@murumets-ee/imports 0.30.0 → 0.31.0

package/dist/parts-schema-DQeMGPBC.mjs

@@ -0,0 +1,2 @@
+import{z as e}from"zod";const t=/^-?\d+(\.\d+)?$/;function n(e,n){let r=e.trim();if(r===``||r.toUpperCase()===`NA`)return null;let{decimal:i,thousands:a}=n,o=r;if(a!==``&&(o=o.split(a).join(``)),i===`,`){if(o.includes(`.`))return NaN;o=o.split(`,`).join(`.`)}else if(o.includes(`,`))return NaN;return t.test(o)?Number.parseFloat(o):NaN}function r(e){return Math.round(e*100)}function i(e){let n=e.trim();if(n===``||n.toUpperCase()===`NA`)return null;let r=n.includes(`,`),i=n.includes(`.`),a;if(r&&i)a=n.replace(/\./g,``).replace(`,`,`.`);else if(r)a=n.replace(`,`,`.`);else if(i)return NaN;else a=n;return t.test(a)?Number.parseFloat(a):NaN}const a=e.discriminatedUnion(`op`,[e.object({op:e.literal(`trim`)}),e.object({op:e.literal(`uppercase`)}),e.object({op:e.literal(`lowercase`)}),e.object({op:e.literal(`stripPrefix`),value:e.string().min(1).max(32)}),e.object({op:e.literal(`stripSuffix`),value:e.string().min(1).max(32)}),e.object({op:e.literal(`defaultIfEmpty`),value:e.string().max(255)}),e.object({op:e.literal(`number`)}),e.object({op:e.literal(`europeanNumber`)})]);function o(e,t,r){let a=e;for(let e of t)if(typeof a==`string`)switch(e.op){case`trim`:a=a.trim();break;case`uppercase`:a=a.toUpperCase();break;case`lowercase`:a=a.toLowerCase();break;case`stripPrefix`:a=a.startsWith(e.value)?a.slice(e.value.length):a;break;case`stripSuffix`:a=a.endsWith(e.value)?a.slice(0,a.length-e.value.length):a;break;case`defaultIfEmpty`:a=a.trim()===``?e.value:a;break;case`number`:a=n(a,r);break;case`europeanNumber`:a=i(a);break}return a}const s=e.object({source:e.string().min(1).max(256),target:e.string().min(1).max(64),transforms:e.array(a).default([])}),c=e.object({decimal:e.enum([`.`,`,`]),thousands:e.enum([``,`.`,`,`,` `])}),l=e.object({targetSchema:e.string().min(1).max(64),numberFormat:c,columnMap:e.array(s)});function u(e,t){let n=new Set(t.fields.map(e=>e.key));for(let r of e.columnMap)if(!n.has(r.target))throw Error(`Mapping target "${r.target}" is not a field of schema "${t.name}" (allowed: ${[...n].join(`, `)})`);let r=e.numberFormat,i=e.columnMap;return async(e,n)=>{let a={};for(let t of i){let n=e[t.source]??``;a[t.target]=o(n,t.transforms,r)}return t.buildRow(a,{importRunId:n.importRunId,rowNumber:n.rowNumber,params:n.params})}}function d(e,t){let n=e.filter(e=>e.active);if(t!==null){let e=n.find(e=>e.brandId===t);if(e)return e}return n.find(e=>e.brandId===null)}function f(e){if(typeof e!=`string`)return null;let t=e.trim();return t===``?null:t}function p(e){return e==null?{kind:`absent`}:typeof e==`number`?Number.isNaN(e)?{kind:`invalid`}:{kind:`value`,value:e}:e.trim()===``?{kind:`absent`}:{kind:`invalid`}}function m(e,t,n){return{kind:`error`,error:{errorType:e,message:t,field:n}}}const h=e.object({brandId:e.string().uuid(),brandSlug:e.string().min(1).max(64),supplierId:e.string().uuid(),supplierDisplayName:e.string().min(1).max(255),importBatchId:e.string().uuid(),supplierCodePrefix:e.string().max(8).regex(/^[A-Z0-9-]+$/,`must be uppercase alphanumeric or "-"`).optional()}),g=new WeakMap;function _(e){let t=g.get(e);if(t)return t;let n=h.parse(e);return g.set(e,n),n}const v=[`de`,`en`,`es`,`fr`,`it`,`nl`,`pt`,`ro`],y=e=>`name_${e}`,b=[{key:`code`,label:`Code`,required:!0,type:`string`},{key:`net_price`,label:`Net price`,required:!1,type:`number`},{key:`gross_price`,label:`Gross price`,required:!1,type:`number`},{key:`list_price`,label:`List price`,required:!1,type:`number`},{key:`currency`,label:`Currency`,required:!1,type:`string`},{key:`barcode`,label:`Barcode`,required:!1,type:`string`},{key:`description1`,label:`Description 1`,required:!1,type:`string`},{key:`description2`,label:`Description 2`,required:!1,type:`string`},{key:`discount_group`,label:`Discount group`,required:!1,type:`string`},{key:`superseding_code`,label:`Superseding code`,required:!1,type:`string`},...v.map(e=>({key:y(e),label:`Name (${e})`,required:!1,type:`string`}))],x=/[\s-]/g;function S(e){return e.toUpperCase().replace(x,``)}const C={name:`parts`,fields:b,buildRow(e,t){let n=_(t.params),i=typeof e.code==`string`?e.code.trim():``;if(!i)return m(`missing_code`,`code is empty`,`code`);let a=S(n.supplierCodePrefix&&i.startsWith(n.supplierCodePrefix)?i.slice(n.supplierCodePrefix.length):i);if(!a)return m(`empty_normalized_code`,`Code "${i}" is empty after normalization`,`code`);let o=f(e.currency)??`EUR`;if(o!==`EUR`)return m(`unsupported_currency`,`Currency "${o}" not supported (EUR-only)`,`currency`);let s=p(e.net_price);if(s.kind===`invalid`)return m(`invalid_net_price`,`net_price is not a number`,`net_price`);let c=p(e.gross_price);if(c.kind===`invalid`)return m(`invalid_gross_price`,`gross_price is not a number`,`gross_price`);let l=p(e.list_price);if(l.kind===`invalid`)return m(`invalid_list_price`,`list_price is not a number`,`list_price`);let u=s.kind===`value`?s.value:null,d=l.kind===`value`?l.value:null,h=u??d;if(h===null)return m(`missing_price`,`no net_price or list_price provided`,`price`);let g=`${a}__${n.supplierId}`;return{kind:`success`,id:g,doc:{doc_id:g,code:i,code_normalized:a,brand_id:n.brandId,brand_slug:n.brandSlug,supplier_id:n.supplierId,supplier_display_name:n.supplierDisplayName,net_price_eur:h,gross_price_eur:c.kind===`value`?c.value:null,currency:`EUR`,barcode:f(e.barcode),name_de:f(e.name_de),name_en:f(e.name_en),name_es:f(e.name_es),name_fr:f(e.name_fr),name_it:f(e.name_it),name_nl:f(e.name_nl),name_pt:f(e.name_pt),name_ro:f(e.name_ro),description1:f(e.description1),description2:f(e.description2),list_price_minor:d===null?null:r(d),discount_group:f(e.discount_group),superseding_code:f(e.superseding_code),purchase_price_minor:null,core_charge_minor:null,core_charge_currency:null,core_charge_refundable:null,core_charge_tax_class:null,core_charge_tax_inclusive:null,import_batch_id:n.importBatchId,imported_at:new Date().toISOString()}}}};export{f as a,l as c,o as d,a as f,r as h,m as i,c as l,n as m,C as n,u as o,i as p,p as r,s,h as t,d as u};
+//# sourceMappingURL=parts-schema-DQeMGPBC.mjs.map

