@murumets-ee/imports 0.0.0-preview-20261014110621

package/dist/read-feed-DsLgRDNR.mjs

@@ -0,0 +1,2 @@
+import{z as e}from"zod";import{createReadStream as t}from"node:fs";import{parse as n}from"csv-parse";function r(e){let t=new Set;for(let n of e)if(n!==``){if(t.has(n))throw Error(`Duplicate header "${n}" after trimming — column names must be unique`);t.add(n)}}async function i(e,n={}){let{hasHeader:r=!0}=n,i=t(e),a=0,o=0;for await(let e of i){for(let t=0;t<e.length;t+=1)e[t]===10&&(a+=1);e.length>0&&(o=e[e.length-1]??0)}return o!==0&&o!==10&&(a+=1),r?Math.max(0,a-1):a}async function*a(e){let{filePath:i,delimiter:a=`	`,hasHeader:o=!0,columns:s,relaxColumnCount:c=!1,quote:l=!1}=e,u=s?Array.from(s):o?e=>{let t=e.map(e=>e.trim());return r(t),t}:!1,d=t(i),f=d.pipe(n({delimiter:a,columns:u,bom:!0,skip_empty_lines:!0,relax_column_count:c,quote:l})),p=0;try{for await(let e of f){p+=1;let t={};if(Array.isArray(e)){let n=e;for(let e=0;e<n.length;e+=1)t[String(e)]=n[e]??``}else for(let[n,r]of Object.entries(e))t[n]=r??``;yield{rowNumber:p,row:t}}}finally{d.destroy()}}const o=e.object({format:e.enum([`csv`,`tsv`,`xlsx`]).optional(),delimiter:e.string().min(1).max(4).optional(),hasHeader:e.boolean().optional(),headerRowIndex:e.number().int().min(0).optional(),sheet:e.string().min(1).max(128).optional(),quote:e.union([e.string().length(1),e.literal(!1)]).optional()}).strict();function s(e){let t=e&&typeof e==`object`&&!Array.isArray(e)?e:{},n=t.feed===void 0||t.feed===null?{}:t.feed;return o.safeParse(n)}function c(e){let t=e.delimiter??(e.format===`csv`?`,`:`	`);return{filePath:e.filePath,delimiter:t,...e.hasHeader!==void 0&&{hasHeader:e.hasHeader},...e.columns!==void 0&&{columns:e.columns},...e.relaxColumnCount!==void 0&&{relaxColumnCount:e.relaxColumnCount},...e.quote!==void 0&&{quote:e.quote}}}async function*l(e){if(e.format===`xlsx`){let{streamXlsx:t}=await import(`./xlsx-reader-D9Ntxm29.mjs`);yield*t({filePath:e.filePath,...e.hasHeader!==void 0&&{hasHeader:e.hasHeader},...e.headerRowIndex!==void 0&&{headerRowIndex:e.headerRowIndex},...e.sheet!==void 0&&{sheet:e.sheet}});return}yield*a(c(e))}export{i as a,r as i,l as n,a as o,s as r,o as t};
+//# sourceMappingURL=read-feed-DsLgRDNR.mjs.map

package/dist/read-feed-DsLgRDNR.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"read-feed-DsLgRDNR.mjs","names":[],"sources":["../src/streaming.ts","../src/read-feed.ts"],"sourcesContent":["/**\n * Tab-delimited / CSV streaming reader. Generic over delimiter so the\n * same path handles `.txt` (tab), `.csv` (comma), and the rare `;`\n * European export dialect.\n *\n * Built on `csv-parse` per giga-test precedent — node-stream-based, low\n * memory, handles UTF-8 BOM (the carmaker feeds are Windows-exported\n * and ship with a BOM that breaks naive split-on-tab parsers).\n *\n * The reader yields `{ rowNumber, row }` pairs where:\n * - `rowNumber` is 1-based and counts the header as row 0.\n * - `row` is `Record<string, string>` keyed by header name. Empty cells\n *   are the empty string, NOT `undefined` — feed transforms test with\n *   `value === ''` consistently.\n *\n * Why this lives in a streaming reader and not inside the transform\n * itself: the transform sees one already-parsed row at a time, never\n * the file. That keeps transform implementations free of I/O concerns\n * and makes them trivially unit-testable with a fixture row map.\n */\n\nimport { createReadStream } from 'node:fs'\nimport { parse } from 'csv-parse'\n\n/**\n * Throw if two (non-empty) header names collide. Headers are the keys the mapping\n * engine matches a `source` against; after trimming, two columns differing only in\n * whitespace would collapse to one key and silently overwrite each other's data.\n * Fail fast instead — the operator fixes the feed or the mapping. Shared by the\n * csv/tsv reader (below) and the xlsx reader. Blank columns are ignored (unmapped).\n */\nexport function assertUniqueHeaders(header: readonly string[]): void {\n  const seen = new Set<string>()\n  for (const h of header) {\n    if (h === '') continue\n    if (seen.has(h)) {\n      throw new Error(`Duplicate header \"${h}\" after trimming — column names must be unique`)\n    }\n    seen.add(h)\n  }\n}\n\n/**\n * Pre-pass: count data rows in a delimited file (excluding the header\n * row when `hasHeader` is true; default `true` to match `streamFeed`).\n *\n * Implementation: stream the file as raw bytes and tally `\\n`. NOT a\n * full parse — escaped newlines inside quoted fields would over-count.\n * That was unconditionally fine while `quote: false` was the only\n * behaviour available; a mapping can now declare RFC 4180 (`quote`),\n * and an embedded newline in a quoted field then makes this an UPPER\n * BOUND rather than a count. It is used only as the progress\n * denominator (`totalRows`), so the visible consequence is a bar that\n * stops short of 100%, never a wrong import. The parsed row count from\n * `streamFeed` is authoritative and is what the run records.\n *\n * Why bytes-only and not csv-parse: a 16 MB feed counts in ~50 ms with\n * `\\n` tally vs ~500 ms with csv-parse parsing. For 500 MB the gap\n * matters — pre-pass needs to be a small fraction of the actual import\n * or the operator perceives the run as \"stuck on pre-pass\" with no UI.\n *\n * Returns the number of DATA rows (header excluded when present).\n * Empty trailing lines (no `\\n` after the final row, or a single\n * trailing `\\n`) are handled identically to how csv-parse treats them\n * with `skip_empty_lines: true`.\n */\nexport async function countDataRows(\n  filePath: string,\n  options: { hasHeader?: boolean } = {},\n): Promise<number> {\n  const { hasHeader = true } = options\n  const stream = createReadStream(filePath)\n  let count = 0\n  let lastByte = 0\n  for await (const chunk of stream as AsyncIterable<Buffer>) {\n    for (let i = 0; i < chunk.length; i += 1) {\n      if (chunk[i] === 0x0a /* \\n */) count += 1\n    }\n    if (chunk.length > 0) lastByte = chunk[chunk.length - 1] ?? 0\n  }\n  // If the file doesn't end with a newline, the final row didn't get\n  // counted by the `\\n` tally — add one. Skipping when the file is\n  // entirely empty (count=0 and lastByte=0).\n  if (lastByte !== 0 && lastByte !== 0x0a) count += 1\n  return hasHeader ? Math.max(0, count - 1) : count\n}\n\nexport interface StreamFeedOptions {\n  /** Path to the file on disk. The PoC uploads land on local disk; S3-keyed reads come later. */\n  filePath: string\n  /**\n   * Single-character field delimiter. Default `\\t` (the carmaker feed\n   * format). Pass `,` for CSV, `;` for some European dialects.\n   */\n  delimiter?: string\n  /**\n   * `true` (default): the first row is the header and column names come\n   * from it. `false`: rows are emitted as positional `{ \"0\": ..., \"1\": ... }`\n   * and the transform reads by index — useful for headerless feeds that\n   * commit to a documented column order.\n   */\n  hasHeader?: boolean\n  /**\n   * Optional explicit column-name list. When provided, takes precedence\n   * over `hasHeader` (header row, if present, is skipped but its values\n   * are ignored). Useful when the upstream header is unstable but the\n   * positional shape isn't.\n   */\n  columns?: ReadonlyArray<string>\n  /**\n   * Forward to `csv-parse` `relax_column_count`. Default `false` —\n   * a row whose column count doesn't match the header surfaces as a\n   * parser error so the transform isn't silently fed truncated data.\n   */\n  relaxColumnCount?: boolean\n  /**\n   * Forward to `csv-parse` `quote`. The character (or `false`) that\n   * encloses fields containing the delimiter. Default `false` — most\n   * supplier feeds in this package's target use case are tab-delimited\n   * exports that contain literal `\"` characters in description fields\n   * (e.g. `LIVRE \"INVERSION D'IMAGE\"` from the Mercedes carmaker feed)\n   * with NO quote-as-field-wrapper convention. csv-parse's library\n   * default `'\"'` then misreads the literal `\"` as an opening quote\n   * and rejects the row with `Invalid Opening Quote`.\n   *\n   * Pass `'\"'` (or any character) to opt back into RFC-4180-style\n   * CSV parsing for genuine quoted-CSV inputs. Pass `'\\''` for\n   * single-quoted dialects.\n   */\n  quote?: string | false\n}\n\nexport interface StreamFeedRow {\n  /** 1-based row number. Header (when present) is row 0; first data row is row 1. */\n  rowNumber: number\n  /**\n   * Cell values keyed by column name (or string-position when\n   * `hasHeader: false` AND no `columns`).\n   *\n   * **Cell-value invariants:**\n   * - Empty cells (`A\\t\\tC`) → `''` (empty string).\n   * - Missing TRAILING cells in `relaxColumnCount: true` mode → the\n   *   key is **absent** from the object, not present-with-`''`. csv-parse\n   *   does not emit keys for short rows. Transforms reading those\n   *   columns get `undefined` from `row['col']` and must handle it\n   *   (`row['col'] ?? ''` is the canonical idiom).\n   * - With the default `relaxColumnCount: false`, short rows reject at\n   *   the parser, so this case never reaches the transform.\n   */\n  row: Record<string, string>\n}\n\n/**\n * Async-iterable over the parsed rows of a delimited file. Use with\n * `for await (const { rowNumber, row } of streamFeed({ filePath, ... }))`.\n *\n * The iterator owns its file descriptor — the `for await` loop closes\n * the underlying stream when it returns or breaks. Aborting mid-stream\n * (`break`, `throw`, signal) is safe; csv-parse propagates the close.\n */\nexport async function* streamFeed(options: StreamFeedOptions): AsyncIterable<StreamFeedRow> {\n  const {\n    filePath,\n    delimiter = '\\t',\n    hasHeader = true,\n    columns,\n    relaxColumnCount = false,\n    quote = false,\n  } = options\n\n  // When the caller provides explicit `columns`, prefer them. When the file\n  // has a header but no explicit `columns`, derive the column names from the\n  // first row via a function so we can TRIM them — real supplier feeds ship\n  // leading/trailing-space headers (R03 Samples A/C: `\" JAEHIND KM-TA\"`), and\n  // an untrimmed key wouldn't match a mapping's `source`. The xlsx reader trims\n  // headers the same way. When neither is true, rows get string-position keys.\n  //\n  // csv-parse's `columns` accepts `ColumnOption[] | boolean | ((record) =>\n  // ColumnOption[])`; `string[]` satisfies `ColumnOption[]`.\n  const columnConfig: string[] | boolean | ((header: string[]) => string[]) = columns\n    ? Array.from(columns)\n    : hasHeader\n      ? (header: string[]) => {\n          const trimmed = header.map((h) => h.trim())\n          assertUniqueHeaders(trimmed)\n          return trimmed\n        }\n      : false\n\n  const stream = createReadStream(filePath)\n  const parser = stream.pipe(\n    parse({\n      delimiter,\n      columns: columnConfig,\n      bom: true,\n      skip_empty_lines: true,\n      // csv-parse defaults to strict column count; opt-in relaxation only.\n      relax_column_count: relaxColumnCount,\n      // Default `quote: false` — see StreamFeedOptions.quote JSDoc.\n      // Carmaker / parts-feed exports contain literal `\"` characters\n      // in description fields and would otherwise reject as\n      // `Invalid Opening Quote`. Genuine RFC-4180 CSV consumers opt\n      // back in by passing `quote: '\"'`.\n      quote,\n    }),\n  )\n\n  let rowNumber = 0\n  try {\n    for await (const rawRow of parser as AsyncIterable<\n      ReadonlyArray<string> | Record<string, string | undefined>\n    >) {\n      rowNumber += 1\n      // csv-parse emits records with string keys when `columns` is\n      // truthy, otherwise an Array. Normalise both shapes to\n      // `Record<string, string>` so transforms can rely on `value === ''`\n      // for missing cells (csv-parse leaves trailing missing cells as\n      // `undefined` when `relax_column_count: true`; this collapses\n      // them to `''` to keep the contract uniform).\n      const row: Record<string, string> = {}\n      if (Array.isArray(rawRow)) {\n        const arr = rawRow as ReadonlyArray<string | undefined>\n        for (let i = 0; i < arr.length; i += 1) {\n          row[String(i)] = arr[i] ?? ''\n        }\n      } else {\n        for (const [k, v] of Object.entries(rawRow)) {\n          row[k] = v ?? ''\n        }\n      }\n      yield { rowNumber, row }\n    }\n  } finally {\n    // Safety: ensure the underlying file descriptor closes even if the\n    // consumer breaks mid-iteration. Node closes streams on\n    // garbage-collection but this makes it deterministic under tests.\n    stream.destroy()\n  }\n}\n","/**\n * Feed reader dispatcher (commerce-demo phase 02, D005). Selects the reader by\n * `format` and yields the uniform `AsyncIterable<{ rowNumber, row }>` the runner\n * consumes — csv/tsv via the existing `streamFeed` (csv-parse), xlsx via the\n * exceljs-backed `streamXlsx`.\n *\n * **Back-compat**: with no `format` (the carmaker worker passes only `filePath`),\n * this defaults to tab-delimited — the original `streamFeed` behaviour — so the\n * existing import path is unchanged.\n *\n * **Bundle safety**: the xlsx reader (and its heavy, Node-only exceljs dep) is\n * `await import()`-ed ONLY for `format: 'xlsx'`, so the csv/tsv path — and anything\n * that statically imports the runner — never pulls exceljs into its module graph.\n */\n\nimport { z } from 'zod'\nimport { type StreamFeedOptions, type StreamFeedRow, streamFeed } from './streaming.js'\n\n/**\n * The reader-config subset safe to persist in `import_run.params.feed` and rehydrate\n * in the worker (commerce-demo PR3). It is exactly the {@link ReadFeedOptions} fields\n * that describe HOW to parse a feed — `filePath` is supplied by the worker after it\n * materialises the upload, never carried in params. Validated at BOTH ends: the\n * commerce run route builds it from a saved `import_mapping`, the worker re-parses it\n * before handing it to {@link readFeed} (defense-in-depth against a tampered row).\n */\nexport const feedReaderConfigSchema = z\n  .object({\n    format: z.enum(['csv', 'tsv', 'xlsx']).optional(),\n    delimiter: z.string().min(1).max(4).optional(),\n    hasHeader: z.boolean().optional(),\n    headerRowIndex: z.number().int().min(0).optional(),\n    sheet: z.string().min(1).max(128).optional(),\n    /**\n     * csv-parse's `quote`. Absent (or `false`) = quotes are ordinary data, which is this\n     * package's long-standing default and what the tab-delimited supplier dumps require. A\n     * single character opts into RFC 4180 for feeds that genuinely quote their fields.\n     *\n     * Persisted deliberately: the two shapes are mutually exclusive on real data (R027 —\n     * each of the three ground-truth feeds hard-fails under the other's setting), so this\n     * is a per-file fact a mapping has to carry. Deriving it from `format` would be the\n     * same class of mistake as a global decimal separator.\n     */\n    quote: z.union([z.string().length(1), z.literal(false)]).optional(),\n  })\n  .strict()\nexport type FeedReaderConfig = z.infer<typeof feedReaderConfigSchema>\n\n/**\n * Rehydrate the reader config a run persisted in `import_run.params.feed`.\n *\n * ONE implementation, because two would drift and drift here is silent: the\n * worker uses it to decide how to READ a feed, and the #842 audit uses it to\n * re-read the same file the same way. An audit that rehydrates differently from\n * the run it is auditing invents corruption, or misses it.\n *\n * ABSENT means \"the carmaker default\", i.e. tab-delimited, which is what an\n * empty config parses to — so a non-object `params`, or `params` with no\n * `feed`, is a real answer rather than a gap. A PRESENT `feed` is validated\n * whatever its shape: a scalar or an array there is a malformed persisted blob,\n * and reading it as \"no config\" would silently import an xlsx run as tsv.\n */\nexport function readPersistedFeedConfig(\n  params: unknown,\n): ReturnType<typeof feedReaderConfigSchema.safeParse> {\n  const record =\n    params && typeof params === 'object' && !Array.isArray(params)\n      ? (params as Record<string, unknown>)\n      : {}\n  // `null` is how \"absent\" round-trips through JSONB, so it defaults with\n  // `undefined`. Everything else present goes to the schema and is rejected on\n  // its merits rather than being flattened into the default.\n  const feed = record.feed === undefined || record.feed === null ? {} : record.feed\n  return feedReaderConfigSchema.safeParse(feed)\n}\n\nexport interface ReadFeedOptions {\n  /** Path to the file on local disk. */\n  filePath: string\n  /** Feed format. Omitted → tab-delimited (tsv), matching the legacy carmaker reader. */\n  format?: 'csv' | 'tsv' | 'xlsx'\n  /** Explicit delimiter override (csv/tsv). Defaults from `format` (`csv` → `,`, else `\\t`). */\n  delimiter?: string\n  /** First row is the header. Default `true`. */\n  hasHeader?: boolean\n  /** 0-based header row index (xlsx). */\n  headerRowIndex?: number\n  /** xlsx worksheet as a 1-based index string. Omit → first sheet. */\n  sheet?: string\n  /** Explicit column-name list (csv/tsv) — see `streamFeed`. */\n  columns?: ReadonlyArray<string>\n  /** Forward to csv-parse `relax_column_count` (csv/tsv). */\n  relaxColumnCount?: boolean\n  /** Forward to csv-parse `quote` (csv/tsv). */\n  quote?: string | false\n}\n\nfunction toStreamFeedOptions(o: ReadFeedOptions): StreamFeedOptions {\n  // No format → tab (the carmaker feed default). `csv` → comma; everything else → tab.\n  const delimiter = o.delimiter ?? (o.format === 'csv' ? ',' : '\\t')\n  return {\n    filePath: o.filePath,\n    delimiter,\n    ...(o.hasHeader !== undefined && { hasHeader: o.hasHeader }),\n    ...(o.columns !== undefined && { columns: o.columns }),\n    ...(o.relaxColumnCount !== undefined && { relaxColumnCount: o.relaxColumnCount }),\n    ...(o.quote !== undefined && { quote: o.quote }),\n  }\n}\n\n/** Stream a feed file's data rows, dispatching on `format`. */\nexport async function* readFeed(options: ReadFeedOptions): AsyncIterable<StreamFeedRow> {\n  if (options.format === 'xlsx') {\n    const { streamXlsx } = await import('./xlsx-reader.js')\n    yield* streamXlsx({\n      filePath: options.filePath,\n      ...(options.hasHeader !== undefined && { hasHeader: options.hasHeader }),\n      ...(options.headerRowIndex !== undefined && { headerRowIndex: options.headerRowIndex }),\n      ...(options.sheet !== undefined && { sheet: options.sheet }),\n    })\n    return\n  }\n  yield* streamFeed(toStreamFeedOptions(options))\n}\n"],"mappings":"qGA+BA,SAAgB,EAAoB,EAAiC,CACnE,IAAM,EAAO,IAAI,IACjB,IAAK,IAAM,KAAK,EACV,OAAM,GACV,IAAI,EAAK,IAAI,CAAC,EACZ,MAAU,MAAM,qBAAqB,EAAE,+CAA+C,EAExF,EAAK,IAAI,CAAC,CAF8E,CAI5F,CA0BA,eAAsB,EACpB,EACA,EAAmC,CAAC,EACnB,CACjB,GAAM,CAAE,YAAY,IAAS,EACvB,EAAS,EAAiB,CAAQ,EACpC,EAAQ,EACR,EAAW,EACf,UAAW,IAAM,KAAS,EAAiC,CACzD,IAAK,IAAI,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EACjC,EAAM,KAAO,KAAe,GAAS,GAEvC,EAAM,OAAS,IAAG,EAAW,EAAM,EAAM,OAAS,IAAM,EAC9D,CAKA,OADI,IAAa,GAAK,IAAa,KAAM,GAAS,GAC3C,EAAY,KAAK,IAAI,EAAG,EAAQ,CAAC,EAAI,CAC9C,CA2EA,eAAuB,EAAW,EAA0D,CAC1F,GAAM,CACJ,WACA,YAAY,IACZ,YAAY,GACZ,UACA,mBAAmB,GACnB,QAAQ,IACN,EAWE,EAAsE,EACxE,MAAM,KAAK,CAAO,EAClB,EACG,GAAqB,CACpB,IAAM,EAAU,EAAO,IAAK,GAAM,EAAE,KAAK,CAAC,EAE1C,OADA,EAAoB,CAAO,EACpB,CACT,EACA,GAEA,EAAS,EAAiB,CAAQ,EAClC,EAAS,EAAO,KACpB,EAAM,CACJ,YACA,QAAS,EACT,IAAK,GACL,iBAAkB,GAElB,mBAAoB,EAMpB,OACF,CAAC,CACH,EAEI,EAAY,EAChB,GAAI,CACF,UAAW,IAAM,KAAU,EAExB,CACD,GAAa,EAOb,IAAM,EAA8B,CAAC,EACrC,GAAI,MAAM,QAAQ,CAAM,EAAG,CACzB,IAAM,EAAM,EACZ,IAAK,IAAI,EAAI,EAAG,EAAI,EAAI,OAAQ,GAAK,EACnC,EAAI,OAAO,CAAC,GAAK,EAAI,IAAM,EAE/B,MACE,IAAK,GAAM,CAAC,EAAG,KAAM,OAAO,QAAQ,CAAM,EACxC,EAAI,GAAK,GAAK,GAGlB,KAAM,CAAE,YAAW,KAAI,CACzB,CACF,QAAU,CAIR,EAAO,QAAQ,CACjB,CACF,CCpNA,MAAa,EAAyB,EACnC,OAAO,CACN,OAAQ,EAAE,KAAK,CAAC,MAAO,MAAO,MAAM,CAAC,CAAC,CAAC,SAAS,EAChD,UAAW,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,EAC7C,UAAW,EAAE,QAAQ,CAAC,CAAC,SAAS,EAChC,eAAgB,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,EACjD,MAAO,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,SAAS,EAW3C,MAAO,EAAE,MAAM,CAAC,EAAE,OAAO,CAAC,CAAC,OAAO,CAAC,EAAG,EAAE,QAAQ,EAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CACpE,CAAC,CAAC,CACD,OAAO,EAiBV,SAAgB,EACd,EACqD,CACrD,IAAM,EACJ,GAAU,OAAO,GAAW,UAAY,CAAC,MAAM,QAAQ,CAAM,EACxD,EACD,CAAC,EAID,EAAO,EAAO,OAAS,IAAA,IAAa,EAAO,OAAS,KAAO,CAAC,EAAI,EAAO,KAC7E,OAAO,EAAuB,UAAU,CAAI,CAC9C,CAuBA,SAAS,EAAoB,EAAuC,CAElE,IAAM,EAAY,EAAE,YAAc,EAAE,SAAW,MAAQ,IAAM,KAC7D,MAAO,CACL,SAAU,EAAE,SACZ,YACA,GAAI,EAAE,YAAc,IAAA,IAAa,CAAE,UAAW,EAAE,SAAU,EAC1D,GAAI,EAAE,UAAY,IAAA,IAAa,CAAE,QAAS,EAAE,OAAQ,EACpD,GAAI,EAAE,mBAAqB,IAAA,IAAa,CAAE,iBAAkB,EAAE,gBAAiB,EAC/E,GAAI,EAAE,QAAU,IAAA,IAAa,CAAE,MAAO,EAAE,KAAM,CAChD,CACF,CAGA,eAAuB,EAAS,EAAwD,CACtF,GAAI,EAAQ,SAAW,OAAQ,CAC7B,GAAM,CAAE,cAAe,MAAM,OAAO,8BACpC,MAAO,EAAW,CAChB,SAAU,EAAQ,SAClB,GAAI,EAAQ,YAAc,IAAA,IAAa,CAAE,UAAW,EAAQ,SAAU,EACtE,GAAI,EAAQ,iBAAmB,IAAA,IAAa,CAAE,eAAgB,EAAQ,cAAe,EACrF,GAAI,EAAQ,QAAU,IAAA,IAAa,CAAE,MAAO,EAAQ,KAAM,CAC5D,CAAC,EACD,MACF,CACA,MAAO,EAAW,EAAoB,CAAO,CAAC,CAChD"}

