feat(recibos): OCR capture for gas butano and municipal predial

Adds four parsers to the statement intake — GAS TIJUANA plus one per municipality, because Tijuana, Rosarito and Ensenada issue three completely different predial documents — and a text-layer fast path for the born-digital invoices the gas company sends. Measured against a new corpus of 14 documents / 29 pages: provider read on 29/29, amount on 26/29, and 21/29 auto-matched against the dev database (22/29 identified). The eight review cases are all legitimate. Five things the corpus forced: - Not every statement is a scan. The gas invoices are born-digital CFDIs whose text layer is exact; rasterising them only loses information (one sample turned `MEDIDOR: VM01014426` into `ar (LTR): 014420`). The new `OcrProvider.textPages` reads the embedded layer via `pdftotext -bbox-layout` — same poppler package as `pdftoppm`, so no new dependency — and OCR stays the fallback for real scans. Poppler's own `<line>` grouping follows text flow rather than the page, so words are regrouped by vertical position; without that, a two-column header leaves every label separated from the value printed beside it. - The clave catastral is not two letters and six digits. Position three is a letter in 15 of the 932 stored claves, and digitising the whole tail mapped a real `MMB01041` to a nonexistent `MM801041`. - Tijuana predial prints no clave at all. Its only identifier is an 8-digit municipal account carried in a 32-digit payment barcode, which the legacy database never held, so it goes in `meterNumber` alongside gas — `accountNumber` holds `DATMEX.predial`, which is not a per-property key and must not be overwritten. Those pages start cold and are taught by the first confirm. - On Rosarito and Ensenada the clave is the primary key, not a fallback: those receipts print nothing else, so a unique hit auto-matches. On a utility bill that merely happens to print one it stays a review hint. - A misread `$` is the dangerous failure. An Ensenada receipt for $2,203.00 OCR'd as `82,203.00`, which would post a charge 37x too large and look ordinary in the ledger. Predial amounts now require a literal `$` and a page that cannot produce one goes to review. The scoped match field is now one exported function rather than three copies of `kind === "GAS" ? ... : ...`, since the lookup, the blank-service fill and the confirm write-back have to agree or a reference gets learned into a column nothing searches. First tests in this package: 23 specs over the parsers and the text-layer reader, every fixture a verbatim OCR excerpt from a real receipt. Adds the jest config they need and a build tsconfig so they stay out of dist. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
feat(recibos): live OCR progress bar on review page
2026-08-01 12:52:20 -07:00 · 2026-08-01 02:43:19 -07:00 · 2026-08-01 02:28:09 -07:00 · 2026-08-01 09:22:56 +00:00 · 2026-08-01 02:21:38 -07:00 · 2026-08-01 09:09:15 +00:00
27 changed files with 1440 additions and 144 deletions
@@ -37,6 +37,16 @@ env:
 jobs:
  build:
    name: Build ${{ matrix.image }}
+    # release.yml pushes the release commit and its tag in a single `git push`,
+    # so Gitea creates two runs for the same commit: one for master, one for the
+    # tag. Only the tag run matters — it is the one that emits the X.Y.Z / X.Y
+    # image tags, and it publishes `latest` and `sha-<short>` too, since it is
+    # the same commit. Skip the branch run rather than racing or cancelling it.
+    # Ordinary pushes to master (any message but `chore(release):`) still build.
+    if: >-
+      github.event_name != 'push' ||
+      startsWith(github.ref, 'refs/tags/') ||
+      !startsWith(github.event.head_commit.message, 'chore(release):')
    runs-on: docker
    container:
      image: docker:27-dind
@@ -152,7 +152,111 @@ jobs:

          # One push for both refs: a commit that lands without its tag builds
          # nothing and looks like a successful release.
-          git push origin "HEAD:master" "refs/tags/v${VERSION}"
+          #
+          # The output is captured because a failing *post-receive* hook does not
+          # fail the push: git prints `remote: error: ...`, updates both refs and
+          # exits 0. That is how v1.0.3 was cut — the hook 500'd, so Gitea never
+          # created the build run, and this step went green anyway.
+          if ! git push origin "HEAD:master" "refs/tags/v${VERSION}" 2>push.log; then
+            cat push.log
+            echo "::error::Push failed. Nothing was released."
+            exit 1
+          fi
+          cat push.log
+
+          if grep -q '^remote: error' push.log; then
+            echo "::warning::The remote's post-receive hook errored. Both refs landed,"
+            echo "::warning::but Gitea most likely created no workflow run for them."
+            echo "::warning::The next step checks and dispatches build.yml if needed."
+          fi
+
+          echo "sha=$(git rev-parse HEAD)" >> "$GITHUB_OUTPUT"
+        id: push
+
+      # Gitea creates workflow runs from the post-receive hook, so a hook error
+      # silently costs you the build: the tag exists, no image is ever published,
+      # and the failure only surfaces later as a 404 when deploy pulls the image.
+      # Confirm the run exists; dispatch it if it does not; fail loudly if that
+      # does not work either.
+      - name: Verify build.yml started
+        env:
+          RELEASE_TOKEN: ${{ secrets.RELEASE_TOKEN }}
+          VERSION: ${{ steps.ver.outputs.version }}
+          SHA: ${{ steps.push.outputs.sha }}
+        run: |
+          node -e '
+            const base = `${process.env.GITHUB_SERVER_URL}/api/v1/repos/${process.env.GITHUB_REPOSITORY}`;
+            const headers = { Authorization: `token ${process.env.RELEASE_TOKEN}` };
+            const sha = process.env.SHA;
+            const tag = `v${process.env.VERSION}`;
+            const sleep = (ms) => new Promise((r) => setTimeout(r, ms));
+
+            // The master push and the tag push carry the SAME commit, so a sha
+            // match alone is not enough: build.yml skips the master run by
+            // design, and that skipped run would satisfy a sha-only check even
+            // if the tag run were never created. When the API reports a ref for
+            // the run, require it to be the tag; when it reports none, fall back
+            // to the sha match rather than failing a release over a field name.
+            const isTagRun = (r) => {
+              const ref = r.head_branch || r.ref || "";
+              return !ref || ref === tag || ref === `refs/tags/${tag}`;
+            };
+
+            const started = async () => {
+              const res = await fetch(`${base}/actions/runs?limit=30`, { headers });
+              if (!res.ok) throw new Error(`runs query failed: HTTP ${res.status}`);
+              const body = await res.json();
+              return (body.workflow_runs || []).some(
+                (r) =>
+                  r.head_sha === sha &&
+                  String(r.path || "").includes("build.yml") &&
+                  isTagRun(r),
+              );
+            };
+
+            // The hook fires synchronously with the push, so a run that is coming
+            // is usually already there; the retries cover a busy instance.
+            const poll = async (attempts) => {
+              for (let i = 0; i < attempts; i++) {
+                if (await started()) return true;
+                await sleep(10_000);
+              }
+              return started();
+            };
+
+            (async () => {
+              if (await poll(3)) {
+                console.log(`build.yml is running for ${sha}.`);
+                return;
+              }
+
+              console.log(`No build.yml run for ${sha}. Dispatching against ${tag}.`);
+              const res = await fetch(
+                `${base}/actions/workflows/build.yml/dispatches`,
+                {
+                  method: "POST",
+                  headers: { ...headers, "Content-Type": "application/json" },
+                  // Must be the tag, not master: metadata-action only emits the
+                  // X.Y.Z and X.Y image tags when the ref is a semver tag. And
+                  // it must be the fully qualified ref — Gitea 404s on `v1.0.3`.
+                  body: JSON.stringify({ ref: `refs/tags/${tag}` }),
+                },
+              );
+              if (!res.ok) console.log(`Dispatch returned HTTP ${res.status}.`);
+
+              if (await poll(3)) {
+                console.log(`build.yml is running for ${sha}.`);
+                return;
+              }
+
+              console.log(`::error::${tag} is pushed but nothing is building it, and`);
+              console.log(`::error::the dispatch did not take. Run "Build and Push Images"`);
+              console.log(`::error::by hand with ref=${tag} (the tag, not master), then`);
+              console.log(`::error::deploy. Check the Gitea server log for the`);
+              console.log(`::error::post-receive error while you are at it.`);
+              process.exit(1);
+            })();
+          '

      - name: Summary
        env:
@@ -0,0 +1,7 @@
+/** @type {import('jest').Config} */
+module.exports = {
+  rootDir: "src",
+  testEnvironment: "node",
+  testRegex: ".*\\.spec\\.ts$",
+  transform: { "^.+\\.ts$": "ts-jest" },
+};
@@ -3,6 +3,7 @@
  "collection": "@nestjs/schematics",
  "sourceRoot": "src",
  "compilerOptions": {
-    "deleteOutDir": true
+    "deleteOutDir": true,
+    "tsConfigPath": "tsconfig.build.json"
  }
 }
@@ -1,6 +1,6 @@
 {
  "name": "@jorgecuadros/api",
-  "version": "1.0.2",
+  "version": "1.0.5",
  "private": true,
  "scripts": {
    "build": "nest build",
@@ -262,9 +262,26 @@ export class OpsService implements OnModuleInit {
   * deploy/scripts/pre-migrate-backup.mjs — the two write into the same volume
   * and both are listed as restore points by this same screen.
   *
-   * --set-gtid-purged=OFF: the production server is the replication SOURCE with
-   * GTID on, so without it every dump embeds SET @@GLOBAL.GTID_PURGED and is
-   * unrestorable onto the very server it came from.
+   * The dumper is probed at runtime rather than assumed. This command runs
+   * inside the API image, whose `mysql-client` is Alpine's — i.e. MariaDB's —
+   * where `mysqldump` is a deprecation-warning shim over `mariadb-dump` that
+   * rejects --set-gtid-purged outright:
+   *   mysqldump: unknown variable 'set-gtid-purged=OFF'
+   * which failed every backup, including the safety backups SYNC and REIMPORT
+   * take first. MariaDB's dumper emits no GTID state unless asked (--gtid), so
+   * there is nothing to suppress there; the flag is passed only when the dumper
+   * on PATH advertises it, and the real binary is called directly only in the
+   * MariaDB case (calling `mariadb-dump` whenever it merely exists would pick
+   * it over a MySQL `mysqldump` earlier in PATH on a host carrying both).
+   *
+   * The probe is a command substitution, not `--help | grep -q`: PIPEFAIL is in
+   * effect and grep closing the pipe early would make a supported flag look
+   * unsupported.
+   *
+   * --set-gtid-purged=OFF (MySQL only): the production server is the
+   * replication SOURCE with GTID on, so without it every dump embeds
+   * SET @@GLOBAL.GTID_PURGED and is unrestorable onto the very server it came
+   * from.
   *
   * The table-count assertion is not belt-and-braces: `gzip -t` passes on the
   * ~372-byte output of a mysqldump that died on its first statement, so a
@@ -277,8 +294,12 @@ export class OpsService implements OnModuleInit {
   */
  private dumpCommand(flags: string, db: string, out: string): string {
    return (
-      `( mysqldump ${flags} --single-transaction --routines --triggers ` +
-      `--no-tablespaces --set-gtid-purged=OFF ${db} | gzip -c > ${out} && ` +
+      `DUMP=mysqldump; GTID=; ` +
+      `case "$(mysqldump --help 2>/dev/null || true)" in ` +
+      `*set-gtid-purged*) GTID=--set-gtid-purged=OFF;; ` +
+      `*) command -v mariadb-dump >/dev/null 2>&1 && DUMP=mariadb-dump;; esac; ` +
+      `( $DUMP ${flags} --single-transaction --routines --triggers ` +
+      `--no-tablespaces $GTID ${db} | gzip -c > ${out} && ` +
      `gzip -t ${out} && ` +
      `TABLAS=$(gunzip -c ${out} | grep -c 'CREATE TABLE') && ` +
      `echo "tablas capturadas: $TABLAS" && ` +
@@ -6,8 +6,10 @@
 * The shipped implementation is self-hosted Tesseract (see tesseract.provider).
 * That choice is evidence-based rather than assumed: run against 46 pages of
 * real scanned CFE, CESPT and Telnor statements, it identified the provider on
- * 46/46 and extracted a usable account reference on 43/46, which is well past
- * the bar for a queue whose whole point is that a human confirms every row. A
+ * 46/46 and extracted a usable account reference on 43/46, and on a later
+ * corpus of 19 scanned municipal predial receipts it read the provider on
+ * 19/19 and an identifier on 18/19 — well past the bar for a queue whose whole
+ * point is that a human confirms every row. A
 * managed document-extraction API (Textract, Document Intelligence, Document
 * AI) fits behind this same interface if per-page accuracy ever proves
 * insufficient, with no schema change — but at 300+ pages/month/company it
@@ -31,10 +33,10 @@ export interface OcrPage {
  /** Full page text, reading order, newline-separated. */
  text: string;
  /**
-   * Word boxes. Needed because two of the three real layouts are *tables* —
-   * the CESPT "RECIBO" prints `No. DE CUENTA` as a column header with the
-   * value in the row beneath it, which line-oriented text cannot associate.
-   * Parsers fall back to geometry for exactly those fields.
+   * Word boxes. Needed because several of the real layouts are *tables* — the
+   * CESPT "RECIBO" prints `No. DE CUENTA` as a column header with the value in
+   * the row beneath it, which line-oriented text cannot associate. Parsers fall
+   * back to geometry for exactly those fields.
   */
  words: OcrWord[];
  /** Mean word confidence across the page, 0..1. */
@@ -48,6 +50,22 @@ export interface OcrProvider {
  renderPages(pdf: Buffer): Promise<Buffer[]>;
  /** OCR a single rendered page image. */
  recognize(pageImage: Buffer): Promise<OcrPage>;
+  /**
+   * Read a PDF's own text layer, one entry per page, `null` where the page has
+   * none worth using.
+   *
+   * Not every statement is a scan. The gas company e-mails born-digital CFDI
+   * invoices whose text is already exact and already positioned — running those
+   * through a rasteriser and a character recogniser can only lose information
+   * (one sample turned `MEDIDOR: VM01014426` into `ar (LTR): 014420`) while
+   * costing about a minute of CPU per page for the privilege. Where the layer
+   * exists it is strictly better input for the same parsers, so it is tried
+   * first and OCR remains the fallback for genuine scans.
+   *
+   * Positions are reported in the same pixel space `recognize` uses, so the
+   * geometric helpers in the parsers work unchanged on either source.
+   */
+  textPages(pdf: Buffer): Promise<(OcrPage | null)[]>;
 }