package/dist/parts-schema-DQeMGPBC.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"parts-schema-DQeMGPBC.mjs","names":[],"sources":["../src/mapping/number-format.ts","../src/mapping/column-transforms.ts","../src/mapping/engine.ts","../src/mapping/target-schema.ts","../src/mapping/parts-schema.ts"],"sourcesContent":["/**\n * Number parsing for the configurable mapping engine (commerce-demo phase 02).\n *\n * Real supplier feeds (R03) disagree on decimal AND thousands separators:\n * `55.02` (dot decimal), `326,88` (comma decimal), `1.234,56` (EU\n * thousands+decimal), `1,234.56` (US), `1 234,56` (space-grouped). The mapping\n * stores an explicit `{ decimal, thousands }` per profile so there is no guessing\n * — `parseNumber` trusts the configured format.\n *\n * Two parsers live here:\n *  - `parseNumber(raw, format)` — the format-driven parser used by the generic\n *    `number` per-column transform.\n *  - `parseEuropeanNumber(raw)` — the carmaker-feed legacy parser (comma decimal,\n *    dot thousands, **bare-dot rejected as ambiguous**). Carries the carmaker feed's\n *    documented format with two intentional broadenings vs the old inline parser:\n *    it trims surrounding whitespace, and returns `null` (not `NaN`) for an empty /\n *    `'NA'` cell so both number ops share absent-vs-malformed semantics. The\n *    carmaker doc-shape suite (27 tests) is the regression net.\n */\n\n/** Separator config for {@link parseNumber}. Stored per `import_mapping`. */\nexport interface NumberFormat {\n  /** Decimal separator. */\n  decimal: '.' | ','\n  /** Thousands grouping separator. `''` = none. */\n  thousands: '' | '.' | ',' | ' '\n}\n\n/** Post-cleanup numeric form: optional sign, digits, optional single decimal. */\nconst STRICT_NUMBER_RE = /^-?\\d+(\\.\\d+)?$/\n\n/**\n * Parse a feed cell into a number using an EXPLICIT `{ decimal, thousands }`\n * format. No ambiguity: thousands separators are stripped, the decimal separator\n * is normalised to `.`, and the result is strict-validated.\n *\n * Returns:\n *  - `null` for an empty / `'NA'` cell (a legitimately-absent value — the caller\n *    decides whether that's allowed for a given target field).\n *  - `NaN` for a non-empty but unparseable cell (malformed — the caller surfaces\n *    a row-level error rather than coercing to 0).\n *  - the parsed number otherwise.\n *\n * A character that is neither a digit, the configured decimal, the configured\n * thousands separator, nor a leading `-` makes the cell unparseable (`NaN`) —\n * e.g. a literal `.` when the format's decimal is `,` and thousands is not `.`.\n */\nexport function parseNumber(raw: string, format: NumberFormat): number | null {\n  const trimmed = raw.trim()\n  if (trimmed === '' || trimmed.toUpperCase() === 'NA') return null\n\n  const { decimal, thousands } = format\n  let s = trimmed\n  // 1. Remove every thousands separator (when one is configured).\n  if (thousands !== '') s = s.split(thousands).join('')\n\n  // 2. Normalise the decimal separator to `.`. After thousands removal, the\n  //    \"other\" separator must not appear — if it does, the cell doesn't match\n  //    the declared format and is unparseable.\n  if (decimal === ',') {\n    if (s.includes('.')) return Number.NaN\n    s = s.split(',').join('.')\n  } else {\n    // decimal === '.'\n    if (s.includes(',')) return Number.NaN\n  }\n\n  if (!STRICT_NUMBER_RE.test(s)) return Number.NaN\n  return Number.parseFloat(s)\n}\n\n/**\n * Convert a major-unit money value (e.g. `55.02` EUR) to integer MINOR units\n * (`5502`). Rounds to the nearest cent to absorb float drift. The commerce-demo\n * PartsDocument money fields (`list_price_minor`, `purchase_price_minor`,\n * `core_charge_minor`) are integer minor units so derived prices (PR4) don't\n * accrue rounding error — see F003 in progress.md.\n */\nexport function toMinorUnits(value: number): number {\n  return Math.round(value * 100)\n}\n\n/**\n * Parse a European-formatted decimal: `,` is the decimal separator, `.` is the\n * (optional) thousands separator. Returns `null` for an empty / `'NA'` cell and\n * `NaN` on unparseable (malformed) input.\n *\n * This is the carmaker-feed legacy parser, preserved from the hardcoded transform.\n * Crucially it **rejects bare-dot** input (`1.98`, `1.234`) as ambiguous rather\n * than guessing — the carmaker feeds are documented as comma-decimal, and a `.`\n * without a `,` could mean a US-format decimal OR a thousands marker. The\n * format-driven {@link parseNumber} doesn't need this guard (its separators are\n * explicit); the carmaker path keeps it.\n *\n * The `null` (absent) vs `NaN` (malformed) split mirrors {@link parseNumber} so\n * the mapping engine treats both number ops uniformly: a target field's `buildRow`\n * reads `null` as \"not provided\" and `NaN` as \"invalid\". This is what lets the\n * reimplemented carmaker transform keep its `GrossPrice 'NA' → null` behaviour\n * (optional field) while still erroring on a malformed `NetPrice` (required).\n *\n * Cases: `'1,98'`→`1.98`, `'1.234,56'`→`1234.56`, `'5'`→`5`, `'-5,00'`→`-5`,\n * empty / `'NA'` → `null`. Rejected (`NaN`): bare-dot (`'1.98'`, `'1.234'`),\n * trailing junk (`'1,98EUR'`), multiple commas (`'1,2,3'`), other non-numeric.\n */\nexport function parseEuropeanNumber(s: string): number | null {\n  const trimmed = s.trim()\n  if (trimmed === '' || trimmed.toUpperCase() === 'NA') return null\n  const hasComma = trimmed.includes(',')\n  const hasDot = trimmed.includes('.')\n  let cleaned: string\n  if (hasComma && hasDot) {\n    // European thousands+decimal: strip every `.`, then swap the (sole) `,` for\n    // `.`. The strict regex below catches multi-comma input.\n    cleaned = trimmed.replace(/\\./g, '').replace(',', '.')\n  } else if (hasComma) {\n    cleaned = trimmed.replace(',', '.')\n  } else if (hasDot) {\n    return Number.NaN\n  } else {\n    cleaned = trimmed\n  }\n  if (!STRICT_NUMBER_RE.test(cleaned)) return Number.NaN\n  return Number.parseFloat(cleaned)\n}\n","/**\n * Per-column transforms for the mapping engine (commerce-demo phase 02, D005).\n *\n * A **bounded, enum-keyed** library of cleaning operations applied to a source\n * cell before it lands on a target field. This is the security model from R02:\n * transforms are declarative operations with Zod-validated params, NEVER\n * user-supplied executable code. The carmaker feed's logic decomposes exactly\n * into `stripPrefix` + `europeanNumber` (+ the schema's code-normalize).\n *\n * Number-producing ops (`number`, `europeanNumber`) are TERMINAL by convention —\n * they convert the string cell into `number | null`; string ops after them are\n * no-ops. The engine reads the result as `string | number | null` and hands it to\n * the target schema's `buildRow`.\n */\n\nimport { z } from 'zod'\nimport { type NumberFormat, parseEuropeanNumber, parseNumber } from './number-format.js'\n\n/** Max length for a strip-affix value — bounds the param, no ReDoS surface. */\nconst MAX_AFFIX_LEN = 32\n\n/**\n * Discriminated union of bounded transform ops. The TS type is INFERRED from\n * this schema so there's a single source of truth + the snapshot is validated at\n * the engine boundary.\n */\nexport const columnTransformSchema = z.discriminatedUnion('op', [\n  z.object({ op: z.literal('trim') }),\n  z.object({ op: z.literal('uppercase') }),\n  z.object({ op: z.literal('lowercase') }),\n  z.object({ op: z.literal('stripPrefix'), value: z.string().min(1).max(MAX_AFFIX_LEN) }),\n  z.object({ op: z.literal('stripSuffix'), value: z.string().min(1).max(MAX_AFFIX_LEN) }),\n  z.object({ op: z.literal('defaultIfEmpty'), value: z.string().max(255) }),\n  z.object({ op: z.literal('number') }),\n  z.object({ op: z.literal('europeanNumber') }),\n])\n\n/** One declarative cleaning operation. */\nexport type ColumnTransform = z.infer<typeof columnTransformSchema>\n\n/** The value a column carries as it moves through the transform pipeline. */\nexport type CellValue = string | number | null\n\n/**\n * Apply a transform pipeline to a raw cell, left-to-right. String ops act on the\n * string value; the `number` / `europeanNumber` ops are terminal (convert to\n * `number | null`, with `NaN` for a malformed numeric — the schema's `buildRow`\n * surfaces that as a row error). Once the value is non-string, later string ops\n * pass through unchanged.\n */\nexport function applyColumnTransforms(\n  raw: string,\n  transforms: readonly ColumnTransform[],\n  format: NumberFormat,\n): CellValue {\n  let value: CellValue = raw\n  for (const t of transforms) {\n    if (typeof value !== 'string') {\n      // A prior number op already produced number|null — string ops are no-ops\n      // past this point (number ops are terminal by convention).\n      continue\n    }\n    switch (t.op) {\n      case 'trim':\n        value = value.trim()\n        break\n      case 'uppercase':\n        value = value.toUpperCase()\n        break\n      case 'lowercase':\n        value = value.toLowerCase()\n        break\n      case 'stripPrefix':\n        value = value.startsWith(t.value) ? value.slice(t.value.length) : value\n        break\n      case 'stripSuffix':\n        value = value.endsWith(t.value) ? value.slice(0, value.length - t.value.length) : value\n        break\n      case 'defaultIfEmpty':\n        value = value.trim() === '' ? t.value : value\n        break\n      case 'number':\n        value = parseNumber(value, format)\n        break\n      case 'europeanNumber':\n        value = parseEuropeanNumber(value)\n        break\n    }\n  }\n  return value\n}\n","/**\n * Mapping engine (commerce-demo phase 02, D005).\n *\n * Turns a declarative `MappingConfig` (which source column → which whitelisted\n * target field, plus per-column transforms + a number format) into a\n * `RowTransform` the existing runner drives. The pipeline per row is:\n *\n *   raw cells → applyColumnTransforms(per column) → MappedRow → schema.buildRow\n *\n * Two whitelists guard the boundary (R02 injection model): the `target_schema`\n * name resolves to a registered `TargetSchema`, and every `column_map[].target`\n * must be one of that schema's fields — enforced here at build time (throws),\n * before any row is processed.\n *\n * `resolveMapping` implements the D002 most-specific ladder used at enqueue time\n * to pick the `(supplier, brand)` mapping over the `(supplier, NULL)` default.\n */\n\nimport { z } from 'zod'\nimport type { RowTransform } from '../transform.js'\nimport { applyColumnTransforms, columnTransformSchema } from './column-transforms.js'\nimport type { MappedRow, TargetSchema } from './target-schema.js'\n\n/** One column-map entry. `target` is checked against the schema whitelist separately. */\nexport const mappingColumnSchema = z.object({\n  /** Source header name (or positional index string for headerless feeds). */\n  source: z.string().min(1).max(256),\n  /** Target field key — must be in the chosen schema's whitelist (enforced in `buildMappingTransform`). */\n  target: z.string().min(1).max(64),\n  /** Ordered per-column cleaning ops. Defaults to none. */\n  transforms: z.array(columnTransformSchema).default([]),\n})\nexport type MappingColumn = z.infer<typeof mappingColumnSchema>\n\n/** Decimal + thousands separators for the `number` transform. */\nexport const numberFormatSchema = z.object({\n  decimal: z.enum(['.', ',']),\n  thousands: z.enum(['', '.', ',', ' ']),\n})\n\n/**\n * The portable mapping config — the subset of an `import_mapping` row the engine\n * needs. The enqueuer SNAPSHOTS this into `import_run.params.mapping` so a run\n * records the exact mapping it used (reproducible) and the transform stays a pure\n * function of `(row, ctx)` with no per-row DB I/O.\n */\nexport const mappingConfigSchema = z.object({\n  targetSchema: z.string().min(1).max(64),\n  numberFormat: numberFormatSchema,\n  columnMap: z.array(mappingColumnSchema),\n})\nexport type MappingConfig = z.infer<typeof mappingConfigSchema>\n\n/**\n * Build a `RowTransform` from a validated mapping config + its target schema.\n *\n * Throws at BUILD time if any `column_map` target is outside the schema's\n * whitelist (deny-by-default, R02) — so a misconfigured mapping fails loudly\n * before a single row is written, rather than silently dropping a column.\n *\n * Generic over the schema's output type, so passing the concrete `partsSchema`\n * yields a `RowTransform<PartsDocument>` (the carmaker path keeps its typing); the\n * registry-resolved `TargetSchema<unknown>` yields `RowTransform<unknown>` (the\n * `commerce:mapping` path, where the output type isn't known until runtime).\n */\nexport function buildMappingTransform<TOut>(\n  config: MappingConfig,\n  schema: TargetSchema<TOut>,\n): RowTransform<TOut> {\n  const allowed = new Set(schema.fields.map((f) => f.key))\n  for (const col of config.columnMap) {\n    if (!allowed.has(col.target)) {\n      throw new Error(\n        `Mapping target \"${col.target}\" is not a field of schema \"${schema.name}\" ` +\n          `(allowed: ${[...allowed].join(', ')})`,\n      )\n    }\n  }\n\n  const format = config.numberFormat\n  const columnMap = config.columnMap\n\n  return async (row, ctx) => {\n    const mapped: MappedRow = {}\n    for (const col of columnMap) {\n      // `source` is read as data from the parsed row map — never used as a key\n      // into framework objects (R02). A missing source column yields `''`.\n      const raw = row[col.source] ?? ''\n      mapped[col.target] = applyColumnTransforms(raw, col.transforms, format)\n    }\n    return schema.buildRow(mapped, {\n      importRunId: ctx.importRunId,\n      rowNumber: ctx.rowNumber,\n      params: ctx.params,\n    })\n  }\n}\n\n/** The minimal shape `resolveMapping` needs — satisfied by an `import_mapping` row or a test fixture. */\nexport interface ResolvableMapping {\n  /** `null` = supplier default (applies to all brands). */\n  brandId: string | null\n  active: boolean\n}\n\n/**\n * Most-specific resolution (D002 ladder): given the candidate mappings for one\n * `(supplier, target_schema)` and the requested brand, return the brand-specific\n * mapping if one exists, else the supplier-default (`brandId === null`). Inactive\n * mappings are skipped. Returns `undefined` when nothing matches.\n */\nexport function resolveMapping<M extends ResolvableMapping>(\n  mappings: readonly M[],\n  brandId: string | null,\n): M | undefined {\n  const active = mappings.filter((m) => m.active)\n  if (brandId !== null) {\n    const specific = active.find((m) => m.brandId === brandId)\n    if (specific) return specific\n  }\n  return active.find((m) => m.brandId === null)\n}\n","/**\n * TargetSchema — the contract a mapping writes INTO (commerce-demo phase 02, D005).\n *\n * A mapping's `column_map` points source columns at a schema's whitelisted target\n * fields; the engine applies the per-column transforms to build a `MappedRow`, then\n * hands it to `schema.buildRow` to produce the typed output document (an ES\n * `PartsDocument` for `parts`, a discount row for `discount_table`).\n *\n * The `fields` list is the SECOND injection-boundary whitelist (the first is the\n * registered schema name): a `column_map` entry whose `target` is not a field key\n * here is rejected at validation time (R02). Two schemas ship in PR2 — `parts` and\n * `discount_table`.\n */\n\nimport type { RowResult } from '../transform.js'\nimport type { CellValue } from './column-transforms.js'\n\n/** One whitelisted target field on a schema. */\nexport interface TargetField {\n  /** Stable target key referenced by `column_map[].target`. */\n  key: string\n  /** Human label for the wizard dropdown (PR3). */\n  label: string\n  /** Whether a mapping for this schema MUST include this target. */\n  required: boolean\n  /** Coarse type — drives the wizard's transform suggestions; `buildRow` does the real coercion. */\n  type: 'string' | 'number'\n}\n\n/** The per-row product of the engine: target field key → cleaned value. */\nexport type MappedRow = Record<string, CellValue>\n\n/** Per-row context handed to `buildRow`. `params` is the opaque per-run params object. */\nexport interface BuildRowContext {\n  importRunId: string\n  rowNumber: number\n  /**\n   * Opaque per-run params (the parts schema reads brand/supplier identity from\n   * it; `discount_table` ignores it). The schema validates the shape it needs.\n   */\n  params: Record<string, unknown>\n}\n\n/** A registered target schema: a whitelist + a row builder. */\nexport interface TargetSchema<TOut> {\n  /** Registered name — also the `import_mapping.target_schema` value. */\n  name: string\n  /** The target-field whitelist (injection boundary). */\n  fields: readonly TargetField[]\n  /** Build a typed output (success) / skip / error from a mapped row. */\n  buildRow(mapped: MappedRow, ctx: BuildRowContext): RowResult<TOut>\n}\n\n// ---------------------------------------------------------------------------\n// Shared `buildRow` helpers (used by every concrete schema).