package/dist/realtime/index.d.mts

@@ -0,0 +1,123 @@
+//#region src/realtime/publish.d.ts
+/**
+ * Typed publisher wrappers around `@murumets-ee/core/realtime.publishEvent`
+ * for the `imports.run.*` topic surface.
+ *
+ * Like queue events, every imports event is admin-only — a future
+ * refactor that opens these to non-admin operators (e.g. an
+ * `imports:view` permission scope) would update {@link ADMIN_SCOPE}
+ * here, not the call sites.
+ *
+ * Publish failures are caught and logged, never thrown — these
+ * wrappers are called from worker state-transition paths and the
+ * upload route, and a missing realtime delivery must NEVER roll back
+ * the database write that triggered it.
+ *
+ * Why dynamic import: `@murumets-ee/core/realtime` carries
+ * `import 'server-only'`, but this module is reachable from the
+ * imports package's entrypoints (e.g. `worker.ts`), which are loaded
+ * by jiti/tsx in `lumi doctor` / `lumi migrate` / `pnpm worker` —
+ * none of which set the `react-server` export condition that turns
+ * `server-only` into the empty stub. A static value-import would
+ * therefore throw at boot. Same dynamic-import pattern as
+ * `packages/queue/src/realtime/publish.ts` and
+ * `packages/ticketing/src/realtime/publish.ts`.
+ *
+ * The `cachedPublish` + `publishSafe` + `_warmRealtimePublisher`
+ * scaffolding here is the third copy of the same shape (queue,
+ * ticketing, imports). Tracked for extraction to a shared
+ * `@murumets-ee/core/realtime/safe-publisher` helper in #201 —
+ * reach for that when it lands instead of copy-pasting a fourth
+ * time.
+ */
+/**
+ * @internal — tests pre-resolve the dynamic import so the first
+ * publish is synchronous, matching the post-warmup behaviour in
+ * production.
+ */
+declare function _warmRealtimePublisher(): Promise<void>;
+/**
+ * Fire after the upload route's outbox txn commits, NOT inside it.
+ * The realtime fanout is best-effort and a failure must not roll
+ * back the row insert.
+ */
+declare function publishImportRunCreated(importRunId: string, fields: {
+  brandSlug?: string;
+  supplierDisplayName?: string;
+  filename: string;
+}): void;
+/**
+ * Fire after the worker writes `status: 'running'` on the row.
+ * Subscribers refetch to move the row from "queued" to "running" in
+ * the Active panel and update the History snapshot.
+ */
+declare function publishImportRunStarted(importRunId: string): void;
+/**
+ * Fire after the worker writes `status: 'succeeded'` + final totals.
+ * Subscribers refetch to move the row from Active → Recent.
+ */
+declare function publishImportRunSucceeded(importRunId: string, totals: {
+  rowsRead: number;
+  rowsSucceeded: number;
+  rowsFailed: number;
+}): void;
+/**
+ * Fire after the worker writes `status: 'failed'` with an error in
+ * `errorSummary`. Subscribers refetch to surface the failure in
+ * Recent.
+ */
+declare function publishImportRunFailed(importRunId: string, reason: string): void;
+//#endregion
+//#region src/realtime/topics.d.ts
+/**
+ * Imports realtime topic registry.
+ *
+ * Topics use the `imports.run.<event>` namespace. Subscribers listen
+ * via `imports.run.**` for everything or narrow per event.
+ *
+ * Payload shapes are intentionally untyped here — callers pass
+ * `Record<string, unknown>` and consumers (the imports admin page)
+ * read fields by name. The point of this layer is "an import_run
+ * row's status changed, refetch" rather than diffing event payloads
+ * against the page's snapshot.
+ *
+ * Why a separate namespace from `queue.job.*`: queue events are
+ * job-shaped (id, type, attempts, …) and admin-only. Imports events
+ * are row-shaped (importRunId, status, brand, supplier, …) and
+ * track the entity lifecycle independent of the underlying queue
+ * mechanics — a future refactor could swap the queue for a different
+ * driver without changing the imports event surface.
+ */
+declare const IMPORTS_TOPICS: {
+  /**
+   * The upload route inserted an `import_run` row + enqueued its
+   * driving queue job. Fired AFTER the outbox txn commits, so a
+   * subscriber's refetch sees the row.
+   */
+  readonly runCreated: "imports.run.created";
+  /**
+   * The worker claimed the row and transitioned its status to
+   * `running` — the streaming reader is open, the first batch is
+   * about to bulk-write. Distinct from `queue.job.claimed` because
+   * imports' state machine flips on the row, not the job.
+   */
+  readonly runStarted: "imports.run.started";
+  /**
+   * The worker finished cleanly — `import_run.status = 'succeeded'`,
+   * `errorSummary` populated, totals locked in. Subscribers refetch
+   * the history slice to bring the row out of the Active panel and
+   * into Recent.
+   */
+  readonly runSucceeded: "imports.run.succeeded";
+  /**
+   * The worker (or `markFailed`) wrote `import_run.status = 'failed'`
+   * with an error in `errorSummary`. Same refetch trigger; the page
+   * differentiates terminal-success vs terminal-failure by reading
+   * the row.
+   */
+  readonly runFailed: "imports.run.failed";
+};
+type ImportsTopic = (typeof IMPORTS_TOPICS)[keyof typeof IMPORTS_TOPICS];
+//#endregion
+export { IMPORTS_TOPICS, type ImportsTopic, _warmRealtimePublisher, publishImportRunCreated, publishImportRunFailed, publishImportRunStarted, publishImportRunSucceeded };
+//# sourceMappingURL=index.d.mts.map

package/dist/realtime/index.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.mts","names":[],"sources":["../../src/realtime/publish.ts","../../src/realtime/topics.ts"],"mappings":";;AAqFA;;;;AAAuD;AAWvD;;;;;;;;;;AAEgF;AAiBhF;;;;AAA2D;AAQ3D;;;;;;;;;;AAEyE;AAezE;;;iBAvDsB,sBAAA,IAA0B,OAAO;AAuDmB;;;;ACxH1E;ADwH0E,iBA5C1D,uBAAA,CACd,WAAA,UACA,MAAA;EAAU,SAAA;EAAoB,mBAAA;EAA8B,QAAA;AAAA;;;;;AChD9D;iBDiEgB,uBAAA,CAAwB,WAAmB;;;ACjEmB;;iBDyE9D,yBAAA,CACd,WAAA,UACA,MAAA;EAAU,QAAA;EAAkB,aAAA;EAAuB,UAAA;AAAA;;;;;;iBAerC,sBAAA,CAAuB,WAAA,UAAqB,MAAc;;;;AAvD1E;;;;AAAuD;AAWvD;;;;;;;;;;AAEgF;AAiBhF;;cC/Fa,cAAA;ED+F8C;AAAA;AAQ3D;;;EAR2D;EASzD;;;;;;EAAA;EAgBc;;;;AAA0D;;EAA1D;;ACxHhB;;;;;;;KA8BY,YAAA,WAAuB,cAAA,eAA6B,cAAc"}

package/dist/realtime/index.mjs

@@ -0,0 +1 @@
+import{a as e,i as t,n,o as r,r as i,t as a}from"../publish-DCXIPwTE.mjs";export{r as IMPORTS_TOPICS,a as _warmRealtimePublisher,n as publishImportRunCreated,i as publishImportRunFailed,t as publishImportRunStarted,e as publishImportRunSucceeded};

package/dist/rolldown-runtime-DK3Fl9T5.mjs

@@ -0,0 +1 @@
+var e=Object.defineProperty,t=(t,n)=>{let r={};for(var i in t)e(r,i,{get:t[i],enumerable:!0});return n||e(r,Symbol.toStringTag,{value:`Module`}),r};export{t};

