# Phenylalanine accumulation in PAH deficiency — HSEF review

> **REDTEAM SYNTHESIS · FINAL · pass-02 · 27 September 2026**
>
> **Revised: Η → Θ.** This edition replaces the one posted earlier the same day. The decisive subclaim (SC2)
> moved because the blocker the earlier edition named is met by a source already in its own ledger, and the
> remaining exception — rare untreated people who escape intellectual disability — blocks the top rung (Ω),
> not Θ. The reasons, and the case for keeping Η, are set out below.
>
> The edition synthesizes **three independent gradings** — A (Θ), B (Η) and a blind run, R2 (Θ) — and two
> adversarial passes by a different model. All inputs are archived unaltered in `../sources/`.
>
> **Sources:** 27 (25 Tier A) · **Load-bearing claims:** 35 · **Verified:** 35 · **Unverified:** 0 ·
> **Contradicted:** 0 · **Gate:** passed
>
> **Adversarial review:** Generated by Sol 6.1; adversarially reviewed by Claude Opus 5.5 (Anthropic) on 2026-10-14. 3 corrections applied (see ADVERSARIAL REVIEW AND APPROVAL). Status: **Approved as revised.**

## In plain terms

Phenylketonuria (PKU) is an inherited condition in which the liver lacks a working enzyme that normally breaks
down phenylalanine, one of the building blocks of dietary protein. Without that enzyme, phenylalanine builds up
in the blood. The claim reviewed here is that this build-up is what causes the severe intellectual disability
seen in people with PKU who are never treated, and that a diet low in phenylalanine, started in the first weeks
of life and kept up through childhood, prevents it.

This review grades the claim Θ (theta), the second-highest level on its scale: the explanation predicts what
happens with high accuracy and leaves only a few exceptions unexplained. The strongest evidence comes from
comparisons that history made possible. Affected brothers and sisters treated early developed far better than
those treated late or not at all. Children did worse the later treatment started and the less tightly
phenylalanine was controlled. Babies without PKU, carried by mothers with PKU and high phenylalanine, were harmed
in step with the mother's levels. A rare minority of untreated people escape intellectual disability; why they
are protected is still unknown, although many of them still show other signs of harm.

Explaining and predicting that protection in advance would move the grade toward the top of the scale; finding
that it reflects something unrelated to phenylalanine would lower it. The grade describes the strength of the
evidence that phenylalanine is the cause. It is not a judgement of anyone's care, and this summary gives no advice
about diet or treatment. The previous edition graded the claim one level lower; the change and its reasons are
recorded openly below.

## AT A GLANCE

| Dimension | Assessment |
|---|---|
| **HSEF level** | **Θ — High predictive accuracy, minimal unresolved exceptions** |
| Confidence | High |
| Verdict | Validated in humans, scope-bounded (`supported`) |
| Role of mechanism | primary driver (step 3) |
| Aggregation | weakest-link, preregistered · SC1 Θ · **SC2 Θ** · SC3 Θ |
| Momentum · Stability · Divergence | Mechanistic only · Stable · Aligned (field framing implies Θ; assessed Θ; gap 0) |
| Previous level | Η (posted 27 September 2026, pass-01) — raised in pass-02 |
| Schema | 5.4 · validator 2.4 · **PASS, 0 fail / 2 disclosed warn** |
| Sources | 27 (25 Tier A) · 35 load-bearing claims · 35 verified |
| Calibration | This edition: unblinded reference case. One blind run (R2) executed and compared after freezing: Θ, inside the sealed band Η–Θ. **Not counted as a completed calibration** pending the owner's decision (see Reviewer notes). |

**The graded claim.** Phenylalanine accumulation arising from PAH deficiency causes the intellectual disability of
untreated phenylketonuria, and dietary phenylalanine restriction begun in the neonatal period and sustained
prevents the severe phenotype.

**What decides the level.** SC2, the primacy subclaim. See *Resolution of SC2* immediately below.

**What would move it to Ω.** A prospectively validated explanation of why some untreated people with severe
biochemical disease keep normal intellect — a measured brain-handling or genetic modifier that predicts outcome in
an independent cohort — together with a necessity-and-sufficiency account of the downstream brain injury.

**What would move it back to Η.** Evidence that protection in untreated classical PKU reflects a
phenylalanine-independent factor that also governs outcome in treated patients, or a well-controlled analysis
showing a residual deficit in optimally treated children that timing, control and nutrition do not explain.

---

## Resolution of SC2 — the decisive subclaim (pass-02)

### Three readings, one rule

Grading A returned **Θ**, grading B **Η**, and the blind run R2 **Θ**. Pass-01 adopted B's Η. Under the
weakest-link rule preregistered in grading B's protocol, SC2 alone sets the aggregate; SC1 and SC3 are Θ in this
edition. The rule is not changed here, and the other two rules return the same letter on this decomposition, so
the outcome is not rule-dependent. Readings are not evidence: the move below rests on named sources, not on two
of three graders agreeing. An independent view (SC2 Θ) was written down after red-teaming R2 and before the prior
edition, pass-01 or B's protocol were opened.

### The case for Ω (upward case first)

Necessity and sufficiency are close to demonstrated at population scale. Removing the exposure early prevents the
severe phenotype: early-treated affected siblings did far better than their late- or never-treated siblings [2],
and register children lost roughly four IQ points for each four weeks' delay in starting treatment and for each
300 µmol/L rise in mean phenylalanine [3]. Adding phenylalanine back provokes deterioration [1,9]. Exposure without
the offspring's own PAH deficiency is sufficient to injure the developing brain: in untreated maternal PKU,
mental retardation, microcephaly and heart disease in the offspring rose with maternal phenylalanine, and 95% of
mothers with blood phenylalanine of 20 mg/dL or more had at least one mentally retarded child [16]; a non-PKU and
a PKU child of one untreated mother were both impaired [27]; and in the prospective collaborative study the
offspring's cognitive index fell as weeks to maternal control increased (r = −0.58) [18]. Grading A and R2 both
argued Ω for the upstream subclaim, and R2 argued that rejecting the whole claim would mean disregarding seven
decades of clinical genetics and newborn screening.

**Rejected for Ω.** Sufficiency across the claimed scope is not deterministic. About one in six or seven untreated
classical patients reached IQ above 70 in the largest sibling series [2]; an international search found 16 living
untreated or late-diagnosed patients with near-normal intellect [20]; across 42 untreated families,
same-genotype siblings differed markedly [21]; two untreated sisters with identical variants and near-identical
blood phenylalanine have full-scale IQs of 88 and 54 [8]. No study has prospectively predicted who is protected.
The candidate explanation — lower brain phenylalanine from different blood–brain handling — rests on three
outcome-selected women in one study [7] and a same-group follow-up in 15 patients whose correlations with
intelligence only approached significance [22]; the 2026 report calls it unconfirmed [8]. The named blocker is the
Ω threshold, "Necessity AND sufficiency demonstrated … no disqualifying contradictory data", and the missing
experiment is a prospective test of a protection modifier.

### The case for Η (as posted)

Two distinct Η cases were posted, and both are kept here.

**Grading B's case** is the protected phenotype above: the same severe biochemistry and genotype coexist with
discordant intellect, the explanation is unvalidated, and people cannot be relabelled "not susceptible" after
their outcome is known. Pass-02 adds evidence that strengthens this case — Ramus et al. [21] show the discordance is
not confined to one family, and Weglage et al. [22] show the brain-handling explanation did not reach significance
for intelligence — and evidence that bears on adult extensions: randomized withdrawal of an enzyme-substitution
drug raised phenylalanine within 8 weeks without significant change in neuropsychiatric or mood endpoints [25],
alongside the noninferior primary endpoint of the 2024 adult loading trial [10].

**Pass-01's case** is residual deficit in early-treated cohorts: children about eight IQ points below revised
norms [3] and arithmetic declining between ages six and twelve [5], with no analysis partitioning that residual
between incomplete control, pre-treatment exposure and a phenylalanine-independent term.

### Resolution: Θ

**Pass-01's blocker is met by a source already in the ledger, as far as observational data can meet it.** Source
[3] reports that the 46 children with the most favourable treatment characteristics had a mean IQ of 112.6 after
standardising for social class, similar to the regression estimate and to revised population norms, and its
authors attribute the residual deficit of early-treated children to "the difficulties in fully controlling the
metabolic abnormality" [3]. The posted resolution quoted the first half of that sentence without its causal clause.
Source [5] reports that children who kept phenylalanine below 900 µmol/L beyond age ten showed no deficits except
in arithmetic, which declined between ages six and twelve in 90% of all the children in the study [5]. The best-treated
subgroup is defined by treatment characteristics, not by outcome, but achieved control is not randomized, so
adherence-linked family factors remain a residual confounder. And the residual is a few IQ points, not the graded
outcome, which is intellectual disability.

**The soft spot, stated plainly:** pass-01's blocker is answered only by observational evidence — a non-randomised
best-treated subgroup in [3] plus the achieved-control comparison in [5] — not by any randomised comparison of
treatment quality.

**B's blocker is a real exception, but a minimal one in the sense Θ allows.** Θ reads "High predictive accuracy
with minimal unresolved exceptions. The mechanism accounts for most major observations in its claimed scope." The
primary literature describes intellectually normal untreated people as very rare [8]. In the largest sibling
series the untreated patients above IQ 70 still showed intellectual impairment [2]; the international series found
that protected patients often had other neurological, psychological or behavioural problems [20]; and the intellectually
normal sister in the 2026 report has white-matter reduction on MRI and increased tendon reflexes [8]. The exception
is to the penetrance of the disability threshold, not to the causal effect of the exposure. It stays in scope and
unexplained, and it is exactly what blocks Ω. Counting it again at Θ would count it twice.

**The maternal-PKU natural experiment** was invoked in the posted Θ case with no source in the ledger. It is now
sourced [16,17,18,27] and enters the independence basis as a fourth qualifying line: exposure without the
offspring's own PAH deficiency, with injury graded by maternal phenylalanine and by the timing of maternal control.
It excludes a PAH-intrinsic brain defect and the affected child's own diet or care as explanations, with
ascertainment errors different from the sibling and register cohorts.

