Summary

• Add vendor_from_file() and vendor_from_io() public API functions that auto-detect the correct Vendor for a file based on its contents
• Make vendor_type optional (None default) in allotrope_from_file, allotrope_from_io, allotrope_model_from_file, allotrope_model_from_io — when None, auto-discovers the vendor
• Add AllotropeVendorNotFoundError exception raised when no parser matches
• Implement sniff() classmethods on all 53 parsers with lightweight file-content checks (sheet names, header patterns, XML root tags, etc.)
• Add discover_vendor_test.py with 309 parametrized tests verifying discovery against all existing test data files
• Add vendor auto-discovery guidance to CLAUDE.md for new parser development

Design

discover_vendor() in parser_factory.py runs a multi-pass detection strategy:

1. Filter by extension — only parsers supporting the file's extension are candidates
2. Sniff pass — call sniff() on each candidate, collect all matches
3. Single match — return immediately
4. Multiple matches — try create_data() on each sniff match to disambiguate (e.g., Gen5 vs Gen5Image both sniff True for .txt, but only one parses successfully)
5. No sniff matches — try create_data() on all candidates as brute-force fallback
6. Sniff matched but parse failed — trust the sniff result (parse failure may be unrelated to discovery, e.g., encoding issues)

Each parser's sniff() is a lightweight classmethod: unique extensions return True, shared extensions check distinctive content (header columns, sheet names, XML root tags, etc.). UTF-16 LE/BE encoded files are handled with BOM detection.

Indistinguishable parser pairs (same file format, either parser works): Gen5/Gen5Image, xPONENT/IntelliFlex, VisionLite/GenesysOnBoard, EchoCherryPick/EchoPlateReformat, QuantStudio/QuantStudioD&A, MSDWorkbench/MethodicalMind.

Test plan

• All 1119 existing parser tests pass
• 309 new discovery tests pass — one per test data file, verifying correct vendor detection
• Lint + mypy clean

🤖 Generated with Claude Code