBvD"}
+{"version":3,"file":"html.d.ts","sourceRoot":"","sources":["../../src/canonical/html.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAgDH;;;;;;GAMG;AACH,wBAAgB,cAAc,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAoBpD;AA2FD;;;;;;;;;;;;;;;;;;;GAmBG;AACH,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAmIpD;AAED,yFAAyF;AACzF,MAAM,WAAW,SAAS;IACtB,GAAG,EAAE,MAAM,EAAE,CAAC;IACd,aAAa,EAAE,OAAO,CAAC;IACvB,eAAe,EAAE,MAAM,CAAC;IACxB,QAAQ,EAAE,OAAO,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,OAAO,CAAC;IAClB,0FAA0F;IAC1F,gBAAgB,EAAE,OAAO,CAAC;CAC7B;AAyBD,4DAA4D;AAC5D,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,GAAG,SAAS,CAqBvD"}

package/dist/canonical/html.js

@@ -57,12 +57,18 @@ const NAMED_ENTITIES = {
     zwnj: '‌',
     zwj: '‍'
 };
-/** Decode HTML entities. Unknown entities are left as written rather than dropped. */
+/**
+ * Decode HTML entities. Unknown entities are left as written rather than dropped.
+ *
+ * Both alternatives are length bounded so the pass stays linear on hostile input: a numeric
+ * reference longer than twelve digits cannot name a code point, and an unbounded `+` there would
+ * backtrack across the whole run of digits that follows an unterminated `&#`.
+ */
 function decodeEntities(input) {
     if (!input.includes('&')) {
         return input;
     }
-    return input.replace(/&(#[xX]?[0-9a-fA-F]+|[a-zA-Z][a-zA-Z0-9]{1,31});/g, (whole, body) => {
+    return input.replace(/&(#[xX]?[0-9a-fA-F]{1,12}|[a-zA-Z][a-zA-Z0-9]{1,31});/g, (whole, body) => {
         if (body.charCodeAt(0) === 35) {
             const isHex = body[1] === 'x' || body[1] === 'X';
             const code = parseInt(isHex ? body.slice(2) : body.slice(1), isHex ? 16 : 10);
@@ -80,26 +86,219 @@ function decodeEntities(input) {
         return named === undefined ? whole : named;
     });
 }
-/** Elements whose content is never user visible text. */
-const INVISIBLE = /<(script|style|head|title|noscript|template)\b[^>]*>[\s\S]*?<\/\1\s*>/gi;
+/** Elements whose content is code rather than prose, and must never reach the projection. */
+const RAW_TEXT_TAGS = new Set(['script', 'style']);
+/** Elements whose content is not rendered in the body, but is still prose if the close tag is lost. */
+const HIDDEN_TAGS = new Set(['head', 'title', 'noscript', 'template']);
 /** Elements that introduce a line break in the text projection. */
-const BLOCK_BREAK = /<\/?(p|div|br|tr|li|h[1-6]|table|blockquote|pre|hr|section|article|ul|ol|dd|dt)\b[^>]*>/gi;
+const BLOCK_BREAK_TAGS = new Set([
+    'p',
+    'div',
+    'br',
+    'tr',
+    'li',
+    'h1',
+    'h2',
+    'h3',
+    'h4',
+    'h5',
+    'h6',
+    'table',
+    'blockquote',
+    'pre',
+    'hr',
+    'section',
+    'article',
+    'ul',
+    'ol',
+    'dd',
+    'dt'
+]);
+/**
+ * Longest markup construct scanned before the opening `<` is given up on.
+ *
+ * Real tags are far shorter. The cap exists so that a document consisting of `<` characters with no
+ * closing delimiter cannot make the scanner re-read the suffix once per `<`.
+ */
+const MAX_TAG_LENGTH = 8192;
+const CLOSE_TAG_CACHE = new Map();
+function closeTagRegex(name) {
+    let regex = CLOSE_TAG_CACHE.get(name);
+    if (!regex) {
+        // Global, so the search can be positioned with lastIndex. Slicing the remainder instead
+        // would copy the rest of the document once per script or style element, which is quadratic
+        // on a body full of them.
+        regex = new RegExp(`</${name}(?:\\s[^>]*)?>`, 'gi');
+        CLOSE_TAG_CACHE.set(name, regex);
+    }
+    return regex;
+}
+function isNameStart(code) {
+    return (code >= 97 && code <= 122) || (code >= 65 && code <= 90);
+}
+function isNameChar(code) {
+    return isNameStart(code) || (code >= 48 && code <= 57) || code === 45 || code === 58 || code === 95;
+}
+/**
+ * End of a tag that opens at `from`, or -1 when no unquoted `>` appears within the cap.
+ *
+ * Quoted attribute values are tracked, so `<a title="a>b">` is one tag rather than a tag plus the
+ * stray text `b">`.
+ */
+function tagEnd(html, from) {
+    const limit = Math.min(html.length, from + MAX_TAG_LENGTH);
+    let quote = 0;
+    for (let i = from; i < limit; i++) {
+        const code = html.charCodeAt(i);
+        if (quote) {
+            if (code === quote) {
+                quote = 0;
+            }
+            continue;
+        }
+        if (code === 34 || code === 39) {
+            quote = code;
+            continue;
+        }
+        if (code === 62) {
+            return i;
+        }
+    }
+    return -1;
+}
 /**
  * Cheap text projection of an HTML body.
  *
  * Good enough for language estimation, marker detection and length statistics. It is not a
  * replacement for the extraction pipeline's DOM walk, and it deliberately keeps blank lines so that
- * paragraph structure survives.
+ * paragraph structure survives. It is emphatically **not** a sanitiser: it produces a non executing
+ * textual view, and says nothing about whether the source markup is safe to render.
+ *
+ * Written as a character scanner rather than a chain of regular expressions because the obvious
+ * `/<[^>]*>/g` form is quadratic on unmatched `<`: every one of them rescans to end of input looking
+ * for a closing delimiter that is not there. Measured at 5.8 seconds on 128k `<` characters, and
+ * doubling the input quadrupled it. This is the projection the malformed HTML fallback uses, so it
+ * is reached by exactly the input that triggers the quadratic case.
+ *
+ * Every code unit is visited a constant number of times. A construct that never closes is charged
+ * to the input it covers: the scanned window is emitted as visible text and the cursor moves past
+ * it, rather than resuming one character later and reading the same suffix again. Emitting it keeps
+ * the projection on the safe side of the project's asymmetry, and matches what the previous regular
+ * expression did with an unterminated trailing tag.
  */
 function stripTagsToText(html) {
-    let text = html.replace(INVISIBLE, ' ');
-    text = text.replace(/<!--[\s\S]*?-->/g, ' ');
-    // Word and Outlook emit downlevel-revealed conditional comments, which are not HTML comments
-    // and survive both the comment strip and the tag strip: <![if !supportLists]>. Left in, they
-    // appear verbatim in the extracted text of every Word authored bullet list.
-    text = text.replace(/<!\[[^\]]*\]>/g, ' ');
-    text = text.replace(BLOCK_BREAK, '\n');
-    text = text.replace(/<[^>]*>/g, '');
+    const out = [];
+    const length = html.length;
+    let i = 0;
+    // Closing delimiter memos. The cursor only moves forward, so once a search for a closer finds
+    // nothing after some position it will find nothing after any later position either. Without
+    // this, a document of unterminated `<!--` openers searches the whole suffix once per opener,
+    // which is the same quadratic shape the tag stripper was rewritten to remove.
+    let noCommentClose = false;
+    const noRawClose = new Set();
+    while (i < length) {
+        const lt = html.indexOf('<', i);
+        if (lt === -1) {
+            out.push(html.slice(i));
+            break;
+        }
+        if (lt > i) {
+            out.push(html.slice(i, lt));
+        }
+        const next = html.charCodeAt(lt + 1);
+        // Comments and declarations.
+        if (next === 33) {
+            if (html.startsWith('<!--', lt)) {
+                const close = noCommentClose ? -1 : html.indexOf('-->', lt + 4);
+                if (close === -1) {
+                    noCommentClose = true;
+                    // Unterminated. Keep the body as text: showing a comment is a cosmetic fault,
+                    // hiding the rest of a message because one comment was never closed is not.
+                    out.push(' ');
+                    i = lt + 4;
+                    continue;
+                }
+                out.push(' ');
+                i = close + 3;
+                continue;
+            }
+            // Doctypes, CDATA, and the downlevel-revealed conditional comments Word emits around
+            // every bullet list: <![if !supportLists]>. Those are not HTML comments and survived
+            // both the comment strip and the tag strip, so they appeared verbatim in extracted text.
+            const close = tagEnd(html, lt + 2);
+            if (close === -1) {
+                out.push(html.slice(lt, Math.min(length, lt + MAX_TAG_LENGTH)));
+                i = Math.min(length, lt + MAX_TAG_LENGTH);
+                continue;
+            }
+            out.push(' ');
+            i = close + 1;
+            continue;
+        }
+        // Processing instructions, including the <?xml ... ?> prologue Word emits.
+        if (next === 63) {
+            const close = tagEnd(html, lt + 2);
+            if (close === -1) {
+                out.push(html.slice(lt, Math.min(length, lt + MAX_TAG_LENGTH)));
+                i = Math.min(length, lt + MAX_TAG_LENGTH);
+                continue;
+            }
+            out.push(' ');
+            i = close + 1;
+            continue;
+        }
+        const closing = next === 47;
+        const nameStart = closing ? lt + 2 : lt + 1;
+        if (!isNameStart(html.charCodeAt(nameStart))) {
+            // Not markup at all: a bare "<" in prose, "a < b", "<3". Visible text.
+            out.push('<');
+            i = lt + 1;
+            continue;
+        }
+        let nameEnd = nameStart;
+        while (nameEnd < length && isNameChar(html.charCodeAt(nameEnd))) {
+            nameEnd++;
+        }
+        const name = html.slice(nameStart, nameEnd).toLowerCase();
+        const close = tagEnd(html, nameEnd);
+        if (close === -1) {
+            out.push(html.slice(lt, Math.min(length, lt + MAX_TAG_LENGTH)));
+            i = Math.min(length, lt + MAX_TAG_LENGTH);
+            continue;
+        }
+        if (!closing && (RAW_TEXT_TAGS.has(name) || HIDDEN_TAGS.has(name))) {
+            let match = null;
+            if (!noRawClose.has(name)) {
+                const regex = closeTagRegex(name);
+                regex.lastIndex = close + 1;
+                match = regex.exec(html);
+                if (!match) {
+                    noRawClose.add(name);
+                }
+            }
+            if (match) {
+                out.push(' ');
+                i = match.index + match[0].length;
+                continue;
+            }
+            if (RAW_TEXT_TAGS.has(name)) {
+                // A script or style that is never closed swallows the rest of the document in a
+                // renderer too, and its body is code, not authored prose.
+                out.push(' ');
+                break;
+            }
+            // head/title/noscript/template with no close tag: drop only the opening tag, so that
+            // whatever follows stays visible.
+            out.push(' ');
+            i = close + 1;
+            continue;
+        }
+        if (BLOCK_BREAK_TAGS.has(name)) {
+            out.push('\n');
+        }
+        i = close + 1;
+    }
+    let text = out.join('');
     text = decodeEntities(text);
     text = text.replace(/\r\n?/g, '\n');
     // Includes U+00A0, which arrives constantly as &nbsp; and must collapse like a normal space.