**Result.** SC2 **Θ**, High confidence. Weakest-link over Θ · Θ · Θ = **Θ**. The adult null results [10,25] test
short-term adult function, not developmental intellectual disability, and are retained as constraints on adult
extensions. This remains a boundary judgement; a reader who counts the protected phenotype as more than minimal
will prefer Η, and that reading is kept in full above.

### What pass-01 wrote (superseded, retained)

*The case for Θ.* Maternal PKU syndrome dissociates genotype from exposure: a fetus without PAH deficiency, exposed
to maternal phenylalanine, sustains developmental injury — which is difficult to explain on any gene-intrinsic
account. Sibling comparisons hold family background more nearly fixed while varying treatment timing. Controlled
phenylalanine add-back alters measurable function in already treated patients. On this reading, the residual
deficits of early-treated cohorts are a property of *incomplete prevention* and belong to SC3, leaving primacy with
minimal unresolved exceptions.

*The case for Η.* Θ requires minimal unresolved exceptions in the predictive picture, and the cohort literature
reports exceptions that are neither rare nor fully explained. Smith, Beasley and Ades report early-treated children
remaining approximately eight points below revised IQ norms, concluding that many continue to suffer a mild degree
of neurological impairment. Azen and colleagues report arithmetic scores declining between ages six and twelve in
ninety per cent of their cohort. Whether those residuals reflect imperfect metabolic control, exposure before
treatment onset, or a component of the phenotype not driven by phenylalanine is **not resolved by the admitted
evidence** — and that third possibility is precisely what the primacy subclaim asserts against.

*Resolution: Η.* The Θ case depends on assigning every residual deficit to SC3. That assignment is plausible but
undemonstrated, and the subclaim whose truth it presupposes is SC2 itself. Until an analysis partitions residual
impairment between incomplete control, pre-treatment exposure and a phenylalanine-independent term, the unresolved
exceptions sit inside the primacy claim and Θ overstates. Η — strong convergent causal evidence,
intervention-sensitive — describes the position accurately. This is a close call and the Θ case is not
unreasonable.

### Recorded dissent at SC1

Grading A and the blind run R2 both assigned **Ω** to SC1 (PAH deficiency causes persistent phenylalanine
elevation on an ordinary diet), on the ground that it approaches a metabolic identity. This edition retains
**Θ**. Pass-02 notes that the earlier stated reason — "Ω is reserved and rare" — is not a framework blocker; the
real reason is that this ledger carries no human necessity-and-sufficiency evidence for SC1 beyond one family [8]
and one mouse editing study [14], and SC1 does not bind under weakest-link. R2's own support for Ω was mis-cited
(its PMID resolves to an editorial) and the intended record says most pathogenic alleles act through misfolding
with residual activity, which contradicts R2's wording that mutations "abolish" the enzyme. The disagreement is
recorded and deferred, not resolved.

---

## The blind run (R2) as a third reading

R2 was described by its coordinator as a blind grading from the sealed grader packet, by a different system; it
reached **Θ** with High confidence. The isolation account is not on file, and R2 used schema fields the packet does
not contain, so it worked with more than the packet. Its letter agrees with this edition; much of its route does not.

**Adopted from R2**, after re-resolving every identifier: the maternal-PKU natural experiment (Lenke and Levy [16];
the international study [17]); the meta-analysis of blood phenylalanine and IQ [19]; the large-neutral-amino-acid
loading experiment, which R2 had left unverified [23]; the Lancet seminar as field framing for Divergence [26]; and
R2's own observation that its amino-acid-depletion, oxidative-stress/myelin and critical-period "rivals" are
sub-mechanisms within phenylalanine injury rather than competitors (they are placed within M2 here).

**Not adopted, with reasons.**
- *Aggregation.* R2 called the claim a conjunction that "requires both to hold", then preregistered scope-matched
  aggregation, which takes the highest matching subclaim. A conjunction is scored on the conjunction, not on its
  easier half. R2's own weakest-link alternative (Ζ) came entirely from a downstream-mechanism subclaim the claim
  does not assert. On this edition's three subclaims R2's levels (Ω, Θ, Θ) give Θ by weakest-link anyway.
- *Generalization subclaim at Η and its Ω blocker.* Both rest on a "63% genotype–phenotype concordance" (hence
  "~37% discordance"). In the PAHdb record R2 intended, 63% is the share of pathogenic alleles that are missense
  mutations. The figure is contradicted and dropped.
- *"IQ below 50 in virtually all" untreated patients.* Contradicted by the sibling series (about one in six or seven
  above IQ 70) [2]; R2's cited sources do not carry the figure.
- *A pharmacological independence line.* The sapropterin and pegvaliase trials R2 counted had blood phenylalanine
  as their primary endpoint [24,25]; cap 4, "Biomarker change without functional/phenotypic change caps at Γ", binds the
  inference from that class alone, and the one randomized pegvaliase test with neuropsychiatric endpoints was null at
  8 weeks [25]. R2's pegvaliase figure ("50–70%") belongs to the open-label programme report, not to a controlled
  comparison.
- *Maternal PKU as the study separating Phe toxicity from amino-acid depletion.* The fetal brain is exposed across
  its own blood–brain barrier, so maternal PKU cannot separate those two sub-mechanisms.
- *Momentum "Decisive work underway".* Of the two gene-therapy programmes R2 named, HMI-102's phase 1/2 trial
  (NCT03952156) is terminated and BMN 307's (NCT04480567) is active but not recruiting (ClinicalTrials.gov, checked
  27 September 2026); both have phenylalanine endpoints and could not move this claim.

**Defects in R2 that are not carried here.** Eight of its seventeen ledger PMIDs resolve to unrelated papers
(a chest-pain study, an NICHD historical note, an editorial, a chromosome-mapping note, a warfarin trial, a
spinocerebellar-ataxia paper and two different PKU papers); one evidence pointer resolves to an X-linked spinal
muscular atrophy paper, and a reviewer note explains the discrepancy with a statement that is false. It shipped
`gate_passed: true` with four unverified claims. Full detail: `redteam/identifier-audit.md` and
`redteam/claim-audit.md`.

---

## Redteam findings

| Pass | Finding | Action |
|---|---|---|
| 01 | Aggregates disagreed by one rung | Traced to SC2 alone; resolved in the open (superseded by pass-02) |
| 01 | Only 7 of 15 sources shared between gradings | Flagged; pass-02 added A's untreated-series source [20] |
| 01 | A asserted Ω at SC1 | Retained Θ; dissent recorded |
| 01 | A carried one ledger row with an empty PMID (Bickel 1953) | Resolved: PMID 13098090 and DOI both present |
| 01 | B lacked all schema 5.3 fields | Migrated using A's field work |
| 01 | A declined calibration designation | B's apparatus adopted |
| 01 | The v5.3 additions caught no error in either package | Recorded as a finding about the schema |
| 02 | Posted SC2 resolution truncated source [3]'s conclusion and omitted its best-treated subgroup | SC2 re-adjudicated; Η → Θ |
| 02 | SC2 carried two different blockers (residual deficit in the JSON governing rule; protected phenotype in B's body) | One adjudication addressing both |
| 02 | Maternal PKU invoked with no ledger source | Sources [16,17,18,27] added; fourth independence line |
| 02 | Divergence used the reviewer's own level as "implied" | Recomputed from field framing [8,26]: implied Θ, Aligned |
| 02 | Evidence toward Η missing from the ledger | Added [21,22,25] |
| 02 | Two US Collaborative Study reports carried different study_ids | Harmonised |
| 02 | Blind run R2: 8 of 17 wrong PMIDs, a contradicted load-bearing figure, a gate contradiction | Not used as base; valid contributions re-sourced; dissent recorded |
| 02 | No lay summary (v5.4) | Added |

**Independent verification (pass-02).** All 15 prior ledger records and all 17 R2 records were resolved in PubMed
and compared on title, first author, journal, year, volume, pages and DOI. The 15 prior records are exact. Every
registered claim was re-read against its record, and the full text of [1], [8], [23] (partial) and [25] (partial)
was read.

**Residual limitations.** This edition is not blind. The sealed band was authored by grading B's author and spans
Η–Θ, so it cannot discriminate between the two readings in dispute. The preregistered protocol followed initial
orientation searches. Europe PMC's correction service was rate-limited during pass-02, so erratum and retraction
status was checked by PubMed publication-type searches instead.

---

# Full review

*The body below is grading B, retained as written; pass-02 appended rows to its tables and added short pointer notes. Where it states the aggregate, SC2, Divergence, the tournament order or source counts, the pass-02 sections above govern.*


## AT A GLANCE

**The causal and preventive claim is strongly supported, with a crucial timing distinction: early phenylalanine control can prevent the severe developmental phenotype, whereas later lowering is not established as a reversal of intellectual disability.**

> **Sources:** 15 primary reports · **Claim entries:** 19 · **Factual premises checked:** 19 · **Unverified:** 0 · **Contradicted:** 0 · **Structural gate:** passed (27 checks; 0 failures; 1 acknowledged unblinded-calibration warning).