\n// ---------------------------------------------------------------------------\n\n/**\n * A nullable text cell → trimmed string, or `null` for empty / non-string. Accepts\n * `undefined` (an UNMAPPED target — indexed access under `noUncheckedIndexedAccess`)\n * and treats it as absent.\n */\nexport function textOrNull(v: CellValue | undefined): string | null {\n  if (typeof v !== 'string') return null\n  const t = v.trim()\n  return t === '' ? null : t\n}\n\n/** The three outcomes a `number` / `europeanNumber` op produces for a cell. */\nexport type NumberOutcome =\n  | { kind: 'value'; value: number }\n  | { kind: 'absent' }\n  | { kind: 'invalid' }\n\n/**\n * Classify a cell that should hold a number. `undefined` (unmapped) / `null` →\n * absent; a real number → value (`NaN` → invalid); a non-empty raw string (no\n * number op declared) → invalid; a blank raw string → absent.\n */\nexport function numberCell(v: CellValue | undefined): NumberOutcome {\n  if (v === undefined || v === null) return { kind: 'absent' }\n  if (typeof v === 'number') {\n    return Number.isNaN(v) ? { kind: 'invalid' } : { kind: 'value', value: v }\n  }\n  return v.trim() === '' ? { kind: 'absent' } : { kind: 'invalid' }\n}\n\n/** Build a `RowResult` error variant (assignable to any `TargetSchema<TOut>`). */\nexport function rowError(errorType: string, message: string, field: string) {\n  return { kind: 'error' as const, error: { errorType, message, field } }\n}\n","/**\n * `parts` target schema — maps a cleaned feed row into the ES `PartsDocument`\n * (commerce-demo phase 02, D005). This is the SHARED assembly + validation logic\n * the reimplemented carmaker transform and every operator-built `parts` mapping\n * run through, so the row taxonomy (code normalization, currency gate, price\n * validation) is uniform across all supplier feeds.\n *\n * **PR2 field coverage.** This schema populates the direct, no-derivation fields:\n * the existing `PartsDocument` columns plus `list_price_minor`, `discount_group`,\n * and `superseding_code` (string/value passthroughs that PR4 / PR6 consume). The\n * DERIVED `purchase_price_minor` (PR4) and all `core_charge_*` fields (PR7) stay\n * null; `core_charge` is deliberately NOT in the PR2 whitelist (deny-by-default —\n * PR7 adds it with its `*S`-row + column ingestion). Replacement-pair and\n * core-charge SIDE EFFECTS (separate entities) are PR6/PR7.\n *\n * **Money.** `list_price_minor` is integer minor units via `toMinorUnits`\n * (F003). The legacy `net_price_eur` stays a major-unit float for back-compat; it\n * holds the direct net price, or — for a two-file LIST feed with no direct net —\n * the list price as a display placeholder until PR4 derives `purchase_price_minor`.\n */\n\nimport type { PartsDocument } from '@murumets-ee/search-elasticsearch'\nimport { z } from 'zod'\nimport { toMinorUnits } from './number-format.js'\nimport {\n  type BuildRowContext,\n  type MappedRow,\n  numberCell,\n  rowError,\n  type TargetField,\n  type TargetSchema,\n  textOrNull,\n} from './target-schema.js'\n\n/**\n * Params the `parts` schema reads from the run — the per-(supplier, brand)\n * identity resolved at enqueue time. Identical to the carmaker transform's params\n * (the carmaker path re-exports this), so brand/supplier injection + the optional\n * supplier code-prefix strip are validated the same way everywhere.\n */\nexport const partsParamsSchema = z.object({\n  brandId: z.string().uuid(),\n  brandSlug: z.string().min(1).max(64),\n  supplierId: z.string().uuid(),\n  supplierDisplayName: z.string().min(1).max(255),\n  importBatchId: z.string().uuid(),\n  /** Supplier code prefix to strip before normalization (e.g. `'ME-'`). Optional. */\n  supplierCodePrefix: z\n    .string()\n    .max(8)\n    .regex(/^[A-Z0-9-]+$/, 'must be uppercase alphanumeric or \"-\"')\n    .optional(),\n})\nexport type PartsParams = z.infer<typeof partsParamsSchema>\n\n/** Per-run memoization of the parsed params (keyed on the `ctx.params` reference). */\nconst paramsCache = new WeakMap<object, PartsParams>()\n\nfunction resolveParams(raw: Record<string, unknown>): PartsParams {\n  const cached = paramsCache.get(raw)\n  if (cached) return cached\n  const parsed = partsParamsSchema.parse(raw)\n  paramsCache.set(raw, parsed)\n  return parsed\n}\n\n/** Per-locale display-name target fields (8 locales). */\nconst NAME_LOCALES = ['de', 'en', 'es', 'fr', 'it', 'nl', 'pt', 'ro'] as const\ntype NameLocale = (typeof NAME_LOCALES)[number]\nconst nameKey = (l: NameLocale): `name_${NameLocale}` => `name_${l}`\n\n/** The `parts` target-field whitelist (PR2 subset — `core_charge`/`purchase_price` are later PRs). */\nconst PARTS_FIELDS: readonly TargetField[] = [\n  { key: 'code', label: 'Code', required: true, type: 'string' },\n  { key: 'net_price', label: 'Net price', required: false, type: 'number' },\n  { key: 'gross_price', label: 'Gross price', required: false, type: 'number' },\n  { key: 'list_price', label: 'List price', required: false, type: 'number' },\n  { key: 'currency', label: 'Currency', required: false, type: 'string' },\n  { key: 'barcode', label: 'Barcode', required: false, type: 'string' },\n  { key: 'description1', label: 'Description 1', required: false, type: 'string' },\n  { key: 'description2', label: 'Description 2', required: false, type: 'string' },\n  { key: 'discount_group', label: 'Discount group', required: false, type: 'string' },\n  { key: 'superseding_code', label: 'Superseding code', required: false, type: 'string' },\n  ...NAME_LOCALES.map(\n    (l): TargetField => ({\n      key: nameKey(l),\n      label: `Name (${l})`,\n      required: false,\n      type: 'string',\n    }),\n  ),\n] as const\n\nconst NORMALIZE_CODE_RE = /[\\s-]/g\n/** Uppercase + dash/whitespace-stripped (D2). `*`/other chars are preserved (e.g. `*S` markers). */\nfunction normalizeCode(s: string): string {\n  return s.toUpperCase().replace(NORMALIZE_CODE_RE, '')\n}\n\n/**\n * The `parts` schema. `buildRow` THROWS on invalid run params (so a misconfigured\n * run surfaces as a transform error, matching the carmaker contract) and returns a\n * typed `RowResult` per row otherwise.\n */\nexport const partsSchema: TargetSchema<PartsDocument> = {\n  name: 'parts',\n  fields: PARTS_FIELDS,\n  buildRow(mapped: MappedRow, ctx: BuildRowContext) {\n    const params = resolveParams(ctx.params)\n\n    // 1. Code (required) → raw `code` + normalized `code_normalized` + doc id.\n    const codeRaw = typeof mapped.code === 'string' ? mapped.code.trim() : ''\n    if (!codeRaw) return rowError('missing_code', 'code is empty', 'code')\n    const stripped =\n      params.supplierCodePrefix && codeRaw.startsWith(params.supplierCodePrefix)\n        ? codeRaw.slice(params.supplierCodePrefix.length)\n        : codeRaw\n    const codeNormalized = normalizeCode(stripped)\n    if (!codeNormalized) {\n      return rowError('empty_normalized_code', `Code \"${codeRaw}\" is empty after normalization`, 'code')\n    }\n\n    // 2. Currency gate (D9 EUR-only). Unmapped/blank defaults to EUR.\n    const currencyText = textOrNull(mapped.currency)\n    const currency = currencyText ?? 'EUR'\n    if (currency !== 'EUR') {\n      return rowError('unsupported_currency', `Currency \"${currency}\" not supported (EUR-only)`, 'currency')\n    }\n\n    // 3. Prices. net/gross/list each: absent → null, invalid → error, value → number.\n    const net = numberCell(mapped.net_price)\n    if (net.kind === 'invalid') return rowError('invalid_net_price', 'net_price is not a number', 'net_price')\n    const gross = numberCell(mapped.gross_price)\n    if (gross.kind === 'invalid') {\n      return rowError('invalid_gross_price', 'gross_price is not a number', 'gross_price')\n    }\n    const list = numberCell(mapped.list_price)\n    if (list.kind === 'invalid') return rowError('invalid_list_price', 'list_price is not a number', 'list_price')\n\n    const netValue = net.kind === 'value' ? net.value : null\n    const listValue = list.kind === 'value' ? list.value : null\n    // At least one price must resolve. `net_price_eur` is non-nullable on the doc,\n    // so it takes the direct net or — for a list-only two-file feed — the list\n    // price as a display placeholder (PR4 derives the true purchase price).\n    const displayPrice = netValue ?? listValue\n    if (displayPrice === null) {\n      return rowError('missing_price', 'no net_price or list_price provided', 'price')\n    }\n\n    const docId = `${codeNormalized}__${params.supplierId}`\n    const doc: PartsDocument = {\n      doc_id: docId,\n      code: codeRaw,\n      code_normalized: codeNormalized,\n      brand_id: params.brandId,\n      brand_slug: params.brandSlug,\n      supplier_id: params.supplierId,\n      supplier_display_name: params.supplierDisplayName,\n      net_price_eur: displayPrice,\n      gross_price_eur: gross.kind === 'value' ? gross.value : null,\n      currency: 'EUR',\n      barcode: textOrNull(mapped.barcode),\n      name_de: textOrNull(mapped.name_de),\n      name_en: textOrNull(mapped.name_en),\n      name_es: textOrNull(mapped.name_es),\n      name_fr: textOrNull(mapped.name_fr),\n      name_it: textOrNull(mapped.name_it),\n      name_nl: textOrNull(mapped.name_nl),\n      name_pt: textOrNull(mapped.name_pt),\n      name_ro: textOrNull(mapped.name_ro),\n      description1: textOrNull(mapped.description1),\n      description2: textOrNull(mapped.description2),\n      // commerce-demo direct fields (PR2): list price → minor units; group + code\n      // passthroughs consumed by PR4 / PR6.\n      list_price_minor: listValue === null ? null : toMinorUnits(listValue),\n      discount_group: textOrNull(mapped.discount_group),\n      superseding_code: textOrNull(mapped.superseding_code),\n      // Derived / later-PR fields stay null in PR2.\n      purchase_price_minor: null,\n      core_charge_minor: null,\n      core_charge_currency: null,\n      core_charge_refundable: null,\n      core_charge_tax_class: null,\n      core_charge_tax_inclusive: null,\n      import_batch_id: params.importBatchId,\n      imported_at: new Date().toISOString(),\n    }\n\n    return { kind: 'success', id: docId, doc }\n  },\n}\n"],"mappings":"wBA6BA,MAAM,EAAmB,kBAkBzB,SAAgB,EAAY,EAAa,EAAqC,CAC5E,IAAM,EAAU,EAAI,KAAK,EACzB,GAAI,IAAY,IAAM,EAAQ,YAAY,IAAM,KAAM,OAAO,KAE7D,GAAM,CAAE,UAAS,aAAc,EAC3B,EAAI,EAOR,GALI,IAAc,KAAI,EAAI,EAAE,MAAM,CAAS,CAAC,CAAC,KAAK,EAAE,GAKhD,IAAY,IAAK,CACnB,GAAI,EAAE,SAAS,GAAG,EAAG,MAAO,KAC5B,EAAI,EAAE,MAAM,GAAG,CAAC,CAAC,KAAK,GAAG,CAC3B,MAEE,GAAI,EAAE,SAAS,GAAG,EAAG,MAAO,KAI9B,OADK,EAAiB,KAAK,CAAC,EACrB,OAAO,WAAW,CAAC,EADY,GAExC,CASA,SAAgB,EAAa,EAAuB,CAClD,OAAO,KAAK,MAAM,EAAQ,GAAG,CAC/B,CAwBA,SAAgB,EAAoB,EAA0B,CAC5D,IAAM,EAAU,EAAE,KAAK,EACvB,GAAI,IAAY,IAAM,EAAQ,YAAY,IAAM,KAAM,OAAO,KAC7D,IAAM,EAAW,EAAQ,SAAS,GAAG,EAC/B,EAAS,EAAQ,SAAS,GAAG,EAC/B,EACJ,GAAI,GAAY,EAGd,EAAU,EAAQ,QAAQ,MAAO,EAAE,CAAC,CAAC,QAAQ,IAAK,GAAG,OAChD,GAAI,EACT,EAAU,EAAQ,QAAQ,IAAK,GAAG,OAC7B,GAAI,EACT,MAAO,SAEP,GAAU,EAGZ,OADK,EAAiB,KAAK,CAAO,EAC3B,OAAO,WAAW,CAAO,EADY,GAE9C,CCxGA,MAOa,EAAwB,EAAE,mBAAmB,KAAM,CAC9D,EAAE,OAAO,CAAE,GAAI,EAAE,QAAQ,MAAM,CAAE,CAAC,EAClC,EAAE,OAAO,CAAE,GAAI,EAAE,QAAQ,WAAW,CAAE,CAAC,EACvC,EAAE,OAAO,CAAE,GAAI,EAAE,QAAQ,WAAW,CAAE,CAAC,EACvC,EAAE,OAAO,CAAE,GAAI,EAAE,QAAQ,aAAa,EAAG,MAAO,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,EAAa,CAAE,CAAC,EACtF,EAAE,OAAO,CAAE,GAAI,EAAE,QAAQ,aAAa,EAAG,MAAO,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,EAAa,CAAE,CAAC,EACtF,EAAE,OAAO,CAAE,GAAI,EAAE,QAAQ,gBAAgB,EAAG,MAAO,EAAE,OAAO,CAAC,CAAC,IAAI,GAAG,CAAE,CAAC,EACxE,EAAE,OAAO,CAAE,GAAI,EAAE,QAAQ,QAAQ,CAAE,CAAC,EACpC,EAAE,OAAO,CAAE,GAAI,EAAE,QAAQ,gBAAgB,CAAE,CAAC,CAC9C,CAAC,EAeD,SAAgB,EACd,EACA,EACA,EACW,CACX,IAAI,EAAmB,EACvB,IAAK,IAAM,KAAK,EACV,UAAO,GAAU,SAKrB,OAAQ,EAAE,GAAV,CACE,IAAK,OACH,EAAQ,EAAM,KAAK,EACnB,MACF,IAAK,YACH,EAAQ,EAAM,YAAY,EAC1B,MACF,IAAK,YACH,EAAQ,EAAM,YAAY,EAC1B,MACF,IAAK,cACH,EAAQ,EAAM,WAAW,EAAE,KAAK,EAAI,EAAM,MAAM,EAAE,MAAM,MAAM,EAAI,EAClE,MACF,IAAK,cACH,EAAQ,EAAM,SAAS,EAAE,KAAK,EAAI,EAAM,MAAM,EAAG,EAAM,OAAS,EAAE,MAAM,MAAM,EAAI,EAClF,MACF,IAAK,iBACH,EAAQ,EAAM,KAAK,IAAM,GAAK,EAAE,MAAQ,EACxC,MACF,IAAK,SACH,EAAQ,EAAY,EAAO,CAAM,EACjC,MACF,IAAK,iBACH,EAAQ,EAAoB,CAAK,EACjC,KACJ,CAEF,OAAO,CACT,CClEA,MAAa,EAAsB,EAAE,OAAO,CAE1C,OAAQ,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,GAAG,EAEjC,OAAQ,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE,EAEhC,WAAY,EAAE,MAAM,CAAqB,CAAC,CAAC,QAAQ,CAAC,CAAC,CACvD,CAAC,EAIY,EAAqB,EAAE,OAAO,CACzC,QAAS,EAAE,KAAK,CAAC,IAAK,GAAG,CAAC,EAC1B,UAAW,EAAE,KAAK,CAAC,GAAI,IAAK,IAAK,GAAG,CAAC,CACvC,CAAC,EAQY,EAAsB,EAAE,OAAO,CAC1C,aAAc,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE,EACtC,aAAc,EACd,UAAW,EAAE,MAAM,CAAmB,CACxC,CAAC,EAeD,SAAgB,EACd,EACA,EACoB,CACpB,IAAM,EAAU,IAAI,IAAI,EAAO,OAAO,IAAK,GAAM,EAAE,GAAG,CAAC,EACvD,IAAK,IAAM,KAAO,EAAO,UACvB,GAAI,CAAC,EAAQ,IAAI,EAAI,MAAM,EACzB,MAAU,MACR,mBAAmB,EAAI,OAAO,8BAA8B,EAAO,KAAK,cACzD,CAAC,GAAG,CAAO,CAAC,CAAC,KAAK,IAAI,EAAE,EACzC,EAIJ,IAAM,EAAS,EAAO,aAChB,EAAY,EAAO,UAEzB,OAAO,MAAO,EAAK,IAAQ,CACzB,IAAM,EAAoB,CAAC,EAC3B,IAAK,IAAM,KAAO,EAAW,CAG3B,IAAM,EAAM,EAAI,EAAI,SAAW,GAC/B,EAAO,EAAI,QAAU,EAAsB,EAAK,EAAI,WAAY,CAAM,CACxE,CACA,OAAO,EAAO,SAAS,EAAQ,CAC7B,YAAa,EAAI,YACjB,UAAW,EAAI,UACf,OAAQ,EAAI,MACd,CAAC,CACH,CACF,CAeA,SAAgB,EACd,EACA,EACe,CACf,IAAM,EAAS,EAAS,OAAQ,GAAM,EAAE,MAAM,EAC9C,GAAI,IAAY,KAAM,CACpB,IAAM,EAAW,EAAO,KAAM,GAAM,EAAE,UAAY,CAAO,EACzD,GAAI,EAAU,OAAO,CACvB,CACA,OAAO,EAAO,KAAM,GAAM,EAAE,UAAY,IAAI,CAC9C,CC3DA,SAAgB,EAAW,EAAyC,CAClE,GAAI,OAAO,GAAM,SAAU,OAAO,KAClC,IAAM,EAAI,EAAE,KAAK,EACjB,OAAO,IAAM,GAAK,KAAO,CAC3B,CAaA,SAAgB,EAAW,EAAyC,CAKlE,OAJI,GAAyB,KAAa,CAAE,KAAM,QAAS,EACvD,OAAO,GAAM,SACR,OAAO,MAAM,CAAC,EAAI,CAAE,KAAM,SAAU,EAAI,CAAE,KAAM,QAAS,MAAO,CAAE,EAEpE,EAAE,KAAK,IAAM,GAAK,CAAE,KAAM,QAAS,EAAI,CAAE,KAAM,SAAU,CAClE,CAGA,SAAgB,EAAS,EAAmB,EAAiB,EAAe,CAC1E,MAAO,CAAE,KAAM,QAAkB,MAAO,CAAE,YAAW,UAAS,OAAM,CAAE,CACxE,CClDA,MAAa,EAAoB,EAAE,OAAO,CACxC,QAAS,EAAE,OAAO,CAAC,CAAC,KAAK,EACzB,UAAW,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE,EACnC,WAAY,EAAE,OAAO,CAAC,CAAC,KAAK,EAC5B,oBAAqB,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,GAAG,EAC9C,cAAe,EAAE,OAAO,CAAC,CAAC,KAAK,EAE/B,mBAAoB,EACjB,OAAO,CAAC,CACR,IAAI,CAAC,CAAC,CACN,MAAM,eAAgB,uCAAuC,CAAC,CAC9D,SAAS,CACd,CAAC,EAIK,EAAc,IAAI,QAExB,SAAS,EAAc,EAA2C,CAChE,IAAM,EAAS,EAAY,IAAI,CAAG,EAClC,GAAI,EAAQ,OAAO,EACnB,IAAM,EAAS,EAAkB,MAAM,CAAG,EAE1C,OADA,EAAY,IAAI,EAAK,CAAM,EACpB,CACT,CAGA,MAAM,EAAe,CAAC,KAAM,KAAM,KAAM,KAAM,KAAM,KAAM,KAAM,IAAI,EAE9D,EAAW,GAAwC,QAAQ,IAG3D,EAAuC,CAC3C,CAAE,IAAK,OAAQ,MAAO,OAAQ,SAAU,GAAM,KAAM,QAAS,EAC7D,CAAE,IAAK,YAAa,MAAO,YAAa,SAAU,GAAO,KAAM,QAAS,EACxE,CAAE,IAAK,cAAe,MAAO,cAAe,SAAU,GAAO,KAAM,QAAS,EAC5E,CAAE,IAAK,aAAc,MAAO,aAAc,SAAU,GAAO,KAAM,QAAS,EAC1E,CAAE,IAAK,WAAY,MAAO,WAAY,SAAU,GAAO,KAAM,QAAS,EACtE,CAAE,IAAK,UAAW,MAAO,UAAW,SAAU,GAAO,KAAM,QAAS,EACpE,CAAE,IAAK,eAAgB,MAAO,gBAAiB,SAAU,GAAO,KAAM,QAAS,EAC/E,CAAE,IAAK,eAAgB,MAAO,gBAAiB,SAAU,GAAO,KAAM,QAAS,EAC/E,CAAE,IAAK,iBAAkB,MAAO,iBAAkB,SAAU,GAAO,KAAM,QAAS,EAClF,CAAE,IAAK,mBAAoB,MAAO,mBAAoB,SAAU,GAAO,KAAM,QAAS,EACtF,GAAG,EAAa,IACb,IAAoB,CACnB,IAAK,EAAQ,CAAC,EACd,MAAO,SAAS,EAAE,GAClB,SAAU,GACV,KAAM,QACR,EACF,CACF,EAEM,EAAoB,SAE1B,SAAS,EAAc,EAAmB,CACxC,OAAO,EAAE,YAAY,CAAC,CAAC,QAAQ,EAAmB,EAAE,CACtD,CAOA,MAAa,EAA2C,CACtD,KAAM,QACN,OAAQ,EACR,SAAS,EAAmB,EAAsB,CAChD,IAAM,EAAS,EAAc,EAAI,MAAM,EAGjC,EAAU,OAAO,EAAO,MAAS,SAAW,EAAO,KAAK,KAAK,EAAI,GACvE,GAAI,CAAC,EAAS,OAAO,EAAS,eAAgB,gBAAiB,MAAM,EAKrE,IAAM,EAAiB,EAHrB,EAAO,oBAAsB,EAAQ,WAAW,EAAO,kBAAkB,EACrE,EAAQ,MAAM,EAAO,mBAAmB,MAAM,EAC9C,CACuC,EAC7C,GAAI,CAAC,EACH,OAAO,EAAS,wBAAyB,SAAS,EAAQ,gCAAiC,MAAM,EAKnG,IAAM,EADe,EAAW,EAAO,QACX,GAAK,MACjC,GAAI,IAAa,MACf,OAAO,EAAS,uBAAwB,aAAa,EAAS,4BAA6B,UAAU,EAIvG,IAAM,EAAM,EAAW,EAAO,SAAS,EACvC,GAAI,EAAI,OAAS,UAAW,OAAO,EAAS,oBAAqB,4BAA6B,WAAW,EACzG,IAAM,EAAQ,EAAW,EAAO,WAAW,EAC3C,GAAI,EAAM,OAAS,UACjB,OAAO,EAAS,sBAAuB,8BAA+B,aAAa,EAErF,IAAM,EAAO,EAAW,EAAO,UAAU,EACzC,GAAI,EAAK,OAAS,UAAW,OAAO,EAAS,qBAAsB,6BAA8B,YAAY,EAE7G,IAAM,EAAW,EAAI,OAAS,QAAU,EAAI,MAAQ,KAC9C,EAAY,EAAK,OAAS,QAAU,EAAK,MAAQ,KAIjD,EAAe,GAAY,EACjC,GAAI,IAAiB,KACnB,OAAO,EAAS,gBAAiB,sCAAuC,OAAO,EAGjF,IAAM,EAAQ,GAAG,EAAe,IAAI,EAAO,aAuC3C,MAAO,CAAE,KAAM,UAAW,GAAI,EAAO,IAAA,CArCnC,OAAQ,EACR,KAAM,EACN,gBAAiB,EACjB,SAAU,EAAO,QACjB,WAAY,EAAO,UACnB,YAAa,EAAO,WACpB,sBAAuB,EAAO,oBAC9B,cAAe,EACf,gBAAiB,EAAM,OAAS,QAAU,EAAM,MAAQ,KACxD,SAAU,MACV,QAAS,EAAW,EAAO,OAAO,EAClC,QAAS,EAAW,EAAO,OAAO,EAClC,QAAS,EAAW,EAAO,OAAO,EAClC,QAAS,EAAW,EAAO,OAAO,EAClC,QAAS,EAAW,EAAO,OAAO,EAClC,QAAS,EAAW,EAAO,OAAO,EAClC,QAAS,EAAW,EAAO,OAAO,EAClC,QAAS,EAAW,EAAO,OAAO,EAClC,QAAS,EAAW,EAAO,OAAO,EAClC,aAAc,EAAW,EAAO,YAAY,EAC5C,aAAc,EAAW,EAAO,YAAY,EAG5C,iBAAkB,IAAc,KAAO,KAAO,EAAa,CAAS,EACpE,eAAgB,EAAW,EAAO,cAAc,EAChD,iBAAkB,EAAW,EAAO,gBAAgB,EAEpD,qBAAsB,KACtB,kBAAmB,KACnB,qBAAsB,KACtB,uBAAwB,KACxB,sBAAuB,KACvB,0BAA2B,KAC3B,gBAAiB,EAAO,cACxB,YAAa,IAAI,KAAK,CAAA,CAAE,YAAY,CAGC,CAAE,CAC3C,CACF"}