package/dist/canonical/html.js.map

@@ -1 +1 @@
-{"version":3,"file":"html.js","sourceRoot":"","sources":["../../src/canonical/html.ts"],"names":[],"mappings":";AAAA;;;;;;;;GAQG;;AAiDH,wCAoBC;AAeD,0CAgBC;AAsCD,0CAqBC;AA7JD,0EAA0E;AAC1E,MAAM,cAAc,GAA2B;IAC3C,GAAG,EAAE,GAAG;IACR,EAAE,EAAE,GAAG;IACP,EAAE,EAAE,GAAG;IACP,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,MAAM,EAAE,GAAG;IACX,IAAI,EAAE,GAAG;IACT,MAAM,EAAE,GAAG;IACX,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,IAAI,EAAE,GAAG;IACT,GAAG,EAAE,GAAG;IACR,KAAK,EAAE,GAAG;IACV,IAAI,EAAE,GAAG;IACT,KAAK,EAAE,GAAG;IACV,GAAG,EAAE,GAAG;IACR,KAAK,EAAE,GAAG;IACV,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,MAAM,EAAE,GAAG;IACX,MAAM,EAAE,GAAG;IACX,MAAM,EAAE,GAAG;IACX,MAAM,EAAE,GAAG;IACX,MAAM,EAAE,GAAG;IACX,MAAM,EAAE,GAAG;IACX,GAAG,EAAE,GAAG;IACR,IAAI,EAAE,GAAG;IACT,GAAG,EAAE,GAAG;CACX,CAAC;AAEF,sFAAsF;AACtF,SAAgB,cAAc,CAAC,KAAa;IACxC,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;QACvB,OAAO,KAAK,CAAC;IACjB,CAAC;IACD,OAAO,KAAK,CAAC,OAAO,CAAC,mDAAmD,EAAE,CAAC,KAAK,EAAE,IAAY,EAAE,EAAE;QAC9F,IAAI,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;YAC5B,MAAM,KAAK,GAAG,IAAI,CAAC,CAAC,CAAC,KAAK,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC;YACjD,MAAM,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;YAC9E,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,IAAI,GAAG,CAAC,IAAI,IAAI,GAAG,QAAQ,EAAE,CAAC;gBACxD,OAAO,KAAK,CAAC;YACjB,CAAC;YACD,IAAI,CAAC;gBACD,OAAO,MAAM,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC;YACtC,CAAC;YAAC,MAAM,CAAC;gBACL,OAAO,KAAK,CAAC;YACjB,CAAC;QACL,CAAC;QACD,MAAM,KAAK,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC;QACnC,OAAO,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC;IAC/C,CAAC,CAAC,CAAC;AACP,CAAC;AAED,yDAAyD;AACzD,MAAM,SAAS,GAAG,yEAAyE,CAAC;AAE5F,mEAAmE;AACnE,MAAM,WAAW,GAAG,2FAA2F,CAAC;AAEhH;;;;;;GAMG;AACH,SAAgB,eAAe,CAAC,IAAY;IACxC,IAAI,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC;IACxC,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,kBAAkB,EAAE,GAAG,CAAC,CAAC;IAC7C,6FAA6F;IAC7F,6FAA6F;IAC7F,4EAA4E;IAC5E,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,gBAAgB,EAAE,GAAG,CAAC,CAAC;IAC3C,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC;IACvC,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;IACpC,IAAI,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC;IAC5B,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;IACpC,6FAA6F;IAC7F,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,eAAe,EAAE,GAAG,CAAC,CAAC;IAC1C,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;IACrC,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;IACvC,OAAO,IAAI,CAAC,IAAI,EAAE,CAAC;AACvB,CAAC;AAcD,MAAM,aAAa,GAA6C;IAC5D,EAAE,EAAE,EAAE,aAAa,EAAE,KAAK,EAAE,8BAA8B,EAAE;IAC5D,EAAE,EAAE,EAAE,uBAAuB,EAAE,KAAK,EAAE,wCAAwC,EAAE;IAChF,EAAE,EAAE,EAAE,YAAY,EAAE,KAAK,EAAE,6BAA6B,EAAE;IAC1D,EAAE,EAAE,EAAE,iBAAiB,EAAE,KAAK,EAAE,kCAAkC,EAAE;IACpE,EAAE,EAAE,EAAE,qBAAqB,EAAE,KAAK,EAAE,qBAAqB,EAAE;IAC3D,EAAE,EAAE,EAAE,oBAAoB,EAAE,KAAK,EAAE,oBAAoB,EAAE;IACzD,EAAE,EAAE,EAAE,sBAAsB,EAAE,KAAK,EAAE,6BAA6B,EAAE;IACpE,EAAE,EAAE,EAAE,oBAAoB,EAAE,KAAK,EAAE,6BAA6B,EAAE;IAClE,EAAE,EAAE,EAAE,sBAAsB,EAAE,KAAK,EAAE,mBAAmB,EAAE;IAC1D,EAAE,EAAE,EAAE,wBAAwB,EAAE,KAAK,EAAE,0BAA0B,EAAE;IACnE,EAAE,EAAE,EAAE,iBAAiB,EAAE,KAAK,EAAE,8BAA8B,EAAE;IAChE,EAAE,EAAE,EAAE,2BAA2B,EAAE,KAAK,EAAE,4BAA4B,EAAE;IACxE,EAAE,EAAE,EAAE,sBAAsB,EAAE,KAAK,EAAE,qBAAqB,EAAE;IAC5D,EAAE,EAAE,EAAE,sBAAsB,EAAE,KAAK,EAAE,kBAAkB,EAAE;IACzD,EAAE,EAAE,EAAE,uBAAuB,EAAE,KAAK,EAAE,gBAAgB,EAAE;IACxD,EAAE,EAAE,EAAE,cAAc,EAAE,KAAK,EAAE,mBAAmB,EAAE;IAClD,EAAE,EAAE,EAAE,cAAc,EAAE,KAAK,EAAE,eAAe,EAAE;IAC9C,EAAE,EAAE,EAAE,kBAAkB,EAAE,KAAK,EAAE,aAAa,EAAE;IAChD,EAAE,EAAE,EAAE,0BAA0B,EAAE,KAAK,EAAE,mCAAmC,EAAE;IAC9E,EAAE,EAAE,EAAE,mBAAmB,EAAE,KAAK,EAAE,8CAA8C,EAAE;CACrF,CAAC;AAEF,4DAA4D;AAC5D,SAAgB,eAAe,CAAC,IAAY;IACxC,MAAM,IAAI,GAAG,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IACjE,MAAM,GAAG,GAAa,EAAE,CAAC;IACzB,KAAK,MAAM,EAAE,EAAE,EAAE,KAAK,EAAE,IAAI,aAAa,EAAE,CAAC;QACxC,IAAI,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;YACnB,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACjB,CAAC;IACL,CAAC;IACD,MAAM,eAAe,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,iBAAiB,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC;IACrE,MAAM,UAAU,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC;IAC3D,OAAO;QACH,GAAG;QACH,aAAa,EAAE,eAAe,GAAG,CAAC;QAClC,eAAe;QACf,QAAQ,EAAE,UAAU,GAAG,CAAC;QACxB,UAAU;QACV,QAAQ,EAAE,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC;QAC9B,0FAA0F;QAC1F,kDAAkD;QAClD,gBAAgB,EAAE,sCAAsC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,eAAe,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,kBAAkB,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM;KAC3I,CAAC;AACN,CAAC"}