 export const OCR_PROVIDER = Symbol("OCR_PROVIDER");
@@ -0,0 +1,67 @@
+import { parseBboxLayout } from "./tesseract.provider";
+
+/**
+ * Shaped like real `pdftotext -bbox-layout` output: the gas invoice lays its
+ * header out as two columns of independent text flows, so poppler puts a label
+ * and the value printed beside it in *different* `<line>` elements. Trusting
+ * that grouping is what left `PERIODO FACTURADO` with no value next to it and
+ * every period field empty on a batch whose text was perfectly readable.
+ */
+function word(x: number, y: number, text: string): string {
+  return `<word xMin="${x}" yMin="${y}" xMax="${x + 20}" yMax="${y + 8}">${text}</word>`;
+}
+
+function doc(...lines: string[]): string {
+  return `<doc><page width="612" height="792">${lines
+    .map((l) => `<flow><block><line>${l}</line></block></flow>`)
+    .join("")}</page></doc>`;
+}
+
+/** Enough words on the page to clear the "is this a real text layer" floor. */
+function padding(): string {
+  return Array.from({ length: 50 }, (_, i) => word(10, 400 + i * 10, `w${i}`)).join("");
+}
+
+describe("parseBboxLayout", () => {
+  it("rejoins a label with the value printed beside it in another flow", () => {
+    const [page] = parseBboxLayout(
+      doc(
+        word(20, 100, "PERIODO") + word(45, 100, "FACTURADO:"),
+        word(300, 100.4, "20260630-20260630"),
+        padding(),
+      ),
+      1,
+    );
+    expect(page).not.toBeNull();
+    expect(page!.text).toContain("PERIODO FACTURADO: 20260630-20260630");
+  });
+
+  it("keeps genuinely separate lines apart", () => {
+    const [page] = parseBboxLayout(
+      doc(word(20, 100, "Cuenta:") + word(80, 100, "0900003463"), word(20, 130, "Nombre:"), padding()),
+      1,
+    );
+    expect(page!.text.split("\n")).toContain("Cuenta: 0900003463");
+    expect(page!.text.split("\n")).toContain("Nombre:");
+  });
+
+  it("scales point coordinates into the render's pixel space", () => {
+    // Word boxes have to land in the same coordinate space tesseract reports,
+    // or the geometric helpers the parsers share silently stop finding values.
+    const [page] = parseBboxLayout(doc(word(72, 144, "X") + padding()), 300 / 72);
+    const x = page!.words.find((w) => w.text === "X")!;
+    expect(x.left).toBeCloseTo(300);
+    expect(x.top).toBeCloseTo(600);
+  });
+
+  it("reports no text layer for a scan carrying a few stray glyphs", () => {
+    expect(parseBboxLayout(doc(word(10, 10, "3") + word(40, 10, "of") + word(60, 10, "5")), 1)).toEqual([
+      null,
+    ]);
+  });
+
+  it("decodes the entities poppler escapes", () => {
+    const [page] = parseBboxLayout(doc(word(10, 10, "A&amp;B") + padding()), 1);
+    expect(page!.text).toContain("A&B");
+  });
+});
@@ -105,6 +105,37 @@ export class TesseractOcrProvider implements OcrProvider {
    });
  }

+  /**
+   * `pdftotext -bbox-layout` — the same poppler package `pdftoppm` comes from,
+   * so this costs no extra dependency in the runtime image.
+   *
+   * A page is only accepted when it carries a real text layer. Scanned PDFs
+   * frequently contain a handful of stray glyphs (a scanner watermark, a page
+   * number stamped by the MFP), and treating those as the page's text would
+   * hand every parser an almost-empty string and silently take OCR out of the
+   * loop — so a floor of MIN_TEXT_WORDS words has to be present before the
+   * layer is believed.
+   */
+  async textPages(pdf: Buffer): Promise<(OcrPage | null)[]> {
+    await this.require();
+    return this.scratch(async (dir) => {
+      const src = join(dir, "in.pdf");
+      await writeFile(src, pdf);
+      const out = join(dir, "out.html");
+      try {
+        await run("pdftotext", ["-bbox-layout", src, out]);
+      } catch (err) {
+        this.logger.warn(
+          `pdftotext failed; falling back to OCR for this file: ${(err as Error).message}`,
+        );
+        return [];
+      }
+      // Points to pixels at the render DPI, so word boxes from either source
+      // land in one coordinate space and `valueUnder`'s thresholds hold.
+      return parseBboxLayout(await readFile(out, "utf8"), this.dpi / 72);
+    });
+  }
+
  async recognize(pageImage: Buffer): Promise<OcrPage> {
    await this.require();
    return this.scratch(async (dir) => {
@@ -136,6 +167,130 @@ export class TesseractOcrProvider implements OcrProvider {
  }
 }