package/dist/parts-schema-RI_ltubl.d.mts

@@ -0,0 +1,260 @@
+import { n as RowResult } from "./transform-dO4z0Ze6.mjs";
+import { PartsDocument } from "@murumets-ee/search-elasticsearch";
+import { z } from "zod";
+
+//#region src/mapping/number-format.d.ts
+/**
+ * Number parsing for the configurable mapping engine (commerce-demo phase 02).
+ *
+ * Real supplier feeds (R03) disagree on decimal AND thousands separators:
+ * `55.02` (dot decimal), `326,88` (comma decimal), `1.234,56` (EU
+ * thousands+decimal), `1,234.56` (US), `1 234,56` (space-grouped). The mapping
+ * stores an explicit `{ decimal, thousands }` per profile so there is no guessing
+ * — `parseNumber` trusts the configured format.
+ *
+ * Two parsers live here:
+ *  - `parseNumber(raw, format)` — the format-driven parser used by the generic
+ *    `number` per-column transform.
+ *  - `parseEuropeanNumber(raw)` — the carmaker-feed legacy parser (comma decimal,
+ *    dot thousands, **bare-dot rejected as ambiguous**). Carries the carmaker feed's
+ *    documented format with two intentional broadenings vs the old inline parser:
+ *    it trims surrounding whitespace, and returns `null` (not `NaN`) for an empty /
+ *    `'NA'` cell so both number ops share absent-vs-malformed semantics. The
+ *    carmaker doc-shape suite (27 tests) is the regression net.
+ */
+/** Separator config for {@link parseNumber}. Stored per `import_mapping`. */
+interface NumberFormat {
+  /** Decimal separator. */
+  decimal: '.' | ',';
+  /** Thousands grouping separator. `''` = none. */
+  thousands: '' | '.' | ',' | ' ';
+}
+/**
+ * Parse a feed cell into a number using an EXPLICIT `{ decimal, thousands }`
+ * format. No ambiguity: thousands separators are stripped, the decimal separator
+ * is normalised to `.`, and the result is strict-validated.
+ *
+ * Returns:
+ *  - `null` for an empty / `'NA'` cell (a legitimately-absent value — the caller
+ *    decides whether that's allowed for a given target field).
+ *  - `NaN` for a non-empty but unparseable cell (malformed — the caller surfaces
+ *    a row-level error rather than coercing to 0).
+ *  - the parsed number otherwise.
+ *
+ * A character that is neither a digit, the configured decimal, the configured
+ * thousands separator, nor a leading `-` makes the cell unparseable (`NaN`) —
+ * e.g. a literal `.` when the format's decimal is `,` and thousands is not `.`.
+ */
+declare function parseNumber(raw: string, format: NumberFormat): number | null;
+/**
+ * Convert a major-unit money value (e.g. `55.02` EUR) to integer MINOR units
+ * (`5502`). Rounds to the nearest cent to absorb float drift. The commerce-demo
+ * PartsDocument money fields (`list_price_minor`, `purchase_price_minor`,
+ * `core_charge_minor`) are integer minor units so derived prices (PR4) don't
+ * accrue rounding error — see F003 in progress.md.
+ */
+declare function toMinorUnits(value: number): number;
+/**
+ * Parse a European-formatted decimal: `,` is the decimal separator, `.` is the
+ * (optional) thousands separator. Returns `null` for an empty / `'NA'` cell and
+ * `NaN` on unparseable (malformed) input.
+ *
+ * This is the carmaker-feed legacy parser, preserved from the hardcoded transform.
+ * Crucially it **rejects bare-dot** input (`1.98`, `1.234`) as ambiguous rather
+ * than guessing — the carmaker feeds are documented as comma-decimal, and a `.`
+ * without a `,` could mean a US-format decimal OR a thousands marker. The
+ * format-driven {@link parseNumber} doesn't need this guard (its separators are
+ * explicit); the carmaker path keeps it.
+ *
+ * The `null` (absent) vs `NaN` (malformed) split mirrors {@link parseNumber} so
+ * the mapping engine treats both number ops uniformly: a target field's `buildRow`
+ * reads `null` as "not provided" and `NaN` as "invalid". This is what lets the
+ * reimplemented carmaker transform keep its `GrossPrice 'NA' → null` behaviour
+ * (optional field) while still erroring on a malformed `NetPrice` (required).
+ *
+ * Cases: `'1,98'`→`1.98`, `'1.234,56'`→`1234.56`, `'5'`→`5`, `'-5,00'`→`-5`,
+ * empty / `'NA'` → `null`. Rejected (`NaN`): bare-dot (`'1.98'`, `'1.234'`),
+ * trailing junk (`'1,98EUR'`), multiple commas (`'1,2,3'`), other non-numeric.
+ */
+declare function parseEuropeanNumber(s: string): number | null;
+//#endregion
+//#region src/mapping/column-transforms.d.ts
+/**
+ * Discriminated union of bounded transform ops. The TS type is INFERRED from
+ * this schema so there's a single source of truth + the snapshot is validated at
+ * the engine boundary.
+ */
+declare const columnTransformSchema: z.ZodDiscriminatedUnion<"op", [z.ZodObject<{
+  op: z.ZodLiteral<"trim">;
+}, "strip", z.ZodTypeAny, {
+  op: "trim";
+}, {
+  op: "trim";
+}>, z.ZodObject<{
+  op: z.ZodLiteral<"uppercase">;
+}, "strip", z.ZodTypeAny, {
+  op: "uppercase";
+}, {
+  op: "uppercase";
+}>, z.ZodObject<{
+  op: z.ZodLiteral<"lowercase">;
+}, "strip", z.ZodTypeAny, {
+  op: "lowercase";
+}, {
+  op: "lowercase";
+}>, z.ZodObject<{
+  op: z.ZodLiteral<"stripPrefix">;
+  value: z.ZodString;
+}, "strip", z.ZodTypeAny, {
+  value: string;
+  op: "stripPrefix";
+}, {
+  value: string;
+  op: "stripPrefix";
+}>, z.ZodObject<{
+  op: z.ZodLiteral<"stripSuffix">;
+  value: z.ZodString;
+}, "strip", z.ZodTypeAny, {
+  value: string;
+  op: "stripSuffix";
+}, {
+  value: string;
+  op: "stripSuffix";
+}>, z.ZodObject<{
+  op: z.ZodLiteral<"defaultIfEmpty">;
+  value: z.ZodString;
+}, "strip", z.ZodTypeAny, {
+  value: string;
+  op: "defaultIfEmpty";
+}, {
+  value: string;
+  op: "defaultIfEmpty";
+}>, z.ZodObject<{
+  op: z.ZodLiteral<"number">;
+}, "strip", z.ZodTypeAny, {
+  op: "number";
+}, {
+  op: "number";
+}>, z.ZodObject<{
+  op: z.ZodLiteral<"europeanNumber">;
+}, "strip", z.ZodTypeAny, {
+  op: "europeanNumber";
+}, {
+  op: "europeanNumber";
+}>]>;
+/** One declarative cleaning operation. */
+type ColumnTransform = z.infer<typeof columnTransformSchema>;
+/** The value a column carries as it moves through the transform pipeline. */
+type CellValue = string | number | null;
+/**
+ * Apply a transform pipeline to a raw cell, left-to-right. String ops act on the
+ * string value; the `number` / `europeanNumber` ops are terminal (convert to
+ * `number | null`, with `NaN` for a malformed numeric — the schema's `buildRow`
+ * surfaces that as a row error). Once the value is non-string, later string ops
+ * pass through unchanged.
+ */
+declare function applyColumnTransforms(raw: string, transforms: readonly ColumnTransform[], format: NumberFormat): CellValue;
+//#endregion
+//#region src/mapping/target-schema.d.ts
+/** One whitelisted target field on a schema. */
+interface TargetField {
+  /** Stable target key referenced by `column_map[].target`. */
+  key: string;
+  /** Human label for the wizard dropdown (PR3). */
+  label: string;
+  /** Whether a mapping for this schema MUST include this target. */
+  required: boolean;
+  /** Coarse type — drives the wizard's transform suggestions; `buildRow` does the real coercion. */
+  type: 'string' | 'number';
+}
+/** The per-row product of the engine: target field key → cleaned value. */
+type MappedRow = Record<string, CellValue>;
+/** Per-row context handed to `buildRow`. `params` is the opaque per-run params object. */
+interface BuildRowContext {
+  importRunId: string;
+  rowNumber: number;
+  /**
+   * Opaque per-run params (the parts schema reads brand/supplier identity from
+   * it; `discount_table` ignores it). The schema validates the shape it needs.
+   */
+  params: Record<string, unknown>;
+}
+/** A registered target schema: a whitelist + a row builder. */
+interface TargetSchema<TOut> {
+  /** Registered name — also the `import_mapping.target_schema` value. */
+  name: string;
+  /** The target-field whitelist (injection boundary). */
+  fields: readonly TargetField[];
+  /** Build a typed output (success) / skip / error from a mapped row. */
+  buildRow(mapped: MappedRow, ctx: BuildRowContext): RowResult<TOut>;
+}
+/**
+ * A nullable text cell → trimmed string, or `null` for empty / non-string. Accepts
+ * `undefined` (an UNMAPPED target — indexed access under `noUncheckedIndexedAccess`)
+ * and treats it as absent.
+ */
+declare function textOrNull(v: CellValue | undefined): string | null;
+/** The three outcomes a `number` / `europeanNumber` op produces for a cell. */
+type NumberOutcome = {
+  kind: 'value';
+  value: number;
+} | {
+  kind: 'absent';
+} | {
+  kind: 'invalid';
+};
+/**
+ * Classify a cell that should hold a number. `undefined` (unmapped) / `null` →
+ * absent; a real number → value (`NaN` → invalid); a non-empty raw string (no
+ * number op declared) → invalid; a blank raw string → absent.
+ */
+declare function numberCell(v: CellValue | undefined): NumberOutcome;
+/** Build a `RowResult` error variant (assignable to any `TargetSchema<TOut>`). */
+declare function rowError(errorType: string, message: string, field: string): {
+  kind: "error";
+  error: {
+    errorType: string;
+    message: string;
+    field: string;
+  };
+};
+//#endregion
+//#region src/mapping/parts-schema.d.ts
+/**
+ * Params the `parts` schema reads from the run — the per-(supplier, brand)
+ * identity resolved at enqueue time. Identical to the carmaker transform's params
+ * (the carmaker path re-exports this), so brand/supplier injection + the optional
+ * supplier code-prefix strip are validated the same way everywhere.
+ */
+declare const partsParamsSchema: z.ZodObject<{
+  brandId: z.ZodString;
+  brandSlug: z.ZodString;
+  supplierId: z.ZodString;
+  supplierDisplayName: z.ZodString;
+  importBatchId: z.ZodString; /** Supplier code prefix to strip before normalization (e.g. `'ME-'`). Optional. */
+  supplierCodePrefix: z.ZodOptional<z.ZodString>;
+}, "strip", z.ZodTypeAny, {
+  supplierId: string;
+  brandId: string;
+  brandSlug: string;
+  supplierDisplayName: string;
+  importBatchId: string;
+  supplierCodePrefix?: string | undefined;
+}, {
+  supplierId: string;
+  brandId: string;
+  brandSlug: string;
+  supplierDisplayName: string;
+  importBatchId: string;
+  supplierCodePrefix?: string | undefined;
+}>;
+type PartsParams = z.infer<typeof partsParamsSchema>;
+/**
+ * The `parts` schema. `buildRow` THROWS on invalid run params (so a misconfigured
+ * run surfaces as a transform error, matching the carmaker contract) and returns a
+ * typed `RowResult` per row otherwise.
+ */
+declare const partsSchema: TargetSchema<PartsDocument>;
+//#endregion
+export { parseEuropeanNumber as _, MappedRow as a, TargetSchema as c, textOrNull as d, CellValue as f, NumberFormat as g, columnTransformSchema as h, BuildRowContext as i, numberCell as l, applyColumnTransforms as m, partsParamsSchema as n, NumberOutcome as o, ColumnTransform as p, partsSchema as r, TargetField as s, PartsParams as t, rowError as u, parseNumber as v, toMinorUnits as y };
+//# sourceMappingURL=parts-schema-RI_ltubl.d.mts.map