+{"version":3,"file":"html.js","sourceRoot":"","sources":["../../src/canonical/html.ts"],"names":[],"mappings":";AAAA;;;;;;;;GAQG;;AAuDH,wCAoBC;AA+GD,0CAmIC;AAsCD,0CAqBC;AAtXD,0EAA0E;AAC1E,MAAM,cAAc,GAA2B;IAC3C,GAAG,EAAE,GAAG;IACR,EAAE,EAAE,GAAG;IACP,EAAE,EAAE,GAAG;IACP,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,MAAM,EAAE,GAAG;IACX,IAAI,EAAE,GAAG;IACT,MAAM,EAAE,GAAG;IACX,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,IAAI,EAAE,GAAG;IACT,GAAG,EAAE,GAAG;IACR,KAAK,EAAE,GAAG;IACV,IAAI,EAAE,GAAG;IACT,KAAK,EAAE,GAAG;IACV,GAAG,EAAE,GAAG;IACR,KAAK,EAAE,GAAG;IACV,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,IAAI,EAAE,GAAG;IACT,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,KAAK,EAAE,GAAG;IACV,MAAM,EAAE,GAAG;IACX,MAAM,EAAE,GAAG;IACX,MAAM,EAAE,GAAG;IACX,MAAM,EAAE,GAAG;IACX,MAAM,EAAE,GAAG;IACX,MAAM,EAAE,GAAG;IACX,GAAG,EAAE,GAAG;IACR,IAAI,EAAE,GAAG;IACT,GAAG,EAAE,GAAG;CACX,CAAC;AAEF;;;;;;GAMG;AACH,SAAgB,cAAc,CAAC,KAAa;IACxC,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;QACvB,OAAO,KAAK,CAAC;IACjB,CAAC;IACD,OAAO,KAAK,CAAC,OAAO,CAAC,wDAAwD,EAAE,CAAC,KAAK,EAAE,IAAY,EAAE,EAAE;QACnG,IAAI,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;YAC5B,MAAM,KAAK,GAAG,IAAI,CAAC,CAAC,CAAC,KAAK,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC;YACjD,MAAM,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;YAC9E,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,IAAI,GAAG,CAAC,IAAI,IAAI,GAAG,QAAQ,EAAE,CAAC;gBACxD,OAAO,KAAK,CAAC;YACjB,CAAC;YACD,IAAI,CAAC;gBACD,OAAO,MAAM,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC;YACtC,CAAC;YAAC,MAAM,CAAC;gBACL,OAAO,KAAK,CAAC;YACjB,CAAC;QACL,CAAC;QACD,MAAM,KAAK,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC;QACnC,OAAO,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC;IAC/C,CAAC,CAAC,CAAC;AACP,CAAC;AAED,6FAA6F;AAC7F,MAAM,aAAa,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC,CAAC;AAEnD,uGAAuG;AACvG,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE,UAAU,CAAC,CAAC,CAAC;AAEvE,mEAAmE;AACnE,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC;IAC7B,GAAG;IACH,KAAK;IACL,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,OAAO;IACP,YAAY;IACZ,KAAK;IACL,IAAI;IACJ,SAAS;IACT,SAAS;IACT,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;CACP,CAAC,CAAC;AAEH;;;;;GAKG;AACH,MAAM,cAAc,GAAG,IAAI,CAAC;AAE5B,MAAM,eAAe,GAAG,IAAI,GAAG,EAAkB,CAAC;AAElD,SAAS,aAAa,CAAC,IAAY;IAC/B,IAAI,KAAK,GAAG,eAAe,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACtC,IAAI,CAAC,KAAK,EAAE,CAAC;QACT,wFAAwF;QACxF,2FAA2F;QAC3F,0BAA0B;QAC1B,KAAK,GAAG,IAAI,MAAM,CAAC,KAAK,IAAI,gBAAgB,EAAE,IAAI,CAAC,CAAC;QACpD,eAAe,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IACrC,CAAC;IACD,OAAO,KAAK,CAAC;AACjB,CAAC;AAED,SAAS,WAAW,CAAC,IAAY;IAC7B,OAAO,CAAC,IAAI,IAAI,EAAE,IAAI,IAAI,IAAI,GAAG,CAAC,IAAI,CAAC,IAAI,IAAI,EAAE,IAAI,IAAI,IAAI,EAAE,CAAC,CAAC;AACrE,CAAC;AAED,SAAS,UAAU,CAAC,IAAY;IAC5B,OAAO,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,EAAE,IAAI,IAAI,IAAI,EAAE,CAAC,IAAI,IAAI,KAAK,EAAE,IAAI,IAAI,KAAK,EAAE,IAAI,IAAI,KAAK,EAAE,CAAC;AACxG,CAAC;AAED;;;;;GAKG;AACH,SAAS,MAAM,CAAC,IAAY,EAAE,IAAY;IACtC,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,EAAE,IAAI,GAAG,cAAc,CAAC,CAAC;IAC3D,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,IAAI,CAAC,GAAG,IAAI,EAAE,CAAC,GAAG,KAAK,EAAE,CAAC,EAAE,EAAE,CAAC;QAChC,MAAM,IAAI,GAAG,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC;QAChC,IAAI,KAAK,EAAE,CAAC;YACR,IAAI,IAAI,KAAK,KAAK,EAAE,CAAC;gBACjB,KAAK,GAAG,CAAC,CAAC;YACd,CAAC;YACD,SAAS;QACb,CAAC;QACD,IAAI,IAAI,KAAK,EAAE,IAAI,IAAI,KAAK,EAAE,EAAE,CAAC;YAC7B,KAAK,GAAG,IAAI,CAAC;YACb,SAAS;QACb,CAAC;QACD,IAAI,IAAI,KAAK,EAAE,EAAE,CAAC;YACd,OAAO,CAAC,CAAC;QACb,CAAC;IACL,CAAC;IACD,OAAO,CAAC,CAAC,CAAC;AACd,CAAC;AAED;;;;;;;;;;;;;;;;;;;GAmBG;AACH,SAAgB,eAAe,CAAC,IAAY;IACxC,MAAM,GAAG,GAAa,EAAE,CAAC;IACzB,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC;IAC3B,IAAI,CAAC,GAAG,CAAC,CAAC;IAEV,8FAA8F;IAC9F,4FAA4F;IAC5F,6FAA6F;IAC7F,8EAA8E;IAC9E,IAAI,cAAc,GAAG,KAAK,CAAC;IAC3B,MAAM,UAAU,GAAG,IAAI,GAAG,EAAU,CAAC;IAErC,OAAO,CAAC,GAAG,MAAM,EAAE,CAAC;QAChB,MAAM,EAAE,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;QAChC,IAAI,EAAE,KAAK,CAAC,CAAC,EAAE,CAAC;YACZ,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;YACxB,MAAM;QACV,CAAC;QACD,IAAI,EAAE,GAAG,CAAC,EAAE,CAAC;YACT,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;QAChC,CAAC;QAED,MAAM,IAAI,GAAG,IAAI,CAAC,UAAU,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;QAErC,6BAA6B;QAC7B,IAAI,IAAI,KAAK,EAAE,EAAE,CAAC;YACd,IAAI,IAAI,CAAC,UAAU,CAAC,MAAM,EAAE,EAAE,CAAC,EAAE,CAAC;gBAC9B,MAAM,KAAK,GAAG,cAAc,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,GAAG,CAAC,CAAC,CAAC;gBAChE,IAAI,KAAK,KAAK,CAAC,CAAC,EAAE,CAAC;oBACf,cAAc,GAAG,IAAI,CAAC;oBACtB,8EAA8E;oBAC9E,4EAA4E;oBAC5E,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;oBACd,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;oBACX,SAAS;gBACb,CAAC;gBACD,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;gBACd,CAAC,GAAG,KAAK,GAAG,CAAC,CAAC;gBACd,SAAS;YACb,CAAC;YACD,qFAAqF;YACrF,qFAAqF;YACrF,yFAAyF;YACzF,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,EAAE,EAAE,GAAG,CAAC,CAAC,CAAC;YACnC,IAAI,KAAK,KAAK,CAAC,CAAC,EAAE,CAAC;gBACf,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,EAAE,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,EAAE,GAAG,cAAc,CAAC,CAAC,CAAC,CAAC;gBAChE,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,EAAE,GAAG,cAAc,CAAC,CAAC;gBAC1C,SAAS;YACb,CAAC;YACD,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACd,CAAC,GAAG,KAAK,GAAG,CAAC,CAAC;YACd,SAAS;QACb,CAAC;QAED,2EAA2E;QAC3E,IAAI,IAAI,KAAK,EAAE,EAAE,CAAC;