> *Pass-02: superseded where it concerns these counts (current: Sources 27 (25 Tier A) · load-bearing claims 35 · verified 35 · unverified 0 · contradicted 0 · gate passed; validator 2.4, 0 fail / 2 disclosed warn) — see the pass-02 sections at the top of this document.*

| Dimension | Assessment |
|---|---|
| Overall HSEF | **Η — Strong convergent causal evidence, intervention-sensitive** |
| Confidence | **High** |
| Role of mechanism | **Primary driver**, within the PAH-deficiency pathway |
| Momentum | **Mechanistic only**, nearest permitted category; active targeted-treatment work does not fit this mature field neatly |
| Stability | **Stable** |
| Divergence | **Aligned**: reviewer-mapped implied Η − assessed Η = 0 |

> *Pass-02: superseded where it concerns the overall level, SC2 and Divergence (now Θ, Θ, and Aligned against a field-framing implied Θ) — see the pass-02 sections at the top of this document.*

**Why this score.** Human treatment-timing contrasts, longitudinal outcomes, specific phenylalanine add-back and controlled animal/genetic experiments converge. The broad causal chain does not yet explain why some untreated people with severe biochemical disease avoid intellectual disability. Those people remain in scope. The narrower early-prevention prediction reaches Θ. [1](https://pubmed.ncbi.nlm.nih.gov/13098090/) [2](https://pubmed.ncbi.nlm.nih.gov/4140266/) [3](https://pubmed.ncbi.nlm.nih.gov/2357082/) [8](https://pubmed.ncbi.nlm.nih.gov/41853686/) [9](https://pubmed.ncbi.nlm.nih.gov/21153445/) [12](https://pubmed.ncbi.nlm.nih.gov/29331172/) [13](https://pubmed.ncbi.nlm.nih.gov/30875376/) [14](https://pubmed.ncbi.nlm.nih.gov/42290942/)

**What the statement does not establish.** It does not establish that every PAH variant or elevation produces intellectual disability, that dietary restriction begun at any age reverses established disability, or that early treatment guarantees identical cognition to unaffected peers. Nutritional adequacy matters; eliminating an essential amino acid is not the proposed intervention. [1](https://pubmed.ncbi.nlm.nih.gov/13098090/) [3](https://pubmed.ncbi.nlm.nih.gov/2357082/) [6](https://pubmed.ncbi.nlm.nih.gov/11264437/) [8](https://pubmed.ncbi.nlm.nih.gov/41853686/) [12](https://pubmed.ncbi.nlm.nih.gov/29331172/) [15](https://pubmed.ncbi.nlm.nih.gov/8194559/)

*Reviewed 2026-09-27, America/Edmonton. User-supplied HSEF package, base schema 5.2 and validator 2.2, with declared local extensions. This is a new, unblinded reference case, not a completed blinded calibration run.*

## Claim and scope

**Original:** “Phenylalanine accumulation from PAH deficiency causes intellectual disability, preventable by dietary restriction.”

**Graded proposition:** “Sustained marked phenylalanine elevation caused by PAH-deficient phenylketonuria is a major cause of developmental intellectual disability, which can be prevented by early, sustained, nutritionally adequate dietary phenylalanine restriction.”

The added timing, exposure and nutritional qualifications are substantive and are disclosed. “Preventable” is interpreted as prevention before major developmental injury; it is not replaced by a claim of later reversal. The causal outcome remains intellectual disability, while studies of IQ shifts, attention, motor function and biochemical markers are labeled as such.

Define the population using biochemical and, where available, genetic evidence of **PAH deficiency before the outcome**. Retain each study's pretreatment severity category. Historical classical-PKU strata commonly involve untreated blood concentrations around or above 1,200 micromol/L; that is a descriptive stratum, not an exact toxicity cutoff or a treatment recommendation. Milder elevations must be analyzed separately. Existing records do not provide a prospectively validated susceptibility classifier that can remove protected untreated people from this population. [6](https://pubmed.ncbi.nlm.nih.gov/11264437/) [7](https://pubmed.ncbi.nlm.nih.gov/9809507/) [8](https://pubmed.ncbi.nlm.nih.gov/41853686/) [15](https://pubmed.ncbi.nlm.nih.gov/8194559/)

For prevention, record actual age at treatment start, subsequent phenylalanine exposure, nutritional adequacy and follow-up. Early infancy is supported as a developmental intervention window; these records do not identify one exact safe deadline for every child. Do not define an adequately treated group solely by its later normal IQ. Achieved metabolic control is a post-treatment variable and must not be mistaken for randomized assignment. [2](https://pubmed.ncbi.nlm.nih.gov/4140266/) [3](https://pubmed.ncbi.nlm.nih.gov/2357082/) [4](https://pubmed.ncbi.nlm.nih.gov/7086586/) [5](https://pubmed.ncbi.nlm.nih.gov/1985428/)

Historical IQ cutoffs help assess intellectual development but do not fully reproduce a contemporary intellectual-disability diagnosis, which also includes adaptive functioning. This limits exact penetrance and prevention-rate estimates. The report supplies no pooled probability that a particular infant will develop disability.

**A result that could fail:** within an independently defined PAH-deficient infant population, verified early and sustained phenylalanine control fails reproducibly to reduce later intellectual disability despite adequate nutrition and comparable baseline disease and other causes. Such a study must be precise enough to exclude the prespecified meaningful effect. Participants cannot be relabeled “not susceptible” or “not true cases” after outcomes are known. This is a falsification condition, not a proposal to withhold effective infant treatment.

## Subclaim score matrix

| ID | Subclaim | Type | Scope | HSEF | Confidence |
|---|---|---|---|---|---|
| SC1 | PAH deficiency impairs phenylalanine metabolism and causes accumulation. | Mechanism | Defined metabolic defect; degree depends on residual activity and exposure | Θ | High |
| SC2 | Developmental phenylalanine accumulation is a major cause of intellectual disability. | Primacy | Biochemically defined PKU; mild and protected untreated phenotypes remain visible | Η | High |
| SC3 | Early, sustained dietary control can prevent the severe developmental phenotype. | Prevention | Early infancy through development, with adequate nutrition; no late-reversal or perfect-cognition guarantee | Θ | High |

> *Pass-02: superseded where it concerns SC2's level (current values: SC1 Θ · SC2 Θ · SC3 Θ) — see the pass-02 sections at the top of this document.*

**Aggregation: weakest-link.** The user asserts the causal chain and dietary prevention jointly, so all three are required. min(Θ, Η, Θ) = **Η**. This rule was recorded before assigning scores, after initial orientation searches. Primary-subclaim aggregation around SC2 and scope-matching to the full causal chain also return Η. No favorable narrower score is substituted for the combined claim.

> *Pass-02: superseded where it concerns the arithmetic (current: weakest-link min(Θ, Θ, Θ) = Θ; primary-subclaim and scope-matched also return Θ) — see the pass-02 sections at the top of this document.*

`prevention` is an explicit local type extension. It is distinct from both a molecular mechanism and reversibility after an established phenotype. Every subclaim has nonempty, equal `claimed_scope` and `scope` fields, plus operational definitions and a possible failed prediction.

## COMPETING MECHANISTIC MODELS

| Rank | Model | HSEF | What the assessment means |
|---|---|---|---|
| 1 | M2: phenylalanine injury modified by development and multiple brain pathways | Η | Broadest coverage; modifiers and complete downstream mediation remain incompletely resolved |
| 2 | M1: PAH-related phenylalanine accumulation causes preventable disability | Η | Strong nominal causal claim, contained within M2 |
| 3 | M3: tyrosine shortage alone explains the neurological phenotype | Γ | Plausible contributor; supplementation evidence does not establish a sufficient replacement explanation |
| 4 | M4: nonspecific nutrition and care explain the apparent dietary rescue | Γ | Some nutritional effects are real, but they do not explain all the Phe-specific evidence |

> *Pass-02: superseded where it concerns model order (nominal M1 now first at Θ, M2 second at Η; scorecard `competing_models`) — see the pass-02 sections at the top of this document.*

M1 and M2 are compatible, not mutually exclusive causal competitors. Rank denotes explanatory coverage, not a posterior probability or justification for a higher grade. The existence of unspecified modifiers is not itself evidence that the modifiers have been explained.

The tyrosine-only explanation is constrained by a trial in which plasma tyrosine rose without better neuropsychological performance. That short trial does not exclude every role for brain amino-acid availability or neurotransmitters during development. In mice, Phe reduction restored monoamines and some behavior without correcting established cognitive deficits, further separating these endpoints. [11](https://pubmed.ncbi.nlm.nih.gov/9579151/) [12](https://pubmed.ncbi.nlm.nih.gov/29331172/)

The nutrition-only explanation has a real experimental observation behind it: a nutrient combination improved an animal recognition-memory task even at high Phe. Nevertheless, isolated Phe add-back and targeted repair of the native PAH allele are not readily explained by extra clinical attention or a generally better diet. The evidence supports nutritional modification alongside Phe causation. It does not show that nutrition without Phe control substitutes for preventing human PKU disability. [1](https://pubmed.ncbi.nlm.nih.gov/13098090/) [9](https://pubmed.ncbi.nlm.nih.gov/21153445/) [13](https://pubmed.ncbi.nlm.nih.gov/30875376/) [14](https://pubmed.ncbi.nlm.nih.gov/42290942/)

No unique experiment can choose between M1 and a broader model that contains M1. The discordant sisters expose an explanatory gap; they do not validate one particular transport, genetic or nutritional explanation. [8](https://pubmed.ncbi.nlm.nih.gov/41853686/)

## HYPOTHESIS STATE EVALUATION FRAMEWORK (HSEF v5.4)

<!-- consensus-consistency-2026-10-09 -->
### Consensus consistency validation — 9 October 2026

This is a transparently unblinded consistency check, not a new peer review or independent blinded audit. The frozen review cutoff remains **2026-09-27**. Its source/claim inventory and grade are preserved; Consensus search results are treated as discovery pointers only.

**Grade:** Θ / High (unchanged). The integrated claim is that high phenylalanine causes the severe untreated PKU phenotype and early dietary control prevents it. The review identifies rare untreated people with severe biochemical disease who escape intellectual disability; this unexplained exception blocks Ω. Adult challenge outcomes inform a narrow exposure-sensitivity subclaim and cannot replace evidence about childhood developmental injury.

**Frozen inventory:** 27 evidence entries and 35 registered claims. Two Consensus search results and two fetched records are retained as raw JSON under `checks/pku-phenylalanine/` in the audit root. Original comparisons below were cross-checked against the review's existing primary-source records; no newly surfaced sources have been added to the original counts.