package/dist/runner-BnFvID1z.mjs

@@ -0,0 +1,2 @@
+import{n as e}from"./read-feed-DsLgRDNR.mjs";import{bulkPartialUpdate as t,bulkUpsert as n}from"@murumets-ee/search-elasticsearch";const r=50,i=5;var a=class{patterns=new Map;maxPatterns;maxSamplesPerPattern;droppedSignatures=0;constructor(e={}){this.maxPatterns=e.maxPatterns??50,this.maxSamplesPerPattern=e.maxSamplesPerPattern??5}addError(e,t,n,r,i){let a=`${t}:${r??`GENERAL`}:${n}`,o=this.patterns.get(a);if(!o){if(this.patterns.size>=this.maxPatterns){this.droppedSignatures+=1;return}o={errorType:t,field:r??null,message:n,count:0,firstOccurrence:e,lastOccurrence:e,samples:[]},this.patterns.set(a,o)}o.count+=1,o.lastOccurrence=e,o.samples.length<this.maxSamplesPerPattern&&o.samples.push({rowNumber:e,rowData:i})}getTotalErrorCount(){let e=0;for(let t of this.patterns.values())e+=t.count;return e}getDistinctPatternCount(){return this.patterns.size}getDroppedSignatureCount(){return this.droppedSignatures}getTopPatterns(){let e=Array.from(this.patterns.values()).sort((e,t)=>t.count-e.count),t=e.reduce((e,t)=>e+t.count,0);return e.map(e=>({errorType:e.errorType,field:e.field,message:e.message,count:e.count,firstOccurrence:e.firstOccurrence,lastOccurrence:e.lastOccurrence,samples:e.samples.slice(),percentage:t>0?e.count/t*100:0}))}snapshot(){return{totalErrors:this.getTotalErrorCount(),distinctPatterns:this.getDistinctPatternCount(),droppedSignatures:this.droppedSignatures,patterns:this.getTopPatterns()}}};function o(e,t){let n=e.totalErrors+t.totalErrors,r=e=>({...e,percentage:n>0?e.count/n*100:0});return{totalErrors:n,distinctPatterns:e.distinctPatterns+t.distinctPatterns,droppedSignatures:e.droppedSignatures+t.droppedSignatures,patterns:[...e.patterns.map(r),...t.patterns.map(r)]}}const s={count:0,nullRate:0,min:null,median:null,max:null},c=5e4;var l=class{count=0;nullCount=0;nonNullSeen=0;samples=[];record(e){if(this.count+=1,e===void 0){this.nullCount+=1;return}if(this.nonNullSeen+=1,this.samples.length<5e4){this.samples.push(e);return}let t=Math.floor(Math.random()*this.nonNullSeen);t<5e4&&(this.samples[t]=e)}snapshot(){if(this.count===0)return{...s};let e=this.nullCount/this.count;if(this.samples.length===0)return{count:this.count,nullRate:e,min:null,median:null,max:null};let t=[...this.samples].sort((e,t)=>e-t),n=t[0]??null,r=t[t.length-1]??null,i=Math.floor(t.length/2),a;if(t.length%2==0){let e=t[i-1],n=t[i];a=e!==void 0&&n!==void 0?(e+n)/2:null}else a=t[i]??null;return{count:this.count,nullRate:e,min:n,median:a,max:r}}};const u=10;function d(e,t){if(e.median===null||t.median===null||e.median<=0||t.median<=0)return null;let n=t.median/e.median;return n>=10||n<=1/10?{ratio:n,prevMedian:e.median,currMedian:t.median}:null}function f(e){if(typeof e!=`object`||!e)return null;let t=e;return typeof t.count!=`number`||typeof t.nullRate!=`number`?null:{count:t.count,nullRate:t.nullRate,min:typeof t.min==`number`?t.min:null,median:typeof t.median==`number`?t.median:null,max:typeof t.max==`number`?t.max:null}}const p=1e3;async function m(r){let{importRunId:i,runLabel:o,params:s,transform:c,feed:u,esClient:d,esIndex:f,batchSize:m=p,onProgress:h,rowLimit:g,signal:_,errorTracker:v=new a,warningTracker:y=new a,priceStats:b=new l,logger:x}=r;if(m<1)throw Error(`batchSize must be >= 1 (got ${m})`);let S=Date.now();x?.info({importRunId:i,runLabel:o,batchSize:m,esIndex:f},`imports.runner: started`);let C=0,w=0,T=0,E=0,D=0,O=0,k=[],A=[],j=[],M=async()=>{if(j.length===0)return;let e=j;j=[];let t=Date.now(),r;try{r=await n(d,{index:f,docs:e.map(({id:e,doc:t})=>({id:e,doc:t})),..._!==void 0&&{signal:_}})}catch(t){let n=_?.aborted??(t instanceof Error&&(t.name===`AbortError`||/abort/i.test(t.message)))?`aborted`:`bulk_request_failed`,r=t instanceof Error?t.message:String(t);for(let{rowNumber:t}of e)v.addError(t,n,r,void 0,null);throw T+=e.length,t}w+=r.succeeded,T+=r.failures.length,O+=1;let a=Date.now()-t,o=(Date.now()-S)/1e3;if(x?.info({importRunId:i,batch:O,flushMs:a,submitted:e.length,succeeded:r.succeeded,failed:r.failures.length,rowsRead:C,rowsPerSecond:o>0?Math.round(C/o):0},`imports.runner: batch flushed`),r.failures.length>0){let t=new Map(e.map(e=>[e.id,e.rowNumber]));for(let e of r.failures){let n=t.get(e.id)??-1;v.addError(n,e.type,e.reason,void 0,{id:e.id})}}let s=new Set(r.failures.map(e=>e.id));for(let t of e)if(!(!t.warnings||t.warnings.length===0)&&!s.has(t.id)){D+=1;for(let e of t.warnings)y.addError(t.rowNumber,e.warningType,e.message,e.field,t.rowData)}for(let t of e)!t.pairs||t.pairs.length===0||s.has(t.id)||k.push(...t.pairs);for(let t of e)s.has(t.id)||b.record(t.priceSample);if(h){let e=(Date.now()-S)/1e3;h({rowsRead:C,rowsSucceeded:w,rowsFailed:T,rowsSkipped:E,rowsWarned:D,batchesCompleted:O,elapsedSeconds:e,rowsPerSecond:e>0?C/e:0,distinctErrorPatterns:v.getDistinctPatternCount()})}},N=m,P=new Map;for await(let{rowNumber:t,row:n}of e(u)){if(_?.aborted||g!==void 0&&C>=g)break;if(C+=1,C===1&&x?.info({importRunId:i,rowNumber:t,sampleKeys:Object.keys(n)},`imports.runner: STREAM_FIRST_ROW`),C%N===0){let e=(Date.now()-S)/1e3;x?.info({importRunId:i,rowsRead:C,rowsSucceeded:w,rowsFailed:T,rowsSkipped:E,pending:j.length,elapsedSec:Math.round(e*10)/10,rowsPerSecond:e>0?Math.round(C/e):0,distinctFirstErrorTypes:P.size},`imports.runner: STREAM_TICK`)}let e={importRunId:i,params:s,runLabel:o,rowNumber:t},r;try{r=await c(n,e)}catch(e){let r=e instanceof Error?e.message:String(e);v.addError(t,`transform_threw`,r,void 0,n),P.has(`transform_threw`)||(P.set(`transform_threw`,{rowNumber:t,message:r}),x?.warn({importRunId:i,rowNumber:t,errorType:`transform_threw`,message:r},`imports.runner: FIRST_ERROR transform_threw`)),T+=1;continue}if(r.kind===`skip`){E+=1,r.patches&&r.patches.length>0&&A.push(...r.patches);continue}if(r.kind===`error`){v.addError(t,r.error.errorType,r.error.message,r.error.field,n),P.has(r.error.errorType)||(P.set(r.error.errorType,{rowNumber:t,message:r.error.message}),x?.warn({importRunId:i,rowNumber:t,errorType:r.error.errorType,field:r.error.field,message:r.error.message},`imports.runner: FIRST_ERROR ${r.error.errorType}`)),T+=1;continue}j.push({id:r.id,doc:r.doc,rowNumber:t,rowData:n,...r.warnings&&r.warnings.length>0&&{warnings:r.warnings},...r.pairs&&r.pairs.length>0&&{pairs:r.pairs},...r.priceSample!==void 0&&{priceSample:r.priceSample}}),j.length>=m&&await M()}await M();let F=0,I=0;for(let e=0;e<A.length;e+=m){if(_?.aborted)throw Error(`imports: run aborted during core-charge attach (phase 2)`);let n=A.slice(e,e+m),r;try{r=await t(d,{index:f,updates:n.map(({id:e,doc:t})=>({id:e,doc:t})),..._!==void 0&&{signal:_}})}catch(e){if(_?.aborted??(e instanceof Error&&(e.name===`AbortError`||/abort/i.test(e.message))))throw e;let t=e instanceof Error?e.message:String(e);for(let e of n)y.addError(-1,`patch_attach_failed`,t,e.orphanWarning?.field,{target:e.id});continue}if(F+=r.succeeded,r.failures.length>0){let e=new Map(n.map(e=>[e.id,e]));for(let t of r.failures){let n=e.get(t.id)?.orphanWarning;t.status===404||t.type===`document_missing_exception`?(I+=1,y.addError(-1,n?.warningType??`orphan_patch`,n?.message??t.reason,n?.field,{target:t.id})):y.addError(-1,`patch_attach_failed`,t.reason,n?.field,{target:t.id})}}}let L=(Date.now()-S)/1e3;return x?.info({importRunId:i,rowsRead:C,rowsSucceeded:w,rowsFailed:T,rowsSkipped:E,batchesCompleted:O,patchesApplied:F,patchesOrphaned:I,totalSec:Math.round(L*10)/10,rowsPerSecond:L>0?Math.round(C/L):0},`imports.runner: finished`),{rowsRead:C,rowsSucceeded:w,rowsFailed:T,rowsSkipped:E,rowsWarned:D,batchesCompleted:O,errors:v.snapshot(),warnings:y.snapshot(),pairs:k,patchesApplied:F,patchesOrphaned:I,priceStats:b.snapshot()}}export{c as a,f as c,a as d,o as f,u as i,r as l,m as n,l as o,s as r,d as s,p as t,i as u};
+//# sourceMappingURL=runner-BnFvID1z.mjs.map