YACd,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,EAAE,EAAE,GAAG,CAAC,CAAC,CAAC;YACnC,IAAI,KAAK,KAAK,CAAC,CAAC,EAAE,CAAC;gBACf,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,EAAE,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,EAAE,GAAG,cAAc,CAAC,CAAC,CAAC,CAAC;gBAChE,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,EAAE,GAAG,cAAc,CAAC,CAAC;gBAC1C,SAAS;YACb,CAAC;YACD,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACd,CAAC,GAAG,KAAK,GAAG,CAAC,CAAC;YACd,SAAS;QACb,CAAC;QAED,MAAM,OAAO,GAAG,IAAI,KAAK,EAAE,CAAC;QAC5B,MAAM,SAAS,GAAG,OAAO,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;QAC5C,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,EAAE,CAAC;YAC3C,uEAAuE;YACvE,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACd,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;YACX,SAAS;QACb,CAAC;QAED,IAAI,OAAO,GAAG,SAAS,CAAC;QACxB,OAAO,OAAO,GAAG,MAAM,IAAI,UAAU,CAAC,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,EAAE,CAAC;YAC9D,OAAO,EAAE,CAAC;QACd,CAAC;QACD,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC,WAAW,EAAE,CAAC;QAE1D,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;QACpC,IAAI,KAAK,KAAK,CAAC,CAAC,EAAE,CAAC;YACf,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,EAAE,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,EAAE,GAAG,cAAc,CAAC,CAAC,CAAC,CAAC;YAChE,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,EAAE,GAAG,cAAc,CAAC,CAAC;YAC1C,SAAS;QACb,CAAC;QAED,IAAI,CAAC,OAAO,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,WAAW,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;YACjE,IAAI,KAAK,GAA2B,IAAI,CAAC;YACzC,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;gBACxB,MAAM,KAAK,GAAG,aAAa,CAAC,IAAI,CAAC,CAAC;gBAClC,KAAK,CAAC,SAAS,GAAG,KAAK,GAAG,CAAC,CAAC;gBAC5B,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;gBACzB,IAAI,CAAC,KAAK,EAAE,CAAC;oBACT,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;gBACzB,CAAC;YACL,CAAC;YACD,IAAI,KAAK,EAAE,CAAC;gBACR,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;gBACd,CAAC,GAAG,KAAK,CAAC,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC;gBAClC,SAAS;YACb,CAAC;YACD,IAAI,aAAa,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;gBAC1B,gFAAgF;gBAChF,0DAA0D;gBAC1D,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;gBACd,MAAM;YACV,CAAC;YACD,qFAAqF;YACrF,kCAAkC;YAClC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACd,CAAC,GAAG,KAAK,GAAG,CAAC,CAAC;YACd,SAAS;QACb,CAAC;QAED,IAAI,gBAAgB,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;YAC7B,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACnB,CAAC;QACD,CAAC,GAAG,KAAK,GAAG,CAAC,CAAC;IAClB,CAAC;IAED,IAAI,IAAI,GAAG,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACxB,IAAI,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC;IAC5B,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;IACpC,6FAA6F;IAC7F,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,eAAe,EAAE,GAAG,CAAC,CAAC;IAC1C,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;IACrC,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;IACvC,OAAO,IAAI,CAAC,IAAI,EAAE,CAAC;AACvB,CAAC;AAcD,MAAM,aAAa,GAA6C;IAC5D,EAAE,EAAE,EAAE,aAAa,EAAE,KAAK,EAAE,8BAA8B,EAAE;IAC5D,EAAE,EAAE,EAAE,uBAAuB,EAAE,KAAK,EAAE,wCAAwC,EAAE;IAChF,EAAE,EAAE,EAAE,YAAY,EAAE,KAAK,EAAE,6BAA6B,EAAE;IAC1D,EAAE,EAAE,EAAE,iBAAiB,EAAE,KAAK,EAAE,kCAAkC,EAAE;IACpE,EAAE,EAAE,EAAE,qBAAqB,EAAE,KAAK,EAAE,qBAAqB,EAAE;IAC3D,EAAE,EAAE,EAAE,oBAAoB,EAAE,KAAK,EAAE,oBAAoB,EAAE;IACzD,EAAE,EAAE,EAAE,sBAAsB,EAAE,KAAK,EAAE,6BAA6B,EAAE;IACpE,EAAE,EAAE,EAAE,oBAAoB,EAAE,KAAK,EAAE,6BAA6B,EAAE;IAClE,EAAE,EAAE,EAAE,sBAAsB,EAAE,KAAK,EAAE,mBAAmB,EAAE;IAC1D,EAAE,EAAE,EAAE,wBAAwB,EAAE,KAAK,EAAE,0BAA0B,EAAE;IACnE,EAAE,EAAE,EAAE,iBAAiB,EAAE,KAAK,EAAE,8BAA8B,EAAE;IAChE,EAAE,EAAE,EAAE,2BAA2B,EAAE,KAAK,EAAE,4BAA4B,EAAE;IACxE,EAAE,EAAE,EAAE,sBAAsB,EAAE,KAAK,EAAE,qBAAqB,EAAE;IAC5D,EAAE,EAAE,EAAE,sBAAsB,EAAE,KAAK,EAAE,kBAAkB,EAAE;IACzD,EAAE,EAAE,EAAE,uBAAuB,EAAE,KAAK,EAAE,gBAAgB,EAAE;IACxD,EAAE,EAAE,EAAE,cAAc,EAAE,KAAK,EAAE,mBAAmB,EAAE;IAClD,EAAE,EAAE,EAAE,cAAc,EAAE,KAAK,EAAE,eAAe,EAAE;IAC9C,EAAE,EAAE,EAAE,kBAAkB,EAAE,KAAK,EAAE,aAAa,EAAE;IAChD,EAAE,EAAE,EAAE,0BAA0B,EAAE,KAAK,EAAE,mCAAmC,EAAE;IAC9E,EAAE,EAAE,EAAE,mBAAmB,EAAE,KAAK,EAAE,8CAA8C,EAAE;CACrF,CAAC;AAEF,4DAA4D;AAC5D,SAAgB,eAAe,CAAC,IAAY;IACxC,MAAM,IAAI,GAAG,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IACjE,MAAM,GAAG,GAAa,EAAE,CAAC;IACzB,KAAK,MAAM,EAAE,EAAE,EAAE,KAAK,EAAE,IAAI,aAAa,EAAE,CAAC;QACxC,IAAI,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;YACnB,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACjB,CAAC;IACL,CAAC;IACD,MAAM,eAAe,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,iBAAiB,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC;IACrE,MAAM,UAAU,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC;IAC3D,OAAO;QACH,GAAG;QACH,aAAa,EAAE,eAAe,GAAG,CAAC;QAClC,eAAe;QACf,QAAQ,EAAE,UAAU,GAAG,CAAC;QACxB,UAAU;QACV,QAAQ,EAAE,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC;QAC9B,0FAA0F;QAC1F,kDAAkD;QAClD,gBAAgB,EAAE,sCAAsC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,eAAe,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,kBAAkB,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM;KAC3I,CAAC;AACN,CAAC"}