**Primary comparisons; already in the frozen review inventory:**

| PMID / DOI | Original primary source and claim relation | Access documented in frozen review | Result and scope |
|---|---|---|---|
| 13098090 / 10.1016/S0140-6736(53)90473-5 | Bickel et al. 1953, #1; human phenylalanine add-back | Complete two-page original scanned paper read; PubMed/Europe PMC bibliographic metadata checked; PubMed has no abstract | A single historical dietary add-back case showed neurologic/behavioral change with altered phenylalanine intake. Supports a specific exposure effect but not neonatal prevention or reversal of established intellectual disability. |
| 21153445 / 10.1007/s10545-010-9253-9 | ten Hoedt et al. 2011, #9; blinded adult challenge | Original-paper abstract and bibliographic/status metadata read in Europe PMC; Entrez/web cross-check | Randomized, double-blind crossover Phe supplementation affected adult mood/sustained attention. Does not recreate developmental exposure or prove childhood ID. |
| 38569786 / 10.1016/j.ajcnut.2023.11.007 | Trepp et al. 2024, #10; adult randomized crossover | Original-paper abstract and bibliographic/status metadata read in Europe PMC; Entrez/web cross-check | In 30 early-treated adults, four weeks of higher Phe was noninferior on the primary working-memory-accuracy outcome; sustained attention was the exception among secondary outcomes. Does not establish childhood or lifelong safety. |
| 4140266 / 10.1016/S0140-6736(74)91873-X | Smith & Wolff 1974, #2; early-versus-late treatment natural-history comparison | Original-paper abstract and bibliographic/status metadata read in indexed records; cross-checked with Entrez/web | Historical sibling/timing evidence supports the importance of early treatment, with ascertainment and historical-care confounding retained. |

The fresh Consensus search surfaced adult challenge evidence and guideline/review synthesis. The primary adult trials sharpen the limitation on generalizing short-term adult endpoints to untreated childhood development; they do not contradict the established developmental-treatment evidence or remove the documented rare exception. Grade remains Θ and original counts remain 27/35.



This dated supplement adds source reconciliation to the original review. Its original narrative, registered scope, source inventory and grades remain historical records. No earlier adversarial approval is attached to this case, so no earlier approval certifies this supplement. New sources described here are supplemental and are not counted in the original inventory.
<!-- end-consensus-consistency -->


### Routing and domain fit

The supplied index has no exact PAH/PKU entry. Rubric 06 was read, but its thresholds address lifespan and aging and explicitly distinguish developmental effects from aging. Applying those endpoints literally would misclassify this inherited developmental disorder.

This review therefore applies the **universal causal ladder**, with a declared developmental-metabolic interpretation: actual intellectual development is the clinical endpoint, not lifespan; acute attention, mouse learning, motor function and metabolite measurements support only their measured endpoints. Threshold quotations in the scorecard come verbatim from the universal framework. They are not presented as quotations from a nonexistent PKU rubric.

### Human causal and preventive evidence

The early-treatment sibling comparison is central because it examines the intellectual outcome the hypothesis names. Smith and Wolff described 28 affected sibling pairs and found better intellectual development with early treatment than with late or absent treatment. Shared family background narrows some alternatives, but ascertainment through an impaired index child, birth order and historical changes in care remain. This is a natural experiment, not a randomized untreated-infant trial. [2](https://pubmed.ncbi.nlm.nih.gov/4140266/)

The much larger UK register reinforces the timing and exposure pattern. It followed 808 of 1,031 early-treated children prospectively and related childhood IQ to treatment start and metabolic control. It also documented residual average deficits and an association between excessively low concentrations and poorer outcome. Its regression coefficients are not randomized causal effect sizes, and successful control cannot be selected retrospectively as proof of a guaranteed preventive effect. [3](https://pubmed.ncbi.nlm.nih.gov/2357082/)

The US programme tests a different developmental stage. At age six, children were randomized to continue or discontinue dietary therapy. The preliminary report found reading and spelling differences, but its adjusted IQ comparison at age eight was 101 versus 98 with P=0.075. That IQ difference must not be called statistically significant. The later age-12 report supports relationships with metabolic history, but its achieved-control groups are observational and are not another independent trial. [4](https://pubmed.ncbi.nlm.nih.gov/7086586/) [5](https://pubmed.ncbi.nlm.nih.gov/1985428/)

Bickel's original paper helps distinguish a Phe-specific effect from general care or diet composition. In an already impaired child, adding Phe back was followed by deterioration, with improvement after its withdrawal; the first add-back was concealed from the mother. This is an informative single-case challenge, not proof that established intellectual disability was cured. The authors themselves left the ultimate developmental outcome open. [1](https://pubmed.ncbi.nlm.nih.gov/13098090/)

### Supporting endpoints and their limits

Two blinded adult trials test Phe exposure more directly while leaving much of the ordinary care context fixed. The smaller trial found worse mood and sustained attention. The larger 2024 trial found its primary working-memory-accuracy endpoint noninferior after four weeks of higher intake, with sustained attention differing among secondary outcomes. Both results remain in the dossier. Neither brief adult experiment reproduces developmental exposure or establishes the incidence of intellectual disability. [9](https://pubmed.ncbi.nlm.nih.gov/21153445/) [10](https://pubmed.ncbi.nlm.nih.gov/38569786/)

Animal studies add intervention control. Low-Phe diets can improve some learning or behavioral measures, while nutrient composition changes some outcomes even without lowering Phe. Adult diet or liver-directed gene therapy improved monoamines and nesting but left severe learning deficits. A 2026 native-allele editing study changed Phe exposure and motor abnormalities without relying on wholesale diet replacement. These studies support specificity, timing and partial reversibility; they do not become human intellectual-disability-prevention trials. [12](https://pubmed.ncbi.nlm.nih.gov/29331172/) [13](https://pubmed.ncbi.nlm.nih.gov/30875376/) [14](https://pubmed.ncbi.nlm.nih.gov/42290942/)

### Exceptions are retained, not defined away

Mild hyperphenylalaninemia is not interchangeable with untreated classical PKU. One cohort with persistent concentrations of 360–600 micromol/L had normal-range average intellectual outcomes. Another, younger cohort spanning concentrations below 900 micromol/L showed average IQ deficits and exposure associations. Their ranges, ages and selection differ. They support caution about treating “any accumulation” as a deterministic intellectual-disability claim; neither establishes a universal no-treatment boundary. [6](https://pubmed.ncbi.nlm.nih.gov/11264437/) [15](https://pubmed.ncbi.nlm.nih.gov/8194559/)

The more difficult exception occurs within severe biochemical disease. Demirbas and colleagues reported two untreated sisters with the same PAH variants, similar repeatedly high blood Phe and discordant intellectual outcomes. Their reported full-scale IQs were 88 and 54, with substantial differences in independent functioning. The work did not establish a biochemical or genetic explanation, did not measure brain Phe, and did not continuously measure infant exposure. It is a meaningful counterexample to an invariably sufficient genotype/blood-level model, not a population estimate and not a refutation of Phe as a major cause. [8](https://pubmed.ncbi.nlm.nih.gov/41853686/)

The older MRS study supplies a candidate explanation: lower brain Phe in three protected untreated individuals. It does not justify a validated susceptibility rule. Participants were studied after their outcomes were known, and the newer discordant-sibling study did not confirm that mechanism in its own subjects. Brain transport must be tested independently rather than inferred from preserved cognition. [7](https://pubmed.ncbi.nlm.nih.gov/9809507/) [8](https://pubmed.ncbi.nlm.nih.gov/41853686/)

### Cap audit

All six universal constraints were assessed for each subclaim. None imposes an overall ceiling on the complete convergent set:

| Constraint | Application here |
|---|---|
| In-vitro-only evidence cannot exceed Γ | Human and in-vivo intervention evidence is present |
| Pure association cannot exceed Β | Register correlations support, but do not replace, natural experiments and specific perturbations |
| Ectopic overexpression caps its own inference | The older gene-transfer arm remains model evidence; native-allele editing and human evidence are separate |
| Biomarker-only improvement cannot establish the disease phenotype | Blood Phe, monoamines, motor outcomes and intellectual development remain distinct |
| Forced-transfer evidence does not establish endogenous propagation | No propagation claim is advanced; oral loading has its actual acute endpoint |
| Familial genetic evidence is scope-restricted | The claim concerns inherited PAH deficiency; it is not generalized to all intellectual disability |

For SC1, impaired conversion and accumulation are the metabolic function being evaluated. That subclaim is not permitted to prove SC2 merely by changing its biomarker. Its higher biochemical score cannot lift the weakest-link clinical aggregate.

### Β/Γ/Δ boundary check

The universal framework permits natural experiments and therapeutic perturbations with relevant outcomes when direct human induction would be unethical. SC2 and SC3 cross the clinical boundary through human intellectual-development contrasts, supported by specific Phe challenges and controlled models. They do not cross it solely because blood Phe falls. SC1 separately crosses its metabolic-functional boundary through human functional findings and targeted correction.

An untreated-infant randomized trial is not required, and the supporting observational studies do not re-cap the complete dossier at Β. Conversely, calling a study an RCT does not make a nonsignificant primary result positive or turn an adherence-selected subgroup into a randomized comparison.

### Threshold ladder and adjudication

| Rung | Universal requirement, condensed here | SC1 | SC2 | SC3 |
|---|---|---|---|---|
| Δ | Relevant causal perturbation | Met | Met | Met |
| Ε | Causality in a defined context | Met | Met | Met |
| Ζ | Convergent lines with different principal errors | Met | Met | Met |
| Η | Intervention-sensitive trajectory with no major direction-of-causation contradiction | Met | Met | Met |
| Θ | High predictive accuracy, minimal unresolved exceptions | Met for metabolic function | **Not established for broad clinical explanatory coverage** | Met for the early-prevention prediction |
| Ω | Necessity AND sufficiency across scope | Not established in this dossier | Not reached after Θ fails; deterministic extensions are contradicted | Not established |

> *Pass-02: superseded where it concerns the SC2 Θ row (now met; the protected untreated phenotype is counted once, at Ω) — see the pass-02 sections at the top of this document.*

For SC2, the unresolved protected untreated phenotype prevents the strongest account of clinical penetrance. This is the same exception-accounting principle applied to the previous H. pylori review: an observed boundary is not automatically an explained one. It is not a demand for perfect penetrance or a particular risk-model statistic.

For SC3, the assessed prediction is narrower: intervention before major developmental injury can prevent the severe untreated phenotype. Smaller residual deficits, later adult task results and failure to reverse already established injury do not test that same prediction. They still constrain broader wording and remain in the record. No claim of universal guarantee is made.