package/dist/runner-BnFvID1z.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"runner-BnFvID1z.mjs","names":[],"sources":["../src/error-tracker.ts","../src/price-stats.ts","../src/runner.ts"],"sourcesContent":["/**\n * Aggregates per-row errors into top-N pattern buckets so a feed of 1M\n * malformed rows surfaces as a handful of actionable signatures rather\n * than a million identical strings.\n *\n * Lifted from giga-test (`backend/src/workers/csv-importer.ts`) and\n * generalized:\n * - Configurable caps so the importer can tune memory bounds per run.\n * - Pure data — no logging side effects, no I/O.\n * - JSON-serialisable output via {@link ErrorTracker.snapshot} for the\n *   `import_run.errorSummary` column.\n *\n * Pattern signature shape: `${errorType}:${field || 'GENERAL'}:${message}`.\n * Same `errorType + field + message` collapses to one bucket; differing\n * messages stay separate. This is intentional: a parser error on column\n * `NetPrice/Discount` (\"invalid number 'NA,5'\") and the same on column\n * `GrossPrice` are operationally distinct even if the parser is the same.\n *\n * Memory bounds: the patterns map is capped at `maxPatterns`. Once full,\n * additional NEW signatures are dropped — known patterns keep accumulating\n * counts. This is the \"top-N most common\" model: rare-but-novel errors\n * past the cap are invisible, but the cap protects against a runaway\n * adversarial feed exploding the map. Sample arrays are independently\n * capped at `maxSamplesPerPattern`.\n */\n\n/**\n * Recursively-defined JSON-serialisable value. Mirrors the shape of\n * `JsonValue` in `@murumets-ee/entity` without taking a dependency on\n * that package — error-tracker is otherwise standalone, and the\n * `ImportRun.errorSummary` JSONB column accepts anything in this shape.\n */\nexport type ImportJsonValue =\n  | string\n  | number\n  | boolean\n  | null\n  | ImportJsonValue[]\n  | { [key: string]: ImportJsonValue }\n\n/** One sample row attached to a pattern. `rowData` is the originally-parsed row. */\nexport interface ErrorSample {\n  rowNumber: number\n  rowData: ImportJsonValue\n}\n\n/** Public shape of an aggregated pattern as returned by {@link ErrorTracker.getTopPatterns}. */\nexport interface ErrorPattern {\n  errorType: string\n  field: string | null\n  message: string\n  count: number\n  firstOccurrence: number\n  lastOccurrence: number\n  samples: ReadonlyArray<ErrorSample>\n  /** `count / totalErrors`, scaled 0..100. `0` when there are zero errors total. */\n  percentage: number\n}\n\nexport interface ErrorTrackerConfig {\n  /** Hard cap on distinct signatures. Defaults to 50. New signatures past the cap are dropped. */\n  maxPatterns?: number\n  /** Hard cap on samples retained per pattern. Defaults to 5. Excess samples are dropped. */\n  maxSamplesPerPattern?: number\n}\n\n/** Default caps — match giga-test for compatibility. */\nexport const DEFAULT_MAX_PATTERNS = 50\nexport const DEFAULT_MAX_SAMPLES_PER_PATTERN = 5\n\ninterface InternalPattern {\n  errorType: string\n  field: string | null\n  message: string\n  count: number\n  firstOccurrence: number\n  lastOccurrence: number\n  samples: ErrorSample[]\n}\n\nexport class ErrorTracker {\n  private readonly patterns = new Map<string, InternalPattern>()\n  private readonly maxPatterns: number\n  private readonly maxSamplesPerPattern: number\n  private droppedSignatures = 0\n\n  constructor(config: ErrorTrackerConfig = {}) {\n    this.maxPatterns = config.maxPatterns ?? DEFAULT_MAX_PATTERNS\n    this.maxSamplesPerPattern = config.maxSamplesPerPattern ?? DEFAULT_MAX_SAMPLES_PER_PATTERN\n  }\n\n  /**\n   * Record one error. Same `(errorType, field, message)` triple bumps the\n   * existing bucket; a new triple opens a new one (subject to {@link maxPatterns}).\n   *\n   * `field` is optional — pass `undefined` for errors not tied to a single\n   * column (e.g. parse errors at row level). Internally normalised to the\n   * literal string `'GENERAL'` so it shares a bucket with other genericised\n   * errors of the same type+message.\n   */\n  addError(\n    rowNumber: number,\n    errorType: string,\n    message: string,\n    field: string | undefined,\n    rowData: ImportJsonValue,\n  ): void {\n    const fieldKey = field ?? 'GENERAL'\n    const signature = `${errorType}:${fieldKey}:${message}`\n\n    let pattern = this.patterns.get(signature)\n    if (!pattern) {\n      if (this.patterns.size >= this.maxPatterns) {\n        this.droppedSignatures += 1\n        return\n      }\n      pattern = {\n        errorType,\n        field: field ?? null,\n        message,\n        count: 0,\n        firstOccurrence: rowNumber,\n        lastOccurrence: rowNumber,\n        samples: [],\n      }\n      this.patterns.set(signature, pattern)\n    }\n\n    pattern.count += 1\n    pattern.lastOccurrence = rowNumber\n    if (pattern.samples.length < this.maxSamplesPerPattern) {\n      pattern.samples.push({ rowNumber, rowData })\n    }\n  }\n\n  /** Total count across every pattern. Counts errors, not patterns. */\n  getTotalErrorCount(): number {\n    let sum = 0\n    for (const p of this.patterns.values()) sum += p.count\n    return sum\n  }\n\n  /** Number of distinct signatures retained in the map (≤ `maxPatterns`). */\n  getDistinctPatternCount(): number {\n    return this.patterns.size\n  }\n\n  /**\n   * Number of NEW signatures dropped because the map was already at\n   * capacity. Surfacing this in the import_run summary tells the operator\n   * \"the top-N was saturated — there's a long tail you're not seeing\".\n   */\n  getDroppedSignatureCount(): number {\n    return this.droppedSignatures\n  }\n\n  /**\n   * Top patterns sorted by descending count, capped at `maxPatterns`.\n   * Stable secondary order is insertion order (Map iteration order is\n   * insertion order; Array.sort is stable in V8).\n   */\n  getTopPatterns(): ErrorPattern[] {\n    const all = Array.from(this.patterns.values()).sort((a, b) => b.count - a.count)\n    const totalErrors = all.reduce((sum, p) => sum + p.count, 0)\n    return all.map((p) => ({\n      errorType: p.errorType,\n      field: p.field,\n      message: p.message,\n      count: p.count,\n      firstOccurrence: p.firstOccurrence,\n      lastOccurrence: p.lastOccurrence,\n      samples: p.samples.slice(),\n      percentage: totalErrors > 0 ? (p.count / totalErrors) * 100 : 0,\n    }))\n  }\n\n  /**\n   * Compact JSON-serialisable snapshot for `import_run.errorSummary`.\n   * Aside from the patterns array, includes the totals so a reader of\n   * just this column doesn't have to re-derive them.\n   */\n  snapshot(): ErrorTrackerSnapshot {\n    return {\n      totalErrors: this.getTotalErrorCount(),\n      distinctPatterns: this.getDistinctPatternCount(),\n      droppedSignatures: this.droppedSignatures,\n      patterns: this.getTopPatterns(),\n    }\n  }\n}\n\nexport interface ErrorTrackerSnapshot {\n  totalErrors: number\n  distinctPatterns: number\n  droppedSignatures: number\n  patterns: ErrorPattern[]\n}\n\n/**\n * Merge two independently-produced `ErrorTrackerSnapshot`s into one — e.g.\n * the runner's per-row `warningSummary` plus a worker-level signal computed\n * AFTER the run completes (F010 / #602: the price-drift anomaly check runs\n * in `worker.ts`, after `runImport` returns, using a fresh single-pattern\n * tracker of its own). Patterns concatenate rather than re-dedup: the two\n * snapshots are assumed to use disjoint `errorType` signature spaces (a\n * worker-level tracker uses a type — `price_anomaly` — the runner's own\n * warning aggregation never emits), so counts sum directly with no risk of\n * double-counting one bucket as two.\n */\nexport function mergeErrorTrackerSnapshots(\n  a: ErrorTrackerSnapshot,\n  b: ErrorTrackerSnapshot,\n): ErrorTrackerSnapshot {\n  const totalErrors = a.totalErrors + b.totalErrors\n  // Each source snapshot computed `percentage` against its OWN `totalErrors`\n  // (see `getTopPatterns` above); the merged total is different, so every\n  // pattern's `percentage` must be recomputed against it — a straight\n  // concat would leave every value stale (e.g. `20/100=20%` staying `20`\n  // once the true merged denominator is `101`). `0` when the merged total\n  // is `0`, matching `getTopPatterns`' own zero-total convention.\n  const rescale = (p: ErrorPattern): ErrorPattern => ({\n    ...p,\n    percentage: totalErrors > 0 ? (p.count / totalErrors) * 100 : 0,\n  })\n  return {\n    totalErrors,\n    distinctPatterns: a.distinctPatterns + b.distinctPatterns,\n    droppedSignatures: a.droppedSignatures + b.droppedSignatures,\n    patterns: [...a.patterns.map(rescale), ...b.patterns.map(rescale)],\n  }\n}\n","/**\n * Price-distribution stats + prior-run anomaly detection (F010 / issue #602).\n *\n * The import pipeline's existing anomaly detection — register-handler.ts's\n * \"6b. Header-presence drift check\" — only catches STRUCTURAL drift: a\n * mapped column vanishing (or getting renamed) from the header row. It does\n * NOT catch VALUE drift: a supplier silently flipping their decimal\n * separator (e.g. `1.234,56` → `1234.56`) or thousands separator WITHOUT\n * touching a single column header. Since the saved\n * `import_mapping.numberFormat` (`{decimal, thousands}`) is unchanged,\n * every price in that run parses as a NUMERICALLY VALID but WRONG value —\n * silently corrupting the catalog's prices. Nothing upstream of this module\n * flags it.\n *\n * This module computes a per-run price-distribution snapshot (`PriceStats`)\n * over `PartsDocument.purchase_price_minor` — the DERIVED cost field,\n * rather than the raw net/list price columns, because both the direct-net\n * AND two-file derivation paths in `mapping/parts-schema.ts` (step 4) funnel\n * through it, so a decimal-format flip on EITHER source column surfaces\n * here regardless of which pricing path a supplier's feed uses.\n *\n * Collection (`PriceStatsCollector`) happens in the SAME streaming pass\n * `runner.ts` already uses to build `totals` — one `.record()` call per\n * ES-accepted row, no second pass over the file, no extra DB round-trip.\n * The prior-run comparison (`detectPriceAnomaly`) happens in `worker.ts`\n * AFTER the run completes, because it needs `AdminClient` to look the\n * prior run up — this module stays DB-agnostic, pure data + pure functions.\n */\n\n/** A run's price-distribution snapshot. Values are in MINOR units (cents), matching `PartsDocument.purchase_price_minor`'s own scale. */\nexport interface PriceStats {\n  /** ES-accepted rows considered — same accounting as `RunImportResult.rowsSucceeded`. */\n  count: number\n  /** Fraction (0..1) of `count` whose price sample was absent (e.g. a list-only row with no resolvable purchase cost — the same rows that carry a `no_purchase_price` warning). `0` when `count` is `0`. */\n  nullRate: number\n  /** Minimum of the non-null samples, in minor units. `null` when there are zero non-null samples. */\n  min: number | null\n  /** Median of the non-null samples, in minor units. `null` when there are zero non-null samples. */\n  median: number | null\n  /** Maximum of the non-null samples, in minor units. `null` when there are zero non-null samples. */\n  max: number | null\n}\n\n/** The all-zero/empty snapshot — a run that produced zero ES-accepted rows. */\nexport const EMPTY_PRICE_STATS: PriceStats = {\n  count: 0,\n  nullRate: 0,\n  min: null,\n  median: null,\n  max: null,\n}\n\n/**\n * Bounded sample cap for the min/median/max computation — the same\n * \"bounded, uniform sample\" philosophy `ErrorTracker` uses top-N patterns\n * for. 50,000 floats is ~400KB, trivial next to the ES bulk-write buffers\n * already in flight for a run this size, and far more than needed for a\n * stable median estimate. A feed with more ES-accepted rows than this cap\n * still gets an EXACT `count` / `nullRate` (those are running counters, not\n * sample-bounded) — only min/median/max become an approximation over a\n * UNIFORM RANDOM sample of `PRICE_SAMPLE_CAP` non-null values, not \"the\n * first N encountered\" (see `record`'s reservoir-sampling note for why\n * order-independence matters specifically for this module's job).\n */\nexport const PRICE_SAMPLE_CAP = 50_000\n\n/**\n * Accumulates price samples during a run's streaming pass. `record` is O(1)\n * (amortized array push below the cap, one RNG draw above it); `snapshot`\n * sorts the retained samples once at the end — cheap relative to the ES\n * bulk-write the run already performs every batch.\n */\nexport class PriceStatsCollector {\n  private count = 0\n  private nullCount = 0\n  private nonNullSeen = 0\n  private readonly samples: number[] = []\n\n  /**\n   * Record one ES-accepted row's price sample. Call exactly once per\n   * accepted row (mirrors the `rowsWarned` / `pairs` accounting in\n   * `runner.ts`'s flush loop) — `undefined` means \"this row had no price to\n   * sample\" and counts toward `nullRate` rather than being skipped.\n   *\n   * Reservoir sampling (Algorithm R) past the cap, not \"first N wins\": each\n   * new non-null value replaces a uniformly-random existing slot with\n   * probability `PRICE_SAMPLE_CAP / nonNullSeen`, giving every non-null\n   * sample seen so far an equal chance of surviving into the final\n   * reservoir. Order-independence matters here specifically — the anomaly\n   * this module exists to catch (a supplier's decimal-format flip) could\n   * start partway through a feed larger than the cap (e.g. a corrupted\n   * re-export appended to an otherwise-normal catalogue), and a first-N\n   * sample would never see a single row past the cutover point, silently\n   * missing exactly the drift F010 was built to detect.\n   */\n  record(value: number | undefined): void {\n    this.count += 1\n    if (value === undefined) {\n      this.nullCount += 1\n      return\n    }\n    this.nonNullSeen += 1\n    if (this.samples.length < PRICE_SAMPLE_CAP) {\n      this.samples.push(value)\n      return\n    }\n    const j = Math.floor(Math.random() * this.nonNullSeen)\n    if (j < PRICE_SAMPLE_CAP) this.samples[j] = value\n  }\n\n  snapshot(): PriceStats {\n    if (this.count === 0) return { ...EMPTY_PRICE_STATS }\n    const nullRate = this.nullCount / this.count\n    if (this.samples.length === 0) {\n      return { count: this.count, nullRate, min: null, median: null, max: null }\n    }\n    const sorted = [...this.samples].sort((a, b) => a - b)\n    const min = sorted[0] ?? null\n    const max = sorted[sorted.length - 1] ?? null\n    const mid = Math.floor(sorted.length / 2)\n    let median: number | null\n    if (sorted.length % 2 === 0) {\n      const lo = sorted[mid - 1]\n      const hi = sorted[mid]\n      median = lo !== undefined && hi !== undefined ? (lo + hi) / 2 : null\n    } else {\n      median = sorted[mid] ?? null\n    }\n    return { count: this.count, nullRate, min, median, max }\n  }\n}\n\n/**\n * Anomaly threshold: a run whose median purchase price is >= 10x higher OR\n * <= 1/10 (0.1x) of the prior successful run's median (for the same\n * transform + supplier + brand) is flagged.\n *\n * Chosen generously above the range a real repricing plausibly covers\n * (even an aggressive markup/discount renegotiation is a low-single-digit\n * multiplier — 2x-3x would already be an extreme outlier for a legitimate\n * price change) but comfortably BELOW the ~100x-1000x magnitude a\n * decimal/thousands-separator flip produces for typical EUR part prices:\n * `\"1.234,56\"` (a European-format €1234.56) misparsed with dot-decimal\n * conventions reads as `1.234,56` → `1234.56`, a ~1000x inflation if the\n * true value was e.g. €1.23; a thousands-separator flip on `\"1234,56\"`\n * (true €1234.56) misparsed as `123456` (dropping the decimal entirely) is\n * a ~100x inflation. 10x leaves a wide safety margin against false\n * positives on genuine price changes while still comfortably catching\n * every realistic decimal/thousands-separator flip.\n */\nexport const PRICE_ANOMALY_THRESHOLD_MULTIPLIER = 10\n\nexport interface PriceAnomaly {\n  /** `curr.median / prev.median`. */\n  ratio: number\n  /** The prior run's median (minor units). */\n  prevMedian: number\n  /** This run's median (minor units). */\n  currMedian: number\n}\n\n/**\n * Compare two runs' price stats for a magnitude-scale anomaly. Returns\n * `null` when there's nothing to compare (either side has no median — zero\n * non-null price samples, e.g. a schema/transform that never emits a price\n * sample, or a run with zero ES-accepted rows) or the ratio falls within\n * the safe band.\n */\nexport function detectPriceAnomaly(prev: PriceStats, curr: PriceStats): PriceAnomaly | null {\n  if (prev.median === null || curr.median === null) return null\n  // A zero/negative median can't produce a meaningful ratio. Shouldn't occur\n  // for real minor-unit prices (`toMinorUnits` on a positive price); this is\n  // defensive-only so a pathological value can't produce Infinity/NaN below.\n  if (prev.median <= 0 || curr.median <= 0) return null\n  const ratio = curr.median / prev.median\n  if (\n    ratio >= PRICE_ANOMALY_THRESHOLD_MULTIPLIER ||\n    ratio <= 1 / PRICE_ANOMALY_THRESHOLD_MULTIPLIER\n  ) {\n    return { ratio, prevMedian: prev.median, currMedian: curr.median }\n  }\n  return null\n}\n\n/**\n * Narrow an `unknown` value (a `field.json()` column read back through\n * `AdminClient`, typed `JsonValue | undefined` on the DTO) to `PriceStats`,\n * or `null` when it doesn't structurally match — e.g. a run written before\n * this field existed, or a run whose transform never emitted a price\n * sample so `min`/`median`/`max` serialised as `null`. Defensive, matching\n * the `params` narrowing pattern already used in `worker.ts`.