+/**
+ * Below this many words a "text layer" is scanner debris, not a document.
+ * The real born-digital samples carry 400+ words a page; the scanned ones
+ * carry none at all, so the exact threshold is not delicate.
+ */
+const MIN_TEXT_WORDS = 40;
+
+const ENTITIES: Record<string, string> = {
+  amp: "&",
+  lt: "<",
+  gt: ">",
+  quot: '"',
+  apos: "'",
+};
+
+function decodeEntities(s: string): string {
+  return s.replace(/&(#x?[0-9a-fA-F]+|[a-z]+);/g, (whole, body: string) => {
+    if (body[0] === "#") {
+      const code =
+        body[1] === "x" || body[1] === "X"
+          ? parseInt(body.slice(2), 16)
+          : parseInt(body.slice(1), 10);
+      return Number.isFinite(code) ? String.fromCodePoint(code) : whole;
+    }
+    return ENTITIES[body] ?? whole;
+  });
+}
+
+/**
+ * Turn `pdftotext -bbox-layout`'s XHTML into one OcrPage per PDF page.
+ *
+ * Parsed with regexes rather than an XML library on purpose: the output is
+ * machine-generated by poppler with a fixed element shape (`page` > `flow` >
+ * `block` > `line` > `word`), and the alternative is a parser dependency in
+ * the API for one file format read in one place. Only `page` and `word` are
+ * consulted — see below for why poppler's own `line` grouping is discarded.
+ *
+ * `confidence` is 1 for every word: these are the document's own characters,
+ * not a recognition guess.
+ */
+export function parseBboxLayout(xhtml: string, scale: number): (OcrPage | null)[] {
+  const pages: (OcrPage | null)[] = [];
+
+  for (const pageMatch of xhtml.matchAll(/<page\b[^>]*>([\s\S]*?)<\/page>/g)) {
+    const words: OcrWord[] = [];
+
+    for (const w of pageMatch[1].matchAll(
+      /<word\s+xMin="([\d.eE+-]+)"\s+yMin="([\d.eE+-]+)"\s+xMax="([\d.eE+-]+)"\s+yMax="([\d.eE+-]+)"\s*>([\s\S]*?)<\/word>/g,
+    )) {
+      const text = decodeEntities(w[5]).trim();
+      if (!text) continue;
+      const left = Number(w[1]) * scale;
+      const top = Number(w[2]) * scale;
+      words.push({
+        text,
+        left,
+        top,
+        width: Number(w[3]) * scale - left,
+        height: Number(w[4]) * scale - top,
+        confidence: 1,
+      });
+    }
+
+    pages.push(
+      words.length >= MIN_TEXT_WORDS
+        ? { text: toVisualRows(words), words, confidence: 1 }
+        : null,
+    );
+  }
+
+  return pages;
+}
+
+/**
+ * Reassemble words into the rows a reader sees, left to right.
+ *
+ * Poppler's own `<line>` grouping cannot be used for this. It groups by text
+ * flow, and these invoices lay their fields out as two columns of independent
+ * flows — so `PERIODO FACTURADO:` and the `20260630-20260630` printed beside
+ * it end up in different `<line>` elements, and every label-then-value pattern
+ * in the parsers misses a value that is plainly there on the page. Regrouping
+ * by vertical position restores the adjacency, and matches what tesseract
+ * hands back for the scanned version of the same layout.
+ *
+ * Rows are cut when a word's vertical centre leaves the band established by
+ * the row's first word, which tolerates the sub-pixel baseline differences
+ * between fonts on one line without merging two genuinely separate lines.
+ */
+function toVisualRows(words: OcrWord[]): string {
+  const centre = (w: OcrWord) => w.top + w.height / 2;
+  const sorted = [...words].sort((a, b) => centre(a) - centre(b) || a.left - b.left);
+
+  const rows: OcrWord[][] = [];
+  let current: OcrWord[] = [];
+  let band = 0;
+
+  for (const w of sorted) {
+    if (!current.length) {
+      current = [w];
+      band = centre(w);
+      continue;
+    }
+    // Half the word's own height: tall headings and body text both sit within
+    // their own line's band, and neither reaches into the next one.
+    if (Math.abs(centre(w) - band) <= Math.max(w.height, current[0].height) / 2) {
+      current.push(w);
+    } else {
+      rows.push(current);
+      current = [w];
+      band = centre(w);
+    }
+  }
+  if (current.length) rows.push(current);
+
+  return rows
+    .map((r) =>
+      [...r]
+        .sort((a, b) => a.left - b.left)
+        .map((w) => w.text)
+        .join(" "),
+    )
+    .join("\n");
+}
+
 /**
 * Turn tesseract's TSV into words plus reassembled text.
 *
@@ -0,0 +1,194 @@
+import type { OcrPage } from "../ocr/ocr.provider";
+import {
+  detectProvider,
+  normalizeCadastralKey,
+  parseStatement,
+} from "./statement-parser";
+
+/**
+ * Every string in this file is a verbatim excerpt of what the OCR engine
+ * actually returned for a real receipt — misreads, dropped spaces, mangled
+ * accents and all. That is the point: these are the specific ways these five
+ * layouts have been observed to fail, and the assertions pin down what the
+ * parser is supposed to do about each one. Inventing clean input here would
+ * test nothing, because clean input was never the problem.
+ */
+function page(text: string): OcrPage {
+  return { text, words: [], confidence: 0.9 };
+}
+
+describe("detectProvider", () => {
+  it("reads a Rosarito predial receipt as predial, not as a water bill", () => {
+    // "Clave Catastral" is also a CESPT structural marker, so a predial page
+    // whose header OCR'd badly must still not be claimed by the CESPT rule.
+    expect(
+      detectProvider(
+        "e | Clave Catastral. KP-128-105 IMPUESTO PREDIAL ea rita\n" +
+          "TASA | VALOR FISCAL | BIMESTRES | INCISO. | IMPUESTO",
+      ),
+    ).toBe("PREDIAL ROSARITO");
+  });
+
+  it("keeps telling the three municipalities apart by their RFC", () => {
+    expect(detectProvider("R.F.C. ATB-541201-KK2")).toBe("PREDIAL TIJUANA");
+    expect(detectProvider("R.F.C. AMP-981201-HJ4")).toBe("PREDIAL ROSARITO");
+    expect(detectProvider("MEN-540301-9J5")).toBe("PREDIAL ENSENADA");
+  });
+
+  it("does not let the CFE rule claim a gas bill over 'PERIODO FACTURADO'", () => {
+    expect(
+      detectProvider("Orden de Facturación: 000009801640\nPERIODO FACTURADO: 20260630-20260630"),
+    ).toBe("GAS TIJUANA");
+  });
+});
+
+describe("normalizeCadastralKey", () => {
+  it("keeps a letter in the third position instead of digitising it", () => {
+    // `MMB01041` is a real key on file; mapping its B to 8 produced a key that
+    // matches no property at all.
+    expect(normalizeCadastralKey("MM-B01-041", [])).toBe("MMB01041");
+  });
+
+  it("repairs the spurious I tesseract inserts into the prefix", () => {
+    expect(normalizeCadastralKey("MIM-200-010", [])).toBe("MM200010");
+  });
+
+  it("digitises confusable glyphs from position four onward", () => {
+    expect(normalizeCadastralKey("KP-1O8-O45", [])).toBe("KP108045");
+  });
+
+  it("flags a prefix it had to truncate", () => {
+    const notes: string[] = [];
+    expect(normalizeCadastralKey("KPX-128-106", notes)).toBe("KP128106");
+    expect(notes).toHaveLength(1);
+  });
+});
+
+describe("parsePredialTijuana", () => {
+  const TIJUANA = page(
+    "Hats | AYUNTAMIENTO DE TIJUANA, BC $2,613.00 23/01/2026\n" +
+      "y) TELEFONO: 973-7000 R.F.C. ATB-541201-KK2\n" +
+      "ER AÑO VALOR FISCAL TASA IMPUESTO |CONCEPTO IMPORTE\n" +
+      "ED ca 2026 1,207,15778 246 2,969.61 1102 - IMPUESTO PREDIAL 2,969.61\n" +
+      "55164964310126000002613000054192\n" +
+      "se 0 O (54427 [a] | TOTALAPAGAR: 2,613.00\n" +
+      "Dc 1097 : FECHA VENCE : 31/ENE/2026",
+  );
+
+  it("splits the payment barcode into account, deadline and amount", () => {
+    const p = parseStatement(TIJUANA);
+    expect(p.provider).toBe("PREDIAL TIJUANA");
+    expect(p.serviceKind).toBe("PROPERTY_TAX");
+    expect(p.accountRef).toBe("55164964");
+    expect(p.amount).toBe(2613);
+    expect(p.dueDate?.toISOString().slice(0, 10)).toBe("2026-01-31");
+    expect(p.period).toBe("2026");
+  });
+
+  it("reads the printed total even when the space in the label is lost", () => {
+    // The real page OCR'd the label as "TOTALAPAGAR:", and it is that reading
+    // that cross-checks the barcode's amount.
+    expect(parseStatement(TIJUANA).crossChecked).toBe(true);
+  });
+
+  it("refuses to trust a barcode the printed total contradicts", () => {
+    const p = parseStatement(
+      page(
+        "R.F.C. ATB-541201-KK2\n" +
+          "55164964310126000002613000054192\n" +
+          "TOTAL A PAGAR: 9,613.00\nFECHA VENCE : 31/ENE/2026",
+      ),
+    );
+    expect(p.crossChecked).toBe(false);
+    expect(p.notes.join(" ")).toContain("no coincide");
+  });
+});
+
+describe("parsePredialRosarito", () => {
+  it("takes the rounded Total, not the Sub Total printed above it", () => {
+    const p = parseStatement(
+      page(
+        "AYUNTAMIENTO MUNICIPAL DE PLAYAS DE ROSARITO, B.C.\n" +
+          "Ce Clave Catastral: + JR-400-008 7 | IMPUESTO PREDIAL\n" +
+          "SUPERFICIE: 228.31 ZONA 30025 “Redondeo IT049 -$0.39 Sub Total $5,409.39\n" +
+          "¿XTEMPORANEO DESPUES DE: 31/01/2026 Elaboro: MGLG\n" +
+          "Total | $5,409.00\n" +
+          "| Periodo por Pagar: 2026/1 2026/6",
+      ),
+    );
+    expect(p.cadastralKey).toBe("JR400008");
+    expect(p.amount).toBe(5409);
+    expect(p.dueDate?.toISOString().slice(0, 10)).toBe("2026-01-31");
+    expect(p.period).toBe("2026");
+  });
+
+  it("is not fooled by the unspaced 'SubTotal' spelling", () => {
+    // This exact page read $9,624.85 off a receipt for $9,625.00 while the
+    // lookbehind still assumed a space.
+    const p = parseStatement(
+      page(
+        "AMP-981201-HJ4 IMPUESTO PREDIAL\n" +
+          "SUPERFICIE. 367.62 ZONA:30151 | Redondco 17049 $0.15 SubTotal $9,624.85\n" +
+          ": Total | $9,625.00",
+      ),
+    );
+    expect(p.amount).toBe(9625);
+  });
+});
+
+describe("parsePredialEnsenada", () => {
+  const totals = (tail: string) =>
+    page(
+      "IMPRESION MAQUINA REGISTRADORA ez | MUNICIPIO DE ENSENADA\n" +
+        "+7] DATOS. DEL.CAUSANTE alta A pe CLAVE MM-200-010 2 CUENTA\n" +
+        `ES g € S| TOTALES 12,744.47 0.00 0.00 324.56 0.00 13,069.03 ${tail} |`,
+    );
+
+  it("reads the paid total off the TOTALES row however the label OCR'd", () => {
+    expect(parseStatement(totals("TOTA LA A $5,797.00")).amount).toBe(5797);
+    expect(parseStatement(totals("orAL: M7 z] $14,414.00")).amount).toBe(14414);
+    expect(parseStatement(totals("| TOTAL: = $6 246.00")).amount).toBe(6246);
+  });
+
+  it("reports no amount rather than one whose $ was misread as an 8", () => {
+    // `TOTAL: A 82,203.00` is a $2,203.00 receipt. Posting $82,203 would look
+    // entirely ordinary in the ledger, so this page must go to review instead.
+    const p = parseStatement(totals("TOTAL: A 82,203.00"));
+    expect(p.amount).toBeNull();
+    expect(p.notes.join(" ")).toContain("capturarlo a mano");
+  });
+
+  it("never falls back to the assessed total on the same row", () => {
+    expect(parseStatement(totals("yo: se TE= 58/4690]")).amount).toBeNull();
+  });
+});
+
+describe("parseGas", () => {
+  const gas = (...cuentas: string[]) =>
+    page(
+      "GTI4608032K2 COMPAÑIA DE GAS DE TIJUANA\n" +
+        "Fecha de Vencimiento: 2026/08/08\n" +
+        cuentas.map((c) => `Cuenta: ${c}`).join("\n") +
+        "\nPERIODO FACTURADO: 20260630-20260630\nTOTAL A PAGAR: $275.82",
+    );
+
+  it("strips the printed leading zero to the stored account number", () => {
+    const p = parseStatement(gas("0900003463", "0900003463", "0900003463"));
+    expect(p.serviceKind).toBe("GAS");
+    expect(p.accountRef).toBe("900003463");
+    expect(p.amount).toBe(275.82);
+    expect(p.dueDate?.toISOString().slice(0, 10)).toBe("2026-08-08");
+    expect(p.period).toBe("2026-06");
+    expect(p.crossChecked).toBe(true);
+  });
+
+  it("takes the majority reading but still sends a disagreement to review", () => {
+    const p = parseStatement(gas("0900003463", "0900003463", "0900003468"));
+    expect(p.accountRef).toBe("900003463");
+    expect(p.crossChecked).toBe(false);
+  });
+
+  it("claims no cross-check from a single printing", () => {
+    expect(parseStatement(gas("0900003463")).crossChecked).toBeNull();
+  });
+});
@@ -7,7 +7,11 @@ import type { OcrPage, OcrWord } from "../ocr/ocr.provider";
 * like with like and never has to know about provider-specific formatting.
 */
 export interface ParsedStatement {
-  /** "CFE" | "CESPT" | "TELNOR", or null when no parser claimed the page. */
+  /**
+   * "CFE" | "CESPT" | "TELNOR" | "GAS TIJUANA" | "PREDIAL TIJUANA" |
+   * "PREDIAL ROSARITO" | "PREDIAL ENSENADA", or null when no parser claimed
+   * the page.
+   */
  provider: string | null;
  serviceKind: ServiceKind | null;
  accountRef: string | null;
@@ -90,6 +94,16 @@ function firstMatch(text: string, patterns: RegExp[]): string | null {
  return null;
 }

+/** Every capture of `pattern` across the page, in order. */
+function allMatches(text: string, pattern: RegExp): string[] {
+  const out: string[] = [];
+  const re = new RegExp(pattern.source, pattern.flags.includes("g") ? pattern.flags : `${pattern.flags}g`);
+  for (const m of text.matchAll(re)) {
+    if (m[1]) out.push(m[1].trim());
+  }
+  return out;
+}
+
 const MONTHS: Record<string, number> = {
  ENE: 0, FEB: 1, MAR: 2, ABR: 3, MAY: 4, JUN: 5,
  JUL: 6, AGO: 7, SEP: 8, OCT: 9, NOV: 10, DIC: 11,
@@ -104,15 +118,16 @@ export function parseDate(raw: string | null | undefined): Date | null {
  let m = s.match(/^(\d{1,2})\/(\d{1,2})\/(\d{4})$/);
  if (m) return utc(+m[3], +m[2] - 1, +m[1]);

-  // 22-JUL-2026  /  22 JUN 26
-  m = s.match(/^(\d{1,2})[-\s]([A-Z]{3})[A-Z]*[-\s](\d{2,4})$/);
+  // 22-JUL-2026  /  22 JUN 26  /  31/ENE/2026 (Tijuana predial)
+  m = s.match(/^(\d{1,2})[-\s/]([A-Z]{3})[A-Z]*[-\s/](\d{2,4})$/);
  if (m && MONTHS[m[2]] !== undefined) {
    const y = m[3].length === 2 ? 2000 + +m[3] : +m[3];
    return utc(y, MONTHS[m[2]], +m[1]);
  }

-  // 2026-07-22 (already normalised, e.g. decoded from a barcode)
-  m = s.match(/^(\d{4})-(\d{2})-(\d{2})$/);
+  // 2026-07-22 (already normalised, e.g. decoded from a barcode) and the
+  // 2026/08/08 the gas bill prints — same field order, different separator.
+  m = s.match(/^(\d{4})[-/](\d{2})[-/](\d{2})$/);
  if (m) return utc(+m[1], +m[2] - 1, +m[3]);

  return null;
@@ -174,9 +189,30 @@ const BRAND: [string, RegExp][] = [
  ["CFE", /comisi[oó]n federal de electricidad|CFE.?contigo|Suministrador de Servicios/i],
  ["CESPT", /CESPT|COMISI[OÓ]N ESTATAL DE SERVICIOS/i],
  ["TELNOR", /TELNOR|TELEFONOS DEL NOROESTE/i],
+  ["GAS TIJUANA", /COMPA[ÑN][IÍ]?A\s*DE\s*GAS\s*DE\s*TIJUANA|bajagas/i],
+  // The municipal RFCs are the single most reliable discriminator on a predial
+  // receipt: they are printed in a clean monospaced run on every layout, they
+  // never change, and they say which of the three city treasuries issued the
+  // page — which the wordmarks alone do not, since a Tijuana receipt also
+  // carries "PLAYAS DE TIJUANA" and a Rosarito one "TIJUANA ENSENADA".
+  ["PREDIAL TIJUANA", /AYUNTAMIENTO\s*DE\s*TIJUANA|ATB.?541201/i],
+  ["PREDIAL ROSARITO", /AYUNTAMIENTO\s*MUNICIPAL\s*DE\s*PLAYAS\s*DE\s*ROSARITO|AMP.?981201|rosarito\.gob/i],
+  ["PREDIAL ENSENADA", /MUNICIPIO\s*DE\s*ENSENADA|MEN.?540301/i],
 ];