### Adversarial steelman

**One rung higher, Θ:** this is a strong alternative. The usual developmental trajectory changes with early treatment; molecularly specific challenges and genetic rescue reduce alternative explanations. A causal claim need not have 100% penetrance, and Θ permits some exceptions. The preventive subclaim is assigned Θ for precisely this reason.

**Why the broad aggregate remains Η:** the scope also includes the causal account of intellectual disability across baseline-defined severe PKU. The unexplained discordance cannot be removed by labeling the protected people nonsusceptible. On the strict interpretation used consistently in this series, the dossier has not demonstrated the breadth of explanation required by Θ for that subclaim. This is a disclosed boundary judgment, not a finding that prevention is doubtful. A reviewer who interprets Θ as population-level causal predictive success with incompletely explained low-penetrance exceptions could defend Θ; the disagreement would be about this threshold, not the main causal evidence.

**One rung lower, Ζ:** historical comparisons, adherence confounding, dietary cointerventions and supporting animal/adult endpoints justify caution. Nevertheless, observed human intellectual development changes with the timing of intervention, and Phe-specific challenges narrow the nonspecific-care explanation. These are trajectory-sensitive findings, supporting Η rather than convergence alone.

**Why not Ω:** high blood Phe and the same PAH genotype do not invariably produce intellectual disability, early treatment does not guarantee every cognitive outcome, and no complete necessity-and-sufficiency account of the downstream injury is established. This does not require Phe to cause every intellectual disability in the general population; it concerns the actual PKU scope.

> *Pass-02: superseded where it concerns the level being steelmanned (the steelman is re-run with Θ assigned) — see the pass-02 sections at the top of this document.*

### Falsification audit and mechanism role

The deterministic extension is contradicted, but the major-cause/prevention proposition is not. The discordance is counted once at the Θ threshold. The adult primary noninferiority result constrains adult extensions, and absent late cognitive rescue constrains reversal. Neither is relabeled as failed neonatal prevention.

**Role: primary driver.** Controlling the exposure before injury changes development; engaging the metabolic/exposure pathway produces functional pathology in relevant systems; continued exposure affects outcome. This supports an upstream causal input in the PAH pathway, not a passive marker. “Rate-limiting” here is an inference about causal contribution, not a measured kinetic constant or deterministic rule for every person.

### Momentum, stability, divergence and change record

**Momentum:** active mechanistic and targeted-treatment work is represented by the 2026 native-allele editing experiment and investigation of discordant untreated siblings. The schema's nearest permitted category is “Mechanistic only,” with a poor fit to a mature field that already prevents clinical disease. No comprehensive trial census or named pending result resolving the next broad threshold is claimed. [8](https://pubmed.ncbi.nlm.nih.gov/41853686/) [14](https://pubmed.ncbi.nlm.nih.gov/42290942/)

**Stability:** Stable, because no specific pending finding was identified that would overturn the established causal/prevention conclusion or demonstrably resolve the next threshold.

**Divergence:** primary papers strongly affirm the metabolic cause while retaining unresolved pathogenesis and discordance. Reviewer-mapped implied Η − assessed Η = 0, Aligned. A paper's strong causal wording is not silently translated into universal sufficiency or an asserted field-wide HSEF score. [8](https://pubmed.ncbi.nlm.nih.gov/41853686/) [12](https://pubmed.ncbi.nlm.nih.gov/29331172/) [14](https://pubmed.ncbi.nlm.nih.gov/42290942/)

> *Pass-02: superseded where it concerns the implied level (recomputed from field framing [8] [26]: implied Θ, assessed Θ, gap 0) — see the pass-02 sections at the top of this document.*

This is a fresh assessment of PAH/PKU, with no prior PAH score supplied. The previous ulcer review supplies the revised method, not a prior biological score. The full causal score is **Η, High confidence**; the closer Η/Θ distinction is an explicit adjudication judgment.

> *Pass-02: superseded where it concerns the full causal score and the change record (current: Θ, High confidence; revised from Η in pass-02, direction raised) — see the pass-02 sections at the top of this document.*

## Evidence inventory and causal tally

The rows below preserve endpoints and limitations. “Decisive” refers to the specified perturbation/function, not necessarily to direct prevention of human intellectual disability.

| Evidence | System | Pivotal perturbation? | What it shows and limits |
|---|---|---|---|
| E1 [1](https://pubmed.ncbi.nlm.nih.gov/13098090/) | human interventional; Moderate | Yes, endpoint-limited | Phenylalanine add-back changes neurological/behavioral state while much of the diet and care context remains fixed. Single case, not neonatal prevention or reversal of intellectual disability. |
| E2 [2](https://pubmed.ncbi.nlm.nih.gov/4140266/) | natural experiment; Strong | Yes, endpoint-limited | Earlier dietary treatment in affected siblings is associated with substantially better intellectual development. Not randomized; ascertainment and birth-order changes in care remain. |
| E3 [3](https://pubmed.ncbi.nlm.nih.gov/2357082/) | human observational; Strong | No | Early-treatment register links actual timing/control to IQ and describes residual deficits. Observational treatment-quality analysis; does not isolate Phe from family/nutritional covariates. |
| E4 [4](https://pubmed.ncbi.nlm.nih.gov/7086586/) | human interventional; Moderate | Yes, endpoint-limited | Randomized continuation/discontinuation tests later childhood dietary dependence. IQ difference is not statistically significant; positive school endpoints are not incident ID. |
| E5 [5](https://pubmed.ncbi.nlm.nih.gov/1985428/) | human observational; Moderate | No | Longer childhood follow-up relates intellectual/achievement outcomes to metabolic history. Achieved-control comparison is not randomized; same US programme as [4]. |
| E6 [6](https://pubmed.ncbi.nlm.nih.gov/11264437/) | human observational; Moderate | No | Persistently mild biochemical elevations can coexist with normal-range intellectual outcomes. Small cohort does not establish a universal safe concentration or current treatment policy. |
| E7 [7](https://pubmed.ncbi.nlm.nih.gov/9809507/) | human interventional; Moderate | No | MRS and an oral load characterize blood–brain phenylalanine transport variation. Cognitive comparison is outcome-selected and not a prospective susceptibility validation. |
| E8 [8](https://pubmed.ncbi.nlm.nih.gov/41853686/) | human observational; Moderate | No | Same PAH variants and similar severe biochemical phenotype occur with discordant cognition. Two people cannot estimate penetrance; brain Phe and continuous infant exposure were unavailable. |
| E9 [9](https://pubmed.ncbi.nlm.nih.gov/21153445/) | human interventional; Moderate | Yes, endpoint-limited | Blinded Phe supplementation affects adult attention and mood. Does not recreate developmental exposure or demonstrate intellectual disability. |
| E10 [10](https://pubmed.ncbi.nlm.nih.gov/38569786/) | human interventional; Strong | No | Randomized adult loading did not worsen the primary working-memory-accuracy outcome beyond the noninferiority margin. Not proof of childhood or lifelong safety; attention outcome differed. |
| E11 [11](https://pubmed.ncbi.nlm.nih.gov/9579151/) | human interventional; Moderate | No | Tyrosine rose without measured neuropsychological benefit. Does not eliminate every developmental or downstream tyrosine-mediated mechanism. |
| E12 [12](https://pubmed.ncbi.nlm.nih.gov/29331172/) | animal model; Strong | Yes, endpoint-limited | Adult lowering by diet/gene therapy separates monoamine and nesting rescue from persistent learning deficits. Biochemical rescue is not cognitive reversal or evidence of failed neonatal prevention. |
| E13 [13](https://pubmed.ncbi.nlm.nih.gov/30875376/) | animal model; Strong | Yes, endpoint-limited | Controlled low-Phe and nutrient diets improve some learning/behavioral endpoints. Incomplete task rescue and benefit of nutrients at high Phe constrain a Phe-only explanation of every readout. |
| E14 [14](https://pubmed.ncbi.nlm.nih.gov/42290942/) | animal model; Strong | Yes, endpoint-limited | Correction of a native pathogenic PAH variant reduces Phe and improves motor phenotype. Motor rescue is not human intellectual-disability prevention. |
| E15 [15](https://pubmed.ncbi.nlm.nih.gov/8194559/) | human observational; Moderate | No | Milder biochemical phenotypes can show IQ shifts despite not uniformly producing severe disability. Different exposure/age cohorts do not yield a universal toxicity threshold. |
| E16 [16](https://pubmed.ncbi.nlm.nih.gov/7421947/) | natural experiment; Strong | Yes | Offspring exposed in utero to high maternal phenylalanine have greatly increased mental retardation, microcephaly and heart disease, correlated with maternal levels. Retrospective survey with literature cases; case-finding bias possible (the authors argue it cannot explain the result); offspring genotype not individually reported in the abstract. |
| E17 [17](https://pubmed.ncbi.nlm.nih.gov/14654658/) | natural experiment; Strong | Yes | Maternal dietary control achieved before or early in pregnancy is associated with optimal offspring outcome; morbidity falls with pre- and periconceptional control. Achieved control is not randomized; outcome here is birth outcome, with cognition reported in the companion report [18]. |
| E18 [18](https://pubmed.ncbi.nlm.nih.gov/10683054/) | natural experiment; Strong | Yes | Offspring cognitive index falls with the number of weeks before maternal metabolic control; offspring of mothers controlled before pregnancy score in the normal range. Observational dose-timing within a prospective study; not randomized; four-year outcome only. |
| E19 [19](https://pubmed.ncbi.nlm.nih.gov/17591452/) | human observational; Moderate | No | Pooled within-study correlations show lower IQ with higher blood phenylalanine in early-treated patients during childhood. Correlational pooling of overlapping cohorts; its stated aim was to validate blood Phe as a trial biomarker; not a causal estimate. |
| E20 [20](https://pubmed.ncbi.nlm.nih.gov/31731404/) | human observational; Moderate | No | An international search found 16 living untreated or late-diagnosed patients with near-normal intellect despite high Phe, most with other neurological, psychological or behavioural problems. A selected case series; cannot estimate frequency; no mechanism established. |
| E21 [21](https://pubmed.ncbi.nlm.nih.gov/10203137/) | human observational; Moderate | No | In untreated PKU, predicted PAH activity relates to intellect but with extensive overlap; same-genotype siblings can differ markedly, not explained by diet. Does not identify the modifier; historical intellectual assessment. |
| E22 [22](https://pubmed.ncbi.nlm.nih.gov/11601498/) | human interventional; Weak | No | Brain Phe and transport kinetics vary widely at similar blood Phe and track white-matter abnormality. Correlations with intelligence only approached significance; small sample; same group as [7], so not an independent replication. |
| E23 [23](https://pubmed.ncbi.nlm.nih.gov/10207169/) | human interventional; Moderate | Yes | In adults with PKU, an oral Phe load raises brain Phe with acute EEG slowing; co-loading other large neutral amino acids blocks both. Six adults, acute endpoint; a mechanism/brain-handling result, not developmental ID. |
| E24 [24](https://pubmed.ncbi.nlm.nih.gov/17693179/) | human interventional; Moderate | No | A pharmacological agent that augments residual PAH activity lowers blood Phe relative to placebo. Primary endpoint was blood Phe only; no cognitive or developmental endpoint — cannot carry SC2 or SC3 (cap 4). |
| E25 [25](https://pubmed.ncbi.nlm.nih.gov/29628378/) | human interventional; Moderate | No | Randomized withdrawal of an enzyme-substitution therapy raised blood Phe within 8 weeks without significant change in neuropsychiatric or mood endpoints. Eight-week adult discontinuation; the authors suggest duration or tools were inadequate; says nothing about developmental ID. |
| E26 [26](https://pubmed.ncbi.nlm.nih.gov/20971365/) | human observational; Weak | No | Field framing: untreated classic PKU leads to profound irreversible disability and early diet prevents most neuropsychological complications. A review; used only for the Divergence implied level, never for an experimental result. |
| E27 [27](https://pubmed.ncbi.nlm.nih.gov/1442884/) | natural experiment; Weak | No | A PKU and a non-PKU child from untreated pregnancies of one mother were both microcephalic and intellectually impaired; fetal PAH offered little protection. Two children; case report. |

