Make 22 weak unit tests assert content (#819 step 3)

[joshfactorial]

Step 3 of the #819 testing audit: tests that asserted existence, success, or a non-zero count instead of content. Test code only. No production behavior changes.

How the candidates were found

A scan of all 1,070 Rust tests flagged 47 with no content assertion of their own. Each was then read together with the helpers it calls:

• 24 OK. The assertions live in a helper, e.g. chimeric_sequence_content.rs's assert_derived, model_parity.rs's byte-for-byte baselines, variants.rs's assert_literal.
• 22 weak.
• 1 needs a decision and is untouched (see below).

The scan cannot see a trivial assert_eq!. One such test, test_runner_max_reads_reduces_processed_count, was found by reading and is included.

Fixed (each with a hand-derived expected value and a mutation of the production code it covers, confirmed applied, failing, then passing on revert)

Tests that could not fail for the reason their name gives

• gen_gc_bias_model::test_all_n_contig_is_skipped: chr1 was 10 bp, below the 100 bp window, so the length guard skipped it before the N path was ever reached. It is now 200 N at a depth different from chr2, and the test asserts the .bins.tsv window counts. Caveat: N handling is guarded twice, and only breaking both fails the test.
• gen_frag_length_model::test_runner_min_reads_zero_skips_filter: its fixture gave the same model whether or not the filter ran. It now has an outlier the MAD ceiling removes at min_reads=2 and keeps at 0, and asserts both models' mean and SD.
• test_transition_matrix_from_tsv: passed with the TSV ignored. It now asserts all four rows.
• balanced_chimeric_offset_survives_a_fragment_no_longer_than_the_read: asserted only off >= 1, on the premise that the floor cannot hold at frag = read. It can, since 37 + 37 ≤ 151, and the test now asserts it on both sides. Removing the upper bound passed the old assertion and fails the new one.
• test_apply_variants: the SNP's REF did not match the sequence, and the het genotypes made the outcome random. With a corrected homozygous fixture it asserts the exact sequence ATGTATGA and CIGAR MMMMDMMMM.
• gen_mut_model::test_runner_with_indels: only asserted mutation_rate > 0. It now asserts variant-type weights of 1/3 each, mutation_rate = 3/13114 (13,114 = H1N1's non-N bases, counted independently), insertion length 2 and deletion length 3. Both indel REFs in the old fixture disagreed with H1N1 and are corrected.
• test_runner_skips_reference_mismatch_variant: asserts the skipped SNP's context (GCT) carries no weight.

Print-only or empty

• generate_fragments: three depth tests printed a number and asserted nothing. They now assert the placed count and depth: 22.5 ± 1.0 (4σ) for fragment coverage, and exactly 10.0 for read depth.
• filter_reads::test_run_configuration: the body was empty with a TODO. Replaced by known-answer tests of create_map_item and RunConfiguration::from.

Exact values instead of is_some() / exists() / is_ok()

• gen_reads config output paths (r1/r2, VCF, BAM)
• test_create_output_file_new, test_check_parent, plus a new missing-parent test
• sample_poisson_extreme_lambda_does_not_overflow, now bounded at ±5σ
• test_prep_file_for_filtering, plus a gzipped-input case
• two config-loading tests

Removed or renamed

• Removed two max_reads smoke tests that could not see max_reads. Three existing tests pin it exactly.
• Renamed test_overwrite_warn to test_overwrite_output_is_accepted. Nothing can capture logs in a unit test, so the warning it was named for is not asserted.

Found, not changed here

• filter-reads output naming (bug, to be filed): it splits the whole path on . and indexes from the end. A plain x.fastq/x.vcf panics (attempt to subtract with overflow, reproduced with the binary), and a dot in a directory name produces a wrong path. The readers themselves handle plain files.
• Decisions for the maintainer (on Testing audit before v4.0.0 #819):
  • SNPs skipped for a REF mismatch still count toward mutation_rate and snp_freq.
  • gen-mut-model does not check indel REFs against the reference.
  • A BED that names no reference contig fails as "Trinuc counts are empty. Unknown error"; test_runner_bed_unknown_contig_succeeds_or_errors_cleanly still accepts any outcome.
  • In long-read mode, balanced_chimeric_offset may return an offset past a short fragment's end.

Evidence

Full workspace suite (1,097 tests), fmt --check and clippy -D warnings pass locally.

Not verified

• The mutation checks were run by delegated agents in isolated worktrees and by hand for this branch's own commit; their reports were reviewed and the diffs read, but not every mutation was re-run independently. Spot-checked: H1N1's 13,114 non-N bases, and the TSV row CDFs.

Refs #819.

🤖 Generated with Claude Code