\n */\nexport function parsePriceStats(value: unknown): PriceStats | null {\n  if (typeof value !== 'object' || value === null) return null\n  const v = value as Record<string, unknown>\n  if (typeof v.count !== 'number' || typeof v.nullRate !== 'number') return null\n  return {\n    count: v.count,\n    nullRate: v.nullRate,\n    min: typeof v.min === 'number' ? v.min : null,\n    median: typeof v.median === 'number' ? v.median : null,\n    max: typeof v.max === 'number' ? v.max : null,\n  }\n}\n","/**\n * One-shot importer: stream rows → transform → batched bulk-write to ES,\n * accumulating per-row errors into `ErrorTracker` and reporting progress\n * to the queue every batch.\n *\n * Per PLAN-ECOMMERCE.md PR 7 (PoC scope):\n * - **Batch size 1000.** Matches giga-test precedent. Configurable for\n *   integration tests that don't want a 1k row floor.\n * - **No resumability and no automatic retries.** The `imports:run` queue\n *   job is registered with `defaultRetries: 0` (see `worker.ts`) so a\n *   failed handler does NOT re-enqueue itself — re-running a multi-batch\n *   import against the same `import_run.id` while the previous attempt\n *   may still be writing is a footgun (duplicate batches, double-counted\n *   progress). Operator retries by creating a NEW `import_run` row.\n * - **No per-supplier transform plugin.** The runner takes a single\n *   `RowTransform<TDoc>` from the registry and applies it to every row;\n *   PR 8 may diverge but only by registering a different transform name.\n * - **Direct `bulkUpsert` into the live aliased index** (D6 alias is\n *   set up by PR 4's `ensureAliasedIndex`; the importer doesn't reindex).\n *\n * Per D21 (sanctioned bulk path): this runner intentionally bypasses\n * AdminClient and entity hooks. Per-batch audit / observability lives\n * on the surrounding `import_run` row + queue progress, NOT per-row.\n */\n\nimport {\n  type BulkIndexResult,\n  bulkPartialUpdate,\n  bulkUpsert,\n  type EsClientLike,\n} from '@murumets-ee/search-elasticsearch'\nimport { ErrorTracker } from './error-tracker.js'\nimport { type PriceStats, PriceStatsCollector } from './price-stats.js'\nimport { type ReadFeedOptions, readFeed } from './read-feed.js'\nimport type {\n  DirectionalPair,\n  DocPatch,\n  RowTransform,\n  RowWarning,\n  TransformContext,\n} from './transform.js'\n\n/** Soft default; chosen to match giga-test. ES bulk requests over ~5MB get split server-side anyway. */\nexport const DEFAULT_BATCH_SIZE = 1000\n\nexport interface RunImportOptions<TDoc> {\n  /** UUID of the `import_run` row driving this run. Forwarded to every transform invocation. */\n  importRunId: string\n  /** Operator-supplied label for the run. Forwarded to the transform context. */\n  runLabel: string\n  /** Opaque per-run params copied from `import_run.params`. */\n  params: Record<string, unknown>\n  /** Transform applied to every parsed row. */\n  transform: RowTransform<TDoc>\n  /** Feed reader options — file path + format (csv/tsv/xlsx) + delimiter/header config. */\n  feed: ReadFeedOptions\n  /** ES client (low-level shape from `@murumets-ee/search-elasticsearch`). */\n  esClient: EsClientLike\n  /** Index alias to write to. Per D6, callers always pass an alias, never a physical index. */\n  esIndex: string\n  /** Rows per `bulkUpsert` call. Default {@link DEFAULT_BATCH_SIZE}. */\n  batchSize?: number\n  /**\n   * Callback invoked after every batch. The handler in `worker.ts`\n   * forwards this to `ctx.updateProgress` for the queue UI; tests\n   * inspect it directly. Synchronous + cheap so a slow callback can't\n   * back-pressure the importer.\n   */\n  onProgress?: (progress: ImportRunProgress) => void\n  /** Optional: stop processing after this many rows. Tests use it; production passes `undefined`. */\n  rowLimit?: number\n  /** Abort signal threaded into the underlying ES client request — cooperative cancel. */\n  signal?: AbortSignal\n  /** Optional ErrorTracker config (caps). Default: top-50 patterns × 5 samples. */\n  errorTracker?: ErrorTracker\n  /**\n   * Optional separate tracker for NON-FATAL row warnings (commerce-demo PR4 —\n   * `no_purchase_price`). Distinct from {@link errorTracker} so warnings never\n   * inflate `totals.failed` / the error summary: a warned row is a SUCCESS whose\n   * doc is indexed. Default: a fresh tracker with the same caps.\n   */\n  warningTracker?: ErrorTracker\n  /**\n   * Optional price-stats collector (F010 / #602) — accumulates a\n   * min/median/max/nullRate distribution over ES-accepted rows' optional\n   * `priceSample`s in the SAME streaming pass that builds `totals`. Default:\n   * a fresh collector. Tests can inject their own to inspect it directly —\n   * mirrors the {@link errorTracker} / {@link warningTracker} DI shape.\n   */\n  priceStats?: PriceStatsCollector\n  /**\n   * Optional structural logger for per-batch telemetry. Receives\n   * `{ batch, ms, succeeded, failed, rowsRead, rowsPerSecond }` after\n   * every flush, plus a one-shot `started` and `finished` event. The\n   * worker forwards `app.logger.child({ pkg: 'imports' })`; tests\n   * leave it undefined.\n   *\n   * Why: without per-batch visibility, a stalled run looks identical\n   * to \"first batch in progress\" forever — the only feedback is the\n   * queue's `progress` jsonb (which writes after each batch) and the\n   * eventual `succeeded`/`failed` job event. Adding this lets the\n   * operator tail the worker log and immediately see whether ES is\n   * rate-limiting, the transform is rejecting every row, or the\n   * stream is actually progressing.\n   */\n  logger?: {\n    info: (data: Record<string, unknown>, msg: string) => void\n    warn: (data: Record<string, unknown>, msg: string) => void\n  }\n}\n\n/**\n * Progress payload written to `toolkit_jobs.progress` after every batch.\n * Caps + flush rules live on the queue's `updateProgress` debounce —\n * callers don't need to throttle.\n */\nexport interface ImportRunProgress {\n  rowsRead: number\n  rowsSucceeded: number\n  rowsFailed: number\n  rowsSkipped: number\n  /** Rows indexed WITH a non-fatal warning (commerce-demo PR4 — `no_purchase_price`). */\n  rowsWarned: number\n  batchesCompleted: number\n  /** Wall-clock seconds since the runner started. */\n  elapsedSeconds: number\n  /** Rows / second, computed at every batch. */\n  rowsPerSecond: number\n  /** Distinct error patterns currently held by the tracker. Saturates at the cap. */\n  distinctErrorPatterns: number\n  /**\n   * Total data-row count from a pre-pass scan, when the worker computed one.\n   * Lets the UI render `processed / total` percent. Absent when the worker\n   * skipped the count (file too large for the count budget, count failed,\n   * etc.) — UI must treat it as optional.\n   */\n  totalRows?: number\n}\n\n/**\n * Final result returned by {@link runImport}. The handler writes these\n * onto the `import_run` row alongside the ErrorTracker snapshot.\n */\nexport interface RunImportResult {\n  /** Total rows read from the file (excludes skipped empty lines). */\n  rowsRead: number\n  /** Rows the transform turned into a successful doc AND the ES cluster acknowledged. */\n  rowsSucceeded: number\n  /**\n   * Rows that the transform rejected (`{ kind: 'error' }`) OR that ES\n   * rejected on bulk-write (per-doc failure). Both are aggregated by\n   * `errorTracker` for the import_run summary.\n   */\n  rowsFailed: number\n  /** Rows that the transform skipped (`{ kind: 'skip' }`) — header noise, blank lines, intentional drop. */\n  rowsSkipped: number\n  /**\n   * Rows that succeeded (doc indexed) but carried ≥1 non-fatal warning —\n   * counted SEPARATELY from `rowsSucceeded`/`rowsFailed` (a warned row is also\n   * counted in `rowsSucceeded`). commerce-demo PR4: `no_purchase_price`.\n   */\n  rowsWarned: number\n  /** Number of `bulkUpsert` calls made. */\n  batchesCompleted: number\n  /** Final value of {@link ErrorTracker.snapshot}. */\n  errors: ReturnType<ErrorTracker['snapshot']>\n  /** Aggregated NON-FATAL warnings (commerce-demo PR4). Empty when no row warned. */\n  warnings: ReturnType<ErrorTracker['snapshot']>\n  /**\n   * Directional supersession pairs harvested as a side-output (commerce-demo PR6,\n   * SD007). Buffered in memory from ES-ACCEPTED rows only (same accounting as\n   * warnings — a row whose doc was rejected by ES contributes no pair). Empty unless\n   * a schema emits them (the `parts` schema, when a row's `superseding_code` is set).\n   * The worker hands these to an injected `pairSink` AFTER the run for the commerce\n   * replacement-merge engine; the runner itself never writes the consumer's tables.\n   */\n  pairs: readonly DirectionalPair[]\n  /**\n   * Doc-patches applied by the phase-2 pass (commerce-demo PR7): a `<code>*S`\n   * core-fee row is not a saleable doc — it emits a {@link DocPatch} that merges its\n   * deposit onto the PARENT part's `core_charge_minor`. The runner buffers these\n   * during the stream (sparse) and applies them via `bulkPartialUpdate` AFTER the ES\n   * pass. `patchesApplied` = targets the cluster merged; `patchesOrphaned` = patches\n   * whose parent genuinely wasn't in the index (a `document_missing`/404 → an\n   * `orphan_core_charge` warning). An INFRA failure (whole-batch transport error, or a\n   * non-404 per-doc failure) is NOT counted as orphaned — it lands in a distinct\n   * `patch_attach_failed` warning bucket so it can't masquerade as a data orphan. Zero\n   * unless a schema emits patches.\n   */\n  patchesApplied: number\n  patchesOrphaned: number\n  /**\n   * Price-distribution snapshot (F010 / #602) accumulated from ES-accepted\n   * rows' `priceSample`s during this same streaming pass. See `price-stats.ts`.\n   * `count` is `0` (all fields `null`/`0`) for a run whose schema never emits\n   * a price sample.\n   */\n  priceStats: PriceStats\n}\n\n/**\n * Apply the runner against a feed file. Stops on rowLimit OR end-of-file\n * OR if `signal` aborts. Throws if the streaming reader / ES client\n * throws — caller (the queue handler) catches that and writes\n * `import_run.status = 'failed'` with the error message in\n * `errorSummary.fatal`.\n */\nexport async function runImport<TDoc>(options: RunImportOptions<TDoc>): Promise<RunImportResult> {\n  const {\n    importRunId,\n    runLabel,\n    params,\n    transform,\n    feed,\n    esClient,\n    esIndex,\n    batchSize = DEFAULT_BATCH_SIZE,\n    onProgress,\n    rowLimit,\n    signal,\n    errorTracker = new ErrorTracker(),\n    warningTracker = new ErrorTracker(),\n    priceStats: priceStatsCollector = new PriceStatsCollector(),\n    logger,\n  } = options\n\n  if (batchSize < 1) {\n    throw new Error(`batchSize must be >= 1 (got ${batchSize})`)\n  }\n\n  const startedAt = Date.now()\n  logger?.info({ importRunId, runLabel, batchSize, esIndex }, 'imports.runner: started')\n  let rowsRead = 0\n  let rowsSucceeded = 0\n  let rowsFailed = 0\n  let rowsSkipped = 0\n  let rowsWarned = 0\n  let batchesCompleted = 0\n\n  // Side-output buffer (commerce-demo PR6, SD007): directional supersession pairs\n  // accumulated from ES-ACCEPTED rows across the whole run. An array append per\n  // sparse superseded row — it never blocks the `bulkUpsert` loop. Handed to the\n  // injected `pairSink` by the worker AFTER the run (read-once / merge-after).\n  const collectedPairs: DirectionalPair[] = []\n\n  // Side-output buffer (commerce-demo PR7): doc-patches from `*S` core-fee rows —\n  // each attaches a core charge onto a DIFFERENT (parent) doc. Collected off the\n  // `skip` result (a `*S` row is never a saleable doc), sparse, applied in a phase-2\n  // `bulkPartialUpdate` after the ES pass (below) — order-independent because ES\n  // `update` is realtime by `_id`.\n  const collectedPatches: DocPatch<TDoc>[] = []\n\n  // Warnings + pairs ride on the pending item (with the raw row for the warning\n  // sample) so they're accounted at FLUSH time — only for rows ES actually accepted.\n  // A warned row is by contract a SUCCESSFUL row, so counting it at push time (before\n  // the bulk response) would wrongly keep a later-ES-rejected row in rowsWarned /\n  // warnings; pairs follow the same accepted-rows-only rule.\n  let pending: Array<{\n    id: string\n    doc: TDoc\n    rowNumber: number\n    rowData: Record<string, string>\n    warnings?: readonly RowWarning[]\n    pairs?: readonly DirectionalPair[]\n    priceSample?: number\n  }> = []\n\n  const flush = async (): Promise<void> => {\n    if (pending.length === 0) return\n    const batch = pending\n    pending = []\n    const flushStartedAt = Date.now()\n    let result: BulkIndexResult\n    try {\n      result = await bulkUpsert<TDoc>(esClient, {\n        index: esIndex,\n        docs: batch.map(({ id, doc }) => ({ id, doc })),\n        ...(signal !== undefined && { signal }),\n      })\n    } catch (err) {\n      // Cluster- or transport-level failure — the whole batch is\n      // unaccounted for. Distinguish abort (operator-driven cancel) from\n      // a real cluster failure so the errorSummary doesn't mislabel a\n      // cancelled run as broken cluster connectivity.\n      const isAbort =\n        signal?.aborted ??\n        (err instanceof Error && (err.name === 'AbortError' || /abort/i.test(err.message)))\n      const errorType = isAbort ? 'aborted' : 'bulk_request_failed'\n      const reason = err instanceof Error ? err.message : String(err)\n      for (const { rowNumber } of batch) {\n        errorTracker.addError(rowNumber, errorType, reason, undefined, null)\n      }\n      rowsFailed += batch.length\n      throw err\n    }\n\n    rowsSucceeded += result.succeeded\n    rowsFailed += result.failures.length\n    batchesCompleted += 1\n\n    const flushMs = Date.now() - flushStartedAt\n    const elapsedSec = (Date.now() - startedAt) / 1000\n    logger?.info(\n      {\n        importRunId,\n        batch: batchesCompleted,\n        flushMs,\n        submitted: batch.length,\n        succeeded: result.succeeded,\n        failed: result.failures.length,\n        rowsRead,\n        rowsPerSecond: elapsedSec > 0 ? Math.round(rowsRead / elapsedSec) : 0,\n      },\n      'imports.runner: batch flushed',\n    )\n\n    if (result.failures.length > 0) {\n      // Map each ES failure back to its source row via `id`. The bulk\n      // response order matches the request order, but ES doesn't promise\n      // that; matching by `id` is the safe path. PoC volume is small\n      // enough that the O(failures × batch) cost is irrelevant.\n      const byId = new Map(batch.map((b) => [b.id, b.rowNumber]))\n      for (const fail of result.failures) {\n        const rowNumber = byId.get(fail.id) ?? -1\n        errorTracker.addError(rowNumber, fail.type, fail.reason, undefined, { id: fail.id })\n      }\n    }\n\n    // Account non-fatal warnings ONLY for rows ES accepted (commerce-demo PR4) — a\n    // warned row is by contract a SUCCESSFUL row, so a row that warned AND then failed\n    // the bulk write must NOT land in rowsWarned/warnings (it's already in rowsFailed).\n    const failedIds = new Set(result.failures.map((f) => f.id))\n    for (const item of batch) {\n      if (!item.warnings || item.warnings.length === 0) continue\n      if (failedIds.has(item.id)) continue\n      rowsWarned += 1\n      for (const w of item.warnings) {\n        warningTracker.addError(item.rowNumber, w.warningType, w.message, w.field, item.rowData)\n      }\n    }\n\n    // Buffer directional pairs for rows ES accepted (commerce-demo PR6, SD007) — same\n    // accepted-rows-only rule as warnings, so a pair from a doc ES rejected is dropped.\n    for (const item of batch) {\n      if (!item.pairs || item.pairs.length === 0) continue\n      if (failedIds.has(item.id)) continue\n      collectedPairs.push(...item.pairs)\n    }\n\n    // Price-distribution sampling (F010 / #602) — same accepted-rows-only rule as\n    // warnings/pairs: a row ES rejected contributes no sample. Every ES-accepted row\n    // records exactly once (`undefined` counts toward `nullRate` rather than being\n    // skipped), so `priceStats.count` tracks `rowsSucceeded` exactly.\n    for (const item of batch) {\n      if (failedIds.has(item.id)) continue\n      priceStatsCollector.record(item.priceSample)\n    }\n\n    if (onProgress) {\n      const elapsedSeconds = (Date.now() - startedAt) / 1000\n      onProgress({\n        rowsRead,\n        rowsSucceeded,\n        rowsFailed,\n        rowsSkipped,\n        rowsWarned,\n        batchesCompleted,\n        elapsedSeconds,\n        rowsPerSecond: elapsedSeconds > 0 ? rowsRead / elapsedSeconds : 0,\n        distinctErrorPatterns: errorTracker.getDistinctPatternCount(),\n      })\n    }\n  }\n\n  // Streaming heartbeat — sample every Nth row so the operator sees the\n  // streamer is alive even when a transform-rejected file would prevent\n  // any flush from firing. Without this, a feed that's 100% rejected\n  // looks identical to \"stuck\" until end-of-file. Sample frequency is\n  // batchSize so the cadence matches the natural batch rhythm.\n  const streamSampleEvery = batchSize\n\n  // First-error capture per error type — surface what's actually rejecting\n  // the rows when 100% of a feed is rejected. Without this, the only signal\n  // is `errorTracker.snapshot()` which only writes to import_run.errorSummary\n  // at the END of the run. Mid-run we want to know NOW.\n  const firstErrorByType = new Map<string, { rowNumber: number; message: string }>()\n\n  for await (const { rowNumber, row } of readFeed(feed)) {\n    if (signal?.aborted) break\n    if (rowLimit !