**Seven pivotal interventions/natural experiments, not seven human ID-prevention trials:** [1,2,4,9,12,13,14]. Only [2] directly contrasts early treatment with late/absent treatment for intellectual development. The other six test specific exposure, later childhood function or animal rescue. The larger observational cohorts contribute replication and timing/severity evidence without becoming randomized interventions. Reports [4] and [5] share the US programme; the UK reports are not treated as independent modalities or pooled participant totals.

**Pass-02 tally.** Eleven records are flagged decisive, representing ten studies: [1,2,4,9,12,13,14], the maternal-PKU survey [16], the Maternal PKU Collaborative Study [17,18] (one study, two reports, counted once) and Pietz [23]. Pietz bears on the brain-handling submechanism only, so the audited tally for SC2/SC3 is nine studies. Only [2] and the maternal studies test the intellectual outcome in humans directly.

## Independence basis

| Evidence line | Sources | Different error addressed | Endpoint status |
|---|---|---|---|
| Human developmental diet intervention and natural experiments | [2](https://pubmed.ncbi.nlm.nih.gov/4140266/) [3](https://pubmed.ncbi.nlm.nih.gov/2357082/) [4](https://pubmed.ncbi.nlm.nih.gov/7086586/) [5](https://pubmed.ncbi.nlm.nih.gov/1985428/) [15](https://pubmed.ncbi.nlm.nih.gov/8194559/) [19](https://pubmed.ncbi.nlm.nih.gov/17591452/) | Affected-sibling and treatment-timing contrasts challenge an inevitable genotype-only brain outcome. Prospective outcomes reduce retrospective outcome recall; family, birth-order and adherence confounding remain. | Childhood intellectual/achievement outcomes; [2] directly compares early treatment against late/absent treatment. Qualifying modality. Pass-02: sources extended with [19]. |
| Specific phenylalanine add-back with the treatment/care context held more nearly fixed | [1](https://pubmed.ncbi.nlm.nih.gov/13098090/) [9](https://pubmed.ncbi.nlm.nih.gov/21153445/) [10](https://pubmed.ncbi.nlm.nih.gov/38569786/) | Isolating added Phe, with blinded crossover in adults, addresses nonspecific nutritional benefit, extra attention and secular changes that could mislead developmental cohorts. | Short-term neurological behavior, mood, attention and working memory; not induced developmental ID. Qualifying modality. |
| Controlled PAH-deficient animal experiments, including dietary manipulation and native-allele repair | [12](https://pubmed.ncbi.nlm.nih.gov/29331172/) [13](https://pubmed.ncbi.nlm.nih.gov/30875376/) [14](https://pubmed.ncbi.nlm.nih.gov/42290942/) | Experimental control and targeted genetic correction address family/adherence confounding and diet-only cointerventions. Human clinical findings supply the species-relevance check. | Metabolism, neurotransmitters, learning tasks, nesting and motor behavior; not a human ID diagnosis. Qualifying modality. |
| Human PAH functional and genotype-discordance observations | [8](https://pubmed.ncbi.nlm.nih.gov/41853686/) [20](https://pubmed.ncbi.nlm.nih.gov/31731404/) [21](https://pubmed.ncbi.nlm.nih.gov/10203137/) | Links the biochemical defect to the specified inherited disorder while exposing that genotype and blood Phe do not uniquely predict cognition. | Biochemical function and existing intellectual phenotype in two siblings. Supporting/boundary row only. Pass-02: sources extended with [20], [21]. |
| Brain-versus-blood phenylalanine physiology | [7](https://pubmed.ncbi.nlm.nih.gov/9809507/) [22](https://pubmed.ncbi.nlm.nih.gov/11601498/) [23](https://pubmed.ncbi.nlm.nih.gov/10207169/) | Direct brain measurements address the assumption that blood Phe fully specifies cerebral exposure. | MRS and transport response, with cognition assessed in outcome-selected individuals. Supporting/boundary row only. Pass-02: sources extended with [22], [23]. |
| Maternal-PKU natural experiment: exposure without offspring PAH deficiency (pass-02) | [16](https://pubmed.ncbi.nlm.nih.gov/7421947/) [17](https://pubmed.ncbi.nlm.nih.gov/14654658/) [18](https://pubmed.ncbi.nlm.nih.gov/10683054/) [27](https://pubmed.ncbi.nlm.nih.gov/1442884/) | A PAH-genotype-intrinsic brain defect, and confounding by the affected child's own diet, care or adherence: the exposed offspring are ordinarily not PAH-deficient [27], yet injury rises with maternal Phe and with delay to maternal control [16,18]. Its ascertainment and adherence errors (maternal case-finding, maternal diet control) differ from those of the sibling and register cohorts. | Offspring intellectual/developmental outcome, microcephaly, congenital anomalies. Qualifying modality. |
| Pharmacological Phe lowering (blind run R2's proposed line) — not qualifying (pass-02) | [24](https://pubmed.ncbi.nlm.nih.gov/17693179/) [25](https://pubmed.ncbi.nlm.nih.gov/29628378/) | Would exclude dietary-compliance and care confounding if a functional endpoint moved; here the primary endpoints are blood Phe [24,25] and the randomized functional secondary endpoints were null at 8 weeks [25]. | Blood Phe; adult short-term neuropsychiatric scales. Not qualifying. |

**Five disclosed rows; three qualifying causal modalities; one direct human developmental-prevention modality.** Different regimens, cohorts and endpoints do not automatically supply independence. Specific blinded Phe add-back earns a separate row because its cointervention and selection errors differ from historical whole-diet comparisons. The animal experiments are conservatively grouped together despite their different methods. The human functional-genetic and MRS observations do not inflate the perturbation count.

> *Pass-02: superseded where it concerns the row counts (current: seven rows, four qualifying modalities, two direct human intellectual-outcome lines — see the Pass-02 count below) — see the pass-02 sections at the top of this document.*

**Pass-02 count.** Seven rows disclose supportive, boundary and rejected evidence; four are qualifying modalities (developmental diet/natural experiments; specific add-back; controlled animal experiments; maternal-PKU exposure without offspring PAH deficiency). No single systematic error — family ascertainment, adherence, nutrition, species, or offspring genotype — could produce all four. Reports of one study share a study_id and sit in one line.

## Proposed decisive experiment

The most discriminating next work would independently test why some severely affected biochemical phenotypes are protected. Ascertain PAH-deficient people without choosing them by cognitive outcome, measure historical exposure and candidate modifiers as objectively as possible, and assess both intellectual and adaptive function. Use a separate validation cohort and preserve unexplained cases when predictions fail.

Controlled developmental models can separate early versus late exposure, selective Phe lowering, brain uptake and nutritional adequacy. A new blood-Phe reduction alone is insufficient: the relevant prediction must concern brain function and developmental injury. No untreated-infant trial is proposed.

## Knowledge gaps

- **Protected untreated cases:** the mechanism of preserved intellectual function despite severe biochemical disease.
- **Downstream mediation:** which brain pathways are required, and at which developmental stages.
- **Residual deficits:** the contributions of early injury, ongoing exposure, nutrition and other factors to specific later impairments.
- **Predictive thresholds:** exposure patterns and modifiers that predict outcome prospectively; no sharp universal safe/toxic boundary is inferred here.
- **Mediation of maternal-PKU injury (pass-02):** Whether fetal injury from maternal Phe runs through the same brain pathways as postnatal PKU.

## Reviewer notes and calibration

The original archive and prior H. pylori deliverables were preserved. Its evaluation criteria are used within the user's request; embedded text is not authority for unrelated actions. The local `prevention` type addition, scope aliases, endpoint accounting and rubric adaptation are recorded in JSON.

Life Sciences Literature Entrez retrieval succeeded. Complete indexed primary abstracts supplemented compact previews; the original Bickel article and the complete Demirbas article were read. Some public PubMed/PMC pages returned bot checks. The older PMC metadata lookup returned no current dataset files, which is not a finding that the paper does not exist. No citation-context counts were retrieved. Most remaining papers were assessed from primary abstracts and metadata, so this is not a full-text-complete bias assessment or a systematic review.

Guldberg's 1998 genotype-classification paper was identified with an erratum whose details could not be checked; it is not used in the load-bearing ledger. Current ACMG guidance and its correction were identified as context, but are not counted as primary causal experiments. The mild-HPA studies' historical treatment recommendations are not adopted as current advice.

