Supported formats

"The same data" looks very different depending on which tool exported it. Rather than publish a list of formats we hope work, this page is generated from the test corpus: 84 files (19 centroid, 35 failure, 4 config, 20 profile, 6 bundle) that both implementations are run against on every change. One canonical board and dataset, re-expressed in each dialect, asserted to produce identical results.

Handled automatically

Structure

Comma, semicolon, tab and pipe delimiters, detected per file. UTF-8 byte order marks, CRLF endings, quoted fields containing the delimiter, and title or metadata rows above the header.

Numbers, dates and encodings

European decimal commas and thousands separators; dd-Mon-yyyy, yyyy/mm/dd, US m/d/y and two-digit years; Excel day serials; Windows-1252 files that would otherwise arrive as mojibake.

Excel workbooks

.xlsx is read directly. Shared strings and rich text are resolved, date-formatted cells are converted from day serials, and the sheet that was read — plus the ones that were not — is named, so data on another tab is visible rather than silently absent.

Column naming

Flexible aliases for designator, coordinate, date, quantity and assembly columns, including headings that carry their unit (Center-X(mil)) and defect columns matched by prefix (Defect Description, Failure Mode). Cells naming several parts (U7, U8) become separate records.

Refused, with a reason

Being permissive about the wrong things is worse than being strict. A placement export, a bill of materials, or a JSON document whose array is not defect records would each invent one "failure" per part, so each is rejected and told why. A recognized file that yields no records is an error, not a quiet success.

Placement / pick-and-place files

Example file	What it exercises	Result
`centroid_comma.csv`	plain comma CSV	parsed
`centroid_tab.txt`	tab-delimited (Altium .txt)	parsed
`centroid_semicolon_eu.csv`	semicolon delimiter with comma decimals (European Excel)	parsed
`centroid_units_mm.csv`	units written into the value (Altium default)	parsed
`centroid_units_mil.csv`	mil units declared in the values	parsed
`centroid_bom.csv`	UTF-8 BOM (Excel save-as)	parsed
`centroid_crlf.csv`	CRLF line endings	parsed
`centroid_preamble.csv`	metadata/title lines above the header	parsed
`centroid_kicad.pos`	KiCad .pos — '#' header, whitespace separated	parsed
`centroid_quoted.csv`	quoted fields containing the delimiter	parsed
`centroid_aliases.csv`	alternative column names	parsed
`centroid_offset_origin.csv`	origin offset far from 0,0 — must not be mistaken for a huge board	parsed
`centroid_negative.csv`	negative coordinates	parsed
`centroid_duplicate.csv`	the same designator twice	parsed
`centroid_header_units.csv`	unit declared in the column heading, not the value (Altium mil)	parsed
`centroid_eagle.mnt`	headerless whitespace report (Eagle .mnt)	parsed
`centroid_blank_coords.csv`	rows with blank / non-numeric coordinates are counted, not silently dropped	parsed
`centroid_empty.csv`	empty file	rejected, with a reason
`centroid_no_columns.csv`	a BOM, not placement data	rejected, with a reason

Failure data

Example file	What it exercises	Result
`failures_ict_comma.csv`	ICT CSV, comma	parsed
`failures_ict_semicolon.csv`	ICT CSV, semicolon	parsed
`failures_ict_tab.txt`	ICT CSV, tab	parsed
`failures_ict_bom_crlf.csv`	ICT CSV, BOM + CRLF	parsed
`failures_ict_titlerow.csv`	title row above the header	parsed
`failures_repair.txt`	pipe-delimited repair log, US dates, free-text defects	parsed
`failures_qms.json`	QMS JSON	parsed
`failures_qms_bare_array.json`	bare top-level JSON array with alternative keys	parsed
`failures_generic_dates.csv`	generic CSV with dd-Mon-yyyy dates and free-text faults	parsed
`failures_multi_refdes.csv`	several designators in one cell	parsed
`failures_pipe_in_cell.csv`	one cell contains pipes — must not hijack the file into the repair-log parser	parsed
`failures_qty.csv`	thousands-separated quantity; an explicit quantity of 0 is not a failure	parsed
`failures_blank_refdes.csv`	rows with no designator are counted as unusable, not dropped in silence	parsed
`failures_qms_bom.json`	QMS JSON saved with a UTF-8 BOM	parsed
`failures_qms_numeric_code.json`	defect codes stored as JSON numbers	parsed
`failures_defect_description.csv`	defect column named 'Defect Description' rather than an exact alias	parsed
`failures_ampersand.csv`	'&'-joined designators, and a 'Failure Mode' defect column	parsed
`failures_padded_refdes.csv`	zero-padded designators are parsed verbatim, not silently renamed	parsed
`failures_multi_assembly.csv`	one export spanning two assemblies	parsed
`failures_ict_datetime.csv`	date-and-time cells in six shapes, one date-only, plus a station column	parsed
`failures_repair_time_station.txt`	repair log with separate TIME and STATION columns (one row has no time)	parsed
`failures_qms_datetime.json`	QMS timestamps with an offset, and a separate time field	parsed
`failures_qms_time_precedence.json`	time in the date cell wins over a time field; U+202F before PM; padded and numeric stations	parsed
`failures_impossible_dates.csv`	day-first and impossible dates (13th month, 30 February) become no date, not a rolled-over one	parsed
`failures_generic_time.csv`	generic CSV with separate date and time columns and a bench column	parsed
`failures_bom.csv`	a BOM would otherwise become one failure per part	rejected, with a reason
`failures_placement.json`	placement data under a generic 'data' key must not pass as a QMS export	rejected, with a reason
`failures_kicad_placement.csv`	KiCad placement CSV dropped into the failure intake	rejected, with a reason
`failures_empty_records.json`	recognized but contains nothing	rejected, with a reason
`failures_no_refdes.csv`	no designator column at all	rejected, with a reason
`failures_workbook.xlsx`	Excel workbook: shared strings, a second sheet, and dates stored as day serials	parsed
`profile_alt_repair.txt`	repair log in the alternate profile's vocabulary	parsed
`profile_alt_ict.csv`	quality-server codes from the alternate profile; unknown codes go to its fallback	parsed
`profile_alt_qms.json`	QMS values as key, label, code, hyphenated key and free text	parsed
`profile_alt_default_codes.csv`	the default profile's codes under the alternate profile	parsed

