# T4 Architecture-B candidate scan: VERDICT (29 August 2026, 8:55 pm ACST)

## Headline

**NON-EMPTY, BUT THE SIGNATURE DOES NOT DISCRIMINATE.** Under the frozen criteria the public
dump returns 38 PRIMARY matches (9 under the strictest frozen intersection). Post-hoc calibration
(results/CALIBRATION.md) shows the frozen stem-anomaly statistic has no power on this dump: among
the 1,066 nodes that pass burst, window and anchor, 38 have stems at or below half expectation and
40 have stems at or above double expectation; the dispersion index is 7.45 against a Poisson value
of 1; across the whole tree 13.7 per cent of long edges sit at half expectation or less. A
half-expectation stem is therefore an ordinary outcome of how the dump counts SNPs (coverage and
platform differences between branches), not a conspicuous anomaly. The matches are the short tail
of a background distribution whose long tail is the same size.

This repeats, on real data, what the 27 August power gate found by simulation: short-stemmed bursts
are routinely produced without any special history. Consequence (4) of the Architecture-B ruling,
"detectability RETURNS: the B-signature is conspicuous, not camouflaged", does not survive contact
with the dump at this resolution. The B-signature may still be real; the public tree cannot see it.

## The matches

38 nodes across eight haplogroups: J 10, E 9, G 6, L 3, T 3, O 3, C 2, H 2. Literature and
geography checks (results/lit-check-*.md, Sonnet subagents, no yfull.com):
- 37 of 38 have between 2 and 10 leaves and no published discussion of their own; their geography
  is inherited from their anchors: five sub-Saharan E-M2, four Arabian E-M35/E-V1008, six West
  Eurasian G, two East Asian C (Han/Korean/Japanese/Dungan), two South Asian H, three L-M27, three
  T-M184, three East Asian O-M175, nine small J1/J2 (mostly Arabian-Mesopotamian-Mediterranean, one
  Central Asian cluster J-Y15598/Z36272).
- The one sized match, **J-Y3165 (104 leaves)**, is identified by phylogeographer.com as the main
  branch of J2b-M205, essentially all living M205 testers: Yemen, Oman, Saudi Arabia, Iraq,
  Levant, Anatolia; 86 per cent of Soqotri; Bronze Age aDNA at Ain Ghazal, Megiddo, Alalakh. Its
  YFull TMRCA is 5300 (4900-5800), which overlaps the window only by its top 26 years, and
  phylogeographer's own TMRCA is about 6000 BP, outside the window. Its stem is 8 SNPs over 1600
  years, a ratio of 0.40, which is unremarkable against the 13.7 per cent background. It matches
  the frozen criteria marginally and its geography is the right region, but selecting it from a
  list of 38 on geography is a post-hoc choice, and its size and Arabian sampling depth are exactly
  the profile the against-interest clause says emptiness would have disfavoured. It is recorded as
  a candidate of interest, not a finding.

## Pre-stated readings

The result is not empty, so the two frozen empty-result readings do not fire. They are recorded
anyway per the freeze: (1) absence would have sharpened "small or unsampled, plausibly
Mesopotamian"; (2) against interest, Arabic-speaking populations are well sampled, so a LARGE
B-clade among them would be disfavoured, the unsampled refuge being Iraq/rural Mesopotamia
specifically. Reading (2) survives in modified form: the only large Arabian match (J-Y3165) is
marginal on window and ordinary on stem, so the well-sampled Arabian record still shows no clearly
distinctive B-clade; the Mesopotamian refuge is untouched by this scan either way.

The pre-stated non-empty reading applies: matches are candidates for examination, not
identifications; no result here proves Noah existed.

## What the test actually established

1. The operationalised B-signature has a base rate of about 1 in 28 eligible nodes in the public
   dump. Any future "we found a short-stemmed burst" claim must clear that base rate first.
2. The dump's raw branch SNP counts are decoupled from YFull's ages (80.6 yr/SNP implied, symmetric
   overdispersion). A stem statistic with power needs the clock-region SNP subset per branch,
   which the dump does not carry, or per-branch coverage data. Neither is available without yfull.com.
   Per-instance confirmation from the E report: the best match by ratio, E-Z1893 (stem 1, ratio
   0.05), is not a distinct node in FTDNA's tree at all (it resolves to E-L515), and E-M4269 and
   E-M4279 sit under intermediate nodes in FTDNA that the YFull dump collapses. Short stems here are
   topology and coverage artefacts, individually and not only in aggregate.
3. J-Z1884 (L858) does not match under any reading: parent CI floors at 5000; stem 60 versus 36
   expected under the search anchor. The ~60-backbone-SNP figure behind the architecture fork is
   reproduced mechanically.
4. Nothing in this result touches the site's published field facts (84, 2-of-84, fairness check).

## Status of the series

The latest entry in the null/unresolved series. Revival condition unchanged: corridor ancient DNA (Cudi ~2900 BC,
Khuzestan ~5000 BC, Mesopotamia 3000-1500 BC), or a per-branch clock-SNP dataset that makes the
stem statistic discriminating.

## Site

NO website change made. If Mike wants the research page to record T4, the line that stays within the result is: the
predicted signature turned out to be common in the public tree at the resolution available, so the
scan neither found nor excluded the Noah line; one Arabian-Mesopotamian clade (J2b-M205's main
branch) matches marginally and is noted, not claimed.

## Records

FROZEN-CRITERIA.md (commit 7cde9be), scripts/scan.py, results/scan_results.json,
results/ITERATION-1.md, results/CALIBRATION.md (post-hoc, labelled), results/lit-check-{J,E,GCH,LTO}.md,
DEVIATIONS.md (#1 rate expectation wrong, formula unchanged).