package/dist/classify/confidence.d.ts

@@ -15,6 +15,18 @@ export interface ConfidenceConfig {
     minAnchorStrength: number;
 }
 export declare const DEFAULT_CONFIDENCE: ConfidenceConfig;
+/**
+ * Validate caller supplied confidence thresholds and fill in the defaults.
+ *
+ * Rejects, never clamps. Clamping hides a configuration mistake and makes the pipeline behave
+ * differently from what the caller asked for, on every message, silently. `NaN` is the case that
+ * matters most: every comparison against it is false, so a `NaN` threshold does not merely shift
+ * behaviour, it makes the visibility bands stop working while everything still looks fine.
+ *
+ * Takes a partial. The previous signature required a complete object, so a caller wanting to move
+ * one threshold had to copy the other two and then own them forever.
+ */
+export declare function validateConfidenceConfig(input?: Partial<ConfidenceConfig>): ConfidenceConfig;
 /**
  * Map a label and confidence to a visibility band.
  *

package/dist/classify/confidence.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"confidence.d.ts","sourceRoot":"","sources":["../../src/classify/confidence.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAE,KAAK,KAAK,EAAE,KAAK,UAAU,EAAE,MAAM,UAAU,CAAC;AAEvD,MAAM,WAAW,gBAAgB;IAC7B,wEAAwE;IACxE,aAAa,EAAE,MAAM,CAAC;IACtB,uEAAuE;IACvE,iBAAiB,EAAE,MAAM,CAAC;IAC1B,iEAAiE;IACjE,iBAAiB,EAAE,MAAM,CAAC;CAC7B;AAED,eAAO,MAAM,kBAAkB,EAAE,gBAIhC,CAAC;AAEF;;;;;;;;GAQG;AACH,wBAAgB,aAAa,CAAC,KAAK,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,EAAE,MAAM,GAAE,gBAAqC,GAAG,UAAU,CAkBzH;AAED;;;;;GAKG;AACH,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,SAAS;IAAE,UAAU,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,UAAU,CAAA;CAAE,EAAE,GAAG,MAAM,CAM9G"}
+{"version":3,"file":"confidence.d.ts","sourceRoot":"","sources":["../../src/classify/confidence.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAsB,KAAK,KAAK,EAAE,KAAK,UAAU,EAAE,MAAM,UAAU,CAAC;AAE3E,MAAM,WAAW,gBAAgB;IAC7B,wEAAwE;IACxE,aAAa,EAAE,MAAM,CAAC;IACtB,uEAAuE;IACvE,iBAAiB,EAAE,MAAM,CAAC;IAC1B,iEAAiE;IACjE,iBAAiB,EAAE,MAAM,CAAC;CAC7B;AAED,eAAO,MAAM,kBAAkB,EAAE,gBAIhC,CAAC;AASF;;;;;;;;;;GAUG;AACH,wBAAgB,wBAAwB,CAAC,KAAK,CAAC,EAAE,OAAO,CAAC,gBAAgB,CAAC,GAAG,gBAAgB,CA6B5F;AAED;;;;;;;;GAQG;AACH,wBAAgB,aAAa,CAAC,KAAK,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,EAAE,MAAM,GAAE,gBAAqC,GAAG,UAAU,CAkBzH;AAED;;;;;GAKG;AACH,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,SAAS;IAAE,UAAU,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,UAAU,CAAA;CAAE,EAAE,GAAG,MAAM,CAU9G"}

package/dist/classify/confidence.js

@@ -8,13 +8,59 @@
  */
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.DEFAULT_CONFIDENCE = void 0;
+exports.validateConfidenceConfig = validateConfidenceConfig;
 exports.visibilityFor = visibilityFor;
 exports.overallConfidence = overallConfidence;
+const types_1 = require("../types");
 exports.DEFAULT_CONFIDENCE = {
     hideThreshold: 0.95,
     collapseThreshold: 0.75,
     minAnchorStrength: 0.85
 };
+function requireUnitInterval(value, option) {
+    if (typeof value !== 'number' || !Number.isFinite(value) || value < 0 || value > 1) {
+        throw new types_1.ConfigurationError(option, 'a finite number between 0 and 1');
+    }
+    return value;
+}
+/**
+ * Validate caller supplied confidence thresholds and fill in the defaults.
+ *
+ * Rejects, never clamps. Clamping hides a configuration mistake and makes the pipeline behave
+ * differently from what the caller asked for, on every message, silently. `NaN` is the case that
+ * matters most: every comparison against it is false, so a `NaN` threshold does not merely shift
+ * behaviour, it makes the visibility bands stop working while everything still looks fine.
+ *
+ * Takes a partial. The previous signature required a complete object, so a caller wanting to move
+ * one threshold had to copy the other two and then own them forever.
+ */
+function validateConfidenceConfig(input) {
+    if (input === undefined || input === null) {
+        return exports.DEFAULT_CONFIDENCE;
+    }
+    if (typeof input !== 'object') {
+        throw new types_1.ConfigurationError('confidence', 'an object');
+    }
+    const config = { ...exports.DEFAULT_CONFIDENCE };
+    if (input.hideThreshold !== undefined) {
+        config.hideThreshold = requireUnitInterval(input.hideThreshold, 'confidence.hideThreshold');
+    }
+    if (input.collapseThreshold !== undefined) {
+        config.collapseThreshold = requireUnitInterval(input.collapseThreshold, 'confidence.collapseThreshold');
+    }
+    if (input.minAnchorStrength !== undefined) {
+        config.minAnchorStrength = requireUnitInterval(input.minAnchorStrength, 'confidence.minAnchorStrength');
+    }
+    // Inverted bands would make `collapse` unreachable and turn every collapse decision into a
+    // hide, which is the direction that loses content.
+    if (config.collapseThreshold > config.hideThreshold) {
+        throw new types_1.ConfigurationError('confidence.collapseThreshold', 'less than or equal to confidence.hideThreshold', 'inverted_thresholds');
+    }
+    // `minAnchorStrength` is deliberately not compared against either threshold. Anchor strength
+    // measures how good the structural evidence is; the thresholds measure how confident the
+    // classification is. They are different dimensions and ordering them would be meaningless.
+    return config;
+}
 /**
  * Map a label and confidence to a visibility band.
  *
@@ -50,10 +96,14 @@ function visibilityFor(label, confidence, config = exports.DEFAULT_CONFIDENCE) {
  * blocks mask one risky decision, and the risky decision is the one that loses content.
  */
 function overallConfidence(decisions) {
-    const risky = decisions.filter(decision => decision.visibility !== 'show');
-    if (!risky.length) {
-        return 1;
+    // A loop, not `Math.min(...)`: spreading a segment array into a call passes one argument per
+    // element and blows V8's argument limit on a large enough message.
+    let lowest = Infinity;
+    for (const decision of decisions) {
+        if (decision.visibility !== 'show' && decision.confidence < lowest) {
+            lowest = decision.confidence;
+        }
     }
-    return Math.min(...risky.map(decision => decision.confidence));
+    return lowest === Infinity ? 1 : lowest;
 }
 //# sourceMappingURL=confidence.js.map

