What

Adds bioassert suggest <file>, the inverse of evaluation: it inspects a file and writes a starting set of suggested assertions to <file>.assertions.txt (or --output FILE) for the author to review and tighten.

Implements the design in specs/suggest.md. The proposals are heuristic guesses from the file's current state, not verified expectations — hence the command name suggest rather than generate.

$ bioassert suggest tests/data/junctions.tsv
Wrote 5 assertions to tests/data/junctions.tsv.assertions.txt

# Generated by bioassert suggest from tests/data/junctions.tsv
# These are guesses from the file's current state; review and adjust before use.

# file
tests/data/junctions.tsv file.exists eq true  # file is present
tests/data/junctions.tsv file.size gt 0B  # file is not empty

# tsv
tests/data/junctions.tsv tsv.columns.count eq 12  # expected 12 columns
tests/data/junctions.tsv tsv.lines.count gte 2  # rows within +/- 50% of 4
tests/data/junctions.tsv tsv.lines.count lte 6

How

• New src/suggest/ module (purely additive; no executor or dispatch code changes):
  • suggestion.rs — the Suggestion output-line type, its render, and the band helper for the +/- 50% bounds.
  • provider.rs — the SuggestionProvider trait and the ordered providers() registry, mirroring the executor dispatch order (file → delimited → bam → fasta).
  • orchestrator.rs — suggest() returning a SuggestResult { suggestions, rendered, warnings }.
  • providers/{file,delimited,bam,fasta}.rs — one provider per family, each reusing that family's existing property functions (exists/get_file_size, column_count/line_count, read_header+counts, read_records+counts).
• Defaults: every file gets file.exists eq true and file.size gt 0B; schema-like counts are pinned with eq (delimited columns, fasta records, bam @SQ); run-to-run quantities get a gte/lte +/- 50% band (delimited rows, fasta total length). Family selection is by extension via a new prefix_for_extension helper.
• Binary wiring: new suggest subcommand (src/cli.rs); output path derives <file>.assertions.txt (resolve_output_file in src/report.rs) and is not overwritten without --force. Exit 0 on success, 2 on a fatal error; provider warnings go to stderr and keep the exit 0.

Notable deviation from the spec

The spec's documented output format uses inline # comments, and the spec also requires the generated file to run with exit 0 — but the grammar's assertion rule ended in ~ EOI with no comment support, so inline-commented output would not parse. To make the documented output valid input (and the round-trip pass), the grammar now accepts an optional trailing # comment (discarded). A # inside a quoted resource/value is consumed by the quoted token first, so it is not treated as a comment.

Testing

• cargo test — 334 passing, 0 failing. New unit tests per provider, the orchestrator, and suggestion rendering; new parser tests for inline comments (incl. no regressions on the existing trailing-token rejection tests).
• New tests/suggest_subcommand_test.rs: insta snapshots for tsv/bam/fasta/unknown-extension, the --force overwrite behaviour, and a round-trip test that suggests a fixture then runs the output and asserts exit 0 — proving every suggested assertion holds against its source file.
• cargo clippy --all-targets clean; cargo fmt --check clean.

Docs

CLAUDE.md (module entry + Suggest command section + new-family provider step + inline-comment note), README.md (Commands table, inline-comment note, "Suggesting a starting set" section), and the design spec at specs/suggest.md.

🤖 Generated with Claude Code