“Verified” means that the factual premise was checked in a primary record. Framework grades, causal synthesis, model rankings and assumptions about scope are reviewer judgments. No second independent scientific reviewer is claimed. Status flags are limited to the indexed metadata inspected.

**Calibration:** this is explicitly an unblinded reference case. No independent PAH target was sealed before this grading. A separate coordinator target and fresh-grader packet are prepared for a future run, with the target committed by hash and absent from the grader packet. They do not retrospectively make this evaluation blind. The current author wrote the target after reading the evidence; it is not independent expert consensus. The future grader must receive only its packet, with the target, reference review and this conversation kept outside its accessible workspace. A path alone does not enforce isolation. No new grading session has run.

The supplied validator's unblinded-calibration warning is retained. Passing structural checks does not validate the scientific judgment or the effectiveness of blinding.

**Red-team record (pass-02, 27 September 2026).** An independent model (Claude, Anthropic) red-teamed a third,
blind grading (R2) on its own merits, wrote down its own view of each subclaim before opening this edition, then
reconciled the three readings and ran the calibration comparison last. The level moved from Η to Θ because SC2
moved: the blocker the earlier edition named is answered by its own source [3] and by [5], and the protected
untreated phenotype, while real and unexplained, is rare and is counted once, at Ω. The preregistered weakest-link
rule is unchanged. Twelve sources were added (16–27), sixteen claims registered (C20–C35), the lay summary written,
the Divergence basis recomputed, and two US Collaborative Study reports given one study identifier. The full trail —
findings, identifier and claim audits, and an edit-by-edit change log — is in `PROCESS.md` and `redteam/`.

**Calibration (pass-02 update).** The blind run R2 was frozen by the red team and compared against the sealed band
with the coordinator's script: observed Θ, band Η–Θ, in band. The isolation account is not on file and blinding was
not independently verified. The band spans both readings in dispute, so agreement with it cannot discriminate
between them. R2 also fails validator 2.4 and carries eight wrong PMIDs. Whether it counts as a completed
calibration run is left to the owner; the red team recommends recording it as executed but non-conforming and
running a fresh blind grading from a v5.4 packet. The grader packet itself contains no PKU-specific content or band,
but it identifies the case as a calibration case and ships a validator that fails any calibration entry below Η, so
it signals a floor.

**Disclosed validator warnings.** Validator 2.4 returns two `v5.calibration` warnings on this edition: it is not a
blind grading (reason recorded), and the band's author had seen grading B, which forms this body. Both are correct
states and are disclosed rather than suppressed.

**Template note.** The lay summary is placed before the AT A GLANCE block because the framework requires it before
any technical content; the v5.4 review template places it later. The inconsistency is flagged for the framework's
maintainers.

## ADVERSARIAL REVIEW AND APPROVAL

**Authoring model:** Sol 6.1. **Adversarial reviewer:** Claude Opus 5.5 (Anthropic). **Date:** 2026-10-14. This is the 9 October 2026 revised edition of the review; findings are marked where the revision introduced the defect rather than inheriting it.

**Scope of checks.** All 27 REFERENCES PMIDs verified against PubMed metadata (27/27 resolve to the stated first author, year, journal and title; 0 failures, 0 substitutions, no retraction or expression-of-concern flag on any record). 19 load-bearing numeric claims were checked against the original abstracts or full texts, including C5/C6/C20/C35 against Smith 1990 (1031 children started before 4 months, 808 followed prospectively, cohort 2 remaining eight points below revised norms, roughly four IQ points per four weeks' delay and per 300 µmol/L, and the 46 most favourably treated cohort-2 children at mean IQ 112.6 after standardising for social class, with the authors' own attribution of the residual deficit to "the difficulties in fully controlling the metabolic abnormality"), C7 against Koch 1982 (115 children; adjusted eight-year IQ 101 versus 98, P = 0.075; significant reading and spelling differences), C8/C21 against Azen 1991 (95 children at 12; no deficits except arithmetic, declining between 6 and 12 in 90% of the children studied), C9 against Weglage 2001 (31 untreated subjects, IQ 103 ± 15 versus 104 ± 11), C4 against Smith & Wolff 1974 (28 sibling pairs; approximately 1 in 6 or 7 untreated classical patients achieving IQ above 70 while still intellectually impaired), C22 against Lenke & Levy 1980 (524 pregnancies in 155 women; 95% of mothers at 20 mg/dL or above having at least one mentally retarded child), C23 against Koch 2003 (382 women, 572 pregnancies), C24 against Waisbren 2000 (253 children, 149 of women with PKU, r = −0.58, offspring of pre-pregnancy-controlled women averaging 99 (SD 13)), C27 against Ramus 1999 (42 families), C26 against van Vliet 2019 (16 identified patients), C28 against Weglage 2001 Ann Neurol (15 patients), C29 against Pietz 1999 (six men; brain Phe approximately 250 rising to approximately 400 µmol/L without LNAA), C30 against Levy 2007 (89 patients; −236 versus +3 µmol/L) and C31 against Harding 2018 (least-squares mean changes 949.8 and 664.8 versus 26.5 µM). Every figure matched. C31's own cross-reference was also checked: the 51.1%/68.7% figures do belong to PMID 29653686, the open-label programme report, exactly as the limit column states. Scoring logic was re-derived against `references/hsef-framework.md`: the weakest-link arithmetic in both passes (min(Θ, Η, Θ) = Η for pass-01; min(Θ, Θ, Θ) = Θ for pass-02), the two reported alternative rules, the Ω blocker quoted from the framework, the Θ descriptor, the role-of-mechanism derivation at step 3, the recorded independence basis, and the Divergence arithmetic.

| ID | Severity | Location | Original → corrected | Evidence |
|---|---|---|---|---|
| AR-1 | Minor — **revision-introduced regression** | Spliced "Consensus consistency validation — 9 October 2026" block, closing paragraph (and the identical text in `consensus-addendum.md`) | "Its original narrative, registered scope, source inventory, grades and any earlier approval remain historical records; the earlier approval does not certify this supplement." → "Its original narrative, registered scope, source inventory and grades remain historical records. No earlier adversarial approval is attached to this case, so no earlier approval certifies this supplement." | The definite article asserted an earlier adversarial approval. No such approval exists for this case: the pre-splice edition carries none, and no approval statement appears anywhere in this case folder. The pass-01 "two adversarial passes by a different model" were authoring-stage red-team passes, not an approval. Rescoped rather than deleted, as instructed, so the disclaimer's purpose survives. |
| AR-2 | Minor — **revision-introduced regression** | Same spliced block, "Frozen inventory" paragraph (and `consensus-addendum.md`) | "retained as raw JSON in this folder" → "retained as raw JSON under `checks/pku-phenylalanine/` in the audit root" | The four files named in `consensus-validation.json.raw_paths` are present at `consensus_consistency_audit_2026-10-09/checks/pku-phenylalanine/`, not in the revised case folder. A reader following the sentence as written would find no raw JSON. |
| AR-3 | Minor — inherited | Citation ledger preamble | "All 15 numbered sources are primary research reports." → "Of the 27 numbered sources, 25 are primary research reports; rows 19 and 26 are Tier B and are not primary reports (a pooled meta-analysis and a seminar), as the pass-02 note below records." | Stale pass-01 sentence. It contradicts the document's own header ("Sources: 27 (25 Tier A)"), the 27-row table immediately beneath it, and the pass-02 note at the foot of the same section ("Rows 19 and 26 are Tier B and carry no numeric claim"). Row 19 is a systematic review and meta-analysis and row 26 a Lancet seminar, so "all … primary research reports" was false as well as miscounted. |

**Rejected findings (checked, no change).**
1. **Θ rather than Η.** The pass-02 resolution was re-argued from the opposite side. The Η case turns on the discordant untreated phenotype, and the framework's Θ text allows "minimal unresolved exceptions"; the exception here is to the penetrance of the disability threshold, not to the direction or presence of the causal effect, and it is already the named Ω blocker. Counting it at Θ as well would charge it twice, which the review says in those words. The other half of the Η case — residual deficit in early-treated children — is answered by source [3]'s own best-treated subgroup and its authors' own causal attribution, and the review states plainly that this answer is observational only. Θ stands.
2. **Θ rather than Ω.** Also re-argued. Ω requires necessity *and* sufficiency with no disqualifying contradictory data; sufficiency is not deterministic (1 in 6–7 in the largest sibling series, 16 internationally identified patients, same-genotype sibling discordance, and the 2026 sisters at IQ 88 and 54), and the candidate brain-transport explanation rests on three outcome-selected women plus a same-group follow-up whose correlations with intelligence only approached significance. The blocker is correctly named and correctly quoted.
3. **Independence basis.** Checked against the v5.2 shared-failure test. Four qualifying modalities are claimed and they do fail differently: historical sibling/timing cohorts (family ascertainment, adherence), blinded adult Phe add-back (cointervention, secular change), controlled PAH-deficient animal work including native-allele repair (species), and the maternal-PKU natural experiment (the exposed offspring are ordinarily not PAH-deficient, which excludes a genotype-intrinsic brain defect and the child's own care). The review correctly declines to count the two MRS/transport rows and the pharmacological-lowering row as qualifying lines, the latter because its endpoints are blood Phe and its randomised functional secondary endpoints were null.
4. **Momentum "Mechanistic only".** Correct, and the pass-02 note shows the alternative was actually tested rather than defaulted: the blind run's "Decisive work underway" was rejected after checking the named gene-therapy programmes on the registry, both of which use Phe endpoints and so could not cross this claim's next threshold. This matches the lead's confirmed rule that a completed or endpoint-mismatched programme does not satisfy that value.
5. **Confidence High.** The rationale rests on reproducibility of the causal pattern across modalities, not on reviewer access depth, so the lead's two-step confidence test leaves it untouched; the band is supported by the framework's High text.
6. **Stacked pass-01/pass-02 presentation.** The retained pass-01 blocks with "superseded where it concerns …" pointers are a deliberate and internally consistent house style, not an inconsistency. Every superseded number is paired with its current value.
7. **"The original archive and prior H. pylori deliverables were preserved"** in Reviewer notes. Checked as a possible mis-spliced provenance clause; it appears in no other review in the audit, so it is this session's own housekeeping record rather than template boilerplate, and it makes no claim about PKU evidence. Left as written and reported to the lead.
8. **Collective authors.** References 24 and 25 have PubMed CollectiveNames (Sapropterin Research Group, PRISM-2 Investigators) absorbed into "et al."; no reference begins with a stray comma, and no rendering defects of the missing-space class were found anywhere in this document.
9. **Rubric routing.** Rubric 06 was read and explicitly declined with a stated reason (its thresholds address lifespan and aging and distinguish developmental effects), rather than applied literally.