== undefined && rowsRead >= rowLimit) break\n    rowsRead += 1\n\n    if (rowsRead === 1) {\n      logger?.info(\n        { importRunId, rowNumber, sampleKeys: Object.keys(row) },\n        'imports.runner: STREAM_FIRST_ROW',\n      )\n    }\n    if (rowsRead % streamSampleEvery === 0) {\n      const elapsedSec = (Date.now() - startedAt) / 1000\n      // Telemetry only — no row content, no error sample text.\n      // PR #262 review (M-5): the prior `firstErrors:\n      // Object.fromEntries(firstErrorByType)` payload included the\n      // raw error message captured from the transform, which for the\n      // carmaker feed embeds supplier codes + price fragments. The\n      // detailed first-error breakdown is already persisted on\n      // `import_run.errorSummary` via `errorTracker.snapshot()` —\n      // operators inspect it there, not in the worker log.\n      logger?.info(\n        {\n          importRunId,\n          rowsRead,\n          rowsSucceeded,\n          rowsFailed,\n          rowsSkipped,\n          pending: pending.length,\n          elapsedSec: Math.round(elapsedSec * 10) / 10,\n          rowsPerSecond: elapsedSec > 0 ? Math.round(rowsRead / elapsedSec) : 0,\n          distinctFirstErrorTypes: firstErrorByType.size,\n        },\n        'imports.runner: STREAM_TICK',\n      )\n    }\n\n    const ctx: TransformContext = { importRunId, params, runLabel, rowNumber }\n    let result: Awaited<ReturnType<typeof transform>>\n    try {\n      result = await transform(row, ctx)\n    } catch (err) {\n      // A throw from the transform is a programmer error — surface it as\n      // a row-level error so the run can continue. (If the bug is\n      // catastrophic, the operator sees the same message repeated and\n      // can stop the run.)\n      const reason = err instanceof Error ? err.message : String(err)\n      errorTracker.addError(rowNumber, 'transform_threw', reason, undefined, row)\n      if (!firstErrorByType.has('transform_threw')) {\n        firstErrorByType.set('transform_threw', { rowNumber, message: reason })\n        // PR #262 review (M-5): drop the `row` payload from the\n        // log line — supplier feeds carry pricing + internal codes.\n        // The full row is on `import_run.errorSummary.samples` for\n        // post-mortem review by operators with view permission.\n        logger?.warn(\n          { importRunId, rowNumber, errorType: 'transform_threw', message: reason },\n          'imports.runner: FIRST_ERROR transform_threw',\n        )\n      }\n      rowsFailed += 1\n      continue\n    }\n\n    if (result.kind === 'skip') {\n      rowsSkipped += 1\n      // A `*S` core-fee row skips the saleable index but may carry a doc-patch\n      // (commerce-demo PR7) — buffer it for the phase-2 attach. Unlike pairs, a\n      // patch has no ES write of its own to gate on, so it's collected here\n      // unconditionally; its success is decided by the phase-2 update.\n      if (result.patches && result.patches.length > 0) {\n        collectedPatches.push(...result.patches)\n      }\n      continue\n    }\n    if (result.kind === 'error') {\n      errorTracker.addError(\n        rowNumber,\n        result.error.errorType,\n        result.error.message,\n        result.error.field,\n        row,\n      )\n      if (!firstErrorByType.has(result.error.errorType)) {\n        firstErrorByType.set(result.error.errorType, {\n          rowNumber,\n          message: result.error.message,\n        })\n        logger?.warn(\n          {\n            importRunId,\n            rowNumber,\n            errorType: result.error.errorType,\n            field: result.error.field,\n            message: result.error.message,\n          },\n          `imports.runner: FIRST_ERROR ${result.error.errorType}`,\n        )\n      }\n      rowsFailed += 1\n      continue\n    }\n\n    // Non-fatal warnings ride on a SUCCESS — the doc is still indexed\n    // (commerce-demo PR4: `no_purchase_price`). Carry them (+ the raw row for the\n    // sample) on the pending item; the flush aggregates them into the separate\n    // `warningTracker` AFTER the bulk response confirms the row was accepted, so a\n    // warned-then-ES-rejected row never inflates rowsWarned / the warning summary.\n    pending.push({\n      id: result.id,\n      doc: result.doc,\n      rowNumber,\n      rowData: row,\n      ...(result.warnings && result.warnings.length > 0 && { warnings: result.warnings }),\n      ...(result.pairs && result.pairs.length > 0 && { pairs: result.pairs }),\n      ...(result.priceSample !== undefined && { priceSample: result.priceSample }),\n    })\n    if (pending.length >= batchSize) {\n      await flush()\n    }\n  }\n\n  await flush()\n\n  // Phase 2 (commerce-demo PR7): attach the buffered `*S` core-charge patches onto\n  // their PARENT docs. Runs AFTER the whole stream so every parent that exists in\n  // this run is already indexed; ES `update` is realtime by `_id`, so the merge\n  // lands regardless of whether the parent streamed before or after its `*S` row\n  // (order-independent). Patches are sparse (only reman parts carry `*S`), but batch\n  // anyway for a giant catalogue. A patch whose parent isn't in the index comes back\n  // as a per-doc failure (`document_missing_exception`) → an `orphan_core_charge`\n  // warning, never a phantom doc, never a silent drop (D004). A cluster failure of\n  // the whole batch is logged as a warning too — the saleable docs already indexed;\n  // only the deferred core-charge attach didn't happen — so we never fail the run\n  // over it.\n  let patchesApplied = 0\n  let patchesOrphaned = 0\n  for (let i = 0; i < collectedPatches.length; i += batchSize) {\n    // Operator-driven cancel BETWEEN batches: THROW rather than break, so a run\n    // cancelled after phase 1 is not marked `succeeded` with core charges left\n    // unattached. Mirrors phase-1's flush, which also re-throws on abort. The queue\n    // handler catches this and marks the run failed.\n    if (signal?.aborted) {\n      throw new Error('imports: run aborted during core-charge attach (phase 2)')\n    }\n    const slice = collectedPatches.slice(i, i + batchSize)\n    let updateResult: BulkIndexResult\n    try {\n      updateResult = await bulkPartialUpdate<TDoc>(esClient, {\n        index: esIndex,\n        updates: slice.map(({ id, doc }) => ({ id, doc })),\n        ...(signal !== undefined && { signal }),\n      })\n    } catch (err) {\n      // An abort surfacing as an AbortError from the ES client is a cancel, NOT an\n      // attach failure — re-throw so the run fails (same as phase 1), never a silent\n      // \"succeeded with patches unapplied\".\n      const isAbort =\n        signal?.aborted ??\n        (err instanceof Error && (err.name === 'AbortError' || /abort/i.test(err.message)))\n      if (isAbort) throw err\n      // Whole-batch transport/cluster failure (ES down, timeout). The parents may\n      // well exist — the attach simply didn't happen — so this is an INFRA failure,\n      // NOT a data orphan: it gets the distinct `patch_attach_failed` bucket and is\n      // NOT counted in patchesOrphaned. Best-effort: the saleable docs already\n      // indexed, and a re-import re-attaches, so we warn + continue rather than fail.\n      const reason = err instanceof Error ? err.message : String(err)\n      for (const patch of slice) {\n        warningTracker.addError(-1, 'patch_attach_failed', reason, patch.orphanWarning?.field, {\n          target: patch.id,\n        })\n      }\n      continue\n    }\n    patchesApplied += updateResult.succeeded\n    if (updateResult.failures.length > 0) {\n      const byId = new Map(slice.map((p) => [p.id, p]))\n      for (const fail of updateResult.failures) {\n        const w = byId.get(fail.id)?.orphanWarning\n        // A missing target (`document_missing_exception` / 404) is a genuine data\n        // orphan — the referenced parent isn't in this import (D004: warn, never a\n        // phantom doc). A stable message (from the patch's orphanWarning) aggregates\n        // orphans; the missing parent id rides in the sample. Any OTHER per-doc\n        // failure (mapper error, etc.) is an infra/attach failure kept in a distinct\n        // bucket so it can't masquerade as a data orphan.\n        const isOrphan = fail.status === 404 || fail.type === 'document_missing_exception'\n        if (isOrphan) {\n          patchesOrphaned += 1\n          warningTracker.addError(\n            -1,\n            w?.warningType ?? 'orphan_patch',\n            w?.message ?? fail.reason,\n            w?.field,\n            {\n              target: fail.id,\n            },\n          )\n        } else {\n          warningTracker.addError(-1, 'patch_attach_failed', fail.reason, w?.field, {\n            target: fail.id,\n          })\n        }\n      }\n    }\n  }\n\n  const totalSec = (Date.now() - startedAt) / 1000\n  logger?.info(\n    {\n      importRunId,\n      rowsRead,\n      rowsSucceeded,\n      rowsFailed,\n      rowsSkipped,\n      batchesCompleted,\n      patchesApplied,\n      patchesOrphaned,\n      totalSec: Math.round(totalSec * 10) / 10,\n      rowsPerSecond: totalSec > 0 ? Math.round(rowsRead / totalSec) : 0,\n    },\n    'imports.runner: finished',\n  )\n\n  return {\n    rowsRead,\n    rowsSucceeded,\n    rowsFailed,\n    rowsSkipped,\n    rowsWarned,\n    batchesCompleted,\n    errors: errorTracker.snapshot(),\n    warnings: warningTracker.snapshot(),\n    pairs: collectedPairs,\n    patchesApplied,\n    patchesOrphaned,\n    priceStats: priceStatsCollector.snapshot(),\n  }\n}\n"],"mappings":"mIAmEA,MAAa,EAAuB,GACvB,EAAkC,EAY/C,IAAa,EAAb,KAA0B,CACxB,SAA4B,IAAI,IAChC,YACA,qBACA,kBAA4B,EAE5B,YAAY,EAA6B,CAAC,EAAG,CAC3C,KAAK,YAAc,EAAO,aAAA,GAC1B,KAAK,qBAAuB,EAAO,sBAAA,CACrC,CAWA,SACE,EACA,EACA,EACA,EACA,EACM,CAEN,IAAM,EAAY,GAAG,EAAU,GADd,GAAS,UACiB,GAAG,IAE1C,EAAU,KAAK,SAAS,IAAI,CAAS,EACzC,GAAI,CAAC,EAAS,CACZ,GAAI,KAAK,SAAS,MAAQ,KAAK,YAAa,CAC1C,KAAK,mBAAqB,EAC1B,MACF,CACA,EAAU,CACR,YACA,MAAO,GAAS,KAChB,UACA,MAAO,EACP,gBAAiB,EACjB,eAAgB,EAChB,QAAS,CAAC,CACZ,EACA,KAAK,SAAS,IAAI,EAAW,CAAO,CACtC,CAEA,EAAQ,OAAS,EACjB,EAAQ,eAAiB,EACrB,EAAQ,QAAQ,OAAS,KAAK,sBAChC,EAAQ,QAAQ,KAAK,CAAE,YAAW,SAAQ,CAAC,CAE/C,CAGA,oBAA6B,CAC3B,IAAI,EAAM,EACV,IAAK,IAAM,KAAK,KAAK,SAAS,OAAO,EAAG,GAAO,EAAE,MACjD,OAAO,CACT,CAGA,yBAAkC,CAChC,OAAO,KAAK,SAAS,IACvB,CAOA,0BAAmC,CACjC,OAAO,KAAK,iBACd,CAOA,gBAAiC,CAC/B,IAAM,EAAM,MAAM,KAAK,KAAK,SAAS,OAAO,CAAC,CAAC,CAAC,MAAM,EAAG,IAAM,EAAE,MAAQ,EAAE,KAAK,EACzE,EAAc,EAAI,QAAQ,EAAK,IAAM,EAAM,EAAE,MAAO,CAAC,EAC3D,OAAO,EAAI,IAAK,IAAO,CACrB,UAAW,EAAE,UACb,MAAO,EAAE,MACT,QAAS,EAAE,QACX,MAAO,EAAE,MACT,gBAAiB,EAAE,gBACnB,eAAgB,EAAE,eAClB,QAAS,EAAE,QAAQ,MAAM,EACzB,WAAY,EAAc,EAAK,EAAE,MAAQ,EAAe,IAAM,CAChE,EAAE,CACJ,CAOA,UAAiC,CAC/B,MAAO,CACL,YAAa,KAAK,mBAAmB,EACrC,iBAAkB,KAAK,wBAAwB,EAC/C,kBAAmB,KAAK,kBACxB,SAAU,KAAK,eAAe,CAChC,CACF,CACF,EAoBA,SAAgB,EACd,EACA,EACsB,CACtB,IAAM,EAAc,EAAE,YAAc,EAAE,YAOhC,EAAW,IAAmC,CAClD,GAAG,EACH,WAAY,EAAc,EAAK,EAAE,MAAQ,EAAe,IAAM,CAChE,GACA,MAAO,CACL,cACA,iBAAkB,EAAE,iBAAmB,EAAE,iBACzC,kBAAmB,EAAE,kBAAoB,EAAE,kBAC3C,SAAU,CAAC,GAAG,EAAE,SAAS,IAAI,CAAO,EAAG,GAAG,EAAE,SAAS,IAAI,CAAO,CAAC,CACnE,CACF,CC1LA,MAAa,EAAgC,CAC3C,MAAO,EACP,SAAU,EACV,IAAK,KACL,OAAQ,KACR,IAAK,IACP,EAca,EAAmB,IAQhC,IAAa,EAAb,KAAiC,CAC/B,MAAgB,EAChB,UAAoB,EACpB,YAAsB,EACtB,QAAqC,CAAC,EAmBtC,OAAO,EAAiC,CAEtC,GADA,KAAK,OAAS,EACV,IAAU,IAAA,GAAW,CACvB,KAAK,WAAa,EAClB,MACF,CAEA,GADA,KAAK,aAAe,EAChB,KAAK,QAAQ,OAAA,IAA2B,CAC1C,KAAK,QAAQ,KAAK,CAAK,EACvB,MACF,CACA,IAAM,EAAI,KAAK,MAAM,KAAK,OAAO,EAAI,KAAK,WAAW,EACjD,EAAA,MAAsB,KAAK,QAAQ,GAAK,EAC9C,CAEA,UAAuB,CACrB,GAAI,KAAK,QAAU,EAAG,MAAO,CAAE,GAAG,CAAkB,EACpD,IAAM,EAAW,KAAK,UAAY,KAAK,MACvC,GAAI,KAAK,QAAQ,SAAW,EAC1B,MAAO,CAAE,MAAO,KAAK,MAAO,WAAU,IAAK,KAAM,OAAQ,KAAM,IAAK,IAAK,EAE3E,IAAM,EAAS,CAAC,GAAG,KAAK,OAAO,CAAC,CAAC,MAAM,EAAG,IAAM,EAAI,CAAC,EAC/C,EAAM,EAAO,IAAM,KACnB,EAAM,EAAO,EAAO,OAAS,IAAM,KACnC,EAAM,KAAK,MAAM,EAAO,OAAS,CAAC,EACpC,EACJ,GAAI,EAAO,OAAS,GAAM,EAAG,CAC3B,IAAM,EAAK,EAAO,EAAM,GAClB,EAAK,EAAO,GAClB,EAAS,IAAO,IAAA,IAAa,IAAO,IAAA,IAAa,EAAK,GAAM,EAAI,IAClE,KACE,GAAS,EAAO,IAAQ,KAE1B,MAAO,CAAE,MAAO,KAAK,MAAO,WAAU,MAAK,SAAQ,KAAI,CACzD,CACF,EAoBA,MAAa,EAAqC,GAkBlD,SAAgB,EAAmB,EAAkB,EAAuC,CAK1F,GAJI,EAAK,SAAW,MAAQ,EAAK,SAAW,MAIxC,EAAK,QAAU,GAAK,EAAK,QAAU,EAAG,OAAO,KACjD,IAAM,EAAQ,EAAK,OAAS,EAAK,OAOjC,OALE,GAAA,IACA,GAAS,EAAA,GAEF,CAAE,QAAO,WAAY,EAAK,OAAQ,WAAY,EAAK,MAAO,EAE5D,IACT,CAUA,SAAgB,EAAgB,EAAmC,CACjE,GAAI,OAAO,GAAU,WAAY,EAAgB,OAAO,KACxD,IAAM,EAAI,EAEV,OADI,OAAO,EAAE,OAAU,UAAY,OAAO,EAAE,UAAa,SAAiB,KACnE,CACL,MAAO,EAAE,MACT,SAAU,EAAE,SACZ,IAAK,OAAO,EAAE,KAAQ,SAAW,EAAE,IAAM,KACzC,OAAQ,OAAO,EAAE,QAAW,SAAW,EAAE,OAAS,KAClD,IAAK,OAAO,EAAE,KAAQ,SAAW,EAAE,IAAM,IAC3C,CACF,CChKA,MAAa,EAAqB,IAoKlC,eAAsB,EAAgB,EAA2D,CAC/F,GAAM,CACJ,cACA,WACA,SACA,YACA,OACA,WACA,UACA,YAAY,EACZ,aACA,WACA,SACA,eAAe,IAAI,EACnB,iBAAiB,IAAI,EACrB,WAAY,EAAsB,IAAI,EACtC,UACE,EAEJ,GAAI,EAAY,EACd,MAAU,MAAM,+BAA+B,EAAU,EAAE,EAG7D,IAAM,EAAY,KAAK,IAAI,EAC3B,GAAQ,KAAK,CAAE,cAAa,WAAU,YAAW,SAAQ,EAAG,yBAAyB,EACrF,IAAI,EAAW,EACX,EAAgB,EAChB,EAAa,EACb,EAAc,EACd,EAAa,EACb,EAAmB,EAMjB,EAAoC,CAAC,EAOrC,EAAqC,CAAC,EAOxC,EAQC,CAAC,EAEA,EAAQ,SAA2B,CACvC,GAAI,EAAQ,SAAW,EAAG,OAC1B,IAAM,EAAQ,EACd,EAAU,CAAC,EACX,IAAM,EAAiB,KAAK,IAAI,EAC5B,EACJ,GAAI,CACF,EAAS,MAAM,EAAiB,EAAU,CACxC,MAAO,EACP,KAAM,EAAM,KAAK,CAAE,KAAI,UAAW,CAAE,KAAI,KAAI,EAAE,EAC9C,GAAI,IAAW,IAAA,IAAa,CAAE,QAAO,CACvC,CAAC,CACH,OAAS,EAAK,CAQZ,IAAM,EAFJ,GAAQ,UACP,aAAe,QAAU,EAAI,OAAS,cAAgB,SAAS,KAAK,EAAI,OAAO,IACtD,UAAY,sBAClC,EAAS,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,EAC9D,IAAK,GAAM,CAAE,eAAe,EAC1B,EAAa,SAAS,EAAW,EAAW,EAAQ,IAAA,GAAW,IAAI,EAGrE,KADA,IAAc,EAAM,OACd,CACR,CAEA,GAAiB,EAAO,UACxB,GAAc,EAAO,SAAS,OAC9B,GAAoB,EAEpB,IAAM,EAAU,KAAK,IAAI,EAAI,EACvB,GAAc,KAAK,IAAI,EAAI,GAAa,IAe9C,GAdA,GAAQ,KACN,CACE,cACA,MAAO,EACP,UACA,UAAW,EAAM,OACjB,UAAW,EAAO,UAClB,OAAQ,EAAO,SAAS,OACxB,WACA,cAAe,EAAa,EAAI,KAAK,MAAM,EAAW,CAAU,EAAI,CACtE,EACA,+BACF,EAEI,EAAO,SAAS,OAAS,EAAG,CAK9B,IAAM,EAAO,IAAI,IAAI,EAAM,IAAK,GAAM,CAAC,EAAE,GAAI,EAAE,SAAS,CAAC,CAAC,EAC1D,IAAK,IAAM,KAAQ,EAAO,SAAU,CAClC,IAAM,EAAY,EAAK,IAAI,EAAK,EAAE,GAAK,GACvC,EAAa,SAAS,EAAW,EAAK,KAAM,EAAK,OAAQ,IAAA,GAAW,CAAE,GAAI,EAAK,EAAG,CAAC,CACrF,CACF,CAKA,IAAM,EAAY,IAAI,IAAI,EAAO,SAAS,IAAK,GAAM,EAAE,EAAE,CAAC,EAC1D,IAAK,IAAM,KAAQ,EACb,MAAC,EAAK,UAAY,EAAK,SAAS,SAAW,IAC3C,GAAU,IAAI,EAAK,EAAE,EACzB,IAAc,EACd,IAAK,IAAM,KAAK,EAAK,SACnB,EAAe,SAAS,EAAK,UAAW,EAAE,YAAa,EAAE,QAAS,EAAE,MAAO,EAAK,OAAO,CAF3E,CAQhB,IAAK,IAAM,KAAQ,EACb,CAAC,EAAK,OAAS,EAAK,MAAM,SAAW,GACrC,EAAU,IAAI,EAAK,EAAE,GACzB,EAAe,KAAK,GAAG,EAAK,KAAK,EAOnC,IAAK,IAAM,KAAQ,EACb,EAAU,IAAI,EAAK,EAAE,GACzB,EAAoB,OAAO,EAAK,WAAW,EAG7C,GAAI,EAAY,CACd,IAAM,GAAkB,KAAK,IAAI,EAAI,GAAa,IAClD,EAAW,CACT,WACA,gBACA,aACA,cACA,aACA,mBACA,iBACA,cAAe,EAAiB,EAAI,EAAW,EAAiB,EAChE,sBAAuB,EAAa,wBAAwB,CAC9D,CAAC,CACH,CACF,EAOM,EAAoB,EAMpB,EAAmB,IAAI,IAE7B,UAAW,GAAM,CAAE,YAAW,SAAS,EAAS,CAAI,EAAG,CAErD,GADI,GAAQ,SACR,IAAa,IAAA,IAAa,GAAY,EAAU,MASpD,GARA,GAAY,EAER,IAAa,GACf,GAAQ,KACN,CAAE,cAAa,YAAW,WAAY,OAAO,KAAK,CAAG,CAAE,EACvD,kCACF,EAEE,EAAW,IAAsB,EAAG,CACtC,IAAM,GAAc,KAAK,IAAI,EAAI,GAAa,IAS9C,GAAQ,KACN,CACE,cACA,WACA,gBACA,aACA,cACA,QAAS,EAAQ,OACjB,WAAY,KAAK,MAAM,EAAa,EAAE,EAAI,GAC1C,cAAe,EAAa,EAAI,KAAK,MAAM,EAAW,CAAU,EAAI,EACpE,wBAAyB,EAAiB,IAC5C,EACA,6BACF,CACF,CAEA,IAAM,EAAwB,CAAE,cAAa,SAAQ,WAAU,WAAU,EACrE,EACJ,GAAI,CACF,EAAS,MAAM,EAAU,EAAK,CAAG,CACnC,OAAS,EAAK,CAKZ,IAAM,EAAS,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,EAC9D,EAAa,SAAS,EAAW,kBAAmB,EAAQ,IAAA,GAAW,CAAG,EACrE,EAAiB,IAAI,iBAAiB,IACzC,EAAiB,IAAI,kBAAmB,CAAE,YAAW,QAAS,CAAO,CAAC,EAKtE,GAAQ,KACN,CAAE,cAAa,YAAW,UAAW,kBAAmB,QAAS,CAAO,EACxE,6CACF,GAEF,GAAc,EACd,QACF,CAEA,GAAI,EAAO,OAAS,OAAQ,CAC1B,GAAe,EAKX,EAAO,SAAW,EAAO,QAAQ,OAAS,GAC5C,EAAiB,KAAK,GAAG,EAAO,OAAO,EAEzC,QACF,CACA,GAAI,EAAO,OAAS,QAAS,CAC3B,EAAa,SACX,EACA,EAAO,MAAM,UACb,EAAO,MAAM,QACb,EAAO,MAAM,MACb,CACF,EACK,EAAiB,IAAI,EAAO,MAAM,SAAS,IAC9C,EAAiB,IAAI,EAAO,MAAM,UAAW,CAC3C,YACA,QAAS,EAAO,MAAM,OACxB,CAAC,EACD,GAAQ,KACN,CACE,cACA,YACA,UAAW,EAAO,MAAM,UACxB,MAAO,EAAO,MAAM,MACpB,QAAS,EAAO,MAAM,OACxB,EACA,+BAA+B,EAAO,MAAM,WAC9C,GAEF,GAAc,EACd,QACF,CAOA,EAAQ,KAAK,CACX,GAAI,EAAO,GACX,IAAK,EAAO,IACZ,YACA,QAAS,EACT,GAAI,EAAO,UAAY,EAAO,SAAS,OAAS,GAAK,CAAE,SAAU,EAAO,QAAS,EACjF,GAAI,EAAO,OAAS,EAAO,MAAM,OAAS,GAAK,CAAE,MAAO,EAAO,KAAM,EACrE,GAAI,EAAO,cAAgB,IAAA,IAAa,CAAE,YAAa,EAAO,WAAY,CAC5E,CAAC,EACG,EAAQ,QAAU,GACpB,MAAM,EAAM,CAEhB,CAEA,MAAM,EAAM,EAaZ,IAAI,EAAiB,EACjB,EAAkB,EACtB,IAAK,IAAI,EAAI,EAAG,EAAI,EAAiB,OAAQ,GAAK,EAAW,CAK3D,GAAI,GAAQ,QACV,MAAU,MAAM,0DAA0D,EAE5E,IAAM,EAAQ,EAAiB,MAAM,EAAG,EAAI,CAAS,EACjD,EACJ,GAAI,CACF,EAAe,MAAM,EAAwB,EAAU,CACrD,MAAO,EACP,QAAS,EAAM,KAAK,CAAE,KAAI,UAAW,CAAE,KAAI,KAAI,EAAE,EACjD,GAAI,IAAW,IAAA,IAAa,CAAE,QAAO,CACvC,CAAC,CACH,OAAS,EAAK,CAOZ,GAFE,GAAQ,UACP,aAAe,QAAU,EAAI,OAAS,cAAgB,SAAS,KAAK,EAAI,OAAO,IACrE,MAAM,EAMnB,IAAM,EAAS,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,EAC9D,IAAK,IAAM,KAAS,EAClB,EAAe,SAAS,GAAI,sBAAuB,EAAQ,EAAM,eAAe,MAAO,CACrF,OAAQ,EAAM,EAChB,CAAC,EAEH,QACF,CAEA,GADA,GAAkB,EAAa,UAC3B,EAAa,SAAS,OAAS,EAAG,CACpC,IAAM,EAAO,IAAI,IAAI,EAAM,IAAK,GAAM,CAAC,EAAE,GAAI,CAAC,CAAC,CAAC,EAChD,IAAK,IAAM,KAAQ,EAAa,SAAU,CACxC,IAAM,EAAI,EAAK,IAAI,EAAK,EAAE,CAAC,EAAE,cAOZ,EAAK,SAAW,KAAO,EAAK,OAAS,8BAEpD,GAAmB,EACnB,EAAe,SACb,GACA,GAAG,aAAe,eAClB,GAAG,SAAW,EAAK,OACnB,GAAG,MACH,CACE,OAAQ,EAAK,EACf,CACF,GAEA,EAAe,SAAS,GAAI,sBAAuB,EAAK,OAAQ,GAAG,MAAO,CACxE,OAAQ,EAAK,EACf,CAAC,CAEL,CACF,CACF,CAEA,IAAM,GAAY,KAAK,IAAI,EAAI,GAAa,IAiB5C,OAhBA,GAAQ,KACN,CACE,cACA,WACA,gBACA,aACA,cACA,mBACA,iBACA,kBACA,SAAU,KAAK,MAAM,EAAW,EAAE,EAAI,GACtC,cAAe,EAAW,EAAI,KAAK,MAAM,EAAW,CAAQ,EAAI,CAClE,EACA,0BACF,EAEO,CACL,WACA,gBACA,aACA,cACA,aACA,mBACA,OAAQ,EAAa,SAAS,EAC9B,SAAU,EAAe,SAAS,EAClC,MAAO,EACP,iBACA,kBACA,WAAY,EAAoB,SAAS,CAC3C,CACF"}