+/**
+ * The predial rules come first because a Rosarito receipt prints "Clave
+ * Catastral" as a boxed label — the very string the CESPT structural rule
+ * looks for — so a page whose municipal header failed to OCR would otherwise
+ * be claimed as a water bill and matched against the wrong column entirely.
+ * "IMPUESTO PREDIAL" appears on all three municipal layouts and on none of the
+ * utility ones, so it is the safe first question to ask.
+ */
 const LAYOUT: [string, RegExp][] = [
+  ["PREDIAL TIJUANA", /IMPUESTO\s*PREDIAL[\s\S]*?(?:CERTIFICACION\s*DE\s*CAJA|PASEO\s*DEL\s*CENTENARIO|PAGA\s*TU\s*PREDIAL)/i],
+  ["PREDIAL ENSENADA", /(?:IMPUESTO\s*PREDIAL[\s\S]*?TRANSPENINSULAR)|(?:IMPRESION\s*MAQUINA\s*REGISTRADORA)/i],
+  ["PREDIAL ROSARITO", /IMPUESTO\s*PREDIAL/i],
+  ["GAS TIJUANA", /Orden\s*de\s*Facturaci[oó]n|FACTOR\s*DE\s*PRESI[OÓ]N|GAS\s*LP/i],
  ["CFE", /NO\.?\s*DE\s*SERVICIO|L[IÍ]MITE\s*DE\s*PAGO|PERIODO\s*FACTURADO/i],
  ["CESPT", /SALDO\s+CORRIENTE|CLAVE\s*CATASTRAL|No\.?\s*DE\s*CUENTA/i],
  ["TELNOR", /Mes\s*de\s*Facturaci[oó]n|Pagar\s*antes\s*de/i],
@@ -364,10 +400,315 @@ function parseTelnor(page: OcrPage): ParsedStatement {
  };
 }

+// --- GAS (Compañía de Gas de Tijuana / bajagas) ------------------------------
+
+/**
+ * These arrive as born-digital CFDI PDFs rather than scans, so the text layer
+ * (see `TesseractOcrProvider.textPages`) usually reads them exactly and the
+ * patterns below only have to be tolerant enough for the scanned case.
+ *
+ * The account number is printed three times — supply address, fiscal data, and
+ * the payment stub at the foot — which is a free cross-check: three readings
+ * that agree are near-certainly right, and any disagreement means one of them
+ * was misread and the page deserves a human glance.
+ *
+ * `Cuenta` is what the matcher compares, not `Contrato`. The migration
+ * recovered gas references out of `PropertyService.notes` into `meterNumber`
+ * and what sat there is the 9-digit account (`900003463`), printed here with a
+ * leading zero as `0900003463`.
+ */
+function parseGas(page: OcrPage): ParsedStatement {
+  const text = page.text;
+  const notes: string[] = [];
+
+  const seen = allMatches(text, /Cuenta\s*[:;.]?\s*([0-9OIlSBD]{6,12})/i).map((s) =>
+    toDigits(s).replace(/^0+/, ""),
+  );
+  const distinct = [...new Set(seen.filter(Boolean))];
+
+  let accountRef: string | null = null;
+  let crossChecked: boolean | null = null;
+  if (distinct.length === 1) {
+    accountRef = distinct[0];
+    if (seen.length > 1) crossChecked = true;
+  } else if (distinct.length > 1) {
+    // Majority wins — the stub and the two address blocks print the same
+    // number, so a single divergent reading is the misread one. It still goes
+    // to review: `crossChecked: false` is what keeps the batch from
+    // auto-matching a number one of three readings disagreed with.
+    const tally = new Map<string, number>();
+    for (const s of seen) tally.set(s, (tally.get(s) ?? 0) + 1);
+    accountRef = [...tally.entries()].sort((a, b) => b[1] - a[1])[0][0];
+    crossChecked = false;
+    notes.push(`el número de cuenta se leyó de ${distinct.length} formas distintas (${distinct.join(", ")})`);
+  }
+
+  const amount = money(
+    firstMatch(text, [
+      /TOTAL\s*A\s*PAGAR\s*[:;.]?\s*\$\s*([\d,]+\.\d{2})/i,
+      /Total\s*a\s*pagar\s*[:;.]?\s*\$\s*([\d,]+\.\d{2})/i,
+    ]),
+  );
+
+  // `20260630-20260630` — the range the bill was cut for. Both ends are the
+  // same reading date on every sample, so the period is reported as the ISO
+  // month rather than a range no ledger row would ever be searched by.
+  const facturado = firstMatch(text, [/PERIODO\s*FACTURADO\s*[:;.]?\s*(\d{8})\s*-\s*\d{8}/i]);
+  const period = facturado ? `${facturado.slice(0, 4)}-${facturado.slice(4, 6)}` : null;
+
+  return {
+    provider: "GAS TIJUANA",
+    serviceKind: "GAS",
+    accountRef: accountRef || null,
+    cadastralKey: null,
+    amount,
+    dueDate: parseDate(
+      firstMatch(text, [/Fecha\s*de\s*Vencimiento\s*[:;.]?\s*(\d{4}\s*\/\s*\d{2}\s*\/\s*\d{2})/i])?.replace(
+        /\s/g,
+        "",
+      ),
+    ),
+    period,
+    crossChecked,
+    notes,
+  };
+}
+
+// --- PREDIAL (municipal property tax) ---------------------------------------
+
+/**
+ * Normalise a printed clave catastral to the eight-character form
+ * `Property.cadastralKey` holds. The municipalities print it grouped
+ * (`KP-128-106`, `MM-B01-041`); the stored value drops the separators
+ * (`KP128106`, `MMB01041`).
+ *
+ * The shape is *not* two letters and six digits, which is the assumption that
+ * has to be resisted here. Across the 932 distinct claves on file, characters
+ * four through eight are digits without exception, but the third is a digit in
+ * 917 of them and one of `A`, `B`, `H`, `T` in the other fifteen. Running the
+ * whole tail through `toDigits` — which maps `B` to `8` — is what turned a real
+ * `MMB01041` into a nonexistent `MM801041`, so only positions four onward get
+ * that treatment and a letter in the third position is kept as printed.
+ *
+ * That leaves a genuine ambiguity at that one position: a `B` there might be a
+ * misread `8`, and 34 stored claves do carry an `8` there against six with a
+ * `B`. It is left as read rather than guessed, because a page that fails to
+ * match lands in the review queue where a human fixes it in seconds, while a
+ * page that matches the wrong property posts a charge to the wrong customer.
+ *
+ * The two-letter prefix is the other fragile part. Tesseract inserts a spurious
+ * `I` into letter pairs with some regularity — a real `MM-200-010` came back as
+ * `MIM-200-010` — so a run longer than two letters has its `I`/`L` dropped
+ * first, which recovers exactly that case. Anything still not two letters is
+ * truncated and flagged, because a wrong prefix silently matches the wrong
+ * property or, more often, nothing at all.
+ */
+export function normalizeCadastralKey(
+  raw: string,
+  notes: string[],
+): string | null {
+  const m = raw.match(/^([A-Za-z|]{2,5})[-\s]?([A-Za-z0-9|]{3})[-\s]?([0-9OIlSBD]{3})$/);
+  if (!m) return null;
+
+  let letters = m[1].toUpperCase().replace(/[^A-Z]/g, "");
+  if (letters.length > 2) {
+    const stripped = letters.replace(/[IL]/g, "");
+    if (stripped.length === 2) {
+      letters = stripped;
+    } else {
+      letters = letters.slice(0, 2);
+      notes.push(`la clave catastral se leyó como "${m[1]}"; se tomó "${letters}"`);
+    }
+  }
+  if (letters.length !== 2) return null;
+
+  const third = m[2][0].toUpperCase();
+  const tail =
+    (/[A-Z]/.test(third) ? third : toDigits(third)) +
+    toDigits(m[2].slice(1)) +
+    toDigits(m[3]);
+
+  return tail.length === 6 ? letters + tail : null;
+}
+
+/** The grouped clave as printed, anchored to its label when one survived OCR. */
+const GROUPED_CLAVE = "[A-Z|]{2,5}-[A-Z0-9OIlSBD]{3}-[0-9OIlSBD]{3}";
+
+function findCadastralKey(text: string, notes: string[]): string | null {
+  const labelled = firstMatch(text, [
+    new RegExp(`Clave\\s*Catastral\\s*[^A-Z0-9]{0,8}(${GROUPED_CLAVE})`, "i"),
+    new RegExp(`CLAVE\\s*[^A-Z0-9]{0,8}(${GROUPED_CLAVE})`, "i"),
+  ]);
+  if (labelled) return normalizeCadastralKey(labelled, notes);
+
+  // Ensenada's label ("CLAVE") lands inside a table header that OCRs into
+  // noise more often than not, so the bare grouped shape is accepted as a
+  // fallback. It is distinctive enough — two letters and two three-character
+  // groups joined by hyphens appears nowhere else on these pages.
+  const bare = firstMatch(text, [new RegExp(`\\b(${GROUPED_CLAVE})\\b`)]);
+  return bare ? normalizeCadastralKey(bare, notes) : null;
+}
+
+/**
+ * Tijuana: a "CERTIFICACIÓN DE CAJA" whose payment barcode is one 32-digit run
+ * of `account(8) + due date(DDMMYY) + amount(9) + folio(9)`, verified against
+ * all five sample pages. Municipal totals are whole pesos (the receipt itself
+ * carries a "Redondeo" line), so the barcode amount needs no decimal point.
+ *
+ * No clave catastral is printed anywhere on this layout — the 8-digit
+ * municipal account is the only identifier, and it is not a number the legacy
+ * database ever held. Until a reviewer confirms one, every Tijuana page lands
+ * in review; confirming teaches the matcher (see `learnAccountRefs`) so the
+ * same property matches itself next year.
+ */
+function parsePredialTijuana(page: OcrPage): ParsedStatement {
+  const text = page.text;
+  const notes: string[] = [];
+
+  const barcode = text.match(/(?<![0-9OIlSBD])([0-9OIlSBD]{32})(?![0-9OIlSBD])/);
+  const printedTotal = money(
+    firstMatch(text, [/TOTAL\s*A?\s*PAGAR\s*[:;.]?\s*\$?\s*([\d,]+\.?\d{0,2})/i]),
+  );
+
+  let accountRef: string | null = null;
+  let amount: number | null = printedTotal;
+  let dueDate: Date | null = null;
+  let crossChecked: boolean | null = null;
+
+  if (barcode) {
+    const run = toDigits(barcode[1]);
+    const d = run.slice(8, 14);
+    const fromBarcode = Number(run.slice(14, 23));
+    accountRef = run.slice(0, 8);
+    dueDate = parseDate(`20${d.slice(4, 6)}-${d.slice(2, 4)}-${d.slice(0, 2)}`);
+    notes.push("cuenta, importe y vencimiento leídos del código de barras");
+
+    if (printedTotal != null) {
+      // Guarding the money, not the account number: the printed total is the
+      // figure a human would key, so when the two disagree one of them is a
+      // misread peso amount and nothing should post unreviewed.
+      crossChecked = Math.abs(printedTotal - fromBarcode) < 0.5;
+      if (!crossChecked) {
+        notes.push(
+          `el total impreso (${printedTotal}) no coincide con el código de barras (${fromBarcode})`,
+        );
+      }
+    }
+    if (amount == null) amount = fromBarcode;
+  }
+
+  if (!dueDate) {
+    dueDate = parseDate(
+      firstMatch(text, [/FECHA\s*VENCE\s*[:;.]?\s*(\d{1,2}\/\w{3}\/\d{4})/i]),
+    );
+  }
+
+  return {
+    provider: "PREDIAL TIJUANA",
+    serviceKind: "PROPERTY_TAX",
+    accountRef: accountRef || null,
+    cadastralKey: null,
+    amount,
+    dueDate,
+    // The fiscal year, which is what the legacy ledger's `period` holds for
+    // predial ("2026" is its single most common value). It is read from the
+    // assessment table's year column, and failing that from the deadline: a
+    // predial bill for year N falls due on 31 January of year N.
+    period:
+      firstMatch(text, [/VALOR\s*FISCAL[\s\S]{0,160}?\b(20\d{2})\b/i]) ??
+      (dueDate ? String(dueDate.getUTCFullYear()) : null),
+    crossChecked,
+    notes,
+  };
+}
+
+/**
+ * Rosarito: a wide "CERTIFICACIÓN DE CAJA" keyed by clave catastral, with no
+ * account number of its own — the clave is the identifier, which is exactly
+ * what `Property.cadastralKey` holds, so these match on the first pass.
+ *
+ * The total is read with a negative lookbehind on "Sub": the receipt prints
+ * `Sub Total $5,409.39` (before the peso rounding) directly above
+ * `Total $5,409.00`, and taking the first "Total" on the page books 39 cents
+ * that the municipality did not charge. The lookbehind allows zero spaces
+ * because the label prints both ways — `Sub Total` on one sample and
+ * `SubTotal` on the next, and the tight one is what slipped past a fixed
+ * `Sub\s` and read $9,624.85 off a receipt for $9,625.00.
+ */
+function parsePredialRosarito(page: OcrPage): ParsedStatement {
+  const notes: string[] = [];
+  const text = page.text;
+
+  return {
+    provider: "PREDIAL ROSARITO",
+    serviceKind: "PROPERTY_TAX",
+    accountRef: null,
+    cadastralKey: findCadastralKey(text, notes),
+    amount: money(firstMatch(text, [/(?<!Sub\s{0,3})Total\s*[|:;.]?\s*\$\s*([\d,]+\.\d{2})/i])),
+    // "EXTEMPORANEO DESPUES DE: 31/01/2026" — the leading E is regularly eaten
+    // by the box rule printed over it, so the anchor starts at "XTEMPORANEO".
+    dueDate: parseDate(
+      firstMatch(text, [/XTEMPOR[AÁ]NEO\s*DESPU[EÉ]S\s*DE\s*[:;.]?\s*(\d{2}\/\d{2}\/\d{4})/i]),
+    ),
+    period: firstMatch(text, [/Periodo\s*por\s*Pagar\s*[:;.]?\s*(20\d{2})/i]),
+    crossChecked: null,
+    notes,
+  };
+}
+
+/**
+ * Ensenada: a dot-matrix "IMPRESION MAQUINA REGISTRADORA" statement, by some
+ * distance the worst-scanning of the three. Matching is by clave catastral.
+ *
+ * The amount is read positionally rather than by label, because the label does
+ * not survive: across five real pages the same word came back as `TOTAL:`,
+ * `TOTA LA A` and `orAL:`. What is stable is the row — the summary line that
+ * starts `TOTALES` carries the assessed figures across it and the amount
+ * actually paid last, at the right margin.
+ *
+ * That last figure must carry a literal `$`. On a real sample the paid total
+ * printed as `TOTAL: A $2,203.00` and OCR'd as `TOTAL: A 82,203.00` — the
+ * dollar sign read as an 8, a mistake that would post a $2,203 charge as
+ * $82,203 and look entirely ordinary in the ledger. Requiring the `$` costs
+ * that page its amount and sends it to review, which is the only acceptable
+ * failure here. The unprefixed figures earlier on the row are deliberately not
+ * a fallback: they are the tax assessed before the early-payment discount, not
+ * what was paid.
+ */
+function parsePredialEnsenada(page: OcrPage): ParsedStatement {
+  const notes: string[] = [];
+  const text = page.text;
+
+  const totalsRow = text.split("\n").find((l) => /TOTALES/i.test(l)) ?? "";
+  const figures = allMatches(totalsRow, /\$\s*(\d[\d,.\s]*\.\d{2})/);
+  const amount = figures.length ? money(figures[figures.length - 1]) : null;
+  if (amount == null) {
+    notes.push("no se pudo leer el importe con certeza; capturarlo a mano");
+  }
+
+  return {
+    provider: "PREDIAL ENSENADA",
+    serviceKind: "PROPERTY_TAX",
+    accountRef: null,
+    cadastralKey: findCadastralKey(text, notes),
+    amount,
+    // This layout prints no payment deadline at all — it is a receipt for a
+    // payment already made at the municipal window.
+    dueDate: null,
+    period: firstMatch(text, [/A[ÑN]O\s*[\s\S]{0,60}?\b(20\d{2})\b/i]),
+    crossChecked: null,
+    notes,
+  };
+}
+
 const PARSERS: Record<string, (page: OcrPage) => ParsedStatement> = {
  CFE: parseCfe,
  CESPT: parseCespt,
  TELNOR: parseTelnor,
+  "GAS TIJUANA": parseGas,
+  "PREDIAL TIJUANA": parsePredialTijuana,
+  "PREDIAL ROSARITO": parsePredialRosarito,
+  "PREDIAL ENSENADA": parsePredialEnsenada,
 };