package/dist/parts-schema-RI_ltubl.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"parts-schema-RI_ltubl.d.mts","names":[],"sources":["../src/mapping/number-format.ts","../src/mapping/column-transforms.ts","../src/mapping/target-schema.ts","../src/mapping/parts-schema.ts"],"mappings":";;;;;;;;;;AAqBA;;;;AAIW;AAsBX;;;;;;;;AAA6D;AA+B7D;AAAA,UAzDiB,YAAA;;EAEf,OAAA;EAuDwC;EArDxC,SAAS;AAAA;;;AA+EkC;;;;AC9E7C;;;;;;;;;;iBDqBgB,WAAA,CAAY,GAAA,UAAa,MAAA,EAAQ,YAAY;;;;;;;;iBA+B7C,YAAA,CAAa,KAAa;;;;;;;;;;;;;;;;;;;;;;;iBA0B1B,mBAAA,CAAoB,CAAS;;;;;AAzDgB;AA+B7D;;cCpDa,qBAAA,EAAqB,CAAA,CAAA,qBAAA,QAAA,CAAA,CAAA,SAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KAYtB,eAAA,GAAkB,CAAA,CAAE,KAAK,QAAQ,qBAAA;;KAGjC,SAAA;;;;;;;;iBASI,qBAAA,CACd,GAAA,UACA,UAAA,WAAqB,eAAA,IACrB,MAAA,EAAQ,YAAA,GACP,SAAA;;;;UCpCc,WAAA;EF6B4C;EE3B3D,GAAA;EF0Dc;EExDd,KAAA;;EAEA,QAAA;EFsDwC;EEpDxC,IAAA;AAAA;;KAIU,SAAA,GAAY,MAAM,SAAS,SAAA;AF0EM;AAAA,UEvE5B,eAAA;EACf,WAAA;EACA,SAAA;EDTW;;;;ECcX,MAAA,EAAQ,MAAM;AAAA;;UAIC,YAAA;;EAEf,IAAA;;EAEA,MAAA,WAAiB,WAAA;;EAEjB,QAAA,CAAS,MAAA,EAAQ,SAAA,EAAW,GAAA,EAAK,eAAA,GAAkB,SAAA,CAAU,IAAA;AAAA;;;;;;iBAY/C,UAAA,CAAW,CAAwB,EAArB,SAAS;;KAO3B,aAAA;EACN,IAAA;EAAe,KAAA;AAAA;EACf,IAAA;AAAA;EACA,IAAA;AAAA;;;;;;iBAOU,UAAA,CAAW,CAAA,EAAG,SAAA,eAAwB,aAAa;;iBASnD,QAAA,CAAS,SAAA,UAAmB,OAAA,UAAiB,KAAA;;;;;;;;;;AFVnB;AA0B1C;;;;AAA6C;AA1BH,cGtC7B,iBAAA,EAAiB,CAAA,CAAA,SAAA;;;;;;;;;;;;;;;;;;;;;;KAalB,WAAA,GAAc,CAAA,CAAE,KAAK,QAAQ,iBAAA;;;;;;cAmD5B,WAAA,EAAa,YAAY,CAAC,aAAA"}

package/dist/plugin.d.mts

@@ -1,4 +1,4 @@
-import { n as FilePathResolver, t as EsClientResolver } from "./worker-BY0JQw39.mjs";
+import { n as FilePathResolver, t as EsClientResolver } from "./worker-ByPOP6Qe.mjs";
 import { ImportsStorageFactory } from "./storage.mjs";
 
 //#region src/plugin.d.ts

package/dist/plugin.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"plugin.d.mts","names":[],"sources":["../src/plugin.ts"],"mappings":";;;;UAqEiB,oBAAA;EA4FoC;;;;;EAtFnD,QAAA,EAAU,gBAAA;;;;;;;EAOV,OAAA;;;;;;;;;;;;;;;EAeA,eAAA,GAAkB,gBAAA;;;;;;;;;;;;;;;;;;EAkBlB,OAAA,GAAU,qBAAA;;;;;;;;;;;;;;;EAeV,aAAA;;;;;;;;;;;;;;;EAeA,oBAAA;;;;;;;;;;;;;EAaA,sBAAA;AAAA;AAAA,iBAGc,OAAA,CAAQ,OAAA,EAAS,oBAAA,GAAoB,IAAA,6BAAA,MAAA;EAAA"}
+{"version":3,"file":"plugin.d.mts","names":[],"sources":["../src/plugin.ts"],"mappings":";;;;UAsEiB,oBAAA;EA4FoC;;;;;EAtFnD,QAAA,EAAU,gBAAA;;;;;;;EAOV,OAAA;;;;;;;;;;;;;;;EAeA,eAAA,GAAkB,gBAAA;;;;;;;;;;;;;;;;;;EAkBlB,OAAA,GAAU,qBAAA;;;;;;;;;;;;;;;EAeV,aAAA;;;;;;;;;;;;;;;EAeA,oBAAA;;;;;;;;;;;;;EAaA,sBAAA;AAAA;AAAA,iBAGc,OAAA,CAAQ,OAAA,EAAS,oBAAA,GAAoB,IAAA,6BAAA,MAAA;EAAA"}

package/dist/plugin.mjs

@@ -1,2 +1,2 @@
-import{a as e,n as t}from"./transform-BzqTvH7B.mjs";import{registerImportsStorageFactory as n}from"./storage.mjs";import{createImportsRunDeadListener as r,createRunImportHandler as i,importsRunJob as a}from"./worker.mjs";import{definePlugin as o}from"@murumets-ee/core";const s=Symbol.for(`@murumets-ee/imports:dead-listener-unsubscribe`);function c(c){let l=c.esIndex??`parts`,u=c.storageBucket===null?null:{bucket:c.storageBucket??`lumi-imports`,orphanTtlHours:c.orphanUploadTtlHours??2,processedRetentionDays:c.processedRetentionDays??7};return c.storage&&n(c.storage),o({name:`@murumets-ee/imports`,server:{entities:[e],init:async n=>{if(!n.plugins.all().some(e=>e.name===`@murumets-ee/queue`)){n.logger.warn(`imports: queue() plugin not in plugins array — imports:run jobs will not be processed`);return}let o=await import(`@murumets-ee/queue/client`),d=n.getClient(e),f=t();o.registerJob(a,i({importRuns:d,transforms:f,esClient:c.esClient,esIndex:l,...c.resolveFilePath!==void 0&&{resolveFilePath:c.resolveFilePath},logger:n.logger.child({pkg:`imports`})}));let p=globalThis[s];p&&p();let m=o.addQueueTerminalListener(r({importRuns:d,logger:n.logger.child({pkg:`imports`})}));if(globalThis[s]=m,u!==null){let{cleanupOrphanedImports:e,cleanupProcessedImports:t}=await import(`./cleanup-DJHXcbpZ.mjs`).then(e=>e.r),{runAsCli:r}=await import(`@murumets-ee/core`),i=n.logger.child({pkg:`imports`,job:`cleanup`});o.registerJob(o.defineJob({name:`imports:cleanup-orphaned-uploads`,description:`Delete toolkit_files rows in the imports bucket with no matching import_run reference and createdAt older than the configured TTL.`,schedule:`@every 1h`}),async()=>{await r(()=>e({app:n,bucket:u.bucket,ttlHours:u.orphanTtlHours,logger:i}))}),o.registerJob(o.defineJob({name:`imports:cleanup-processed-uploads`,description:`Delete toolkit_files rows for import_runs in terminal status (succeeded/failed/cancelled) whose finishedAt is older than the configured retention window. The import_run row itself is preserved.`,schedule:`@every 1h`}),async()=>{await r(()=>t({app:n,bucket:u.bucket,retentionDays:u.processedRetentionDays,logger:i}))})}n.logger.info({esIndex:l,cleanup:u===null?{enabled:!1}:{enabled:!0,...u}},`Imports plugin initialized`)}}})}export{c as imports};
+import{c as e,n as t,o as n}from"./transform-DloItB80.mjs";import{registerImportsStorageFactory as r}from"./storage.mjs";import{createImportsRunDeadListener as i,createRunImportHandler as a,importsRunJob as o}from"./worker.mjs";import{definePlugin as s}from"@murumets-ee/core";const c=Symbol.for(`@murumets-ee/imports:dead-listener-unsubscribe`);function l(l){let u=l.esIndex??`parts`,d=l.storageBucket===null?null:{bucket:l.storageBucket??`lumi-imports`,orphanTtlHours:l.orphanUploadTtlHours??2,processedRetentionDays:l.processedRetentionDays??7};return l.storage&&r(l.storage),s({name:`@murumets-ee/imports`,server:{entities:[e,n],init:async n=>{if(!n.plugins.all().some(e=>e.name===`@murumets-ee/queue`)){n.logger.warn(`imports: queue() plugin not in plugins array — imports:run jobs will not be processed`);return}let r=await import(`@murumets-ee/queue/client`),s=n.getClient(e),f=t();r.registerJob(o,a({importRuns:s,transforms:f,esClient:l.esClient,esIndex:u,...l.resolveFilePath!==void 0&&{resolveFilePath:l.resolveFilePath},logger:n.logger.child({pkg:`imports`})}));let p=globalThis[c];p&&p();let m=r.addQueueTerminalListener(i({importRuns:s,logger:n.logger.child({pkg:`imports`})}));if(globalThis[c]=m,d!==null){let{cleanupOrphanedImports:e,cleanupProcessedImports:t}=await import(`./cleanup-CFo7Pw0t.mjs`).then(e=>e.r),{runAsCli:i}=await import(`@murumets-ee/core`),a=n.logger.child({pkg:`imports`,job:`cleanup`});r.registerJob(r.defineJob({name:`imports:cleanup-orphaned-uploads`,description:`Delete toolkit_files rows in the imports bucket with no matching import_run reference and createdAt older than the configured TTL.`,schedule:`@every 1h`}),async()=>{await i(()=>e({app:n,bucket:d.bucket,ttlHours:d.orphanTtlHours,logger:a}))}),r.registerJob(r.defineJob({name:`imports:cleanup-processed-uploads`,description:`Delete toolkit_files rows for import_runs in terminal status (succeeded/failed/cancelled) whose finishedAt is older than the configured retention window. The import_run row itself is preserved.`,schedule:`@every 1h`}),async()=>{await i(()=>t({app:n,bucket:d.bucket,retentionDays:d.processedRetentionDays,logger:a}))})}n.logger.info({esIndex:u,cleanup:d===null?{enabled:!1}:{enabled:!0,...d}},`Imports plugin initialized`)}}})}export{l as imports};
 //# sourceMappingURL=plugin.mjs.map