package/dist/schema-registry-DB5aIZMN.mjs

@@ -0,0 +1,2 @@
+import{a as e,i as t,r as n}from"./parts-schema-QqLci_Pl.mjs";const r={name:`discount_table`,fields:[{key:`discount_group`,label:`Discount group`,required:!0,type:`string`},{key:`discount_pct`,label:`Discount %`,required:!0,type:`number`}],buildRow(n,r){let i=typeof n.discount_group==`string`?n.discount_group.trim():``;if(i===``)return e(`missing_discount_group`,`discount_group is empty`,`discount_group`);let a=t(n.discount_pct);return a.kind===`invalid`?e(`invalid_discount_pct`,`discount_pct is not a number`,`discount_pct`):a.kind===`absent`?e(`missing_discount_pct`,`discount_pct is empty`,`discount_pct`):a.value<0||a.value>100?e(`discount_pct_out_of_range`,`discount_pct ${a.value} is outside 0–100`,`discount_pct`):{kind:`success`,id:i,doc:{discountGroup:i,discountPct:a.value}}}},i={[n.name]:n,[r.name]:r};function a(e){return i[e]}function o(){return Object.keys(i)}function s(){return Object.values(i).map(e=>({name:e.name,fields:e.fields}))}export{r as i,a as n,o as r,s as t};
+//# sourceMappingURL=schema-registry-DB5aIZMN.mjs.map

package/dist/schema-registry-DB5aIZMN.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"schema-registry-DB5aIZMN.mjs","names":[],"sources":["../src/mapping/discount-table-schema.ts","../src/mapping/schema-registry.ts"],"sourcesContent":["/**\n * `discount_table` target schema — the non-parts feed kind (commerce-demo phase\n * 02 / 03). A discount table (R03 Sample B: `Code,Discount`) maps to\n * `{ discountGroup, discountPct }` rows.\n *\n * PR2 ships the schema (whitelist + `buildRow`) so the engine can parse + validate\n * a discount feed; the PERSISTENCE sink (upserting `supplier_discount_group` rows\n * + the two-file purchase-price derivation it feeds) lands in PR4. The `id` is the\n * discount group token so a re-import idempotently overwrites.\n */\n\nimport {\n  type BuildRowContext,\n  type MappedRow,\n  numberCell,\n  rowError,\n  type TargetField,\n  type TargetSchema,\n} from './target-schema.js'\n\n/** One parsed discount-table row: a group token + the purchase discount % off list. */\nexport interface DiscountTableRow {\n  discountGroup: string\n  discountPct: number\n}\n\nconst DISCOUNT_TABLE_FIELDS: readonly TargetField[] = [\n  { key: 'discount_group', label: 'Discount group', required: true, type: 'string' },\n  { key: 'discount_pct', label: 'Discount %', required: true, type: 'number' },\n]\n\nexport const discountTableSchema: TargetSchema<DiscountTableRow> = {\n  name: 'discount_table',\n  fields: DISCOUNT_TABLE_FIELDS,\n  // params unused — a discount table is supplier-scoped at the run level, not per row.\n  buildRow(mapped: MappedRow, _ctx: BuildRowContext) {\n    // Group tokens are single tokens INCLUDING `'0'` (R03 Sample B) — a non-empty\n    // string check, so `'0'` is a valid group.\n    const group = typeof mapped.discount_group === 'string' ? mapped.discount_group.trim() : ''\n    if (group === '') {\n      return rowError('missing_discount_group', 'discount_group is empty', 'discount_group')\n    }\n\n    const pct = numberCell(mapped.discount_pct)\n    if (pct.kind === 'invalid') {\n      return rowError('invalid_discount_pct', 'discount_pct is not a number', 'discount_pct')\n    }\n    if (pct.kind === 'absent') {\n      return rowError('missing_discount_pct', 'discount_pct is empty', 'discount_pct')\n    }\n    if (pct.value < 0 || pct.value > 100) {\n      return rowError(\n        'discount_pct_out_of_range',\n        `discount_pct ${pct.value} is outside 0–100`,\n        'discount_pct',\n      )\n    }\n\n    return { kind: 'success', id: group, doc: { discountGroup: group, discountPct: pct.value } }\n  },\n}\n","/**\n * Registry of the concrete target schemas (commerce-demo phase 02, D005).\n *\n * The engine resolves a mapping's `target_schema` name to its `TargetSchema` here.\n * An unregistered name is rejected (deny-by-default — the first injection-boundary\n * whitelist; the second is each schema's field set).\n */\n\nimport { discountTableSchema } from './discount-table-schema.js'\nimport { partsSchema } from './parts-schema.js'\nimport type { TargetField, TargetSchema } from './target-schema.js'\n\n/**\n * `TargetSchema<TOut>` is covariant in `TOut` (it appears only in `buildRow`'s\n * return), so a concrete schema is assignable to `TargetSchema<unknown>` — the\n * shape callers that don't know the output type (the `commerce:mapping` transform)\n * consume. Callers that DO know the type (the carmaker wrapper) import the concrete\n * schema directly to keep `RowTransform<PartsDocument>`.\n */\nconst SCHEMAS: Readonly<Record<string, TargetSchema<unknown>>> = {\n  [partsSchema.name]: partsSchema,\n  [discountTableSchema.name]: discountTableSchema,\n}\n\n/** Resolve a registered schema by name, or `undefined` if not registered. */\nexport function getTargetSchema(name: string): TargetSchema<unknown> | undefined {\n  return SCHEMAS[name]\n}\n\n/** All registered schema names (for the wizard's schema picker, PR3). */\nexport function listTargetSchemas(): string[] {\n  return Object.keys(SCHEMAS)\n}\n\n/** Operator-facing summary of a target schema: its name + the field whitelist (PR3 wizard). */\nexport interface TargetSchemaSummary {\n  name: string\n  fields: readonly TargetField[]\n}\n\n/**\n * Describe every registered target schema (name + field whitelist) for the PR3\n * mapping-wizard UI. Returns only metadata (no `buildRow`), so the wizard's RSC page\n * can ship it across the client boundary as plain JSON without dragging the engine\n * into the client bundle.\n */\nexport function describeTargetSchemas(): TargetSchemaSummary[] {\n  return Object.values(SCHEMAS).map((s) => ({ name: s.name, fields: s.fields }))\n}\n"],"mappings":"8DA+BA,MAAa,EAAsD,CACjE,KAAM,iBACN,OAAQ,CANR,CAAE,IAAK,iBAAkB,MAAO,iBAAkB,SAAU,GAAM,KAAM,QAAS,EACjF,CAAE,IAAK,eAAgB,MAAO,aAAc,SAAU,GAAM,KAAM,QAAS,CAKnE,EAER,SAAS,EAAmB,EAAuB,CAGjD,IAAM,EAAQ,OAAO,EAAO,gBAAmB,SAAW,EAAO,eAAe,KAAK,EAAI,GACzF,GAAI,IAAU,GACZ,OAAO,EAAS,yBAA0B,0BAA2B,gBAAgB,EAGvF,IAAM,EAAM,EAAW,EAAO,YAAY,EAe1C,OAdI,EAAI,OAAS,UACR,EAAS,uBAAwB,+BAAgC,cAAc,EAEpF,EAAI,OAAS,SACR,EAAS,uBAAwB,wBAAyB,cAAc,EAE7E,EAAI,MAAQ,GAAK,EAAI,MAAQ,IACxB,EACL,4BACA,gBAAgB,EAAI,MAAM,mBAC1B,cACF,EAGK,CAAE,KAAM,UAAW,GAAI,EAAO,IAAK,CAAE,cAAe,EAAO,YAAa,EAAI,KAAM,CAAE,CAC7F,CACF,ECzCM,EAA2D,EAC9D,EAAY,MAAO,GACnB,EAAoB,MAAO,CAC9B,EAGA,SAAgB,EAAgB,EAAiD,CAC/E,OAAO,EAAQ,EACjB,CAGA,SAAgB,GAA8B,CAC5C,OAAO,OAAO,KAAK,CAAO,CAC5B,CAcA,SAAgB,GAA+C,CAC7D,OAAO,OAAO,OAAO,CAAO,CAAC,CAAC,IAAK,IAAO,CAAE,KAAM,EAAE,KAAM,OAAQ,EAAE,MAAO,EAAE,CAC/E"}

package/dist/storage-resolver.d.mts

@@ -0,0 +1,32 @@
+import { r as FilePathResolver } from "./worker-CRH_PjKO.mjs";
+
+//#region src/storage-resolver.d.ts
+/**
+ * Resolve a storage key to a path the streaming reader can open.
+ *
+ * Two paths:
+ *
+ * 1. **Local-disk shortcut** — when the configured adapter implements
+ *    the optional `getLocalPath(key)` method (LocalDiskAdapter does;
+ *    R2Adapter does not), the bytes are already on a filesystem
+ *    visible to this process. Return that path directly with **no
+ *    cleanup** (the file is the canonical storage object — the
+ *    cleanup callback's job is to remove the *copy* we made into
+ *    /tmp; with no copy there's nothing to remove). This eliminates
+ *    the upload→R2→download→tmpfile round trip on single-VPS / dev
+ *    setups where it's pure overhead — a 137 MB feed that took ~25s
+ *    to round-trip completes in the time it takes to `fs.access` the
+ *    path.
+ *
+ * 2. **Remote-storage path** (R2/S3/etc.) — download the object into
+ *    a tmpfile and return that. The caller is required to await the
+ *    `cleanup` so the tmpfile doesn't outlive the run.
+ *
+ * Lazy-imports `@murumets-ee/storage` + `@murumets-ee/core` so
+ * deployments that don't ingest from remote storage never load the
+ * R2 client.
+ */
+declare const storageResolveFilePath: FilePathResolver;
+//#endregion
+export { storageResolveFilePath };
+//# sourceMappingURL=storage-resolver.d.mts.map

package/dist/storage-resolver.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"storage-resolver.d.mts","names":[],"sources":["../src/storage-resolver.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;cA6Da,sBAAA,EAAwB,gBAgDpC"}

package/dist/storage-resolver.mjs

@@ -0,0 +1,2 @@
+import{getImportsStorage as e}from"./storage.mjs";import{promises as t}from"node:fs";import{tmpdir as n}from"node:os";import{join as r}from"node:path";const i=async i=>{let a=await e(),o=await a.getAdapterLocalPath(i);if(o)return{localPath:o};let{body:s}=await a.download(i),c=Buffer.isBuffer(s)?s:Buffer.from(await new Response(s).arrayBuffer()),l=i.split(`/`).pop()??`feed`,u=r(n(),`imports-${crypto.randomUUID()}-${l}`);return await t.writeFile(u,c),{localPath:u,cleanup:async()=>{try{await t.unlink(u)}catch(e){if(e.code!==`ENOENT`)throw e}}}};export{i as storageResolveFilePath};
+//# sourceMappingURL=storage-resolver.mjs.map

package/dist/storage-resolver.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"storage-resolver.mjs","names":["fs"],"sources":["../src/storage-resolver.ts"],"sourcesContent":["/**\n * Storage-backed `resolveFilePath` for the imports plugin. Downloads\n * an object out of `@murumets-ee/storage` to a tmpfile so the queue\n * worker can stream it via `node:fs.createReadStream`.\n *\n * Wire into the imports plugin like this:\n *\n * ```ts\n * import { imports } from '@murumets-ee/imports/plugin'\n * import { storageResolveFilePath } from '@murumets-ee/imports/storage-resolver'\n *\n * imports({\n *   esClient: () => esClient,\n *   resolveFilePath: storageResolveFilePath,\n * })\n * ```\n *\n * Why this lives here rather than in the route or in the storage\n * package:\n * - The route shouldn't know about the consumer's runner; it just\n *   persists a key on `import_run.filePath`.\n * - The storage package is generic (no opinion on imports).\n * - The imports worker is the natural download point: it owns the\n *   run's lifecycle and can guarantee tmpfile cleanup.\n *\n * The helper is in a separate subpath so consumers that use a\n * different upload sink (local disk, S3 with a custom adapter, etc.)\n * don't pull in `@murumets-ee/storage` transitively.\n */\n\nimport { promises as fs } from 'node:fs'\nimport { tmpdir } from 'node:os'\nimport { join } from 'node:path'\nimport { getImportsStorage } from './storage.js'\nimport type { FilePathResolver } from './worker.js'\n\n/**\n * Resolve a storage key to a path the streaming reader can open.\n *\n * Two paths:\n *\n * 1. **Local-disk shortcut** — when the configured adapter implements\n *    the optional `getLocalPath(key)` method (LocalDiskAdapter does;\n *    R2Adapter does not), the bytes are already on a filesystem\n *    visible to this process. Return that path directly with **no\n *    cleanup** (the file is the canonical storage object — the\n *    cleanup callback's job is to remove the *copy* we made into\n *    /tmp; with no copy there's nothing to remove). This eliminates\n *    the upload→R2→download→tmpfile round trip on single-VPS / dev\n *    setups where it's pure overhead — a 137 MB feed that took ~25s\n *    to round-trip completes in the time it takes to `fs.access` the\n *    path.\n *\n * 2. **Remote-storage path** (R2/S3/etc.) — download the object into\n *    a tmpfile and return that. The caller is required to await the\n *    `cleanup` so the tmpfile doesn't outlive the run.\n *\n * Lazy-imports `@murumets-ee/storage` + `@murumets-ee/core` so\n * deployments that don't ingest from remote storage never load the\n * R2 client.\n */\nexport const storageResolveFilePath: FilePathResolver = async (storageKey) => {\n  // Consult the imports-specific storage handle (registered by\n  // `imports({ storage: ... })`). Falls back to the global\n  // `getStorageConfig()` client when no per-imports factory was\n  // wired — preserves the original single-storage behaviour for\n  // consumers that haven't migrated yet.\n  const storage = await getImportsStorage()\n\n  // Local-disk shortcut: the adapter is on the same filesystem.\n  // `getAdapterLocalPath` is the public hook on `StorageClient` that\n  // forwards to `adapter.getLocalPath(key)` when the adapter\n  // implements it; absence of that method is the signal to fall\n  // through to the remote-download path below.\n  const directPath = await storage.getAdapterLocalPath(storageKey)\n  if (directPath) {\n    return { localPath: directPath }\n  }\n\n  const { body } = await storage.download(storageKey)\n\n  // `DownloadResult.body` is `Buffer | ReadableStream<Uint8Array>` per\n  // storage's adapter contract. Normalise to Buffer for the simple\n  // tmpfile-write path; the carmaker feeds top out at ~tens of MB,\n  // well within memory.\n  const buffer = Buffer.isBuffer(body)\n    ? body\n    : Buffer.from(await new Response(body).arrayBuffer())\n\n  // Suffix carries the basename hint (last storage-key segment) so\n  // operator-readable temp paths help debugging without leaking the\n  // full original filename through the FS.\n  const suffix = storageKey.split('/').pop() ?? 'feed'\n  const localPath = join(tmpdir(), `imports-${crypto.randomUUID()}-${suffix}`)\n  await fs.writeFile(localPath, buffer)\n\n  return {\n    localPath,\n    cleanup: async () => {\n      try {\n        await fs.unlink(localPath)\n      } catch (err) {\n        // ENOENT is fine — file already gone (OS cleanup, manual\n        // intervention, etc.). Anything else, rethrow so the worker's\n        // best-effort cleanup logs a warning.\n        if ((err as NodeJS.ErrnoException).code !== 'ENOENT') throw err\n      }\n    },\n  }\n}\n"],"mappings":"uJA6DA,MAAa,EAA2C,KAAO,IAAe,CAM5E,IAAM,EAAU,MAAM,EAAkB,EAOlC,EAAa,MAAM,EAAQ,oBAAoB,CAAU,EAC/D,GAAI,EACF,MAAO,CAAE,UAAW,CAAW,EAGjC,GAAM,CAAE,QAAS,MAAM,EAAQ,SAAS,CAAU,EAM5C,EAAS,OAAO,SAAS,CAAI,EAC/B,EACA,OAAO,KAAK,MAAM,IAAI,SAAS,CAAI,CAAC,CAAC,YAAY,CAAC,EAKhD,EAAS,EAAW,MAAM,GAAG,CAAC,CAAC,IAAI,GAAK,OACxC,EAAY,EAAK,EAAO,EAAG,WAAW,OAAO,WAAW,EAAE,GAAG,GAAQ,EAG3E,OAFA,MAAMA,EAAG,UAAU,EAAW,CAAM,EAE7B,CACL,YACA,QAAS,SAAY,CACnB,GAAI,CACF,MAAMA,EAAG,OAAO,CAAS,CAC3B,OAAS,EAAK,CAIZ,GAAK,EAA8B,OAAS,SAAU,MAAM,CAC9D,CACF,CACF,CACF"}