 const EMPTY: ParsedStatement = {
@@ -32,12 +32,27 @@ export interface MatchResult {
 * person. Names are displayed for the reviewer to sanity-check, and are never
 * an input to matching.
 */
-@Injectable()
-export class StatementMatcherService {
-  constructor(private readonly prisma: PrismaService) {}
-
-  /** Which PropertyService column a given kind's statements actually print. */
-  private fieldFor(kind: ServiceKind): "accountNumber" | "meterNumber" | null {
+/**
+ * Which `PropertyService` column a given kind's statements actually print.
+ *
+ * Exported because the same answer governs three places that must agree: the
+ * lookup here, the blank-service fill on review, and the write-back on confirm.
+ * When they disagree, a reference gets learned into a column nothing searches,
+ * and the same page returns to the review queue every month forever.
+ *
+ * `meterNumber` is doing double duty for the two kinds whose printed reference
+ * DATMEX never held in `accountNumber`:
+ *   - GAS, where the number lived in free-text notes, and
+ *   - PROPERTY_TAX, where `accountNumber` holds DATMEX.predial — a 3-4 digit
+ *     office file number that is neither unique nor printed on any statement.
+ *     The Tijuana municipal receipt prints an 8-digit account and no clave
+ *     catastral at all, so it needs a column of its own; overwriting the legacy
+ *     predial numbers to make room would destroy the only link back to the
+ *     original records.
+ */
+export function scopedRefField(
+  kind: ServiceKind,
+): "accountNumber" | "meterNumber" | null {
  switch (kind) {
    case "ELECTRIC": // CFE "NO. DE SERVICIO" -> DATMEX.rpu
    case "WATER": // CESPT "Cuenta" / "No. DE CUENTA" -> DATMEX.agua
@@ -45,17 +60,18 @@ export class StatementMatcherService {
    case "FEDERAL_ZONE":
    case "CABLE":
      return "accountNumber";
-      case "GAS": // no account column in DATMEX; the number lived in notes
+    case "GAS": // bajagas "Cuenta" -> recovered from notes into meterNumber
+    case "PROPERTY_TAX": // Tijuana's 8-digit municipal account
      return "meterNumber";
-      // PROPERTY_TAX deliberately has no scoped column: what its
-      // accountNumber holds is DATMEX.predial, which is neither unique nor
-      // printed on any statement. Predial bills match on the clave catastral
-      // alone — see matchByCadastralKey.
    default:
      return null;
  }
 }

+@Injectable()
+export class StatementMatcherService {
+  constructor(private readonly prisma: PrismaService) {}
+
  async match(parsed: ParsedStatement, expectedKind: ServiceKind): Promise<MatchResult> {
    const kind = parsed.serviceKind ?? expectedKind;

@@ -68,33 +84,39 @@ export class StatementMatcherService {
      );
    }

-    const field = this.fieldFor(kind);
+    const field = scopedRefField(kind);

    if (field && parsed.accountRef) {
      const hit = await this.byServiceField(kind, field, parsed.accountRef);
      if (hit) return hit;
    }

-    // Secondary key. The clave catastral is printed on CESPT bills as well as
-    // predial ones, so it rescues a page whose account number did not OCR —
-    // which happened on real samples, where the clave read cleanly and the
-    // account number did not.
+    // The clave catastral is printed on CESPT bills as well as predial ones, so
+    // it rescues a page whose account number did not OCR — which happened on
+    // real samples, where the clave read cleanly and the account number did
+    // not. On the Rosarito and Ensenada predial layouts it is not a rescue at
+    // all but the only identifier the receipt carries, so a unique hit there is
+    // as good as any account-number match and is treated as one.
    if (parsed.cadastralKey) {
-      const hit = await this.byCadastralKey(kind, parsed.cadastralKey);
+      const primary = kind === "PROPERTY_TAX" && !parsed.accountRef;
+      const hit = await this.byCadastralKey(kind, parsed.cadastralKey, primary);
      if (hit) return hit;
    }

    if (!field && !parsed.cadastralKey) {
+      return this.unmatched(`no hay campo de búsqueda definido para ${kind}`);
+    }
+    if (!parsed.accountRef && !parsed.cadastralKey) {
      return this.unmatched(
        kind === "PROPERTY_TAX"
-          ? "el predial sólo se puede identificar por clave catastral y no se leyó ninguna"
-          : `no hay campo de búsqueda definido para ${kind}`,
+          ? "no se leyó ni la clave catastral ni la cuenta municipal"
+          : "no se pudo leer la referencia de la cuenta",
      );
    }
    return this.unmatched(
      parsed.accountRef
        ? `no se encontró ningún servicio de ${kind} con la referencia ${parsed.accountRef}`
-        : "no se pudo leer la referencia de la cuenta",
+        : `no se encontró ninguna propiedad con la clave catastral ${parsed.cadastralKey}`,
    );
  }

@@ -145,6 +167,8 @@ export class StatementMatcherService {
  private async byCadastralKey(
    kind: ServiceKind,
    key: string,
+    /** True when the clave is the identifier the statement was issued against. */
+    primary: boolean,
  ): Promise<MatchResult | null> {
    const props = await this.prisma.property.findMany({
      where: { cadastralKey: key },
@@ -174,15 +198,21 @@ export class StatementMatcherService {
      };
    }

-    // The clave identifies the property with certainty, but it is a *secondary*
-    // key: it was not the number the statement was issued against. Left for
-    // review so the confirm also teaches the matcher the account number, rather
-    // than the same page needing the fallback again next month.
+    // When the clave is the *secondary* key — a utility bill that also happens
+    // to print it — the page is left for review, because the clave was not the
+    // number the statement was issued against and confirming is what teaches
+    // the matcher the account number for next month. When it is the primary key
+    // (Rosarito and Ensenada predial, which print nothing else), a unique hit
+    // is a real match and there is no second number to learn.
    return {
      propertyServiceId: candidates[0].propertyServiceId ?? null,
      customerId: candidates[0].customerId,
-      note: `identificado por clave catastral ${key}; confirme para registrar también el número de cuenta`,
-      confident: false,
+      note: primary
+        ? `coincidencia exacta por clave catastral ${key}`
+        : `identificado por clave catastral ${key}; confirme para registrar también el número de cuenta`,
+      // A clave with no service row of the right kind behind it still needs a
+      // human: there is nothing to attach the posting to.
+      confident: primary && candidates[0].propertyServiceId != null,
      candidates,
    };
  }
@@ -16,7 +16,7 @@ import { BillingService } from "../billing/billing.service";
 import type { UploadedFileLike } from "../storage/upload-file";
 import { OCR_PROVIDER, type OcrProvider } from "./ocr/ocr.provider";
 import { parseStatement } from "./parsers/statement-parser";
-import { StatementMatcherService } from "./statement-matcher.service";
+import { StatementMatcherService, scopedRefField } from "./statement-matcher.service";
 import type { ConfirmBatchDto, ReviewDocumentDto } from "./statement.dto";

 /**
@@ -127,14 +127,23 @@ export class StatementsService {
      await this.storage.put(sourceKey, file.buffer, "application/pdf");

      const pages = await this.ocr.renderPages(file.buffer);
-      for (const image of pages) {
+      // Page images are still rendered and stored for every file, text layer or
+      // not: the review screen shows the reviewer the page, and "what the
+      // parser read" is only checkable against a picture of the paper.
+      const textLayer = await this.ocr.textPages(file.buffer).catch(() => []);
+
+      for (const [index, image] of pages.entries()) {
        pageNumber += 1;
        const storageKey = `statement/${batchId}/page-${pageNumber}.png`;
        await this.storage.put(storageKey, image, "image/png");

        try {
-          const ocr = await this.ocr.recognize(image);
+          const embedded = textLayer[index] ?? null;
+          const ocr = embedded ?? (await this.ocr.recognize(image));
          const parsed = parseStatement(ocr);
+          if (embedded) {
+            parsed.notes.unshift("texto leído del PDF original, sin OCR");
+          }
          const match = await this.matcher.match(parsed, serviceKind);

          const notes = [...parsed.notes, match.note].filter(Boolean);
@@ -295,8 +304,8 @@ export class StatementsService {
        where: { id: doc.batchId },
        select: { serviceKind: true },
      });
-      if (batch) {
-        const field = batch.serviceKind === "GAS" ? "meterNumber" : "accountNumber";
+      const field = batch && scopedRefField(batch.serviceKind);
+      if (batch && field) {
        const blank = await this.prisma.propertyService.findMany({
          where: {
            kind: batch.serviceKind,
@@ -434,7 +443,8 @@ export class StatementsService {
    docs: { matchedPropertyServiceId: string | null; extractedAccountRef: string | null }[],
    kind: ServiceKind,
  ) {
-    const field = kind === "GAS" ? "meterNumber" : "accountNumber";
+    const field = scopedRefField(kind);
+    if (!field) return;
    for (const d of docs) {
      if (!d.matchedPropertyServiceId || !d.extractedAccountRef) continue;
      await this.prisma.propertyService.updateMany({
@@ -0,0 +1,4 @@
+{
+  "extends": "./tsconfig.json",
+  "exclude": ["node_modules", "dist", "**/*.spec.ts"]
+}
@@ -1,6 +1,6 @@
 {
  "name": "@jorgecuadros/web",
-  "version": "1.0.2",
+  "version": "1.0.5",
  "private": true,
  "scripts": {
    "dev": "next dev -p 4500",
@@ -837,6 +837,46 @@ button {
  display: inline-block;
 }

