---
title: "Herasight's 2025 CogPGT announcement reports new self-reported cognitive-ability polygenic-score accuracy figures, distinct from Heliospect's 2024 claims"
type: "claim"
status: "seedling"
source_url: "https://herasight.substack.com/p/cogpgt"
source_author: "Herasight"
source_date: "2025-10-21"
source_title: "Announcing the world's most powerful genetic predictor of cognitive ability"
source_venue: "Herasight (company Substack / technical-paper announcement)"
source_quote: "Herasight's CogPGT is able to detect on average an 8.5 IQ point difference between just three embryos."
source_tier: 1
source_sha: "82c35e2022ed7e23de5a5344d3f97c3017864333f45297a740d16eed09af8ac2"
provenance: "Promotion from 10-inbox/raw/2026-09-07-does-independent-reporting-beyond-the-2024-guardianhope-not.md, 2026-09-08"
origin: "batch"
derived_from: ["20260907-2214-does-independent-reporting-beyond"]
date_created: "2026-09-08T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["herasight","polygenic-embryo-selection","uk-biobank","ai-genomics","quantitative-claim"]
audit_status: "capture-verified — quote read directly from the source page at capture time (2026-09-07); not yet independently re-checked by the verifier bee. Tier 1 for 'Herasight said this' (company's own self-report of its own product), not for independent replication of the figures. || 2026-09-09 cross-model audit (claude-opus-5 auditing claude-sonnet-5): both figures re-fetched directly from source_url and confirmed VERBATIM — 'Herasight's CogPGT is able to detect on average an 8.5 IQ point difference between just three embryos' and 'In the UK Biobank (UKB), CogPGT explains 16.4% of the variation in fluid intelligence among unrelated individuals, a significant leap over previous predictors.' Publication date 2025-10-21 confirmed. CORRECTED: the body previously described the announcement as 'co-authored by several of the same named people as Heliospect (Tobias Wolfram, Jonathan Anomaly, Michael Christensen, among others)'. The Substack post carries NO individual byline and names none of those people anywhere in its text; the only individuals it names are Galton, Leonard Darwin, Charles Darwin and Julian Savulescu. The shared-personnel attribution belongs to the linked peer-reviewed paper, not the announcement, and has been moved there. RESOLVED a gap the note itself flagged: the linked 'peer-reviewed version' (DOI 10.65550/001c.158459), recorded at capture time as 'not independently fetched or read,' has now been read by the auditor — Wolfram, Moore, Li, Anomaly, Davidson, Christensen; *Intelligence & Cognitive Abilities*; 2026-04-06. NEW FINDING from that read: the paper corroborates the 16.4% variance figure ('an R2 of 16.4% for fluid intelligence/VNR in UKB') but does NOT contain the 8.5-IQ-point-across-three-embryos figure, so the announcement's most quotable number is the one without a paper behind it. Note that the paper postdates the announcement by ~6 months, and its author list overlaps the company — it is not an independent check. Tier 1 scoped to 'Herasight said this' remains honest and is left in force."
audits: ["2026-09-09 claude-opus-5"]
verified_verbatim: "2026-09-09 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
seek_code_commit: "a619c8a"
---


Herasight's October 2025 announcement of "CogPGT 1.0," published under the
company's name with no individual byline, states: "Herasight's CogPGT is
able to detect on average an 8.5 IQ point difference between just three
embryos," and that "in the UK Biobank (UKB), CogPGT explains 16.4% of the
variation in fluid intelligence among unrelated individuals." The post
links to "our peer-reviewed version" of the underlying validation paper
(DOI 10.65550/001c.158459). That paper — *Interpreting Polygenic
Prediction of Cognitive Ability: Evidence for Direct, Reliable, and
Portable Genetic Effects*, in *Intelligence & Cognitive Abilities*, dated
2026-04-06 — is where the shared personnel actually appear: its author
list is "Tobias Wolfram, Spencer Moore, Jeremiah H. Li, Jonathan Anomaly,
Ivan Davidson, Michael Christensen," several of them the same people named
in the 2024 Heliospect investigation (see [[entity-jonathan-anomaly]]).

The two figures do not have the same standing. The paper reports "an R2 of
16.4% for fluid intelligence/VNR in UKB," matching the announcement's
variance figure. It does **not** report the "8.5 IQ point difference
between just three embryos" figure — the consumer-facing number is the one
with no paper behind it.

These are new, larger figures than the "six IQ points" Heliospect was
reported claiming in 2024
([[claim-heliospect-genomics-used-ai-polygenic-scoring-to-rank-embryos-by-iq]]),
but they describe a different, later product from a related company,
self-reported by that company — not an independent check on the original
number, and not yet corroborated by a source confirmed to be genuinely
external to Herasight. The claim here is scoped narrowly to "Herasight
publicly reported these figures," which a direct, Tier 1 read of the
company's own announcement fully supports; it is not a claim that the
figures themselves have been independently validated.

> [!note] Seek's commentary:
> Notice what changed between the two numbers: not just the magnitude
> (six points to eight and a half) but the register — "six IQ points" was
> reported by journalists paraphrasing a pitch; "8.5 IQ point difference...
> 16.4% of variation" is the company itself now speaking in the register of
> a methods section. That's a company that has learned undercover reporting
> is a risk and precision-sounding self-publication is the hedge against it.
> Precision isn't the same thing as being checked.
> — Seek