package/dist/storage.d.mts

@@ -0,0 +1,31 @@
+import { StorageClient } from "@murumets-ee/storage";
+
+//#region src/storage.d.ts
+/** Lazy storage-client factory. May be sync or async. */
+type ImportsStorageFactory = () => StorageClient | Promise<StorageClient>;
+/**
+ * Register a factory for the imports-only `StorageClient`. Called by
+ * `imports({ storage: ... })` during plugin init. Subsequent calls
+ * overwrite — the last `imports()` plugin instance wins, which
+ * matches every other plugin's "second registration replaces first"
+ * behaviour.
+ *
+ * Resets the resolved-client cache so a re-registration in dev (HMR
+ * re-evaluating `lumi.config.ts`) doesn't keep handing out a stale
+ * client built from the previous factory.
+ */
+declare function registerImportsStorageFactory(factory: ImportsStorageFactory): void;
+/**
+ * Resolve the imports `StorageClient`. Returns a cached promise after
+ * the first call within a process. When no factory was registered (no
+ * `imports({ storage })` config), falls back to the global storage
+ * default — keeps existing consumers working without a config bump.
+ */
+declare function getImportsStorage(): Promise<StorageClient>;
+/**
+ * @internal — tests reset the registration to a clean state.
+ */
+declare function _resetImportsStorageRegistration(): void;
+//#endregion
+export { ImportsStorageFactory, _resetImportsStorageRegistration, getImportsStorage, registerImportsStorageFactory };
+//# sourceMappingURL=storage.d.mts.map

package/dist/storage.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"storage.d.mts","names":[],"sources":["../src/storage.ts"],"mappings":";;;;KA2BY,qBAAA,SAA8B,aAAA,GAAgB,OAAA,CAAQ,aAAA;AA+DlE;;;;AAAgD;;;;;;;AAAhD,iBAhCgB,6BAAA,CAA8B,OAA8B,EAArB,qBAAqB;;;;;;;iBAW5D,iBAAA,IAAqB,OAAO,CAAC,aAAA;;;;iBAqB7B,gCAAA"}

package/dist/storage.mjs

@@ -0,0 +1,2 @@
+const e=Symbol.for(`@murumets-ee/imports:storage-factory`),t=Symbol.for(`@murumets-ee/imports:storage-promise`);function n(e){return globalThis[e]}function r(e,t){globalThis[e]=t}function i(n){r(e,n),r(t,void 0)}function a(){let i=n(t);if(i)return i;let a=n(e),o=Promise.resolve(a?a():s()).catch(e=>{throw r(t,void 0),e});return r(t,o),o}function o(){r(e,void 0),r(t,void 0)}async function s(){let{createStorageClient:e}=await import(`@murumets-ee/storage`),{getStorageConfig:t}=await import(`@murumets-ee/storage/plugin`),{getApp:n}=await import(`@murumets-ee/core`);return e(t(),{app:n()})}export{o as _resetImportsStorageRegistration,a as getImportsStorage,i as registerImportsStorageFactory};
+//# sourceMappingURL=storage.mjs.map

package/dist/storage.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"storage.mjs","names":[],"sources":["../src/storage.ts"],"sourcesContent":["/**\n * Per-feature storage handle for the imports flow.\n *\n * Imports lets the consumer plug in a dedicated `StorageClient` (e.g.\n * a local-disk adapter for single-VPS deployments where worker + web\n * share a filesystem and the R2 round-trip is pure overhead) without\n * disturbing the global `getStorageConfig()` default that media,\n * ticketing attachments, and friends keep using.\n *\n * Wiring: the imports plugin's `init` hook stashes the consumer's\n * `storage` factory on a `globalThis` Symbol; `getImportsStorage()`\n * reads it back and caches the resolved client. When no factory is\n * registered (the legacy back-compat path), the function falls\n * through to `createStorageClient(getStorageConfig())` — so existing\n * consumers that only ever wired one storage layer continue to work\n * unchanged.\n *\n * Why a Symbol on `globalThis` and not module state: matches the\n * `@murumets-ee/storage/plugin` pattern. Turbopack can duplicate the\n * same module across chunks, so module-level singletons can leak into\n * separate-instance pairs that disagree about which adapter to use.\n * The Symbol survives that.\n */\n\nimport type { StorageClient } from '@murumets-ee/storage'\n\n/** Lazy storage-client factory. May be sync or async. */\nexport type ImportsStorageFactory = () => StorageClient | Promise<StorageClient>\n\nconst IMPORTS_STORAGE_FACTORY_KEY = Symbol.for('@murumets-ee/imports:storage-factory')\n// Cache the in-flight Promise rather than the resolved client. Two\n// concurrent first-callers both miss the cache, but only the first\n// creates a Promise; the others await the same one. PR #262 review\n// (M-9): caching the resolved value has a TOCTOU window where two\n// callers each invoke the factory and the loser's client is\n// silently discarded — but any in-flight use that captured the\n// loser keeps reading/writing through it.\nconst IMPORTS_STORAGE_PROMISE_KEY = Symbol.for('@murumets-ee/imports:storage-promise')\n\nfunction readGlobal<T>(key: symbol): T | undefined {\n  return (globalThis as Record<symbol, unknown>)[key] as T | undefined\n}\n\nfunction writeGlobal(key: symbol, value: unknown): void {\n  ;(globalThis as Record<symbol, unknown>)[key] = value\n}\n\n/**\n * Register a factory for the imports-only `StorageClient`. Called by\n * `imports({ storage: ... })` during plugin init. Subsequent calls\n * overwrite — the last `imports()` plugin instance wins, which\n * matches every other plugin's \"second registration replaces first\"\n * behaviour.\n *\n * Resets the resolved-client cache so a re-registration in dev (HMR\n * re-evaluating `lumi.config.ts`) doesn't keep handing out a stale\n * client built from the previous factory.\n */\nexport function registerImportsStorageFactory(factory: ImportsStorageFactory): void {\n  writeGlobal(IMPORTS_STORAGE_FACTORY_KEY, factory)\n  writeGlobal(IMPORTS_STORAGE_PROMISE_KEY, undefined)\n}\n\n/**\n * Resolve the imports `StorageClient`. Returns a cached promise after\n * the first call within a process. When no factory was registered (no\n * `imports({ storage })` config), falls back to the global storage\n * default — keeps existing consumers working without a config bump.\n */\nexport function getImportsStorage(): Promise<StorageClient> {\n  const cached = readGlobal<Promise<StorageClient>>(IMPORTS_STORAGE_PROMISE_KEY)\n  if (cached) return cached\n\n  const factory = readGlobal<ImportsStorageFactory>(IMPORTS_STORAGE_FACTORY_KEY)\n  const promise = Promise.resolve(factory ? factory() : defaultStorage()).catch(\n    (err: unknown): never => {\n      // Allow a retry on the next call if the first build failed (e.g.\n      // missing env, transient app-not-ready). Without this, a one-off\n      // failure pins a rejected promise into the cache forever.\n      writeGlobal(IMPORTS_STORAGE_PROMISE_KEY, undefined)\n      throw err\n    },\n  )\n  writeGlobal(IMPORTS_STORAGE_PROMISE_KEY, promise)\n  return promise\n}\n\n/**\n * @internal — tests reset the registration to a clean state.\n */\nexport function _resetImportsStorageRegistration(): void {\n  writeGlobal(IMPORTS_STORAGE_FACTORY_KEY, undefined)\n  writeGlobal(IMPORTS_STORAGE_PROMISE_KEY, undefined)\n}\n\n/**\n * Back-compat fallback — build a storage client from the global\n * `getStorageConfig()` (the old single-storage shape). Lazy-imports so\n * deployments that fully wire imports to a different adapter never\n * load the default factory's transitive dependencies.\n */\nasync function defaultStorage(): Promise<StorageClient> {\n  const { createStorageClient } = await import('@murumets-ee/storage')\n  const { getStorageConfig } = await import('@murumets-ee/storage/plugin')\n  const { getApp } = await import('@murumets-ee/core')\n  return createStorageClient(getStorageConfig(), { app: getApp() })\n}\n"],"mappings":"AA6BA,MAAM,EAA8B,OAAO,IAAI,sCAAsC,EAQ/E,EAA8B,OAAO,IAAI,sCAAsC,EAErF,SAAS,EAAc,EAA4B,CACjD,OAAQ,WAAuC,EACjD,CAEA,SAAS,EAAY,EAAa,EAAsB,CACrD,WAAwC,GAAO,CAClD,CAaA,SAAgB,EAA8B,EAAsC,CAClF,EAAY,EAA6B,CAAO,EAChD,EAAY,EAA6B,IAAA,EAAS,CACpD,CAQA,SAAgB,GAA4C,CAC1D,IAAM,EAAS,EAAmC,CAA2B,EAC7E,GAAI,EAAQ,OAAO,EAEnB,IAAM,EAAU,EAAkC,CAA2B,EACvE,EAAU,QAAQ,QAAQ,EAAU,EAAQ,EAAI,EAAe,CAAC,CAAC,CAAC,MACrE,GAAwB,CAKvB,MADA,EAAY,EAA6B,IAAA,EAAS,EAC5C,CACR,CACF,EAEA,OADA,EAAY,EAA6B,CAAO,EACzC,CACT,CAKA,SAAgB,GAAyC,CACvD,EAAY,EAA6B,IAAA,EAAS,EAClD,EAAY,EAA6B,IAAA,EAAS,CACpD,CAQA,eAAe,GAAyC,CACtD,GAAM,CAAE,uBAAwB,MAAM,OAAO,wBACvC,CAAE,oBAAqB,MAAM,OAAO,+BACpC,CAAE,UAAW,MAAM,OAAO,qBAChC,OAAO,EAAoB,EAAiB,EAAG,CAAE,IAAK,EAAO,CAAE,CAAC,CAClE"}

package/dist/transform-Cu5YEoOl.mjs

@@ -0,0 +1,2 @@
+import{t as e}from"./rolldown-runtime-DK3Fl9T5.mjs";import{i as t,n}from"./engine-B0ExE4jL.mjs";import{n as r,r as i}from"./schema-registry-DB5aIZMN.mjs";import{behavior as a,defineEntity as o,field as s,searchable as c}from"@murumets-ee/entity";import{z as l}from"zod";import{sql as u}from"drizzle-orm";var d=e({IMPORT_RUN_STATUSES:()=>f,ImportRun:()=>p});const f=[`pending`,`running`,`succeeded`,`failed`,`cancelled`],p=o({name:`import_run`,fields:{label:s.text({required:!0,maxLength:255}),status:s.select({options:f,default:`pending`,indexed:!0}),filePath:s.text({required:!0,maxLength:2048}),transformName:s.text({required:!0,maxLength:128,indexed:!0}),params:s.json(),totals:s.json(),errorSummary:s.json(),warningSummary:s.json(),priceStats:s.json(),startedAt:s.date(),finishedAt:s.date(),queueJobId:s.text({maxLength:64,indexed:!0})},behaviors:[a.auditable(),c({fields:[`label`,`transformName`,`queueJobId`,`status`,`filePath`],projection:e=>({id:e.id,label:e.label,description:`${e.transformName} · ${e.status}`})})],scope:`global`,admin:{group:`imports`,label:`Import runs`,labelSingular:`Import run`,icon:`upload`,hideFromMenu:!0}}),m={decimal:`.`,thousands:``},h=l.array(n).max(512),g=l.strictObject({columns:l.array(l.string().min(1).max(256)).min(1).max(256),rows:l.array(l.record(l.string().max(500))).max(5)});function _(e){if(e.headerSnapshot!==void 0&&e.headerSnapshot!==null){let t=g.safeParse(e.headerSnapshot);if(!t.success)throw Error(`Invalid headerSnapshot: ${t.error.issues[0]?.message??`bad shape`}`);let n=JSON.stringify(e.headerSnapshot).length;if(n>1e5)throw Error(`headerSnapshot too large (${n} > 100000 serialized chars) — trim sample rows/cells`)}let n=typeof e.targetSchema==`string`?e.targetSchema:``,a=r(n);if(!a)throw Error(`Unknown target schema "${n}" (registered: ${i().join(`, `)})`);let o=t.safeParse(e.numberFormat);if(!o.success)throw Error(`Invalid numberFormat: ${o.error.issues[0]?.message??`bad shape`}`);let s=h.safeParse(e.columnMap);if(!s.success)throw Error(`Invalid columnMap: ${s.error.issues[0]?.message??`bad shape`}`);let c=JSON.stringify(e.columnMap).length;if(c>1e5)throw Error(`columnMap too large (${c} > 100000 serialized chars)`);let l=new Set(a.fields.map(e=>e.key)),u=new Set,d=new Set;for(let e of s.data){if(!l.has(e.target))throw Error(`Mapping target "${e.target}" is not a field of schema "${a.name}" (allowed: ${[...l].join(`, `)})`);u.has(e.target)&&d.add(e.target),u.add(e.target)}if(d.size>0)throw Error(`Mapping for schema "${a.name}" maps a target more than once: ${[...d].join(`, `)}`);let f=new Set(s.data.map(e=>e.target)),p=a.fields.filter(e=>e.required&&!f.has(e.key)).map(e=>e.key);if(p.length>0)throw Error(`Mapping for schema "${a.name}" is missing required target(s): ${p.join(`, `)}`)}const v={name:`imports:import-mapping-validation`,hooks:{beforeCreate:async e=>(_({targetSchema:e.targetSchema,numberFormat:e.numberFormat??m,columnMap:e.columnMap??[],headerSnapshot:e.headerSnapshot}),y(e)),beforeUpdate:async(e,t,n)=>{let r=await n?.loadCurrent?.()??{};return _({targetSchema:`targetSchema`in t?t.targetSchema:r.targetSchema,numberFormat:`numberFormat`in t?t.numberFormat:r.numberFormat??m,columnMap:`columnMap`in t?t.columnMap:r.columnMap??[],headerSnapshot:`headerSnapshot`in t?t.headerSnapshot:void 0}),y(t)}}};function y(e){return`quote`in e&&e.quote===``?{...e,quote:null}:e}const b=[`parts`,`discount_table`],x=[`csv`,`tsv`,`xlsx`],S=o({name:`import_mapping`,scope:`global`,fields:{supplierId:s.reference({entity:`supplier`,required:!0,indexed:!0}),brandId:s.reference({entity:`brand`,indexed:!0}),targetSchema:s.select({options:b,required:!0,indexed:!0}),name:s.text({required:!0,maxLength:255}),format:s.select({options:x,required:!0}),delimiter:s.text({maxLength:4}),hasHeader:s.boolean({default:!0}),headerRowIndex:s.number({integer:!0,min:0,default:0}),sheet:s.text({maxLength:128}),quote:s.text({maxLength:1}),numberFormat:s.json({default:{decimal:`.`,thousands:``}}),columnMap:s.json({default:[]}),headerSnapshot:s.json(),active:s.boolean({default:!0,indexed:!0})},behaviors:[a.auditable(),c({fields:[`name`,`targetSchema`,`format`],projection:e=>({id:e.id,label:e.name,description:`${e.targetSchema} · ${e.format}`})}),{name:`imports:import-mapping-unique`,constraints:(e,t)=>{if(t!==`main`)return[];let n=e.supplierId,r=e.brandId,i=e.targetSchema;return!n||!r||!i?[]:[{kind:`uniqueIndex`,name:`uq_import_mapping_supplier_brand_schema`,on:[n,r,i],where:u`${r} IS NOT NULL`},{kind:`uniqueIndex`,name:`uq_import_mapping_supplier_schema_default`,on:[n,i],where:u`${r} IS NULL`}]}},{name:`imports:import-mapping-quote`,constraints:(e,t)=>{if(t!==`main`)return[];let n=e.quote;return n?[{kind:`checkConstraint`,name:`ck_import_mapping_quote_single_char`,expression:u`${n} IS NULL OR length(${n}) = 1`}]:[]}},v],admin:{group:`imports`,label:`Import mappings`,labelSingular:`Import mapping`,icon:`table`,hideFromMenu:!0}});var C=class{entries=new Map;register(e,t){if(this.entries.has(e))throw Error(`Transform "${e}" is already registered — two plugins cannot register the same transform name.`);this.entries.set(e,t)}set(e,t){this.entries.set(e,t)}get(e){return this.entries.get(e)}has(e){return this.entries.has(e)}list(){return Array.from(this.entries.keys()).sort()}clear(){this.entries.clear()}};const w=Symbol.for(`@murumets-ee/imports:transforms`);function T(){let e=globalThis,t=e[w];return t||(t=new C,e[w]=t),t}function E(e,t){T().set(e,t)}export{b as a,v as c,d,x as i,f as l,T as n,S as o,E as r,_ as s,C as t,p as u};
+//# sourceMappingURL=transform-Cu5YEoOl.mjs.map