+/* Upload progress (Operaciones ingest) */
+.upload-progress {
+  display: flex;
+  flex-direction: column;
+  gap: 6px;
+  padding: 4px 0 8px;
+}
+.progress-track {
+  position: relative;
+  overflow: hidden;
+  height: 8px;
+  border-radius: 999px;
+  background: var(--paper-2);
+}
+.progress-fill {
+  height: 100%;
+  border-radius: 999px;
+  background: var(--brand-500);
+  transition: width 0.2s linear;
+}
+.progress-indeterminate .progress-fill {
+  width: 40% !important;
+  animation: progress-slide 1.2s var(--ease-out-quart) infinite;
+}
+@keyframes progress-slide {
+  0% {
+    transform: translateX(-100%);
+  }
+  100% {
+    transform: translateX(250%);
+  }
+}
+.upload-progress-stats {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 12px;
+  font-size: 12px;
+  color: var(--muted);
+}
+
@keyframes shimmer {
  0% {
    background-position: -420px 0;
@@ -1,6 +1,6 @@
 "use client";

-import { useCallback, useEffect, useRef, useState } from "react";
+import { Fragment, useCallback, useEffect, useRef, useState } from "react";
 import { AppShell } from "@/components/AppShell";
 import { useCan } from "@/lib/abilities";
 import {
@@ -20,6 +20,7 @@ import {
  startOpsJob,
  uploadIngest,
 } from "@/lib/api";
+import type { UploadProgress } from "@/lib/api";
 import type {
  BackupFile,
  IngestFile,
@@ -56,6 +57,7 @@ function Operaciones() {
  const [confirm, setConfirm] = useState<ConfirmState>(null);
  const [confirmText, setConfirmText] = useState("");
  const [uploading, setUploading] = useState<string | null>(null);
+  const [progress, setProgress] = useState<UploadProgress | null>(null);
  const [starting, setStarting] = useState(false);

  const fileInputs = useRef<Record<string, HTMLInputElement | null>>({});
@@ -119,14 +121,16 @@ function Operaciones() {
    setError(null);
    setNotice(null);
    setUploading(name);
+    setProgress(null);
    try {
-      await uploadIngest(name, file);
+      await uploadIngest(name, file, setProgress);
      setNotice(`${name} cargado.`);
      refreshLists();
    } catch (e) {
      setError((e as Error)?.message ?? "No se pudo cargar el archivo.");
    } finally {
      setUploading(null);
+      setProgress(null);
      const input = fileInputs.current[name];
      if (input) input.value = "";
    }
@@ -243,7 +247,8 @@ function Operaciones() {
            </thead>
            <tbody>
              {(ingest ?? []).map((f) => (
-                <tr key={f.name}>
+                <Fragment key={f.name}>
+                  <tr>
                    <td className="mono">{f.name}</td>
                    <td>
                      <span className={`badge ${f.present ? "badge-positive" : "badge-negative"}`}>
@@ -274,6 +279,7 @@ function Operaciones() {
                          <button
                            className="btn btn-ghost"
                            type="button"
+                            disabled={uploading === f.name}
                            onClick={() => handleDeleteIngest(f.name)}
                          >
                            Eliminar
@@ -282,6 +288,14 @@ function Operaciones() {
                      </div>
                    </td>
                  </tr>
+                  {uploading === f.name && (
+                    <tr>
+                      <td colSpan={5}>
+                        <UploadProgressBar progress={progress} />
+                      </td>
+                    </tr>
+                  )}
+                </Fragment>
              ))}
            </tbody>
          </table>
@@ -496,6 +510,61 @@ function Operaciones() {
  );
 }

+/** "1:05" / "0:09" — remaining time, coarse on purpose. */
+function formatEta(seconds: number): string {
+  const s = Math.max(0, Math.round(seconds));
+  if (s >= 3600) {
+    const h = Math.floor(s / 3600);
+    const m = Math.round((s % 3600) / 60);
+    return `${h} h ${m} min`;
+  }
+  return `${Math.floor(s / 60)}:${String(s % 60).padStart(2, "0")}`;
+}
+
+/**
+ * Live upload readout. The bar tracks bytes handed to the network; once those
+ * are all sent the server still has to write the file, so the tail of the
+ * upload reads "Procesando…" instead of sitting at 100%.
+ */
+function UploadProgressBar({ progress }: { progress: UploadProgress | null }) {
+  const pct = progress?.fraction != null ? Math.round(progress.fraction * 100) : null;
+  return (
+    <div className="upload-progress">
+      <div
+        className={`progress-track${pct === null ? " progress-indeterminate" : ""}`}
+        role="progressbar"
+        aria-valuemin={0}
+        aria-valuemax={100}
+        aria-valuenow={pct ?? undefined}
+      >
+        <div className="progress-fill" style={{ width: `${pct ?? 100}%` }} />
+      </div>
+      <div className="upload-progress-stats mono">
+        {progress === null ? (
+          "Preparando…"
+        ) : progress.finishing ? (
+          `Procesando en el servidor… (${formatBytes(progress.total)} enviados)`
+        ) : (
+          <>
+            {pct !== null && <strong>{pct}%</strong>}
+            {progress.total > 0 && (
+              <span>
+                {formatBytes(progress.loaded)} / {formatBytes(progress.total)}
+              </span>
+            )}
+            {progress.bytesPerSecond > 0 && (
+              <span>{formatBytes(progress.bytesPerSecond)}/s</span>
+            )}
+            {progress.secondsRemaining !== null && progress.bytesPerSecond > 0 && (
+              <span>faltan {formatEta(progress.secondsRemaining)}</span>
+            )}
+          </>
+        )}
+      </div>
+    </div>
+  );
+}
+
 function OpTile({
  title,
  desc,
@@ -132,9 +132,7 @@ function BatchReview({ id }: { id: string }) {
      {error && <div className="state-box state-error">{error}</div>}

      {processing && (
-        <div className="state-box">
-          Leyendo los recibos… esta pantalla se actualiza sola.
-        </div>
+        <ProcessingBanner docsLength={docs.length} pendingOcr={batch.byStatus.PENDING_OCR ?? 0} />
      )}

      <SummaryCard batch={batch} readyCount={readyCount} />
@@ -170,6 +168,55 @@ const STATUS_LABEL_BATCH: Record<string, string> = {
  FAILED: "Falló",
 };

+/**
+ * Live readout while OCR is running. The backend tells us how many pages are
+ * still PENDING_OCR, so we can show real progress instead of "loading…". When
+ * the docs list hasn't caught up to the upload yet (total === 0) we fall back
+ * to the indeterminate bar.
+ */
+function ProcessingBanner({
+  docsLength,
+  pendingOcr,
+}: {
+  docsLength: number;
+  pendingOcr: number;
+}) {
+  const done = Math.max(docsLength - pendingOcr, 0);
+  const pct =
+    docsLength > 0 ? Math.min(100, Math.round((done / docsLength) * 100)) : null;
+  return (
+    <div className="card" style={{ padding: 16 }}>
+      <div className="upload-progress" style={{ padding: 0 }}>
+        <div
+          className={`progress-track${pct === null ? " progress-indeterminate" : ""}`}
+          role="progressbar"
+          aria-valuemin={0}
+          aria-valuemax={100}
+          aria-valuenow={pct ?? undefined}
+        >
+          <div className="progress-fill" style={{ width: `${pct ?? 100}%` }} />
+        </div>
+        <div className="upload-progress-stats">
+          {pct === null ? (
+            <span>Leyendo los recibos…</span>
+          ) : (
+            <>
+              <strong>{pct}%</strong>
+              <span>
+                {done} de {docsLength} página(s) leídas
+              </span>
+              {pendingOcr > 0 && <span>{pendingOcr} en cola</span>}
+            </>
+          )}
+          <span style={{ marginLeft: "auto" }}>
+            Esta pantalla se actualiza sola.
+          </span>
+        </div>
+      </div>
+    </div>
+  );
+}
+
 function SummaryCard({
  batch,
  readyCount,
@@ -28,9 +28,15 @@ import type { ServiceKind, StatementBatch, StatementBatchStatus } from "@/lib/ty
 */

 /** The kinds the parsers actually recognise today. */
-const SUPPORTED: ServiceKind[] = ["ELECTRIC", "WATER", "TELEPHONE"];
+const SUPPORTED: ServiceKind[] = [
+  "ELECTRIC",
+  "WATER",
+  "TELEPHONE",
+  "GAS",
+  "PROPERTY_TAX",
+];
 /** Uploadable, but every page will land in review until a parser learns it. */
-const OTHER_KINDS: ServiceKind[] = ["GAS", "PROPERTY_TAX", "FEDERAL_ZONE", "CABLE"];
+const OTHER_KINDS: ServiceKind[] = ["FEDERAL_ZONE", "CABLE"];

 const STATUS_LABEL: Record<StatementBatchStatus, string> = {
  UPLOADED: "Recibido",
@@ -803,38 +803,117 @@ export function listIngest(): Promise<IngestFile[]> {
  return apiFetch<IngestFile[]>("/ops/ingest");
 }

+/** Live upload stats reported to `uploadFile`'s `onProgress` callback. */
+export type UploadProgress = {
+  loaded: number;
+  /** 0 when the browser can't compute the request size. */
+  total: number;
+  /** 0..1, or null when `total` is unknown. */
+  fraction: number | null;
+  /** Smoothed transfer rate. */
+  bytesPerSecond: number;
+  /** null until a rate and a total are both known. */
+  secondsRemaining: number | null;
+  /** True once the bytes are sent and we're waiting on the server's reply. */
+  finishing: boolean;
+};
+
 /**
 * Multipart upload — not JSON, so it bypasses apiFetch's Content-Type. `path`
 * is API-relative (may include a query string); `filename` overrides the part
- * name sent to the server.
+ * name sent to the server. Uses XHR rather than fetch because fetch has no way
+ * to report request-body progress.
 */
-export async function uploadFile(
+export function uploadFile(
  path: string,
  file: File,
  filename?: string,
+  onProgress?: (p: UploadProgress) => void,
 ): Promise<unknown> {
  const body = new FormData();
  body.append("file", file, filename ?? file.name);
-  const res = await fetch(`${API_ORIGIN}${path}`, {
-    method: "POST",
-    credentials: "include",
-    body,
+
+  return new Promise((resolve, reject) => {
+    const xhr = new XMLHttpRequest();
+    xhr.open("POST", `${API_ORIGIN}${path}`);
+    xhr.withCredentials = true;
+
+    if (onProgress) {
+      // Exponentially smoothed rate — raw per-chunk deltas jump around too
+      // much to read.
+      let lastAt = performance.now();
+      let lastLoaded = 0;
+      let rate = 0;
+
+      xhr.upload.onprogress = (e) => {
+        const now = performance.now();
+        const dt = (now - lastAt) / 1000;
+        if (dt >= 0.15) {
+          const sample = (e.loaded - lastLoaded) / dt;
+          rate = rate === 0 ? sample : rate * 0.7 + sample * 0.3;
+          lastAt = now;
+          lastLoaded = e.loaded;
+        }
+        const total = e.lengthComputable ? e.total : 0;
+        onProgress({
+          loaded: e.loaded,
+          total,
+          fraction: total ? e.loaded / total : null,
+          bytesPerSecond: rate,
+          secondsRemaining:
+            total && rate > 0 ? (total - e.loaded) / rate : null,
+          finishing: false,
        });
-  if (!res.ok) {
-    let message = `Error ${res.status}`;
-    try {
-      const b = await res.json();
-      if (b?.message) message = b.message;
-    } catch {
-      /* ignore */
-    }
-    throw new ApiError(res.status, message);
-  }
-  return res.status === 204 ? undefined : res.json().catch(() => undefined);
+      };
+      // Bytes are out the door; the server still has to write the file.
+      xhr.upload.onload = () => {
+        onProgress({
+          loaded: file.size,
+          total: file.size,
+          fraction: 1,
+          bytesPerSecond: rate,
+          secondsRemaining: 0,
+          finishing: true,
+        });
+      };
    }