package/dist/plugin.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"plugin.mjs","names":[],"sources":["../src/plugin.ts"],"sourcesContent":["/**\n * Imports plugin — registers the `import_run` entity and (when ES + queue\n * are wired) registers the `imports:run` job handler.\n *\n * Per PLAN-ECOMMERCE.md PR 7 (PoC scope): plumbing only. The bespoke\n * admin pages (upload form, parts search) live in PR 8a. Per-feed\n * transforms (carmaker, …) live in PR 8 and self-register against the\n * process-global transform registry — this plugin's init does NOT\n * import any transform.\n *\n * @example\n * ```ts\n * import { defineLumiConfig } from '@murumets-ee/core'\n * import { imports } from '@murumets-ee/imports/plugin'\n * import { Client } from '@elastic/elasticsearch'\n *\n * const es = new Client({ node: process.env.ES_URL })\n *\n * export default defineLumiConfig({\n *   plugins: [\n *     // queue() and the carmaker-transform plugin must be present too.\n *     queue(),\n *     imports({ esClient: () => es, esIndex: 'parts' }),\n *   ],\n * })\n * ```\n */\n\nimport { definePlugin } from '@murumets-ee/core'\n// **Why no static `@murumets-ee/core/clients` import in this file.**\n//\n// `@murumets-ee/core/clients` begins with `import 'server-only'`, which\n// throws when this plugin module is loaded at `lumi.config.ts`\n// evaluation time by the CLI (`lumi migrate`/`seed`/`doctor` use jiti,\n// which doesn't apply the `react-server` export condition).\n// Static-importing would propagate that throw all the way through\n// `lumi.config.ts` → CLI startup, blocking every CLI invocation.\n//\n// **Today (after PLAN-DECLARATIVE-PLUGINS PR 2):** plugin `init`\n// reaches `AdminClient` via the framework-provided `app.getClient(\n// entity)` accessor — no `createAdminClient` call, no dynamic import\n// either. The remaining `@murumets-ee/core/clients` dependency lives\n// in `./admin/cleanup.js`, which IS still dynamic-imported inside\n// `init` and which still benefits from this rationale. The pattern\n// matches `getDb()` in `@murumets-ee/queue/admin.ts`.\nimport { ImportRun } from './entities/import-run.js'\nimport { type ImportsStorageFactory, registerImportsStorageFactory } from './storage.js'\nimport { getTransformRegistry } from './transform.js'\nimport {\n  createImportsRunDeadListener,\n  createRunImportHandler,\n  type EsClientResolver,\n  type FilePathResolver,\n  importsRunJob,\n} from './worker.js'\n\n/**\n * `Symbol.for`-keyed slot for the active dead-listener unsubscribe\n * handle. Symbol.for so an HMR-replaced module instance still finds\n * the previous unsubscribe (`Symbol`-by-reference would skip the\n * lookup across module-evaluation boundaries). Same pattern as\n * `transform.ts`'s registry slot and `storage.ts`'s factory slot.\n */\nconst IMPORTS_DEAD_LISTENER_KEY = Symbol.for('@murumets-ee/imports:dead-listener-unsubscribe')\n\ninterface GlobalThisWithUnsubscribe {\n  [IMPORTS_DEAD_LISTENER_KEY]?: () => void\n}\n\nexport interface ImportsPluginOptions {\n  /**\n   * Elasticsearch client resolver. Lazy so consumers can construct\n   * the client inside a route initialiser instead of at plugin-init\n   * time. Required — without it, the worker can't bulk-write.\n   */\n  esClient: EsClientResolver\n  /**\n   * ES index alias to write to. Defaults to `'parts'` (the\n   * `PARTS_INDEX_ALIAS` re-exported from `@murumets-ee/search-elasticsearch`,\n   * inlined here to avoid a runtime import for a one-line constant).\n   * Per D6 always pass an alias, never a physical index.\n   */\n  esIndex?: string\n  /**\n   * Resolve `import_run.filePath` to a local FS path before streaming.\n   * **Required when uploads land in remote storage** (R2/S3/etc.) —\n   * without it, the worker hands the storage key to `createReadStream`\n   * and crashes with `ENOENT`.\n   *\n   * Typical wiring downloads the storage object to a tmpfile and\n   * returns its path; the optional `cleanup` callback runs after the\n   * run finishes (success or failure).\n   *\n   * Omit this option for the original on-disk PoC setup where the\n   * upload route writes directly to a local directory and persists\n   * the absolute path on `import_run.filePath`.\n   */\n  resolveFilePath?: FilePathResolver\n  /**\n   * Optional dedicated `StorageClient` factory for imports. When\n   * provided, the upload route (in `@murumets-ee/commerce`) and the\n   * worker resolver consult this client instead of the global\n   * `getStorageConfig()` default — so a deployment can pin imports\n   * to a local-disk adapter (single-VPS / dev) while media,\n   * ticketing attachments, and the rest of the app keep using the\n   * default R2 (or S3) adapter.\n   *\n   * Lazy / async so consumers can defer adapter construction until\n   * the first import — useful when the local-disk root is\n   * environment-dependent or when the factory needs `getApp()` to\n   * have resolved.\n   *\n   * Omit to keep the legacy single-storage behaviour (everything on\n   * the global default).\n   */\n  storage?: ImportsStorageFactory\n  /**\n   * Bucket name to scope the orphan-cleanup queue job. Rows in\n   * `toolkit_files` with this bucket value AND no matching\n   * `import_run.filePath` AND `createdAt < cutoff` get garbage-\n   * collected on each cleanup tick. Defaults to `'lumi-imports'` —\n   * the convention `apps/admin-playground/lib/imports-storage.ts`\n   * uses by default. Override if your deployment pins a different\n   * bucket name via `IMPORTS_STORAGE_BUCKET` or equivalent.\n   *\n   * Set to `null` to DISABLE the cleanup job entirely (e.g. for\n   * deployments that handle their own retention via S3 lifecycle\n   * rules). Disabling without an alternative will leak storage\n   * over time as failed uploads accumulate.\n   */\n  storageBucket?: string | null\n  /**\n   * Hours after which an unreferenced `toolkit_files` row in the\n   * imports bucket is considered orphaned and deleted by the\n   * cleanup job. Default 2h — long enough that a normal\n   * upload-then-register round-trip never wipes a fresh upload,\n   * short enough that storage waste from failed uploads doesn't\n   * accumulate.\n   *\n   * The upload→register flow happens client-side automatically (no\n   * \"operator clicks register\" step in the current UI), so the TTL\n   * just needs to cover network blips + retries. Bumping the TTL\n   * is safe if your operators commonly upload then wait before\n   * processing.\n   */\n  orphanUploadTtlHours?: number\n  /**\n   * Days after which a terminal-status `import_run`'s source file is\n   * deleted from `toolkit_files`. Default 7 days — gives operators a\n   * reprocess window if a transform bug surfaces post-success, while\n   * keeping storage costs bounded. Set to 0 for instant cleanup at\n   * the next cleanup tick after `import_run.status` becomes terminal\n   * (`succeeded` / `failed` / `cancelled`).\n   *\n   * The `import_run` row itself is preserved — the audit trail and\n   * History panel still show the past run; only the source feed\n   * file is removed.\n   */\n  processedRetentionDays?: number\n}\n\nexport function imports(options: ImportsPluginOptions) {\n  const esIndex = options.esIndex ?? 'parts'\n  // Resolve the cleanup config eagerly. `storageBucket === null` is the\n  // explicit opt-out path; `undefined` falls back to the convention.\n  // Two cleanup jobs share the same bucket but have independent TTLs:\n  // orphans (no import_run ref) churn fast (2h); processed runs\n  // (terminal status) get a longer reprocess window (7d).\n  const cleanupConfig =\n    options.storageBucket === null\n      ? null\n      : {\n          bucket: options.storageBucket ?? 'lumi-imports',\n          orphanTtlHours: options.orphanUploadTtlHours ?? 2,\n          processedRetentionDays: options.processedRetentionDays ?? 7,\n        }\n\n  // Register the storage factory eagerly — runs at `imports()` factory\n  // call time during `lumi.config.ts` evaluation, NOT inside the\n  // `init` hook. The hook only fires at `createApp()` time (Next.js\n  // request boot / worker boot), but the factory needs to be visible\n  // to consumers that read it at module-load (e.g. the commerce\n  // route's lazy `getStorageClient` thunk closes over a value that\n  // resolves the moment a request comes in). Late registration in\n  // `init` would race the first request.\n  if (options.storage) {\n    registerImportsStorageFactory(options.storage)\n  }\n\n  return definePlugin({\n    name: '@murumets-ee/imports',\n    server: {\n      entities: [ImportRun],\n      init: async (app) => {\n        // Queue PACKAGE being importable doesn't mean the queue() PLUGIN\n        // is in the consumer's plugins array. Without it, registerJob()\n        // succeeds but no worker ever picks up the job. Probe and warn.\n        const queuePluginPresent = app.plugins.all().some((p) => p.name === '@murumets-ee/queue')\n        if (!queuePluginPresent) {\n          app.logger.warn(\n            'imports: queue() plugin not in plugins array — imports:run jobs will not be processed',\n          )\n          return\n        }\n\n        // Queue client stays dynamic-imported: we already early-returned\n        // above when the queue() plugin isn't in the consumer's plugins\n        // array, so static-importing here would force-load the queue\n        // client module in deployments that don't enable the worker.\n        // `@murumets-ee/core/clients` is NO LONGER imported here —\n        // `app.getClient(entity)` gives us a typed, app-bound\n        // AdminClient without the F1+F2 foot-guns from\n        // PLAN-DECLARATIVE-PLUGINS.md.\n        const queueClientModule = await import('@murumets-ee/queue/client')\n\n        const importRuns = app.getClient(ImportRun)\n        const transforms = getTransformRegistry()\n\n        queueClientModule.registerJob(\n          importsRunJob,\n          createRunImportHandler({\n            importRuns,\n            transforms,\n            esClient: options.esClient,\n            esIndex,\n            ...(options.resolveFilePath !== undefined && {\n              resolveFilePath: options.resolveFilePath,\n            }),\n            logger: app.logger.child({ pkg: 'imports' }),\n          }),\n        )\n\n        // Subscribe to the queue's terminal-state event stream so a\n        // worker that crashes mid-import — and whose `imports:run` job\n        // ultimately reaches `dead` without the handler reaching its\n        // own `markFailed` path — flips the mirror `import_run` row to\n        // `failed`. This replaces what would otherwise be a per-domain\n        // periodic sweep; the queue's terminal notifier is the generic\n        // primitive (graphile-worker's `job:failed` event, BullMQ's\n        // `failed` listener — same shape).\n        //\n        // The listener is multi-subscriber, so we coexist with the\n        // notifications plugin's admin-email handler without either of\n        // us clobbering the other. Filtering on `event.jobType` keeps\n        // this listener cheap on jobs we don't own.\n        //\n        // **HMR**: each re-init of this `init` hook constructs a new\n        // listener instance. The notifier registry stores listeners\n        // by reference, so without explicit cleanup an HMR cycle would\n        // STACK duplicate listeners — `findMany`+`update` per dead\n        // event growing linearly with the number of HMR rounds. Stash\n        // the unsubscribe on a globalThis Symbol-keyed slot (same\n        // pattern as `transform.ts` / `storage.ts`) and call it before\n        // re-subscribing so HMR is idempotent.\n        const previousUnsubscribe = (globalThis as GlobalThisWithUnsubscribe)[\n          IMPORTS_DEAD_LISTENER_KEY\n        ]\n        if (previousUnsubscribe) previousUnsubscribe()\n        const unsubscribe = queueClientModule.addQueueTerminalListener(\n          createImportsRunDeadListener({\n            importRuns,\n            logger: app.logger.child({ pkg: 'imports' }),\n          }),\n        )\n        ;(globalThis as GlobalThisWithUnsubscribe)[IMPORTS_DEAD_LISTENER_KEY] = unsubscribe\n\n        // Storage-cleanup jobs. Two passes, same bucket:\n        //   1. Orphan sweep — uploads whose register handshake never\n        //      completed (no `import_run` reference). 2h TTL.\n        //   2. Processed sweep — uploads whose `import_run` is terminal\n        //      (succeeded / failed / cancelled). 7-day reprocess\n        //      window then drop.\n        // `defineJob({ schedule })` declares the cadence; the queue\n        // worker reads the metadata at startup, materialises a\n        // `toolkit_scheduled_jobs` row, and its cron loop fires the\n        // registered handler each tick. No explicit\n        // `client.schedule()` call needed — the worker owns the\n        // schedule, not the handler. Same helper-signature shape as\n        // ticketing's cleanup.\n        if (cleanupConfig !== null) {\n          const { cleanupOrphanedImports, cleanupProcessedImports } = await import(\n            './admin/cleanup.js'\n          )\n          const { runAsCli } = await import('@murumets-ee/core')\n          const cleanupLogger = app.logger.child({ pkg: 'imports', job: 'cleanup' })\n\n          queueClientModule.registerJob(\n            queueClientModule.defineJob({\n              name: 'imports:cleanup-orphaned-uploads',\n              description:\n                'Delete toolkit_files rows in the imports bucket with no matching import_run reference and createdAt older than the configured TTL.',\n              schedule: '@every 1h',\n            }),\n            async () => {\n              // `runAsCli` is load-bearing today: the queue worker's\n              // handler context (see `packages/queue/src/handler-context.ts`)\n              // tracks the job type but NOT a user / RequestContext.\n              // Any AdminClient call reachable from the cleanup\n              // helpers — currently `createAdminClient(ImportRun).getTable()`\n              // for column-metadata resolution — runs without a user\n              // unless we enter a CLI scope. The synthetic CLI user\n              // is the right attribution for scheduled-job work\n              // (matches ticketing's `cleanupOrphanedUploads` shape).\n              await runAsCli(() =>\n                cleanupOrphanedImports({\n                  app,\n                  bucket: cleanupConfig.bucket,\n                  ttlHours: cleanupConfig.orphanTtlHours,\n                  logger: cleanupLogger,\n                }),\n              )\n            },\n          )\n\n          queueClientModule.registerJob(\n            queueClientModule.defineJob({\n              name: 'imports:cleanup-processed-uploads',\n              description:\n                'Delete toolkit_files rows for import_runs in terminal status (succeeded/failed/cancelled) whose finishedAt is older than the configured retention window. The import_run row itself is preserved.',\n              schedule: '@every 1h',\n            }),\n            async () => {\n              await runAsCli(() =>\n                cleanupProcessedImports({\n                  app,\n                  bucket: cleanupConfig.bucket,\n                  retentionDays: cleanupConfig.processedRetentionDays,\n                  logger: cleanupLogger,\n                }),\n              )\n            },\n          )\n        }\n\n        app.logger.info(\n          {\n            esIndex,\n            // Stable shape regardless of enabled/disabled so structured-\n            // log queries (`cleanup.enabled = false`) don't have to\n            // handle a string fallback.\n            cleanup:\n              cleanupConfig === null ? { enabled: false } : { enabled: true, ...cleanupConfig },\n          },\n          'Imports plugin initialized',\n        )\n      },\n    },\n  })\n}\n"],"mappings":"8QA+DA,MAAM,EAA4B,OAAO,IAAI,gDAAgD,EAkG7F,SAAgB,EAAQ,EAA+B,CACrD,IAAM,EAAU,EAAQ,SAAW,QAM7B,EACJ,EAAQ,gBAAkB,KACtB,KACA,CACE,OAAQ,EAAQ,eAAiB,eACjC,eAAgB,EAAQ,sBAAwB,EAChD,uBAAwB,EAAQ,wBAA0B,CAC5D,EAcN,OAJI,EAAQ,SACV,EAA8B,EAAQ,OAAO,EAGxC,EAAa,CAClB,KAAM,uBACN,OAAQ,CACN,SAAU,CAAC,CAAS,EACpB,KAAM,KAAO,IAAQ,CAKnB,GAAI,CADuB,EAAI,QAAQ,IAAI,CAAC,CAAC,KAAM,GAAM,EAAE,OAAS,oBAC9C,EAAG,CACvB,EAAI,OAAO,KACT,uFACF,EACA,MACF,CAUA,IAAM,EAAoB,MAAM,OAAO,6BAEjC,EAAa,EAAI,UAAU,CAAS,EACpC,EAAa,EAAqB,EAExC,EAAkB,YAChB,EACA,EAAuB,CACrB,aACA,aACA,SAAU,EAAQ,SAClB,UACA,GAAI,EAAQ,kBAAoB,IAAA,IAAa,CAC3C,gBAAiB,EAAQ,eAC3B,EACA,OAAQ,EAAI,OAAO,MAAM,CAAE,IAAK,SAAU,CAAC,CAC7C,CAAC,CACH,EAwBA,IAAM,EAAuB,WAC3B,GAEE,GAAqB,EAAoB,EAC7C,IAAM,EAAc,EAAkB,yBACpC,EAA6B,CAC3B,aACA,OAAQ,EAAI,OAAO,MAAM,CAAE,IAAK,SAAU,CAAC,CAC7C,CAAC,CACH,EAgBA,GAfC,WAA0C,GAA6B,EAepE,IAAkB,KAAM,CAC1B,GAAM,CAAE,yBAAwB,2BAA4B,MAAM,OAChE,yBACD,CAAA,KAAA,GAAA,EAAA,CAAA,EACK,CAAE,YAAa,MAAM,OAAO,qBAC5B,EAAgB,EAAI,OAAO,MAAM,CAAE,IAAK,UAAW,IAAK,SAAU,CAAC,EAEzE,EAAkB,YAChB,EAAkB,UAAU,CAC1B,KAAM,mCACN,YACE,qIACF,SAAU,WACZ,CAAC,EACD,SAAY,CAUV,MAAM,MACJ,EAAuB,CACrB,MACA,OAAQ,EAAc,OACtB,SAAU,EAAc,eACxB,OAAQ,CACV,CAAC,CACH,CACF,CACF,EAEA,EAAkB,YAChB,EAAkB,UAAU,CAC1B,KAAM,oCACN,YACE,oMACF,SAAU,WACZ,CAAC,EACD,SAAY,CACV,MAAM,MACJ,EAAwB,CACtB,MACA,OAAQ,EAAc,OACtB,cAAe,EAAc,uBAC7B,OAAQ,CACV,CAAC,CACH,CACF,CACF,CACF,CAEA,EAAI,OAAO,KACT,CACE,UAIA,QACE,IAAkB,KAAO,CAAE,QAAS,EAAM,EAAI,CAAE,QAAS,GAAM,GAAG,CAAc,CACpF,EACA,4BACF,CACF,CACF,CACF,CAAC,CACH"}
