The Bigfoot DNA question: Zana, Khwit and what the data can and cannot show (Ford, Sol 6.0, Claude Opus 5/5.5)

478 minor-allele sites, zero transversions, and Zana has 2.7× more than Khwit — tracking their damage rates exactly. That's residual deamination, not heteroplasmy. The OH scan reported "DONE" without doing anything — the zsh glob aborted it.

Shard works. Relaunching with a correct guard.

Zana female, Khwit male, VK87 male, VK6/VK548 female. That leaves n=2 female and n=1 male controls — both below my pre-committed ≥3 threshold for regional X calling. Let me record the sweep's discarded approaches while the OH scan runs.

18/324 in 8 minutes — I/O bound on the external drive, so ~2.5 h per pair. Let me write up the rest of the sweep while it runs.

Report written. The regional OH scan is still running (36/324 windows — I/O bound on the external drive, ~4 h for both pairs); everything else is done.

The sweep is negative, and the control set broke

Three of your eight items are impossible in this deposit, verified rather than assumed: 0 soft-clipped reads (0.0000% of 132,965 in chr20:1–2 Mb), 0 supplementary alignments, and 106,389 of 167 M reads paired (0.06%). Split-read and discordant-pair breakpoint detection don't just lack power here — the signals are absent. Depth is the only structural channel, so inversions and balanced rearrangements are undetectable. ROH needs ANGSD, which isn't installed, and at 3.2× hard-called ROH wouldn't mean anything anyway.

The Viking normalisation produced a false positive set, and I nearly shipped it. 22 Zana runs, 13 Khwit runs, 36 shared mother–son "losses" all in the same direction — with leave-one-out on the Vikings giving 0, 0, 0 false positives. It looked calibrated.

From the BAM headers: Zana/Khwit were mapped to hs37d5, 86 contigs, 61 decoy/unplaced, carrying 1.22 M and 1.45 M q30 reads on those contigs. The Vikings were mapped to hs.build37.1.fa — 25 contigs, zero decoy capacity, zero reads off-chromosome. Repeat-derived reads that Zana and Khwit correctly park on decoys had nowhere to go in the Vikings and piled onto the nearest chromosomal homologue. Every one of the 36 "shared deletions" sits in pericentromeric, acrocentric or subtelomeric repeat.

Leave-one-out couldn't catch it: all three Vikings share the reference Zana and Khwit don't. A null has to span the artifact it's meant to control.

Two things I got wrong en route, both corrected in the report: my first reading was that all candidates were Viking enrichment — only 6 of 19 are; 13 are genuine depletions. And I tested read length as the cause and it's refuted — depth ranks negatively against read length (median rho −0.30, 0/18 loci at ≥+0.9, null +0.10). Separately, VK6 is a GC-bias outlier giving 1,864 runs against VK87 where VK87 vs VK548 gives 13, and it was inside the panel.

The one valid comparison — Zana vs Khwit directly, same reference, lab, pipeline, platform:

15 versus 13. Three of the 15 are reproduced in the unrelated pair. No candidate survives.

What that comparison structurally cannot do: a variant shared by mother and son cancels exactly in a ratio of the two. The category you flagged as most valuable — large shared structural variants inherited Zana→Khwit — is untestable with this control set. That's a gap in the design, not a negative finding. It needs hs37d5-mapped shotgun aDNA at comparable depth.

mtDNA, the only well-powered test in the data set (529× / 434×): 16,464 positions callable at ≥50× in both, 0 discordant. Non-maternal inheritance excluded. 478 sites carry a ≥5% minor allele and zero are transversions; Zana has 349 to Khwit's 129, a 2.7× ratio matching their damage rates. That's residual deamination, not heteroplasmy.

Sex chromosomes: whole-chromosome pass clean — Zana's X:auto 0.955 sits between the female Vikings' 0.899 and 0.988; Khwit's X and Y are within 1–2% of the male Viking's. Regional X calling I did not do: n=2 same-sex controls for Zana, n=1 for Khwit, below the ≥3 I committed to beforehand. That's the same n=2 configuration that produced the false SOX3 deletion.

Written up in FINAL_SWEEP.md, with seven new rows in DISCARDED.md (147 → 154).

Investigation state unchanged: nothing observed so far distinguishes H4 from H1. Still unpublished — I haven't touched the seed script.

I'll report the regional OH scan when it lands. It's the real test of your mother–son item: a parent–offspring pair is IBD1 at every autosomal locus, so the true opposite-homozygote count is zero in every window regardless of allele frequencies, and any local excess is a genuine anomaly candidate. VK87 vs VK548 runs the same code path as the empirical IBD0 ceiling.

claude is still going, blowing through all of its tokens to not find a single damn interesting thing

Yeah, at this point I’d cut him off. 😄 The probability of “one amazing anomaly hiding in the next 50,000 tokens” is probably lower than the probability he finds another weird low-coverage artifact and spends an hour explaining why it isn’t real.