-export function uploadIngest(name: string, file: File): Promise<unknown> {
-  return uploadFile(`/ops/ingest/${encodeURIComponent(name)}`, file, name);
+    xhr.onload = () => {
+      let parsed: unknown;
+      try {
+        parsed = xhr.responseText ? JSON.parse(xhr.responseText) : undefined;
+      } catch {
+        parsed = undefined;
+      }
+      if (xhr.status >= 200 && xhr.status < 300) {
+        resolve(parsed);
+        return;
+      }
+      const message =
+        (parsed as { message?: string } | undefined)?.message ??
+        `Error ${xhr.status}`;
+      reject(new ApiError(xhr.status, message));
+    };
+    xhr.onerror = () =>
+      reject(new ApiError(0, "Fallo de red durante la carga."));
+    xhr.onabort = () => reject(new ApiError(0, "Carga cancelada."));
+    xhr.ontimeout = () => reject(new ApiError(0, "Tiempo de carga agotado."));
+
+    xhr.send(body);
+  });
+}
+
+export function uploadIngest(
+  name: string,
+  file: File,
+  onProgress?: (p: UploadProgress) => void,
+): Promise<unknown> {
+  return uploadFile(
+    `/ops/ingest/${encodeURIComponent(name)}`,
+    file,
+    name,
+    onProgress,
+  );
 }

 export function deleteIngest(name: string): Promise<unknown> {
@@ -47,6 +47,13 @@ ENV NODE_ENV=production
 # That breaks the pre-migrate deploy backup AND the whole "Operaciones" admin
 # panel (backup, restore, sync, re-import all shell out to these binaries).
 #
+# mdbtools-utils, NOT mdbtools. Alpine splits the project: `mdbtools` is the
+# shared library only, and the command-line tools migration/extract.py shells out
+# to (`mdb-tables`, `mdb-export`) are in the -utils subpackage. Installing the
+# wrong one builds fine and fails at run time — the re-import in the "Operaciones"
+# panel dies with:
+#   RuntimeError: mdbtools not found on PATH (need mdb-tables and mdb-export)
+#
 # tesseract-ocr + tesseract-ocr-data-spa + poppler-utils drive the statement
 # OCR intake (RECEIPT_CAPTURE_SPEC §2): poppler's `pdftoppm` rasterises each
 # scanned page and tesseract reads it, with the Spanish traineddata for the
@@ -55,7 +62,7 @@ ENV NODE_ENV=production
 # dependency. If they are absent the API still boots — the statements module
 # reports itself unavailable and only that feature is disabled — but statement
 # ingest is the point of shipping them.
-RUN apk add --no-cache python3 mdbtools mysql-client mariadb-connector-c openssl \
+RUN apk add --no-cache python3 mdbtools-utils mysql-client mariadb-connector-c openssl \
      tesseract-ocr tesseract-ocr-data-spa poppler-utils \
 && apk add --no-cache --virtual .pybuild python3-dev build-base \
 && rm -rf /var/cache/apk/*
@@ -294,10 +294,19 @@ a warning, which is what local development wants.
 Two more things the panel's dumps now do, for the same reasons the pre-migrate
 backup does them (see `deploy/scripts/pre-migrate-backup.mjs`):

- **`--set-gtid-purged=OFF`.** galactus is the replication *source* with GTID
-  on, so without this every dump embeds `SET @@GLOBAL.GTID_PURGED` and cannot be
-  restored onto the server it came from — which is precisely what the restore
-  screen exists to do.
+- **`--set-gtid-purged=OFF`, but only when the dumper supports it.** galactus is
+  the replication *source* with GTID on, so on a MySQL client this flag is what
+  keeps every dump from embedding `SET @@GLOBAL.GTID_PURGED` and becoming
+  unrestorable onto the server it came from — which is precisely what the
+  restore screen exists to do. The panel, however, dumps from *inside the API
+  container*, where Alpine's `mysql-client` is MariaDB's: there `mysqldump` is a
+  shim over `mariadb-dump`, the flag does not exist, and passing it failed every
+  backup with `mysqldump: unknown variable 'set-gtid-purged=OFF'`. So the panel
+  probes `mysqldump --help` and passes the flag only if it is advertised,
+  invoking `mariadb-dump` directly otherwise (MariaDB writes no GTID state
+  unless asked with `--gtid`, so there is nothing to suppress). The pre-migrate
+  backup keeps the flag unconditionally — it runs in a real `mysql:8.4` image,
+  not in the API container.
 - **`set -o pipefail` and a `CREATE TABLE` count.** `mysqldump | gzip` reports
  gzip's exit status, and a `mysqldump` that dies on its first statement still
  produces a ~372-byte perfectly valid archive that passes `gzip -t`. Without
@@ -134,7 +134,8 @@ single-movement form.
 > **BUILT — 2026-08-01.** Implemented and verified end to end against real
 > scanned statements. `apps/api/src/statements/` holds the module: a swappable
 > `OcrProvider` seam with a self-hosted Tesseract implementation, per-provider
-> parsers for CFE / CESPT / Telnor, a scoped matcher, and a review queue that
+> parsers for CFE / CESPT / Telnor / gas / predial, a scoped matcher, and a
+> review queue that
 > posts through `BillingService.createBatch` with `source: "OCR"`. Web:
 > the "Captura automática (OCR)" tab of the Captura screen (upload + batch
 > list) and `/recibos/:id` (review queue with the page image beside the
@@ -207,6 +208,54 @@ single-movement form.
 > each bill (`9`, `405`, `406`); Tesseract read `405` as `205`. Handwriting is
 > a review hint at best and is deliberately not an input to matching.

+> **EXTENDED — gas and predial, 2026-08-01.** A second corpus (14 documents,
+> 29 pages: five municipal predial batches and ten gas invoices) added four
+> parsers — `GAS TIJUANA` plus one per municipality, because Tijuana, Rosarito
+> and Ensenada issue three completely different documents. End to end against
+> the dev database that is **21/29 auto-matched, 22/29 identified**, with the
+> provider read on 29/29 and an amount on 26/29.
+>
+> The eight review cases are all legitimate: five Tijuana pages whose municipal
+> account is not yet on file (see below), one clave not in the book, one page
+> too poorly scanned to read a clave at all, and one gas account shared by two
+> services. Excluding the structural Tijuana case, that is 21/24.
+>
+> **Five things this corpus proved:**
+>
+> 1. **Not every statement is a scan.** The gas company sends born-digital CFDI
+>    invoices whose text layer is exact. Rasterising and re-recognising those
+>    can only lose information — one sample turned `MEDIDOR: VM01014426` into
+>    `ar (LTR): 014420` — so `OcrProvider.textPages` reads the embedded layer
+>    first (`pdftotext -bbox-layout`, same poppler package as `pdftoppm`) and
+>    OCR stays the fallback for real scans. Page images are still rendered and
+>    stored either way, because the reviewer needs to see the paper.
+> 2. **The clave catastral is not two letters and six digits.** Positions four
+>    through eight are digits in all 932 stored claves, but the third is a
+>    letter in fifteen of them (`MMB01041`, `CGH52121`). Digitising the whole
+>    tail maps that `B` to an `8` and produces a key matching no property.
+> 3. **Tijuana predial prints no clave catastral at all.** Its only identifier
+>    is an 8-digit municipal account, carried in a 32-digit payment barcode
+>    (`account(8) + DDMMYY + amount(9) + folio(9)`) that the legacy database
+>    never held. It goes in `PROPERTY_TAX.meterNumber` — the same column gas
+>    uses, and for the same reason: `accountNumber` holds `DATMEX.predial`,
+>    which is not a per-property key and overwriting it would destroy the only
+>    link back to the original records. So Tijuana pages start cold and are
+>    taught by the first confirm, exactly like gas.
+> 4. **On Rosarito and Ensenada the clave is the primary key, not a fallback.**
+>    Those receipts print nothing else, so a unique clave hit there is a real
+>    match and auto-matches; on a utility bill that merely happens to print one
+>    it stays a review hint, as before.
+> 5. **A misread `$` is the dangerous failure, not a missing one.** An Ensenada
+>    receipt for `$2,203.00` OCR'd as `82,203.00` — the dollar sign read as an
+>    8, which would post a charge 37× too large and look entirely ordinary in
+>    the ledger. Every predial amount therefore requires a literal `$`, and a
+>    page that cannot produce one reports no amount and goes to review. Two of
+>    the 29 pages take that path, which is the correct outcome for both.
+>
+> Regression cover for all of the above lives in
+> `statement-parser.spec.ts` and `tesseract.provider.spec.ts`; every fixture in
+> them is a verbatim OCR excerpt from a real receipt.
+
 ### Motivation (from the meeting)

 Each utility company (CFE, water, phone, gas...) sends 300+ individual
@@ -330,8 +379,8 @@ Per the meeting notes' own field list:
 | Agua — Número de cuenta | `WATER` | `accountNumber` | `AGUA` | ✅ populated today |
 | Zona Fed — Número de Zona Federal | `FEDERAL_ZONE` | `accountNumber` | `ZFED` | ✅ populated today |
 | Tel — Número de teléfono | `TELEPHONE` *(new)* | `accountNumber` | `Property.phone1/2/3` (currently on `Property`, not `PropertyService`) | ⚠️ schema gap — see below |
-| Impuesto — Clave Catastral | `PROPERTY_TAX` | `accountNumber` | migrated from `PREDIAL`, **not** `CLAVE` | ⚠️ needs verification — see below |
-| Gas — Número de medidor | `GAS` | `meterNumber` | not populated — folded into free-text `notes` today | ⚠️ data gap — see below |
+| Impuesto — Clave Catastral | `PROPERTY_TAX` | `Property.cadastralKey`, plus `meterNumber` for Tijuana's municipal account | `CLAVE`; `PREDIAL` is left on `accountNumber` and never matched against | ✅ built — see the 2026-08-01 extension note |
+| Gas — Número de medidor | `GAS` | `meterNumber` | not populated — folded into free-text `notes` today | ✅ 160/334 recovered from `notes` |

 Confidence rule of thumb once a field is confirmed populated, tune after
 seeing real statements:
@@ -33,7 +33,7 @@ from pathlib import Path
 import boto3
 from botocore.config import Config

-from dbenv import connect, load_env
+from dbenv import connect, require, setting
 from extract import sanitize_column_name as san

 csv.field_size_limit(300_000_000)
@@ -102,12 +102,16 @@ def main():
    only = set(x.strip() for x in args.tables.split(",") if x.strip())
    sources = [s for s in SOURCES if not only or s["key"] in only]

-    env = load_env(args.env)
+    # Process environment first, deploy/.env.<env> second, with the same
+    # credential aliases the API uses — the "Operaciones" re-import runs this
+    # inside the API container, which has S3_ENDPOINT / MINIO_ROOT_* injected
+    # and no deploy/ directory at all.
    s3 = boto3.client(
-        "s3", endpoint_url=env["S3_ENDPOINT"],
-        aws_access_key_id=env["MINIO_ROOT_USER"], aws_secret_access_key=env["MINIO_ROOT_PASSWORD"],
+        "s3", endpoint_url=require(args.env, "S3_ENDPOINT"),
+        aws_access_key_id=require(args.env, "S3_ACCESS_KEY", "MINIO_ROOT_USER"),
+        aws_secret_access_key=require(args.env, "S3_SECRET_KEY", "MINIO_ROOT_PASSWORD"),
        config=Config(signature_version="s3v4"), region_name="us-east-1")
-    bucket = env["S3_BUCKET"]
+    bucket = setting(args.env, "S3_BUCKET") or "jorgecuadros-documents"

    conn = connect(args.env)
    cur = conn.cursor()
@@ -32,29 +32,53 @@ REPO = Path(__file__).resolve().parents[1]


 def load_env(env: str) -> dict:
+    """deploy/.env.<env> parsed to a dict, or {} when the file is absent.
+
+    Absent is normal, not an error: the API container runs these scripts with
+    DATABASE_URL / S3_* injected as real environment variables and ships no
+    deploy/ directory. Use `setting()` / `require()` rather than this — they
+    layer the process environment on top, which is what actually resolves."""
    f = REPO / "deploy" / f".env.{env}"
    if not f.exists():
-        raise SystemExit(
-            f"missing {f} — deploy the '{env}' DB stack and write its .env first "
-            f"(see dbenv.py header)."
-        )
+        return {}
    out = {}
    for line in f.read_text().splitlines():
        line = line.strip()
        if line and not line.startswith("#") and "=" in line:
            k, v = line.split("=", 1)
            out[k] = v
-    if "DATABASE_URL" not in out:
-        raise SystemExit(f"{f} has no DATABASE_URL")
    return out