package/dist/classify/confidence.js.map

@@ -1 +1 @@
-{"version":3,"file":"confidence.js","sourceRoot":"","sources":["../../src/classify/confidence.ts"],"names":[],"mappings":";AAAA;;;;;;GAMG;;;AA4BH,sCAkBC;AAQD,8CAMC;AA/CY,QAAA,kBAAkB,GAAqB;IAChD,aAAa,EAAE,IAAI;IACnB,iBAAiB,EAAE,IAAI;IACvB,iBAAiB,EAAE,IAAI;CAC1B,CAAC;AAEF;;;;;;;;GAQG;AACH,SAAgB,aAAa,CAAC,KAAY,EAAE,UAAkB,EAAE,SAA2B,0BAAkB;IACzG,QAAQ,KAAK,EAAE,CAAC;QACZ,KAAK,eAAe;YAChB,IAAI,UAAU,IAAI,MAAM,CAAC,aAAa,EAAE,CAAC;gBACrC,OAAO,MAAM,CAAC;YAClB,CAAC;YACD,OAAO,UAAU,IAAI,MAAM,CAAC,iBAAiB,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,MAAM,CAAC;QACxE,KAAK,mBAAmB;YACpB,OAAO,UAAU,IAAI,MAAM,CAAC,iBAAiB,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,MAAM,CAAC;QACxE,KAAK,WAAW,CAAC;QACjB,KAAK,YAAY;YACb,OAAO,UAAU,IAAI,MAAM,CAAC,iBAAiB,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,MAAM,CAAC;QACxE,KAAK,cAAc,CAAC;QACpB,KAAK,aAAa,CAAC;QACnB,KAAK,SAAS,CAAC;QACf;YACI,OAAO,MAAM,CAAC;IACtB,CAAC;AACL,CAAC;AAED;;;;;GAKG;AACH,SAAgB,iBAAiB,CAAC,SAAoE;IAClG,MAAM,KAAK,GAAG,SAAS,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC,QAAQ,CAAC,UAAU,KAAK,MAAM,CAAC,CAAC;IAC3E,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC;QAChB,OAAO,CAAC,CAAC;IACb,CAAC;IACD,OAAO,IAAI,CAAC,GAAG,CAAC,GAAG,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC,CAAC;AACnE,CAAC"}
+{"version":3,"file":"confidence.js","sourceRoot":"","sources":["../../src/classify/confidence.ts"],"names":[],"mappings":";AAAA;;;;;;GAMG;;;AAqCH,4DA6BC;AAWD,sCAkBC;AAQD,8CAUC;AA/GD,oCAA2E;AAW9D,QAAA,kBAAkB,GAAqB;IAChD,aAAa,EAAE,IAAI;IACnB,iBAAiB,EAAE,IAAI;IACvB,iBAAiB,EAAE,IAAI;CAC1B,CAAC;AAEF,SAAS,mBAAmB,CAAC,KAAc,EAAE,MAAc;IACvD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC,IAAI,KAAK,GAAG,CAAC,EAAE,CAAC;QACjF,MAAM,IAAI,0BAAkB,CAAC,MAAM,EAAE,iCAAiC,CAAC,CAAC;IAC5E,CAAC;IACD,OAAO,KAAK,CAAC;AACjB,CAAC;AAED;;;;;;;;;;GAUG;AACH,SAAgB,wBAAwB,CAAC,KAAiC;IACtE,IAAI,KAAK,KAAK,SAAS,IAAI,KAAK,KAAK,IAAI,EAAE,CAAC;QACxC,OAAO,0BAAkB,CAAC;IAC9B,CAAC;IACD,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QAC5B,MAAM,IAAI,0BAAkB,CAAC,YAAY,EAAE,WAAW,CAAC,CAAC;IAC5D,CAAC;IAED,MAAM,MAAM,GAAqB,EAAE,GAAG,0BAAkB,EAAE,CAAC;IAC3D,IAAI,KAAK,CAAC,aAAa,KAAK,SAAS,EAAE,CAAC;QACpC,MAAM,CAAC,aAAa,GAAG,mBAAmB,CAAC,KAAK,CAAC,aAAa,EAAE,0BAA0B,CAAC,CAAC;IAChG,CAAC;IACD,IAAI,KAAK,CAAC,iBAAiB,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,iBAAiB,GAAG,mBAAmB,CAAC,KAAK,CAAC,iBAAiB,EAAE,8BAA8B,CAAC,CAAC;IAC5G,CAAC;IACD,IAAI,KAAK,CAAC,iBAAiB,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,iBAAiB,GAAG,mBAAmB,CAAC,KAAK,CAAC,iBAAiB,EAAE,8BAA8B,CAAC,CAAC;IAC5G,CAAC;IAED,2FAA2F;IAC3F,mDAAmD;IACnD,IAAI,MAAM,CAAC,iBAAiB,GAAG,MAAM,CAAC,aAAa,EAAE,CAAC;QAClD,MAAM,IAAI,0BAAkB,CAAC,8BAA8B,EAAE,gDAAgD,EAAE,qBAAqB,CAAC,CAAC;IAC1I,CAAC;IAED,6FAA6F;IAC7F,yFAAyF;IACzF,2FAA2F;IAC3F,OAAO,MAAM,CAAC;AAClB,CAAC;AAED;;;;;;;;GAQG;AACH,SAAgB,aAAa,CAAC,KAAY,EAAE,UAAkB,EAAE,SAA2B,0BAAkB;IACzG,QAAQ,KAAK,EAAE,CAAC;QACZ,KAAK,eAAe;YAChB,IAAI,UAAU,IAAI,MAAM,CAAC,aAAa,EAAE,CAAC;gBACrC,OAAO,MAAM,CAAC;YAClB,CAAC;YACD,OAAO,UAAU,IAAI,MAAM,CAAC,iBAAiB,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,MAAM,CAAC;QACxE,KAAK,mBAAmB;YACpB,OAAO,UAAU,IAAI,MAAM,CAAC,iBAAiB,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,MAAM,CAAC;QACxE,KAAK,WAAW,CAAC;QACjB,KAAK,YAAY;YACb,OAAO,UAAU,IAAI,MAAM,CAAC,iBAAiB,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,MAAM,CAAC;QACxE,KAAK,cAAc,CAAC;QACpB,KAAK,aAAa,CAAC;QACnB,KAAK,SAAS,CAAC;QACf;YACI,OAAO,MAAM,CAAC;IACtB,CAAC;AACL,CAAC;AAED;;;;;GAKG;AACH,SAAgB,iBAAiB,CAAC,SAAoE;IAClG,6FAA6F;IAC7F,mEAAmE;IACnE,IAAI,MAAM,GAAG,QAAQ,CAAC;IACtB,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;QAC/B,IAAI,QAAQ,CAAC,UAAU,KAAK,MAAM,IAAI,QAAQ,CAAC,UAAU,GAAG,MAAM,EAAE,CAAC;YACjE,MAAM,GAAG,QAAQ,CAAC,UAAU,CAAC;QACjC,CAAC;IACL,CAAC;IACD,OAAO,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC;AAC5C,CAAC"}

package/dist/classify/reconcile.d.ts

@@ -10,43 +10,51 @@
  * The plain part is still classified independently, and its verdict is used two ways: to confirm,
  * which raises nothing but costs nothing, and to contradict, which caps confidence so that a
  * disputed region is collapsed rather than hidden.
+ *
+ * Selection runs *before* classification, and that ordering is load bearing. A message can hold
+ * several independent body sections, and the classifier has to see all of them as one linear
+ * sequence for its positional rules to mean anything. Choosing the representation first is what
+ * makes that possible, and it is possible because selection needs only length, anchor presence and
+ * whether segmentation degraded. None of those are labels. See
+ * `docs/adr-001-document-global-classification.md`.
  */
 import { type Representation } from '../types';
+import { type Block } from '../segment/types';
 import { type BlockDecision } from './rules';
-export interface RepresentationResult {
+/** One representation of one body section, before any classification has run. */
+export interface RepresentationInfo {
     representation: Representation;
-    decisions: BlockDecision[];
+    blocks: Block[];
     /** Total characters of authored content, used to spot a stub representation. */
     contentChars: number;
     /** True when the representation contains at least one structural anchor. */
     hasAnchor: boolean;
-    /** Blocks matched against known history, if any was supplied. */
-    matches?: readonly {
-        blockIndex: number;
-        containment: number;
-        exact: boolean;
-        isParent: boolean;
-    }[];
+    /** True when segmentation failed and the representation is one unclassified block. */
+    degraded: boolean;
 }
-export interface ReconcileOutcome {
-    primary: RepresentationResult;
-    secondary: RepresentationResult | null;
-    disagrees: boolean;
+export interface PrimaryChoice {
+    primary: RepresentationInfo;
+    /** The other representation of the same section, when it is worth cross checking against. */
+    secondary: RepresentationInfo | null;
 }
 /**
- * Choose which representation drives the output.
+ * Choose which representation of one section drives the output.
  *
  * HTML wins by default when both exist, because it is where the structure survives. Plain text
  * wins when the HTML is a stub, which happens with clients that send a token HTML wrapper around a
  * text body, and when only the plain part has an anchor.
  */
-export declare function pickPrimary(text: RepresentationResult | null, html: RepresentationResult | null): ReconcileOutcome;
+export declare function pickPrimary(text: RepresentationInfo | null, html: RepresentationInfo | null): PrimaryChoice;
 /**
- * Apply cross representation evidence to the primary decisions.
+ * Apply cross representation evidence to one section's decisions, in place.
  *
  * The only adjustment made is downward. Raising confidence because the other representation agrees
  * would let two copies of the same mistake reinforce each other, and the cost of a wrong hide is
  * much higher than the cost of a collapse that could have been a hide.
+ *
+ * Scoped to one section on purpose. The comparison only means something between two renderings of
+ * the same content; measuring one section against another section's alternative would compare
+ * unrelated things.
  */
-export declare function reconcile(outcome: ReconcileOutcome, collapseCeiling?: number): BlockDecision[];
+export declare function reconcileSection(primaryRepresentation: Representation, primaryDecisions: readonly BlockDecision[], secondaryDecisions: readonly BlockDecision[] | null, collapseCeiling?: number): void;
 //# sourceMappingURL=reconcile.d.ts.map

