Use this check before investigating a release-package discrepancy. It compares your selected export records; it does not approve a board or a production order.
- Select the BOM and placement UTF-8 CSV/TSV files deliberately. Review the headers and first three parsed rows. Nothing is uploaded or automatically scanned.
- Choose each reference header. Map value/footprint only when their literal column meanings are comparable; leave them Not compared otherwise. Do not map two meanings to the same column.
- Declare each export’s revision, variant, side and filters, or explicitly write Unknown. Choose comma/space BOM grouping. Enable quantity only if one reference means one physical part and select the per-reference quantity column.
- Compare and inspect paired, one-sided, repeated and conflicting references. Follow the original parsed record-start lines back to the source; duplicate evidence remains complete.
- Prepare complete JSON locally, wait for its byte count and SHA-256, then click Download. Filtering and 20-row pagination affect only the screen. Keep the downloaded file if you need evidence after closing this page.
| Choice | Setting | Meaning and tradeoff |
|---|
| Reference column | Required in each export | Individual ASCII references start with a letter, then letters/digits/underscore; maximum40 characters. BOM grouping uses your explicit separator. |
| Value / footprint | Optional; Not compared by default | Exact trimmed CSV text, no electrical equivalence. Empty/omitted fields remain unobserved. Each mapped field≤512UTF-16 units. |
| Quantity | Off by default | Only one part per reference. Example R1,R2 requires quantity2; a build quantity for100boards is not interchangeable. |
| Declared scope | Required text for both files | Your statement is retained but not verified. Different variants/sides/filters can explain one-sided rows; do not infer that explanation automatically. |
| Prepare / download | Two separate actions | Preparation revalidates the exact selected bytes/settings in a dedicated local Worker; saving requires a further click. |
- Reference column
Setting: Required in each export
Individual ASCII references start with a letter, then letters/digits/underscore; maximum40 characters. BOM grouping uses your explicit separator.
- Value / footprint
Setting: Optional; Not compared by default
Exact trimmed CSV text, no electrical equivalence. Empty/omitted fields remain unobserved. Each mapped field≤512UTF-16 units.
- Quantity
Setting: Off by default
Only one part per reference. Example R1,R2 requires quantity2; a build quantity for100boards is not interchangeable.
- Declared scope
Setting: Required text for both files
Your statement is retained but not verified. Different variants/sides/filters can explain one-sided rows; do not infer that explanation automatically.
- Prepare / download
Setting: Two separate actions
Preparation revalidates the exact selected bytes/settings in a dedicated local Worker; saving requires a further click.
Illustrative 3-reference example — not a device or manufacturing measurement
BOM contains grouped R1,R2 with quantity2. Placement contains R1 and R3. With comma grouping and explicit one-part-per-reference quantity checking: R1 is paired, R2 occurs only in the selected BOM, R3 only in the selected placement; the grouped quantity agrees. Leave unmapped value/footprint as unobserved. This cannot prove a missing physical part.
Limits and processing
Local JavaScript/Web Crypto in your browser; no server processing, paid inference or GPU requirement. Requires browser module Workers, SHA-256 and Intl.Segmenter for shortened long previews. Each selected file≤400,000bytes and100,000UTF-16 units;≤500data rows and32unique nonempty headers≤200UTF-16 units. Each input≤5,000expanded references and2,000,000expanded mappedUTF-16 units. Core reports and prepared JSON each≤16,000,000UTF-8 bytes. Preview only3rows, shortened at whole grapheme boundaries within120UTF-16 units. No encoding guess, replacement decode or remote fallback. Files must be rectangular UTF-8 CSV/TSV; ranges and #subparts are refused.
Recovery without losing the valid selection
A rejected or cancelled file inspection clears only that unfinished slot; reselect it using the nearby action. Comparison cancellation retains inspected files/maps but clears the pending result. Export preparation failure/cancellation keeps the completed comparison and valid selections, so prepare again. Any source, mapping, quantity or scope change invalidates the comparison and prepared download; compare again. Clear, closing or refreshing releases in-memory inputs; the tool does not save them automatically. Earlier explicit downloads remain on your device. Cancellation during native file reading or CPU serialization has not been accepted on every device.
Questions before using the report
Does a one-sided reference mean a missing component?
No. It means that reference is absent from one selected export. Different scope, revision, filters or variants are possible causes, but none is inferred or verified.
Does equal value or footprint prove compatibility?
No. It proves only agreement of the mapped, trimmed text. Coordinates, units, origin, rotation, clearance, electrical equivalence and physical assembly are not checked.
Why are JSON fields and source labels still English or mixed scripts?
JSON uses the shared action’s fixed schema and original data. Interface language does not translate reference IDs, filenames, headers, labels or original literal values. Machine keys/statuses and canonical report limitations are retained.
Where do selected files go?
Only browser memory and the dedicated local Worker. No filesystem scan, remote service, input URL or automatic storage. Only an explicit Download saves the prepared JSON; your browser manages that file.
Can an assistant use the same check?
The selected-byte/expectedSHA/policy and prepared-byte/digest actions share one validated contract. This private browser workflow is accepted only within its recorded tests; installed AI-host/MCP, every device, mother-tongue review and public release are not certified.