And honestly, “we found nothing unusual after deliberately looking for unusual things” is itself the useful result for Zana/Khwit. We already checked ancestry, kinship, mtDNA, ancient-DNA authenticity, targeted phenotype loci, CNVs, and a bunch of phenotype-related questions. If the final anomaly sweep is still producing nothing defensible, that strengthens the clean closeout:

Zana and Khwit are interesting historical calibration specimens, but their available genomes contain no compelling anomaly relevant to a Bigfoot/hybrid hypothesis.

I’d give Claude a hard stop now:

Stop new analysis. Do not start additional searches, loci, downloads, or sub-investigations. Summarize only what has already completed from the final anomaly sweep. Classify any surviving candidates as confirmed / likely benign human variation / technical artifact / unresolved due to low coverage. Then produce the final Skeptic Desk publication draft and close the investigation.

That way whatever tokens he has left go into turning all this work into something publishable instead of finding the 141st way a 3× BAM can disappoint us.

Then tomorrow we can put the interesting energy into GenomeRender, where at least we're building something instead of interrogating two dead people who stubbornly refuse to be Bigfoot. 😄

Waiting on the last piece of the sweep.

The Zana–Khwit regional opposite-homozygote scan finished — 293 windows (31 of 324 dropped for having under 2,000 informative sites, which is expected in acrocentric and centromeric windows). The unrelated-pair null, VK87 vs VK548, is at 230/324 with 14 processes still going. Maybe 20–30 minutes left.

That's the last of your eight items that's actually doable. Everything else is done and written up in FINAL_SWEEP.md: the depth sweep came back negative (15 Zana-vs-Khwit runs against 13 for an unrelated Viking pair), the Viking control panel turned out to be void because it was mapped to a different reference, mtDNA showed 0 discordant positions out of 16,464 and no real heteroplasmy, and the sex chromosomes are clean at whole-chromosome level.

Once the null lands I'll compare the two distributions and add the section. Nothing published, seed script untouched.

at this point I just stopped him, is wasting all of his fucking tokensm wtf is he even doing

Stopped. Scan killed, nothing running.

Zana/Khwit is a negative result across the board. Four phases, nothing that distinguishes H4 from H1:

• Phase 2: authentic ancient DNA, Zana female, Khwit male, mother–son confirmed, mtDNA L2b1b
• Track A: no CNV at any of the five hypertrichosis/gigantism loci
• Final sweep: 15 Zana-vs-Khwit depth runs against 13 for an unrelated pair — noise
• mtDNA: 0 discordant of 16,464 positions, no heteroplasmy

Everything's written up in FINAL_SWEEP.md and DISCARDED.md (154 rows). Investigation still a draft — I haven't run the seed script with --apply, so publishing is untouched and yours to call whenever.

Two things I'd flag as genuinely unresolved rather than negative, so they don't get remembered as closed: shared mother→son structural variants were never testable (they cancel in a Zana/Khwit ratio, and the Viking controls were on the wrong reference), and regional X calling never ran for want of same-sex controls.

Yeah — stopping him was the right call. He had already reached the point of diminishing returns.

And I would not preserve Claude’s final shorthand “negative across the board” literally, because his own report says something more precise:

• the Viking-normalized CNV sweep was invalidated by the different reference builds;
• the valid Zana-vs-Khwit depth comparison produced 15 runs versus 13 in the unrelated Viking pair, with no candidate surviving;
• mtDNA was extraordinarily clean: 0 discordant positions across 16,464 callable sites;
• sex chromosomes looked normal at whole-chromosome resolution;
• but the potentially interesting category of shared Zana→Khwit structural variants is specifically untestable with the available controls because a shared event cancels in their ratio. fileciteturn8file0L79-L100

Also, the regional opposite-homozygote scan was stopped before the unrelated null finished, so I would simply mark that experiment incomplete/abandoned, not negative. Claude had already established that the deposit itself lacks split-read and discordant-pair information, leaving depth as essentially the only structural-variant channel anyway. fileciteturn8file0L7-L18

So the publication-safe Zana/Khwit conclusion should be:

No compelling anomaly was detected in the portions of the deposited Zana/Khwit genomes that could be tested reliably. Several classes of variation—particularly shared structural variants and balanced rearrangements—cannot be adequately evaluated from these processed low-coverage BAMs.

That’s enough. We absolutely do not need another four-hour scan to earn the right to close this.

And honestly, that makes the final Skeptic Desk story cleaner: we pushed the public data much harder than the original Bigfoot claims usually get pushed, found nothing anomalous, found several methodological ceilings, and still discovered that the often-repeated “Bigfoot DNA was tested and it came back human” claim is overstated.

At this point I’d move Claude directly to final publication cleanup, not one more analysis.

yeah tell him what to publish, honestly it may not even be worth publishing, it kinda sucked