package/dist/classify/reconcile.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"reconcile.d.ts","sourceRoot":"","sources":["../../src/classify/reconcile.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAe,KAAK,cAAc,EAAE,MAAM,UAAU,CAAC;AAC5D,OAAO,EAAE,KAAK,aAAa,EAAE,MAAM,SAAS,CAAC;AAE7C,MAAM,WAAW,oBAAoB;IACjC,cAAc,EAAE,cAAc,CAAC;IAC/B,SAAS,EAAE,aAAa,EAAE,CAAC;IAC3B,gFAAgF;IAChF,YAAY,EAAE,MAAM,CAAC;IACrB,4EAA4E;IAC5E,SAAS,EAAE,OAAO,CAAC;IACnB,iEAAiE;IACjE,OAAO,CAAC,EAAE,SAAS;QAAE,UAAU,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,OAAO,CAAC;QAAC,QAAQ,EAAE,OAAO,CAAA;KAAE,EAAE,CAAC;CACvG;AAED,MAAM,WAAW,gBAAgB;IAC7B,OAAO,EAAE,oBAAoB,CAAC;IAC9B,SAAS,EAAE,oBAAoB,GAAG,IAAI,CAAC;IACvC,SAAS,EAAE,OAAO,CAAC;CACtB;AAgBD;;;;;;GAMG;AACH,wBAAgB,WAAW,CAAC,IAAI,EAAE,oBAAoB,GAAG,IAAI,EAAE,IAAI,EAAE,oBAAoB,GAAG,IAAI,GAAG,gBAAgB,CA8BlH;AAED;;;;;;GAMG;AACH,wBAAgB,SAAS,CAAC,OAAO,EAAE,gBAAgB,EAAE,eAAe,SAAM,GAAG,aAAa,EAAE,CA8B3F"}
+{"version":3,"file":"reconcile.d.ts","sourceRoot":"","sources":["../../src/classify/reconcile.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAEH,OAAO,EAAe,KAAK,cAAc,EAAE,MAAM,UAAU,CAAC;AAC5D,OAAO,EAAE,KAAK,KAAK,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,EAAE,KAAK,aAAa,EAAE,MAAM,SAAS,CAAC;AAE7C,iFAAiF;AACjF,MAAM,WAAW,kBAAkB;IAC/B,cAAc,EAAE,cAAc,CAAC;IAC/B,MAAM,EAAE,KAAK,EAAE,CAAC;IAChB,gFAAgF;IAChF,YAAY,EAAE,MAAM,CAAC;IACrB,4EAA4E;IAC5E,SAAS,EAAE,OAAO,CAAC;IACnB,sFAAsF;IACtF,QAAQ,EAAE,OAAO,CAAC;CACrB;AAED,MAAM,WAAW,aAAa;IAC1B,OAAO,EAAE,kBAAkB,CAAC;IAC5B,6FAA6F;IAC7F,SAAS,EAAE,kBAAkB,GAAG,IAAI,CAAC;CACxC;AAgBD;;;;;;GAMG;AACH,wBAAgB,WAAW,CAAC,IAAI,EAAE,kBAAkB,GAAG,IAAI,EAAE,IAAI,EAAE,kBAAkB,GAAG,IAAI,GAAG,aAAa,CAiD3G;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,gBAAgB,CAC5B,qBAAqB,EAAE,cAAc,EACrC,gBAAgB,EAAE,SAAS,aAAa,EAAE,EAC1C,kBAAkB,EAAE,SAAS,aAAa,EAAE,GAAG,IAAI,EACnD,eAAe,SAAM,GACtB,IAAI,CAiCN"}

package/dist/classify/reconcile.js

@@ -11,15 +11,22 @@
  * The plain part is still classified independently, and its verdict is used two ways: to confirm,
  * which raises nothing but costs nothing, and to contradict, which caps confidence so that a
  * disputed region is collapsed rather than hidden.
+ *
+ * Selection runs *before* classification, and that ordering is load bearing. A message can hold
+ * several independent body sections, and the classifier has to see all of them as one linear
+ * sequence for its positional rules to mean anything. Choosing the representation first is what
+ * makes that possible, and it is possible because selection needs only length, anchor presence and
+ * whether segmentation degraded. None of those are labels. See
+ * `docs/adr-001-document-global-classification.md`.
  */
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.pickPrimary = pickPrimary;
-exports.reconcile = reconcile;
-/** Fraction of characters classified as hidden or collapsed history. */
-function historyShare(result) {
+exports.reconcileSection = reconcileSection;
+/** Fraction of characters classified as history. */
+function historyShare(decisions) {
     let history = 0;
     let total = 0;
-    for (const decision of result.decisions) {
+    for (const decision of decisions) {
         const chars = decision.block.content.length;
         total += chars;
         if (decision.label === 'REPLY_HISTORY') {
@@ -29,7 +36,7 @@ function historyShare(result) {
     return total > 0 ? history / total : 0;
 }
 /**
- * Choose which representation drives the output.
+ * Choose which representation of one section drives the output.
  *
  * HTML wins by default when both exist, because it is where the structure survives. Plain text
  * wins when the HTML is a stub, which happens with clients that send a token HTML wrapper around a
@@ -37,10 +44,10 @@ function historyShare(result) {
  */
 function pickPrimary(text, html) {
     if (text && !html) {
-        return { primary: text, secondary: null, disagrees: false };
+        return { primary: text, secondary: null };
     }
     if (html && !text) {
-        return { primary: html, secondary: null, disagrees: false };
+        return { primary: html, secondary: null };
     }
     if (!text || !html) {
         throw new Error('reconcile called with no representations');
@@ -58,30 +65,56 @@ function pickPrimary(text, html) {
         primary = text;
         secondary = html;
     }
-    // Disagreement is measured on the share of content treated as history, not on block counts,
-    // because the two representations rarely segment into the same number of blocks.
-    const disagrees = Math.abs(historyShare(primary) - historyShare(secondary)) > 0.25;
-    return { primary, secondary, disagrees };
+    // Degrading is not the same as winning.
+    //
+    // A representation that could not be segmented is one undifferentiated block. It knows nothing
+    // about quote structure, so letting it drive the output turns a message whose plain text part
+    // classified perfectly well into a single UNKNOWN blob. That did not happen before only by
+    // accident: the degraded HTML block carried no content at all, so the stub test above caught
+    // it. Giving that block its real projected text removes the accident, which is why the rule has
+    // to be written down. Stated symmetrically, so it stays correct if the text segmenter ever
+    // grows a degraded path of its own.
+    const isStub = (result) => (result === text ? textIsStub : htmlIsStub);
+    if (primary.degraded && !secondary.degraded && !isStub(secondary)) {
+        const displaced = primary;
+        primary = secondary;
+        secondary = displaced;
+    }
+    // A degraded secondary has no opinion to contradict with. Its single block is UNKNOWN by
+    // construction, so measuring it against the primary would manufacture a disagreement out of the
+    // fact that it never got to disagree.
+    if (secondary.degraded) {
+        return { primary, secondary: null };
+    }
+    return { primary, secondary };
 }
 /**
- * Apply cross representation evidence to the primary decisions.
+ * Apply cross representation evidence to one section's decisions, in place.
  *
  * The only adjustment made is downward. Raising confidence because the other representation agrees
  * would let two copies of the same mistake reinforce each other, and the cost of a wrong hide is
  * much higher than the cost of a collapse that could have been a hide.
+ *
+ * Scoped to one section on purpose. The comparison only means something between two renderings of
+ * the same content; measuring one section against another section's alternative would compare
+ * unrelated things.
  */
-function reconcile(outcome, collapseCeiling = 0.9) {
-    const { primary, secondary, disagrees } = outcome;
-    const reason = primary.representation === 'html' ? 'html_primary_representation' : 'text_primary_representation';
-    for (const decision of primary.decisions) {
+function reconcileSection(primaryRepresentation, primaryDecisions, secondaryDecisions, collapseCeiling = 0.9) {
+    const reason = primaryRepresentation === 'html' ? 'html_primary_representation' : 'text_primary_representation';
+    for (const decision of primaryDecisions) {
         if (!decision.reasons.includes(reason)) {
             decision.reasons.push(reason);
         }
     }
-    if (!secondary || !disagrees) {
-        return primary.decisions;
+    if (!secondaryDecisions) {
+        return;
+    }
+    // Disagreement is measured on the share of content treated as history, not on block counts,
+    // because the two representations rarely segment into the same number of blocks.
+    if (Math.abs(historyShare(primaryDecisions) - historyShare(secondaryDecisions)) <= 0.25) {
+        return;
     }
-    for (const decision of primary.decisions) {
+    for (const decision of primaryDecisions) {
         if (decision.label !== 'REPLY_HISTORY') {
             continue;
         }
@@ -95,6 +128,5 @@ function reconcile(outcome, collapseCeiling = 0.9) {
             decision.reasons.push('representations_disagree');
         }
     }
-    return primary.decisions;
 }
 //# sourceMappingURL=reconcile.js.map