+{"version":3,"file":"plugin.mjs","names":[],"sources":["../src/plugin.ts"],"sourcesContent":["/**\n * Imports plugin — registers the `import_run` entity and (when ES + queue\n * are wired) registers the `imports:run` job handler.\n *\n * Per PLAN-ECOMMERCE.md PR 7 (PoC scope): plumbing only. The bespoke\n * admin pages (upload form, parts search) live in PR 8a. Per-feed\n * transforms (carmaker, …) live in PR 8 and self-register against the\n * process-global transform registry — this plugin's init does NOT\n * import any transform.\n *\n * @example\n * ```ts\n * import { defineLumiConfig } from '@murumets-ee/core'\n * import { imports } from '@murumets-ee/imports/plugin'\n * import { Client } from '@elastic/elasticsearch'\n *\n * const es = new Client({ node: process.env.ES_URL })\n *\n * export default defineLumiConfig({\n *   plugins: [\n *     // queue() and the carmaker-transform plugin must be present too.\n *     queue(),\n *     imports({ esClient: () => es, esIndex: 'parts' }),\n *   ],\n * })\n * ```\n */\n\nimport { definePlugin } from '@murumets-ee/core'\n// **Why no static `@murumets-ee/core/clients` import in this file.**\n//\n// `@murumets-ee/core/clients` begins with `import 'server-only'`, which\n// throws when this plugin module is loaded at `lumi.config.ts`\n// evaluation time by the CLI (`lumi migrate`/`seed`/`doctor` use jiti,\n// which doesn't apply the `react-server` export condition).\n// Static-importing would propagate that throw all the way through\n// `lumi.config.ts` → CLI startup, blocking every CLI invocation.\n//\n// **Today (after PLAN-DECLARATIVE-PLUGINS PR 2):** plugin `init`\n// reaches `AdminClient` via the framework-provided `app.getClient(\n// entity)` accessor — no `createAdminClient` call, no dynamic import\n// either. The remaining `@murumets-ee/core/clients` dependency lives\n// in `./admin/cleanup.js`, which IS still dynamic-imported inside\n// `init` and which still benefits from this rationale. The pattern\n// matches `getDb()` in `@murumets-ee/queue/admin.ts`.\nimport { ImportMapping } from './entities/import-mapping.js'\nimport { ImportRun } from './entities/import-run.js'\nimport { type ImportsStorageFactory, registerImportsStorageFactory } from './storage.js'\nimport { getTransformRegistry } from './transform.js'\nimport {\n  createImportsRunDeadListener,\n  createRunImportHandler,\n  type EsClientResolver,\n  type FilePathResolver,\n  importsRunJob,\n} from './worker.js'\n\n/**\n * `Symbol.for`-keyed slot for the active dead-listener unsubscribe\n * handle. Symbol.for so an HMR-replaced module instance still finds\n * the previous unsubscribe (`Symbol`-by-reference would skip the\n * lookup across module-evaluation boundaries). Same pattern as\n * `transform.ts`'s registry slot and `storage.ts`'s factory slot.\n */\nconst IMPORTS_DEAD_LISTENER_KEY = Symbol.for('@murumets-ee/imports:dead-listener-unsubscribe')\n\ninterface GlobalThisWithUnsubscribe {\n  [IMPORTS_DEAD_LISTENER_KEY]?: () => void\n}\n\nexport interface ImportsPluginOptions {\n  /**\n   * Elasticsearch client resolver. Lazy so consumers can construct\n   * the client inside a route initialiser instead of at plugin-init\n   * time. Required — without it, the worker can't bulk-write.\n   */\n  esClient: EsClientResolver\n  /**\n   * ES index alias to write to. Defaults to `'parts'` (the\n   * `PARTS_INDEX_ALIAS` re-exported from `@murumets-ee/search-elasticsearch`,\n   * inlined here to avoid a runtime import for a one-line constant).\n   * Per D6 always pass an alias, never a physical index.\n   */\n  esIndex?: string\n  /**\n   * Resolve `import_run.filePath` to a local FS path before streaming.\n   * **Required when uploads land in remote storage** (R2/S3/etc.) —\n   * without it, the worker hands the storage key to `createReadStream`\n   * and crashes with `ENOENT`.\n   *\n   * Typical wiring downloads the storage object to a tmpfile and\n   * returns its path; the optional `cleanup` callback runs after the\n   * run finishes (success or failure).\n   *\n   * Omit this option for the original on-disk PoC setup where the\n   * upload route writes directly to a local directory and persists\n   * the absolute path on `import_run.filePath`.\n   */\n  resolveFilePath?: FilePathResolver\n  /**\n   * Optional dedicated `StorageClient` factory for imports. When\n   * provided, the upload route (in `@murumets-ee/commerce`) and the\n   * worker resolver consult this client instead of the global\n   * `getStorageConfig()` default — so a deployment can pin imports\n   * to a local-disk adapter (single-VPS / dev) while media,\n   * ticketing attachments, and the rest of the app keep using the\n   * default R2 (or S3) adapter.\n   *\n   * Lazy / async so consumers can defer adapter construction until\n   * the first import — useful when the local-disk root is\n   * environment-dependent or when the factory needs `getApp()` to\n   * have resolved.\n   *\n   * Omit to keep the legacy single-storage behaviour (everything on\n   * the global default).\n   */\n  storage?: ImportsStorageFactory\n  /**\n   * Bucket name to scope the orphan-cleanup queue job. Rows in\n   * `toolkit_files` with this bucket value AND no matching\n   * `import_run.filePath` AND `createdAt < cutoff` get garbage-\n   * collected on each cleanup tick. Defaults to `'lumi-imports'` —\n   * the convention `apps/admin-playground/lib/imports-storage.ts`\n   * uses by default. Override if your deployment pins a different\n   * bucket name via `IMPORTS_STORAGE_BUCKET` or equivalent.\n   *\n   * Set to `null` to DISABLE the cleanup job entirely (e.g. for\n   * deployments that handle their own retention via S3 lifecycle\n   * rules). Disabling without an alternative will leak storage\n   * over time as failed uploads accumulate.\n   */\n  storageBucket?: string | null\n  /**\n   * Hours after which an unreferenced `toolkit_files` row in the\n   * imports bucket is considered orphaned and deleted by the\n   * cleanup job. Default 2h — long enough that a normal\n   * upload-then-register round-trip never wipes a fresh upload,\n   * short enough that storage waste from failed uploads doesn't\n   * accumulate.\n   *\n   * The upload→register flow happens client-side automatically (no\n   * \"operator clicks register\" step in the current UI), so the TTL\n   * just needs to cover network blips + retries. Bumping the TTL\n   * is safe if your operators commonly upload then wait before\n   * processing.\n   */\n  orphanUploadTtlHours?: number\n  /**\n   * Days after which a terminal-status `import_run`'s source file is\n   * deleted from `toolkit_files`. Default 7 days — gives operators a\n   * reprocess window if a transform bug surfaces post-success, while\n   * keeping storage costs bounded. Set to 0 for instant cleanup at\n   * the next cleanup tick after `import_run.status` becomes terminal\n   * (`succeeded` / `failed` / `cancelled`).\n   *\n   * The `import_run` row itself is preserved — the audit trail and\n   * History panel still show the past run; only the source feed\n   * file is removed.\n   */\n  processedRetentionDays?: number\n}\n\nexport function imports(options: ImportsPluginOptions) {\n  const esIndex = options.esIndex ?? 'parts'\n  // Resolve the cleanup config eagerly. `storageBucket === null` is the\n  // explicit opt-out path; `undefined` falls back to the convention.\n  // Two cleanup jobs share the same bucket but have independent TTLs:\n  // orphans (no import_run ref) churn fast (2h); processed runs\n  // (terminal status) get a longer reprocess window (7d).\n  const cleanupConfig =\n    options.storageBucket === null\n      ? null\n      : {\n          bucket: options.storageBucket ?? 'lumi-imports',\n          orphanTtlHours: options.orphanUploadTtlHours ?? 2,\n          processedRetentionDays: options.processedRetentionDays ?? 7,\n        }\n\n  // Register the storage factory eagerly — runs at `imports()` factory\n  // call time during `lumi.config.ts` evaluation, NOT inside the\n  // `init` hook. The hook only fires at `createApp()` time (Next.js\n  // request boot / worker boot), but the factory needs to be visible\n  // to consumers that read it at module-load (e.g. the commerce\n  // route's lazy `getStorageClient` thunk closes over a value that\n  // resolves the moment a request comes in). Late registration in\n  // `init` would race the first request.\n  if (options.storage) {\n    registerImportsStorageFactory(options.storage)\n  }\n\n  return definePlugin({\n    name: '@murumets-ee/imports',\n    server: {\n      entities: [ImportRun, ImportMapping],\n      init: async (app) => {\n        // Queue PACKAGE being importable doesn't mean the queue() PLUGIN\n        // is in the consumer's plugins array. Without it, registerJob()\n        // succeeds but no worker ever picks up the job. Probe and warn.\n        const queuePluginPresent = app.plugins.all().some((p) => p.name === '@murumets-ee/queue')\n        if (!queuePluginPresent) {\n          app.logger.warn(\n            'imports: queue() plugin not in plugins array — imports:run jobs will not be processed',\n          )\n          return\n        }\n\n        // Queue client stays dynamic-imported: we already early-returned\n        // above when the queue() plugin isn't in the consumer's plugins\n        // array, so static-importing here would force-load the queue\n        // client module in deployments that don't enable the worker.\n        // `@murumets-ee/core/clients` is NO LONGER imported here —\n        // `app.getClient(entity)` gives us a typed, app-bound\n        // AdminClient without the F1+F2 foot-guns from\n        // PLAN-DECLARATIVE-PLUGINS.md.\n        const queueClientModule = await import('@murumets-ee/queue/client')\n\n        const importRuns = app.getClient(ImportRun)\n        const transforms = getTransformRegistry()\n\n        queueClientModule.registerJob(\n          importsRunJob,\n          createRunImportHandler({\n            importRuns,\n            transforms,\n            esClient: options.esClient,\n            esIndex,\n            ...(options.resolveFilePath !== undefined && {\n              resolveFilePath: options.resolveFilePath,\n            }),\n            logger: app.logger.child({ pkg: 'imports' }),\n          }),\n        )\n\n        // Subscribe to the queue's terminal-state event stream so a\n        // worker that crashes mid-import — and whose `imports:run` job\n        // ultimately reaches `dead` without the handler reaching its\n        // own `markFailed` path — flips the mirror `import_run` row to\n        // `failed`. This replaces what would otherwise be a per-domain\n        // periodic sweep; the queue's terminal notifier is the generic\n        // primitive (graphile-worker's `job:failed` event, BullMQ's\n        // `failed` listener — same shape).\n        //\n        // The listener is multi-subscriber, so we coexist with the\n        // notifications plugin's admin-email handler without either of\n        // us clobbering the other. Filtering on `event.jobType` keeps\n        // this listener cheap on jobs we don't own.\n        //\n        // **HMR**: each re-init of this `init` hook constructs a new\n        // listener instance. The notifier registry stores listeners\n        // by reference, so without explicit cleanup an HMR cycle would\n        // STACK duplicate listeners — `findMany`+`update` per dead\n        // event growing linearly with the number of HMR rounds. Stash\n        // the unsubscribe on a globalThis Symbol-keyed slot (same\n        // pattern as `transform.ts` / `storage.ts`) and call it before\n        // re-subscribing so HMR is idempotent.\n        const previousUnsubscribe = (globalThis as GlobalThisWithUnsubscribe)[\n          IMPORTS_DEAD_LISTENER_KEY\n        ]\n        if (previousUnsubscribe) previousUnsubscribe()\n        const unsubscribe = queueClientModule.addQueueTerminalListener(\n          createImportsRunDeadListener({\n            importRuns,\n            logger: app.logger.child({ pkg: 'imports' }),\n          }),\n        )\n        ;(globalThis as GlobalThisWithUnsubscribe)[IMPORTS_DEAD_LISTENER_KEY] = unsubscribe\n\n        // Storage-cleanup jobs. Two passes, same bucket:\n        //   1. Orphan sweep — uploads whose register handshake never\n        //      completed (no `import_run` reference). 2h TTL.\n        //   2. Processed sweep — uploads whose `import_run` is terminal\n        //      (succeeded / failed / cancelled). 7-day reprocess\n        //      window then drop.\n        // `defineJob({ schedule })` declares the cadence; the queue\n        // worker reads the metadata at startup, materialises a\n        // `toolkit_scheduled_jobs` row, and its cron loop fires the\n        // registered handler each tick. No explicit\n        // `client.schedule()` call needed — the worker owns the\n        // schedule, not the handler. Same helper-signature shape as\n        // ticketing's cleanup.\n        if (cleanupConfig !== null) {\n          const { cleanupOrphanedImports, cleanupProcessedImports } = await import(\n            './admin/cleanup.js'\n          )\n          const { runAsCli } = await import('@murumets-ee/core')\n          const cleanupLogger = app.logger.child({ pkg: 'imports', job: 'cleanup' })\n\n          queueClientModule.registerJob(\n            queueClientModule.defineJob({\n              name: 'imports:cleanup-orphaned-uploads',\n              description:\n                'Delete toolkit_files rows in the imports bucket with no matching import_run reference and createdAt older than the configured TTL.',\n              schedule: '@every 1h',\n            }),\n            async () => {\n              // `runAsCli` is load-bearing today: the queue worker's\n              // handler context (see `packages/queue/src/handler-context.ts`)\n              // tracks the job type but NOT a user / RequestContext.\n              // Any AdminClient call reachable from the cleanup\n              // helpers — currently `createAdminClient(ImportRun).getTable()`\n              // for column-metadata resolution — runs without a user\n              // unless we enter a CLI scope. The synthetic CLI user\n              // is the right attribution for scheduled-job work\n              // (matches ticketing's `cleanupOrphanedUploads` shape).\n              await runAsCli(() =>\n                cleanupOrphanedImports({\n                  app,\n                  bucket: cleanupConfig.bucket,\n                  ttlHours: cleanupConfig.orphanTtlHours,\n                  logger: cleanupLogger,\n                }),\n              )\n            },\n          )\n\n          queueClientModule.registerJob(\n            queueClientModule.defineJob({\n              name: 'imports:cleanup-processed-uploads',\n              description:\n                'Delete toolkit_files rows for import_runs in terminal status (succeeded/failed/cancelled) whose finishedAt is older than the configured retention window. The import_run row itself is preserved.',\n              schedule: '@every 1h',\n            }),\n            async () => {\n              await runAsCli(() =>\n                cleanupProcessedImports({\n                  app,\n                  bucket: cleanupConfig.bucket,\n                  retentionDays: cleanupConfig.processedRetentionDays,\n                  logger: cleanupLogger,\n                }),\n              )\n            },\n          )\n        }\n\n        app.logger.info(\n          {\n            esIndex,\n            // Stable shape regardless of enabled/disabled so structured-\n            // log queries (`cleanup.enabled = false`) don't have to\n            // handle a string fallback.\n            cleanup:\n              cleanupConfig === null ? { enabled: false } : { enabled: true, ...cleanupConfig },\n          },\n          'Imports plugin initialized',\n        )\n      },\n    },\n  })\n}\n"],"mappings":"qRAgEA,MAAM,EAA4B,OAAO,IAAI,gDAAgD,EAkG7F,SAAgB,EAAQ,EAA+B,CACrD,IAAM,EAAU,EAAQ,SAAW,QAM7B,EACJ,EAAQ,gBAAkB,KACtB,KACA,CACE,OAAQ,EAAQ,eAAiB,eACjC,eAAgB,EAAQ,sBAAwB,EAChD,uBAAwB,EAAQ,wBAA0B,CAC5D,EAcN,OAJI,EAAQ,SACV,EAA8B,EAAQ,OAAO,EAGxC,EAAa,CAClB,KAAM,uBACN,OAAQ,CACN,SAAU,CAAC,EAAW,CAAa,EACnC,KAAM,KAAO,IAAQ,CAKnB,GAAI,CADuB,EAAI,QAAQ,IAAI,CAAC,CAAC,KAAM,GAAM,EAAE,OAAS,oBAC9C,EAAG,CACvB,EAAI,OAAO,KACT,uFACF,EACA,MACF,CAUA,IAAM,EAAoB,MAAM,OAAO,6BAEjC,EAAa,EAAI,UAAU,CAAS,EACpC,EAAa,EAAqB,EAExC,EAAkB,YAChB,EACA,EAAuB,CACrB,aACA,aACA,SAAU,EAAQ,SAClB,UACA,GAAI,EAAQ,kBAAoB,IAAA,IAAa,CAC3C,gBAAiB,EAAQ,eAC3B,EACA,OAAQ,EAAI,OAAO,MAAM,CAAE,IAAK,SAAU,CAAC,CAC7C,CAAC,CACH,EAwBA,IAAM,EAAuB,WAC3B,GAEE,GAAqB,EAAoB,EAC7C,IAAM,EAAc,EAAkB,yBACpC,EAA6B,CAC3B,aACA,OAAQ,EAAI,OAAO,MAAM,CAAE,IAAK,SAAU,CAAC,CAC7C,CAAC,CACH,EAgBA,GAfC,WAA0C,GAA6B,EAepE,IAAkB,KAAM,CAC1B,GAAM,CAAE,yBAAwB,2BAA4B,MAAM,OAChE,yBACD,CAAA,KAAA,GAAA,EAAA,CAAA,EACK,CAAE,YAAa,MAAM,OAAO,qBAC5B,EAAgB,EAAI,OAAO,MAAM,CAAE,IAAK,UAAW,IAAK,SAAU,CAAC,EAEzE,EAAkB,YAChB,EAAkB,UAAU,CAC1B,KAAM,mCACN,YACE,qIACF,SAAU,WACZ,CAAC,EACD,SAAY,CAUV,MAAM,MACJ,EAAuB,CACrB,MACA,OAAQ,EAAc,OACtB,SAAU,EAAc,eACxB,OAAQ,CACV,CAAC,CACH,CACF,CACF,EAEA,EAAkB,YAChB,EAAkB,UAAU,CAC1B,KAAM,oCACN,YACE,oMACF,SAAU,WACZ,CAAC,EACD,SAAY,CACV,MAAM,MACJ,EAAwB,CACtB,MACA,OAAQ,EAAc,OACtB,cAAe,EAAc,uBAC7B,OAAQ,CACV,CAAC,CACH,CACF,CACF,CACF,CAEA,EAAI,OAAO,KACT,CACE,UAIA,QACE,IAAkB,KAAO,CAAE,QAAS,EAAM,EAAI,CAAE,QAAS,GAAM,GAAG,CAAc,CACpF,EACA,4BACF,CACF,CACF,CACF,CAAC,CACH"}

