dev: standalone compare_tables.py (ASV-table jaccard/count equivalence)

[cjfields]

Summary

Pulls the ASV-table equivalence check that bench_pooled.py currently only emits inside its WFA sweep CSV (jaccard_vs_nw / count_frac_vs_nw) into a small standalone script, so any two-or-more seqtab(_nochim).json runs can be compared directly — nw vs wfa arms, k5 vs k7, pool modes — without re-running a sweep.

load_table / table_equiv are kept byte-for-byte identical to bench_pooled.py, so the numbers match that CSV exactly (verified on the PacBio pseudo sweep: e0 jaccard 0.9499 / count_frac 0.7309, e30 0.9989).

Output

baseline: nw = .../nw/seqtab_nochim.json  (1901 ASVs)

label    ASVs  jaccard  count_frac  +only  -only identical
----------------------------------------------------------
  e30    1899   0.9989      0.9747      0      2     False
   e0    1989   0.9499      0.7309     94      6     False

• count_frac = the CSV's count_frac_vs_nw (fraction of shared ASVs with identical per-sample counts).
• +only / -only = ASVs the comparison has that the baseline lacks / vice-versa (surfaces the over-call signature — e0 invents 94 ASVs the NW table lacks).
• --abundance N restricts to ASVs with total count >= N (mirrors MIN_ABUNDANCE).

Complements compare_asvs.py (the richer abundance-stratified churn + nearest-neighbour Hamming characterizer); this is the quick jaccard/count equivalence check.

Test plan

• Reproduces the sweep CSV jaccard/count_frac exactly on tmp/pb_k7_pseudo.
• Both --abundance N and --abundance=N forms parse correctly.

🤖 Generated with Claude Code