+def setting(env: str, *keys: str):
+    """First non-empty value for `keys`, process environment first, then
+    deploy/.env.<env>. Several keys = fallback aliases (S3_ACCESS_KEY then
+    MINIO_ROOT_USER, as apps/api/src/storage/storage.service.ts does)."""
+    fromfile = load_env(env)
+    for k in keys:
+        v = os.environ.get(k) or fromfile.get(k)
+        if v:
+            return v
+    return None
+
+
+def require(env: str, *keys: str) -> str:
+    v = setting(env, *keys)
+    if not v:
+        raise SystemExit(
+            f"missing {' / '.join(keys)} — set it in the environment, or deploy the "
+            f"'{env}' stack and write {REPO / 'deploy' / f'.env.{env}'} "
+            f"(see dbenv.py header)."
+        )
+    return v
+
+
 def database_url(env: str) -> str:
    """Target DB URL. A DATABASE_URL in the process environment wins over
    deploy/.env.<env> — this is how the API container (which has its own
    DATABASE_URL and no deploy/.env files) drives a re-import against its own
    database."""
-    return os.environ.get("DATABASE_URL") or load_env(env)["DATABASE_URL"]
+    return require(env, "DATABASE_URL")


 def connect(env: str):
@@ -1,6 +1,6 @@
 {
  "name": "jorgecuadros-platform",
-  "version": "1.0.2",
+  "version": "1.0.5",
  "private": true,
  "workspaces": [
    "apps/*",
@@ -1,6 +1,6 @@
 {
  "name": "@jorgecuadros/database",
-  "version": "1.0.2",
+  "version": "1.0.5",
  "private": true,
  "main": "generated/client/index.js",
  "types": "generated/client/index.d.ts",
Author	SHA1	Message	Date
rmancinasandClaude Opus 5	d6501f1d74	feat(recibos): OCR capture for gas butano and municipal predial Build and Push Images / Build jorgecuadros-web (push) Successful in 1m50s Details Build and Push Images / Build jorgecuadros-api (push) Successful in 2m8s Details Adds four parsers to the statement intake — GAS TIJUANA plus one per municipality, because Tijuana, Rosarito and Ensenada issue three completely different predial documents — and a text-layer fast path for the born-digital invoices the gas company sends. Measured against a new corpus of 14 documents / 29 pages: provider read on 29/29, amount on 26/29, and 21/29 auto-matched against the dev database (22/29 identified). The eight review cases are all legitimate. Five things the corpus forced: - Not every statement is a scan. The gas invoices are born-digital CFDIs whose text layer is exact; rasterising them only loses information (one sample turned `MEDIDOR: VM01014426` into `ar (LTR): 014420`). The new `OcrProvider.textPages` reads the embedded layer via `pdftotext -bbox-layout` — same poppler package as `pdftoppm`, so no new dependency — and OCR stays the fallback for real scans. Poppler's own `<line>` grouping follows text flow rather than the page, so words are regrouped by vertical position; without that, a two-column header leaves every label separated from the value printed beside it. - The clave catastral is not two letters and six digits. Position three is a letter in 15 of the 932 stored claves, and digitising the whole tail mapped a real `MMB01041` to a nonexistent `MM801041`. - Tijuana predial prints no clave at all. Its only identifier is an 8-digit municipal account carried in a 32-digit payment barcode, which the legacy database never held, so it goes in `meterNumber` alongside gas — `accountNumber` holds `DATMEX.predial`, which is not a per-property key and must not be overwritten. Those pages start cold and are taught by the first confirm. - On Rosarito and Ensenada the clave is the primary key, not a fallback: those receipts print nothing else, so a unique hit auto-matches. On a utility bill that merely happens to print one it stays a review hint. - A misread `$` is the dangerous failure. An Ensenada receipt for $2,203.00 OCR'd as `82,203.00`, which would post a charge 37x too large and look ordinary in the ledger. Predial amounts now require a literal `$` and a page that cannot produce one goes to review. The scoped match field is now one exported function rather than three copies of `kind === "GAS" ? ... : ...`, since the lookup, the blank-service fill and the confirm write-back have to agree or a reference gets learned into a column nothing searches. First tests in this package: 23 specs over the parsers and the text-layer reader, every fixture a verbatim OCR excerpt from a real receipt. Adds the jest config they need and a build tsconfig so they stay out of dist. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-08-01 12:52:20 -07:00
rmancinas	216309190c	feat(recibos): live OCR progress bar on review page Build and Push Images / Build jorgecuadros-web (push) Successful in 1m51s Details Build and Push Images / Build jorgecuadros-api (push) Successful in 2m35s Details Backend already returns per-status counts via byStatus; render a real progress bar (X% / N de M / en cola) while PENDING_OCR pages remain, using the existing progress-track CSS. Falls back to indeterminate when no docs have been reported yet.	2026-08-01 02:43:19 -07:00
rmancinasandClaude Opus 5	e589bda28b	ci(build): skip the redundant master build when a release is cut Build and Push Images / Build jorgecuadros-api (push) Canceled after 1m10s Details Build and Push Images / Build jorgecuadros-web (push) Canceled after 1m8s Details release.yml pushes the release commit and its tag in a single `git push`, so Gitea created two build.yml runs for the same commit. Only the tag run matters: it emits the X.Y.Z and X.Y image tags, and since it is the same commit it publishes `latest` and `sha-<short>` as well. The master run was pure duplicate work that had to be waited out or cancelled by hand. Guard the build job with an `if` that skips a branch push whose head commit message starts with `chore(release):`. Ordinary pushes to master are unaffected, and tag pushes and manual dispatches always build. The skipped master run keeps the release commit's sha, which would have let release.yml's "Verify build.yml started" check go green on it alone even if the tag run were never created — the exact failure that check exists to catch. It now also requires the run's ref to be the tag, falling back to the sha match only when the API reports no ref. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-08-01 02:28:09 -07:00
gitea-actions	98f7aa8a2d	chore(release): v1.0.5 Build and Push Images / Build jorgecuadros-api (push) Canceled after 0s Details Build and Push Images / Build jorgecuadros-web (push) Canceled after 0s Details Cut by rmancinas via the "Cut release" workflow. Pushing the tag triggers build.yml; deploy separately with tag=1.0.5.	2026-08-01 09:22:56 +00:00
rmancinasandClaude Opus 5	898cf48c80	fix(migration): re-import died on the last step because blob_extract required deploy/.env.prod Every transform resolves its target through dbenv.database_url(), which lets a DATABASE_URL in the process environment win — that is how the API container drives a re-import against its own database with no deploy/ directory present. blob_extract.py was the one step that bypassed it and called load_env() directly for the MinIO credentials, so the "Operaciones" re-import loaded all the data and then exited 1 on: missing /repo/deploy/.env.prod — deploy the 'prod' DB stack and write its .env first Give the S3 settings the same resolution as the DB URL: load_env() now returns {} for an absent file, and setting()/require() layer the process environment on top of it. blob_extract reads S3_ENDPOINT / S3_BUCKET and accepts either S3_ACCESS_KEY/S3_SECRET_KEY or MINIO_ROOT_USER/MINIO_ROOT_PASSWORD, matching the fallback order in storage.service.ts and the vars the api service already sets in deploy/galactus/jorgecuadros-app.compose.yml. A genuinely missing setting still fails fast, now naming the variable. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-08-01 02:21:38 -07:00
gitea-actions	70fe425043	chore(release): v1.0.4 Build and Push Images / Build jorgecuadros-api (push) Successful in 2m2s Details Build and Push Images / Build jorgecuadros-web (push) Successful in 1m55s Details Cut by rmancinas via the "Cut release" workflow. Pushing the tag triggers build.yml; deploy separately with tag=1.0.4.	2026-08-01 09:09:15 +00:00
rmancinasandClaude Opus 5	567b033c46	fix(docker): re-import failed because the Access CLI tools were never installed Build and Push Images / Build jorgecuadros-web (push) Successful in 1m37s Details Build and Push Images / Build jorgecuadros-api (push) Successful in 3m22s Details The API image installed Alpine's `mdbtools` package, which ships only the shared library. The command-line tools that migration/extract.py actually shells out to -- `mdb-tables` and `mdb-export` -- are in the separate `mdbtools-utils` subpackage, so the build succeeded and the re-import in the "Operaciones" admin panel failed at run time with: RuntimeError: mdbtools not found on PATH (need mdb-tables and mdb-export) Install `mdbtools-utils` instead; it pulls the library in as a dependency. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-08-01 02:07:22 -07:00
rmancinasandClaude Opus 5	1934470d53	ci(release): dispatch the fallback build with a fully qualified ref Gitea's workflow dispatch API 404s on a bare `v1.0.3` and accepts only `refs/tags/v1.0.3`, so the fallback added in `fdbe9fd` would have failed the release instead of rescuing it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-08-01 02:00:23 -07:00
rmancinasandClaude Opus 5	fdbe9fdb88	ci(release): fail the release when the build never starts Gitea creates workflow runs from the post-receive hook. When that hook errors the refs still land, git prints `remote: error: Internal Server Error` and exits 0 — a post-receive failure does not fail a push. v1.0.3 was cut exactly that way: tag pushed, no build run created, no images published, and the release step green. It surfaced two steps later as a 404 when the deploy tried to pull 1.0.3. Capture the push output and warn on `remote: error`, then verify a build.yml run actually exists for the new commit, dispatching it against the tag if not. Fail the release if that does not take either, so a release that publishes nothing is red instead of green. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-08-01 01:58:10 -07:00
gitea-actions	e082113640	chore(release): v1.0.3 Cut by rmancinas via the "Cut release" workflow. Pushing the tag triggers build.yml; deploy separately with tag=1.0.3.	2026-08-01 08:45:42 +00:00
rmancinasandClaude Opus 5	860d483bad	fix(ops): backup failed on the MariaDB client shipped in the API image Build and Push Images / Build jorgecuadros-api (push) Successful in 2m7s Details Build and Push Images / Build jorgecuadros-web (push) Successful in 2m7s Details Every backup on galactus died with: mysqldump: unknown variable 'set-gtid-purged=OFF' respaldo incompleto eliminado Alpine's mysql-client is MariaDB's, so `mysqldump` inside the API container is a shim over `mariadb-dump`, which has no --set-gtid-purged. That took out BACKUP and, because they take a safety dump first, SYNC and REIMPORT too. Probe `mysqldump --help` and pass the flag only when it is advertised, calling `mariadb-dump` directly otherwise — MariaDB writes no GTID state unless asked with --gtid, so there is nothing to suppress. Testing whether mariadb-dump merely exists would be wrong: on a host carrying both clients it would shadow a perfectly good MySQL mysqldump. The probe uses a command substitution rather than `--help \| grep -q` because PIPEFAIL is in effect for these commands and grep closing the pipe early would report a supported flag as unsupported. pre-migrate-backup.mjs is unaffected — it dumps from a real mysql:8.4 image, not from the API container. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-08-01 01:43:25 -07:00
rmancinasandClaude Opus 5	783ec83464	feat(ops): show upload percent, speed and ETA for ingest files Build and Push Images / Build jorgecuadros-web (push) Successful in 1m46s Details Build and Push Images / Build jorgecuadros-api (push) Successful in 3m3s Details The ingest upload used fetch(), which cannot report request-body progress, so the only feedback was a static "Cargando…" label — no way to tell a stalled 2 GB upload from a working one. Switch uploadFile() to XMLHttpRequest and expose an optional onProgress callback reporting loaded/total bytes, a smoothed transfer rate and a remaining-time estimate. The Operaciones ingest table renders a progress bar row under the file being uploaded. Once the bytes are all sent the server still has to write the file, so that tail reads "Procesando en el servidor…" rather than parking at 100%. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-08-01 01:38:22 -07:00