package/dist/runner-DvKNfQJO.mjs

@@ -0,0 +1,2 @@
+import{bulkUpsert as e}from"@murumets-ee/search-elasticsearch";import{createReadStream as t}from"node:fs";import{parse as n}from"csv-parse";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`}:${n}`,o=this.patterns.get(a);if(!o){if(this.patterns.size>=this.maxPatterns){this.droppedSignatures+=1;return}o={errorType:t,field:r??null,message:n,count:0,firstOccurrence:e,lastOccurrence:e,samples:[]},this.patterns.set(a,o)}o.count+=1,o.lastOccurrence=e,o.samples.length<this.maxSamplesPerPattern&&o.samples.push({rowNumber:e,rowData:i})}getTotalErrorCount(){let e=0;for(let t of this.patterns.values())e+=t.count;return e}getDistinctPatternCount(){return this.patterns.size}getDroppedSignatureCount(){return this.droppedSignatures}getTopPatterns(){let e=Array.from(this.patterns.values()).sort((e,t)=>t.count-e.count),t=e.reduce((e,t)=>e+t.count,0);return e.map(e=>({errorType:e.errorType,field:e.field,message:e.message,count:e.count,firstOccurrence:e.firstOccurrence,lastOccurrence:e.lastOccurrence,samples:e.samples.slice(),percentage:t>0?e.count/t*100:0}))}snapshot(){return{totalErrors:this.getTotalErrorCount(),distinctPatterns:this.getDistinctPatternCount(),droppedSignatures:this.droppedSignatures,patterns:this.getTopPatterns()}}};function o(e){let t=new Set;for(let n of e)if(n!==``){if(t.has(n))throw Error(`Duplicate header "${n}" after trimming — column names must be unique`);t.add(n)}}async function s(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*c(e){let{filePath:r,delimiter:i=`	`,hasHeader:a=!0,columns:s,relaxColumnCount:c=!1,quote:l=!1}=e,u=s?Array.from(s):a?e=>{let t=e.map(e=>e.trim());return o(t),t}:!1,d=t(r),f=d.pipe(n({delimiter:i,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()}}function l(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*u(e){if(e.format===`xlsx`){let{streamXlsx:t}=await import(`./xlsx-reader-B2_F37jA.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*c(l(e))}const d=1e3;async function f(t){let{importRunId:n,runLabel:r,params:i,transform:o,feed:s,esClient:c,esIndex:l,batchSize:f=d,onProgress:p,rowLimit:m,signal:h,errorTracker:g=new a,logger:_}=t;if(f<1)throw Error(`batchSize must be >= 1 (got ${f})`);let v=Date.now();_?.info({importRunId:n,runLabel:r,batchSize:f,esIndex:l},`imports.runner: started`);let y=0,b=0,x=0,S=0,C=0,w=[],T=async()=>{if(w.length===0)return;let t=w;w=[];let r=Date.now(),i;try{i=await e(c,{index:l,docs:t.map(({id:e,doc:t})=>({id:e,doc:t})),...h!==void 0&&{signal:h}})}catch(e){let n=h?.aborted??(e instanceof Error&&(e.name===`AbortError`||/abort/i.test(e.message)))?`aborted`:`bulk_request_failed`,r=e instanceof Error?e.message:String(e);for(let{rowNumber:e}of t)g.addError(e,n,r,void 0,null);throw x+=t.length,e}b+=i.succeeded,x+=i.failures.length,C+=1;let a=Date.now()-r,o=(Date.now()-v)/1e3;if(_?.info({importRunId:n,batch:C,flushMs:a,submitted:t.length,succeeded:i.succeeded,failed:i.failures.length,rowsRead:y,rowsPerSecond:o>0?Math.round(y/o):0},`imports.runner: batch flushed`),i.failures.length>0){let e=new Map(t.map(e=>[e.id,e.rowNumber]));for(let t of i.failures){let n=e.get(t.id)??-1;g.addError(n,t.type,t.reason,void 0,{id:t.id})}}if(p){let e=(Date.now()-v)/1e3;p({rowsRead:y,rowsSucceeded:b,rowsFailed:x,rowsSkipped:S,batchesCompleted:C,elapsedSeconds:e,rowsPerSecond:e>0?y/e:0,distinctErrorPatterns:g.getDistinctPatternCount()})}},E=f,D=new Map;for await(let{rowNumber:e,row:t}of u(s)){if(h?.aborted||m!==void 0&&y>=m)break;if(y+=1,y===1&&_?.info({importRunId:n,rowNumber:e,sampleKeys:Object.keys(t)},`imports.runner: STREAM_FIRST_ROW`),y%E===0){let e=(Date.now()-v)/1e3;_?.info({importRunId:n,rowsRead:y,rowsSucceeded:b,rowsFailed:x,rowsSkipped:S,pending:w.length,elapsedSec:Math.round(e*10)/10,rowsPerSecond:e>0?Math.round(y/e):0,distinctFirstErrorTypes:D.size},`imports.runner: STREAM_TICK`)}let a={importRunId:n,params:i,runLabel:r,rowNumber:e},s;try{s=await o(t,a)}catch(r){let i=r instanceof Error?r.message:String(r);g.addError(e,`transform_threw`,i,void 0,t),D.has(`transform_threw`)||(D.set(`transform_threw`,{rowNumber:e,message:i}),_?.warn({importRunId:n,rowNumber:e,errorType:`transform_threw`,message:i},`imports.runner: FIRST_ERROR transform_threw`)),x+=1;continue}if(s.kind===`skip`){S+=1;continue}if(s.kind===`error`){g.addError(e,s.error.errorType,s.error.message,s.error.field,t),D.has(s.error.errorType)||(D.set(s.error.errorType,{rowNumber:e,message:s.error.message}),_?.warn({importRunId:n,rowNumber:e,errorType:s.error.errorType,field:s.error.field,message:s.error.message},`imports.runner: FIRST_ERROR ${s.error.errorType}`)),x+=1;continue}w.push({id:s.id,doc:s.doc,rowNumber:e}),w.length>=f&&await T()}await T();let O=(Date.now()-v)/1e3;return _?.info({importRunId:n,rowsRead:y,rowsSucceeded:b,rowsFailed:x,rowsSkipped:S,batchesCompleted:C,totalSec:Math.round(O*10)/10,rowsPerSecond:O>0?Math.round(y/O):0},`imports.runner: finished`),{rowsRead:y,rowsSucceeded:b,rowsFailed:x,rowsSkipped:S,batchesCompleted:C,errors:g.snapshot()}}export{s as a,i as c,o as i,a as l,f as n,c as o,u as r,r as s,d as t};
+//# sourceMappingURL=runner-DvKNfQJO.mjs.map

package/dist/runner-DvKNfQJO.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"runner-DvKNfQJO.mjs","names":[],"sources":["../src/error-tracker.ts","../src/streaming.ts","../src/read-feed.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 * 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's acceptable for the supplier-feed use case because we default\n * `quote: false` (every `\\n` IS a row boundary). For genuine RFC-4180\n * CSV inputs with embedded newlines, treat the result as an upper\n * bound; the actual parsed row count from `streamFeed` is authoritative.\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 { streamFeed, type StreamFeedOptions, type StreamFeedRow } from './streaming.js'\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","/**\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 { bulkUpsert, type BulkIndexResult, type EsClientLike } from '@murumets-ee/search-elasticsearch'\nimport { ErrorTracker } from './error-tracker.js'\nimport { readFeed, type ReadFeedOptions } from './read-feed.js'\nimport type { RowTransform, TransformContext } 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 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  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  /** Number of `bulkUpsert` calls made. */\n  batchesCompleted: number\n  /** Final value of {@link ErrorTracker.snapshot}. */\n  errors: ReturnType<ErrorTracker['snapshot']>\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    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(\n    { importRunId, runLabel, batchSize, esIndex },\n    'imports.runner: started',\n  )\n  let rowsRead = 0\n  let rowsSucceeded = 0\n  let rowsFailed = 0\n  let rowsSkipped = 0\n  let batchesCompleted = 0\n\n  let pending: Array<{ id: string; doc: TDoc; rowNumber: number }> = []\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    if (onProgress) {\n      const elapsedSeconds = (Date.now() - startedAt) / 1000\n      onProgress({\n        rowsRead,\n        rowsSucceeded,\n        rowsFailed,\n        rowsSkipped,\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({ importRunId, rowNumber, sampleKeys: Object.keys(row) }, 'imports.runner: STREAM_FIRST_ROW')\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      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    pending.push({ id: result.id, doc: result.doc, rowNumber })\n    if (pending.length >= batchSize) {\n      await flush()\n    }\n  }\n\n  await flush()\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      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    batchesCompleted,\n    errors: errorTracker.snapshot(),\n  }\n}\n"],"mappings":"4IAmEA,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,EC9JA,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,CAuBA,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,CCrMA,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,CCjCA,MAAa,EAAqB,IA6GlC,eAAsB,EAAgB,EAA2D,CAC/F,GAAM,CACJ,cACA,WACA,SACA,YACA,OACA,WACA,UACA,YAAY,EACZ,aACA,WACA,SACA,eAAe,IAAI,EACnB,UACE,EAEJ,GAAI,EAAY,EACd,MAAU,MAAM,+BAA+B,EAAU,EAAE,EAG7D,IAAM,EAAY,KAAK,IAAI,EAC3B,GAAQ,KACN,CAAE,cAAa,WAAU,YAAW,SAAQ,EAC5C,yBACF,EACA,IAAI,EAAW,EACX,EAAgB,EAChB,EAAa,EACb,EAAc,EACd,EAAmB,EAEnB,EAA+D,CAAC,EAE9D,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,CAEA,GAAI,EAAY,CACd,IAAM,GAAkB,KAAK,IAAI,EAAI,GAAa,IAClD,EAAW,CACT,WACA,gBACA,aACA,cACA,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,MAMpD,GALA,GAAY,EAER,IAAa,GACf,GAAQ,KAAK,CAAE,cAAa,YAAW,WAAY,OAAO,KAAK,CAAG,CAAE,EAAG,kCAAkC,EAEvG,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,EACf,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,CAEA,EAAQ,KAAK,CAAE,GAAI,EAAO,GAAI,IAAK,EAAO,IAAK,WAAU,CAAC,EACtD,EAAQ,QAAU,GACpB,MAAM,EAAM,CAEhB,CAEA,MAAM,EAAM,EAEZ,IAAM,GAAY,KAAK,IAAI,EAAI,GAAa,IAe5C,OAdA,GAAQ,KACN,CACE,cACA,WACA,gBACA,aACA,cACA,mBACA,SAAU,KAAK,MAAM,EAAW,EAAE,EAAI,GACtC,cAAe,EAAW,EAAI,KAAK,MAAM,EAAW,CAAQ,EAAI,CAClE,EACA,0BACF,EAEO,CACL,WACA,gBACA,aACA,cACA,mBACA,OAAQ,EAAa,SAAS,CAChC,CACF"}

package/dist/storage-resolver.d.mts

@@ -1,4 +1,4 @@
-import { n as FilePathResolver } from "./worker-BY0JQw39.mjs";
+import { n as FilePathResolver } from "./worker-ByPOP6Qe.mjs";
 
 //#region src/storage-resolver.d.ts
 /**