package/dist/classify/reconcile.js.map

@@ -1 +1 @@
-{"version":3,"file":"reconcile.js","sourceRoot":"","sources":["../../src/classify/reconcile.ts"],"names":[],"mappings":";AAAA;;;;;;;;;;;;GAYG;;AA2CH,kCA8BC;AASD,8BA8BC;AA1FD,wEAAwE;AACxE,SAAS,YAAY,CAAC,MAA4B;IAC9C,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,QAAQ,IAAI,MAAM,CAAC,SAAS,EAAE,CAAC;QACtC,MAAM,KAAK,GAAG,QAAQ,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC;QAC5C,KAAK,IAAI,KAAK,CAAC;QACf,IAAI,QAAQ,CAAC,KAAK,KAAK,eAAe,EAAE,CAAC;YACrC,OAAO,IAAI,KAAK,CAAC;QACrB,CAAC;IACL,CAAC;IACD,OAAO,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;AAC3C,CAAC;AAED;;;;;;GAMG;AACH,SAAgB,WAAW,CAAC,IAAiC,EAAE,IAAiC;IAC5F,IAAI,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC;QAChB,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;IAChE,CAAC;IACD,IAAI,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC;QAChB,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;IAChE,CAAC;IACD,IAAI,CAAC,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC;QACjB,MAAM,IAAI,KAAK,CAAC,0CAA0C,CAAC,CAAC;IAChE,CAAC;IAED,6EAA6E;IAC7E,MAAM,UAAU,GAAG,IAAI,CAAC,YAAY,GAAG,CAAC,GAAG,IAAI,CAAC,YAAY,CAAC;IAC7D,MAAM,UAAU,GAAG,IAAI,CAAC,YAAY,GAAG,CAAC,GAAG,IAAI,CAAC,YAAY,CAAC;IAE7D,IAAI,OAAO,GAAG,IAAI,CAAC;IACnB,IAAI,SAAS,GAAG,IAAI,CAAC;IACrB,IAAI,UAAU,IAAI,CAAC,UAAU,EAAE,CAAC;QAC5B,OAAO,GAAG,IAAI,CAAC;QACf,SAAS,GAAG,IAAI,CAAC;IACrB,CAAC;SAAM,IAAI,CAAC,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC,SAAS,IAAI,CAAC,UAAU,EAAE,CAAC;QAC1D,OAAO,GAAG,IAAI,CAAC;QACf,SAAS,GAAG,IAAI,CAAC;IACrB,CAAC;IAED,4FAA4F;IAC5F,iFAAiF;IACjF,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,YAAY,CAAC,OAAO,CAAC,GAAG,YAAY,CAAC,SAAS,CAAC,CAAC,GAAG,IAAI,CAAC;IAEnF,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,SAAS,EAAE,CAAC;AAC7C,CAAC;AAED;;;;;;GAMG;AACH,SAAgB,SAAS,CAAC,OAAyB,EAAE,eAAe,GAAG,GAAG;IACtE,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,SAAS,EAAE,GAAG,OAAO,CAAC;IAClD,MAAM,MAAM,GAAW,OAAO,CAAC,cAAc,KAAK,MAAM,CAAC,CAAC,CAAC,6BAA6B,CAAC,CAAC,CAAC,6BAA6B,CAAC;IAEzH,KAAK,MAAM,QAAQ,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;QACvC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC;YACrC,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAClC,CAAC;IACL,CAAC;IAED,IAAI,CAAC,SAAS,IAAI,CAAC,SAAS,EAAE,CAAC;QAC3B,OAAO,OAAO,CAAC,SAAS,CAAC;IAC7B,CAAC;IAED,KAAK,MAAM,QAAQ,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;QACvC,IAAI,QAAQ,CAAC,KAAK,KAAK,eAAe,EAAE,CAAC;YACrC,SAAS;QACb,CAAC;QACD,IAAI,QAAQ,CAAC,OAAO,CAAC,QAAQ,CAAC,wBAAwB,CAAC,EAAE,CAAC;YACtD,iFAAiF;YACjF,iEAAiE;YACjE,SAAS;QACb,CAAC;QACD,QAAQ,CAAC,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,UAAU,EAAE,eAAe,CAAC,CAAC;QACrE,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,QAAQ,CAAC,0BAA0B,CAAC,EAAE,CAAC;YACzD,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;QACtD,CAAC;IACL,CAAC;IAED,OAAO,OAAO,CAAC,SAAS,CAAC;AAC7B,CAAC"}
+{"version":3,"file":"reconcile.js","sourceRoot":"","sources":["../../src/classify/reconcile.ts"],"names":[],"mappings":";AAAA;;;;;;;;;;;;;;;;;;;GAmBG;;AA6CH,kCAiDC;AAaD,4CAsCC;AAzHD,oDAAoD;AACpD,SAAS,YAAY,CAAC,SAAmC;IACrD,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;QAC/B,MAAM,KAAK,GAAG,QAAQ,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC;QAC5C,KAAK,IAAI,KAAK,CAAC;QACf,IAAI,QAAQ,CAAC,KAAK,KAAK,eAAe,EAAE,CAAC;YACrC,OAAO,IAAI,KAAK,CAAC;QACrB,CAAC;IACL,CAAC;IACD,OAAO,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;AAC3C,CAAC;AAED;;;;;;GAMG;AACH,SAAgB,WAAW,CAAC,IAA+B,EAAE,IAA+B;IACxF,IAAI,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC;QAChB,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;IAC9C,CAAC;IACD,IAAI,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC;QAChB,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;IAC9C,CAAC;IACD,IAAI,CAAC,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC;QACjB,MAAM,IAAI,KAAK,CAAC,0CAA0C,CAAC,CAAC;IAChE,CAAC;IAED,6EAA6E;IAC7E,MAAM,UAAU,GAAG,IAAI,CAAC,YAAY,GAAG,CAAC,GAAG,IAAI,CAAC,YAAY,CAAC;IAC7D,MAAM,UAAU,GAAG,IAAI,CAAC,YAAY,GAAG,CAAC,GAAG,IAAI,CAAC,YAAY,CAAC;IAE7D,IAAI,OAAO,GAAG,IAAI,CAAC;IACnB,IAAI,SAAS,GAAG,IAAI,CAAC;IACrB,IAAI,UAAU,IAAI,CAAC,UAAU,EAAE,CAAC;QAC5B,OAAO,GAAG,IAAI,CAAC;QACf,SAAS,GAAG,IAAI,CAAC;IACrB,CAAC;SAAM,IAAI,CAAC,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC,SAAS,IAAI,CAAC,UAAU,EAAE,CAAC;QAC1D,OAAO,GAAG,IAAI,CAAC;QACf,SAAS,GAAG,IAAI,CAAC;IACrB,CAAC;IAED,wCAAwC;IACxC,EAAE;IACF,+FAA+F;IAC/F,8FAA8F;IAC9F,2FAA2F;IAC3F,6FAA6F;IAC7F,gGAAgG;IAChG,2FAA2F;IAC3F,oCAAoC;IACpC,MAAM,MAAM,GAAG,CAAC,MAA0B,EAAE,EAAE,CAAC,CAAC,MAAM,KAAK,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC;IAC3F,IAAI,OAAO,CAAC,QAAQ,IAAI,CAAC,SAAS,CAAC,QAAQ,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,EAAE,CAAC;QAChE,MAAM,SAAS,GAAG,OAAO,CAAC;QAC1B,OAAO,GAAG,SAAS,CAAC;QACpB,SAAS,GAAG,SAAS,CAAC;IAC1B,CAAC;IAED,yFAAyF;IACzF,gGAAgG;IAChG,sCAAsC;IACtC,IAAI,SAAS,CAAC,QAAQ,EAAE,CAAC;QACrB,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;IACxC,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC;AAClC,CAAC;AAED;;;;;;;;;;GAUG;AACH,SAAgB,gBAAgB,CAC5B,qBAAqC,EACrC,gBAA0C,EAC1C,kBAAmD,EACnD,eAAe,GAAG,GAAG;IAErB,MAAM,MAAM,GAAW,qBAAqB,KAAK,MAAM,CAAC,CAAC,CAAC,6BAA6B,CAAC,CAAC,CAAC,6BAA6B,CAAC;IAExH,KAAK,MAAM,QAAQ,IAAI,gBAAgB,EAAE,CAAC;QACtC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC;YACrC,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAClC,CAAC;IACL,CAAC;IAED,IAAI,CAAC,kBAAkB,EAAE,CAAC;QACtB,OAAO;IACX,CAAC;IAED,4FAA4F;IAC5F,iFAAiF;IACjF,IAAI,IAAI,CAAC,GAAG,CAAC,YAAY,CAAC,gBAAgB,CAAC,GAAG,YAAY,CAAC,kBAAkB,CAAC,CAAC,IAAI,IAAI,EAAE,CAAC;QACtF,OAAO;IACX,CAAC;IAED,KAAK,MAAM,QAAQ,IAAI,gBAAgB,EAAE,CAAC;QACtC,IAAI,QAAQ,CAAC,KAAK,KAAK,eAAe,EAAE,CAAC;YACrC,SAAS;QACb,CAAC;QACD,IAAI,QAAQ,CAAC,OAAO,CAAC,QAAQ,CAAC,wBAAwB,CAAC,EAAE,CAAC;YACtD,iFAAiF;YACjF,iEAAiE;YACjE,SAAS;QACb,CAAC;QACD,QAAQ,CAAC,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,UAAU,EAAE,eAAe,CAAC,CAAC;QACrE,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,QAAQ,CAAC,0BAA0B,CAAC,EAAE,CAAC;YACzD,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;QACtD,CAAC;IACL,CAAC;AACL,CAAC"}

package/dist/cli/annotate.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"annotate.d.ts","sourceRoot":"","sources":["../../src/cli/annotate.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AASH,OAAO,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAC;AAM9C,OAAO,EAAoC,KAAK,UAAU,EAAE,MAAM,QAAQ,CAAC;AA0D3E,wBAAsB,eAAe,CAAC,IAAI,EAAE,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,CAwLvE;AAED;;;;;;GAMG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM,CA2DzE"}
+{"version":3,"file":"annotate.d.ts","sourceRoot":"","sources":["../../src/cli/annotate.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AASH,OAAO,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAC;AAM9C,OAAO,EAAyC,KAAK,UAAU,EAAE,MAAM,QAAQ,CAAC;AA0DhF,wBAAsB,eAAe,CAAC,IAAI,EAAE,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,CA2LvE;AAED;;;;;;GAMG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM,CA2DzE"}