Criticality configuration

Example file	What it exercises	Result
`config_comma.csv`	config with mixed-case values	parsed
`config_semicolon.csv`	semicolon config with alias column names	parsed
`config_empty.csv`	empty config	rejected, with a reason
`config_no_refdes.csv`	config with no designator column	rejected, with a reason

Customer profiles

What a defect means is configuration, not code: each manufacturer's failure types, entry-station hotkeys, repair vocabulary, defect codes, shift times and component classes live in one profile file that both implementations read. The corpus includes a second profile that differs from the default in every section, with failure files parsed under it, so nothing can quietly assume the default. Profiles that contradict themselves are refused with a reason.

Example file	What it exercises	Result
`profile_alt.json`	a second profile differing from the default in every section	parsed
`profile_minimal.json`	only a taxonomy: rules and codes empty, shifts and classes from the default	parsed
`profile_bad_schema.json`	a JSON file that is not a profile	rejected, with a reason
`profile_bad_hotkey.json`	two failure types share a hotkey (case-insensitive)	rejected, with a reason
`profile_bad_rule.json`	a keyword rule naming the default's type, not this profile's	rejected, with a reason
`profile_bad_code.json`	a defect code naming a type the profile does not have	rejected, with a reason
`profile_bad_fallback.json`	fallback that is not one of the profile's types	rejected, with a reason
`profile_bad_timezone.json`	unknown IANA time zone	rejected, with a reason
`profile_bad_overlap.json`	a shift overlapping another (crossing midnight)	rejected, with a reason
`profile_bad_prefix.json`	one refdes prefix in two component classes	rejected, with a reason
`profile_bad_color.json`	a colour that is not #rrggbb	rejected, with a reason
`profile_bad_hotkey_newline.json`	a hotkey with a trailing newline is not a single key	rejected, with a reason
`profile_bad_fallback_type.json`	a fallback given as a list, not a type key	rejected, with a reason
`profile_bad_version.json`	version true is not version 1	rejected, with a reason
`profile_bad_recent.json`	a recent period as long as the shortest classified window	rejected, with a reason
`profile_bad_share.json`	a share above 1	rejected, with a reason
`profile_bad_threshold.json`	a day count given as true	rejected, with a reason
`profile_bad_window_days.json`	a window length past the calendar	rejected, with a reason
`profile_bad_dup_code.json`	two defect codes that are the same once normalized	rejected, with a reason
`profile_bom.json`	profile saved with a UTF-8 byte-order mark	parsed

Data bundles

Records kept on a station — entered at the bench or brought in from another machine — export as one portable file with the kept boards and the customer profile. Importing one merges by record, so importing the same bundle twice adds nothing; rows that do not validate are counted and named, never dropped quietly or half-applied.

Example file	What it exercises	Result
`bundle_records.json`	records for three board revisions, mixed date/datetime, nine unusable rows (one is demo data, two name days that do not exist)	parsed
`bundle_with_profile.json`	a bundle carrying its customer profile; records keep that profile's types	parsed
`bundle_loose.json`	version written as 1.0 and boards as null — read the same by both pipelines	parsed
`bundle_newer.json`	a bundle from a newer version is refused, not half-read	rejected, with a reason
`bundle_not_a_bundle.json`	a board package is not a data bundle	rejected, with a reason
`bundle_bad_profile.json`	a bundle whose profile contradicts itself	rejected, with a reason

Something not on this list?

Reading a new export dialect is a small, contained change — the adapters are deliberately isolated from everything downstream. Send a sample (with anything sensitive removed) to info@engsecsolutions.com and it will be added to this corpus, which means it stays working.