[Paururo]

The tutorial says its numbers come from an actual run. They came from runs in July, and today's code gives other ones: the case counts moved, and after #58 the QC line's pseudogene_count counts AMAP's own calls, not Mycobrowser's 13. The same numbers were copied to the front page, the outputs page and the custom-reference page.

What was run

The tutorial's own commands, on 18 October 2026, with the code of 947a99d (Nextflow 26.10.2, paururo/amap:latest, the same digest as Docker Hub's), each with --skip_functional_annotation added:

• step 5, --validate_only, as written (it never reaches eggNOG): 8 s;
• step 6 on CDC1551 (NC_002755.2, the tutorial's URL): 10 tasks, 2 min 16 s on an M-series Mac under emulation;
• step 9, Sakai against K-12 with --ref_annotation ref.gbk --ref_genome ref.fna: 10 tasks, 5 min 47 s.

The genomes are the benchmark's copies of the same four NCBI files, fetched with the same URLs this morning; a HEAD request on each URL still answers 200 with a Last-Modified older than that copy.

GET_OUTPUTS only runs after eggNOG, so the QC line comes from GET_OUTPUTS' own script, rendered from modules/local/get.nf with the renderer tests/test_sample_ids.py uses, run in the container with Nextflow's docker command on that run's FIX_CDS output, with an eggNOG table that has its header and no rows. None of the six columns the tutorial shows depends on eggNOG, and the tutorial now says how the line was made.

Numbers

	July	Now
CDC1551 case2	2816	2813
CDC1551 case4	1097	1100
CDC1551 case3, case1_initial, rv_missing, total_cds, case1_removed, case1_kept	24, 141, 97, 4077, 130, 11	same
CDC1551 total_cds (QC)	4020	4014
CDC1551 pseudogene_count	13	123
CDC1551 coding_density_pct	91.5	91.3
Sakai case2	3325	3324
Sakai case3	34	31
Sakai case4	537	507
Sakai case1_kept	1356	1387
Sakai rv_missing	102	133
Sakai final annotation	5244 genes	5335 genes
K-12 proteins, Bakta CDS, false positives	4318, 4296, 142	same

CDC1551 differs from the benchmark's (2811, 1101, 142) by one or two because the tutorial's FASTA keeps NCBI's "complete sequence" description, which since #62 makes Bakta read it as circular.

Also from the run

• The --validate_only verdict never reaches the console; it is in results/pipeline_info/validation.log, and tail -n 8 of it is the tutorial's excerpt byte for byte. The tutorial now says so.
• The count of proteins extracted from a custom reference is only in PREPARE_REFERENCE's .command.out, not printed as the warning said; the tutorial gives the grep that finds it.
• ls -F so that the listing's slashes are real, amap_report.html in it, and which entries a run without eggNOG lacks (this run: no DUPLICATIONS/, PANGENE/ or final_report.html, as the outputs page says). Without eggNOG the run still writes amap_report.html, so "what you lose" now names final_report.html.
• "would otherwise surface twenty minutes in" is now "minutes in": Bakta, where a bad genome shows, finished 2 minutes into this run.

Not changed

• "work/ ... for one genome is a few GB": 93 MB for CDC1551 without eggNOG. The eggNOG database is built in --eggnog_db, outside work/, so a full run is probably far below that, but only a run with eggNOG can say.
• The eggNOG sizes (60 GB core, about 3 GB index) come from eggnog.nf, not from a run.
• The note on the two protein files says *-AMAP.clean.faa drops pseudogenes; in this run it keeps 42 of the 123 (export_clean_proteome.py compares the FASTA ID with simple_id). That is a code fix, being worked on separately.

Checks

tools/build_docs.py rebuilt the pages (and docs/about/changelog.html); build_docs.py --check and tools/check_site.py pass with Python 3.11. The case figures on the tutorial and outputs pages redraw from the new tables. A trial merge with #64 conflicts only in the generated docs/about/changelog.html: rebuild it with tools/build_docs.py after merging either one.