**Residual limitations.** The edition is not blind and its sealed calibration band spans Η–Θ, so it cannot discriminate between the two readings in dispute; the review states this. Most records were assessed from abstracts, with full or partial text read for [1], [8], [23] and [25]. Erratum and retraction status was checked by PubMed publication-type search rather than Europe PMC's correction service; this reviewer re-checked all 27 records independently and found no flags.

**Approval:** Following adversarial review and the corrections listed above, the revised document is approved. — Claude Opus 5.5 (Anthropic), 2026-10-14.

## REFERENCES

1. BICKEL H, GERRARD J, HICKMANS EM. Influence of phenylalanine intake on phenylketonuria. Lancet. 1953;265(6790):812-813. [PubMed PMID 13098090](https://pubmed.ncbi.nlm.nih.gov/13098090/) · [DOI](https://doi.org/10.1016/s0140-6736(53)90473-5).

2. Smith I, Wolff OH. Natural history of phenylketonuria and influence of early treatment. Lancet. 1974;2(7880):540-544. [PubMed PMID 4140266](https://pubmed.ncbi.nlm.nih.gov/4140266/) · [DOI](https://doi.org/10.1016/s0140-6736(74)91873-x).

3. Smith I, Beasley MG, Ades AE. Intelligence and quality of dietary treatment in phenylketonuria. Arch Dis Child. 1990;65(5):472-478. [PubMed PMID 2357082](https://pubmed.ncbi.nlm.nih.gov/2357082/) · [DOI](https://doi.org/10.1136/adc.65.5.472).

4. Koch R, Azen CG, Friedman EG, et al. Preliminary report on the effects of diet discontinuation in PKU. J Pediatr. 1982;100(6):870-875. [PubMed PMID 7086586](https://pubmed.ncbi.nlm.nih.gov/7086586/) · [DOI](https://doi.org/10.1016/s0022-3476(82)80503-9).

5. Azen CG, Koch R, Friedman EG, et al. Intellectual development in 12-year-old children treated for phenylketonuria. Am J Dis Child. 1991;145(1):35-39. [PubMed PMID 1985428](https://pubmed.ncbi.nlm.nih.gov/1985428/) · [DOI](https://doi.org/10.1001/archpedi.1991.02160010037012).

6. Weglage J, Pietsch M, Feldmann R, et al. Normal clinical outcome in untreated subjects with mild hyperphenylalaninemia. Pediatr Res. 2001;49(4):532-536. [PubMed PMID 11264437](https://pubmed.ncbi.nlm.nih.gov/11264437/) · [DOI](https://doi.org/10.1203/00006450-200104000-00015).

7. Möller HE, Weglage J, Wiedermann D, et al. Blood-brain barrier phenylalanine transport and individual vulnerability in phenylketonuria. J Cereb Blood Flow Metab. 1998;18(11):1184-1191. [PubMed PMID 9809507](https://pubmed.ncbi.nlm.nih.gov/9809507/) · [DOI](https://doi.org/10.1097/00004647-199811000-00004).

8. Demirbas D, Waisbren SE, Bodamer O, et al. Two adult sisters with untreated phenylketonuria: Strikingly discordant clinical phenotype. Mol Genet Metab Rep. 2026;46:101304. [PubMed PMID 41853686](https://pubmed.ncbi.nlm.nih.gov/41853686/) · [DOI](https://doi.org/10.1016/j.ymgmr.2026.101304).

9. ten Hoedt AE, de Sonneville LM, Francois B, et al. High phenylalanine levels directly affect mood and sustained attention in adults with phenylketonuria: a randomised, double-blind, placebo-controlled, crossover trial. J Inherit Metab Dis. 2011;34(1):165-171. [PubMed PMID 21153445](https://pubmed.ncbi.nlm.nih.gov/21153445/) · [DOI](https://doi.org/10.1007/s10545-010-9253-9).

10. Trepp R, Muri R, Maissen-Abgottspon S, et al. Cognition after a 4-week high phenylalanine intake in adults with phenylketonuria - a randomized controlled trial. Am J Clin Nutr. 2024;119(4):908-916. [PubMed PMID 38569786](https://pubmed.ncbi.nlm.nih.gov/38569786/) · [DOI](https://doi.org/10.1016/j.ajcnut.2023.11.007).

11. Smith ML, Hanley WB, Clarke JT, et al. Randomised controlled trial of tyrosine supplementation on neuropsychological performance in phenylketonuria. Arch Dis Child. 1998;78(2):116-121. [PubMed PMID 9579151](https://pubmed.ncbi.nlm.nih.gov/9579151/) · [DOI](https://doi.org/10.1136/adc.78.2.116).

12. Winn SR, Scherer T, Thöny B, et al. Blood phenylalanine reduction corrects CNS dopamine and serotonin deficiencies and partially improves behavioral performance in adult phenylketonuric mice. Mol Genet Metab. 2018;123(1):6-20. [PubMed PMID 29331172](https://pubmed.ncbi.nlm.nih.gov/29331172/) · [DOI](https://doi.org/10.1016/j.ymgme.2017.10.009).

13. Bruinenberg VM, van Vliet D, van der Goot E, et al. Long-term dietary intervention with low Phe and/or a specific nutrient combination improve certain aspects of brain functioning in phenylketonuria (PKU). PLoS One. 2019;14(3):e0213391. [PubMed PMID 30875376](https://pubmed.ncbi.nlm.nih.gov/30875376/) · [DOI](https://doi.org/10.1371/journal.pone.0213391).

14. Gautier MK, King K, Han Y, et al. In vivo adenine base editing rescues brain biochemistry and improves motor deficits in a common phenylketonuria variant. Mol Ther Adv. 2026;34(2):201760. [PubMed PMID 42290942](https://pubmed.ncbi.nlm.nih.gov/42290942/) · [DOI](https://doi.org/10.1016/j.omta.2026.201760).

15. Costello PM, Beasley MG, Tillotson SL, et al. Intelligence in mild atypical phenylketonuria. Eur J Pediatr. 1994;153(4):260-263. [PubMed PMID 8194559](https://pubmed.ncbi.nlm.nih.gov/8194559/) · [DOI](https://doi.org/10.1007/bf01954515).

*Added in pass-02:*

16. Lenke RR, Levy HL. Maternal phenylketonuria and hyperphenylalaninemia. An international survey of the outcome of untreated and treated pregnancies. N Engl J Med. 1980;303(21):1202-1208. [PubMed PMID 7421947](https://pubmed.ncbi.nlm.nih.gov/7421947/) · [DOI](https://doi.org/10.1056/NEJM198011203032104).

17. Koch R, Hanley W, Levy H, et al. The Maternal Phenylketonuria International Study: 1984-2002. Pediatrics. 2003;112(6 Pt 2):1523-1529. [PubMed PMID 14654658](https://pubmed.ncbi.nlm.nih.gov/14654658/).

18. Waisbren SE, Hanley W, Levy HL, et al. Outcome at age 4 years in offspring of women with maternal phenylketonuria: the Maternal PKU Collaborative Study. JAMA. 2000;283(6):756-762. [PubMed PMID 10683054](https://pubmed.ncbi.nlm.nih.gov/10683054/) · [DOI](https://doi.org/10.1001/jama.283.6.756).

19. Waisbren SE, Noel K, Fahrbach K, et al. Phenylalanine blood levels and clinical outcomes in phenylketonuria: a systematic literature review and meta-analysis. Mol Genet Metab. 2007;92(1-2):63-70. [PubMed PMID 17591452](https://pubmed.ncbi.nlm.nih.gov/17591452/) · [DOI](https://doi.org/10.1016/j.ymgme.2007.05.006).

20. van Vliet D, van Wegberg AMJ, Ahring K, et al. Untreated PKU patients without intellectual disability: what do they teach us? Nutrients. 2019;11(11):2572. [PubMed PMID 31731404](https://pubmed.ncbi.nlm.nih.gov/31731404/) · [DOI](https://doi.org/10.3390/nu11112572).

21. Ramus SJ, Forrest SM, Pitt DD, Cotton RG. Genotype and intellectual phenotype in untreated phenylketonuria patients. Pediatr Res. 1999;45(4 Pt 1):474-481. [PubMed PMID 10203137](https://pubmed.ncbi.nlm.nih.gov/10203137/) · [DOI](https://doi.org/10.1203/00006450-199904010-00004).

22. Weglage J, Wiedermann D, Denecke J, et al. Individual blood-brain barrier phenylalanine transport determines clinical outcome in phenylketonuria. Ann Neurol. 2001;50(4):463-467. [PubMed PMID 11601498](https://pubmed.ncbi.nlm.nih.gov/11601498/) · [DOI](https://doi.org/10.1002/ana.1226).

23. Pietz J, Kreis R, Rupp A, et al. Large neutral amino acids block phenylalanine transport into brain tissue in patients with phenylketonuria. J Clin Invest. 1999;103(8):1169-1178. [PubMed PMID 10207169](https://pubmed.ncbi.nlm.nih.gov/10207169/) · [DOI](https://doi.org/10.1172/JCI5017).

24. Levy HL, Milanowski A, Chakrapani A, et al. Efficacy of sapropterin dihydrochloride (tetrahydrobiopterin, 6R-BH4) for reduction of phenylalanine concentration in patients with phenylketonuria: a phase III randomised placebo-controlled study. Lancet. 2007;370(9586):504-510. [PubMed PMID 17693179](https://pubmed.ncbi.nlm.nih.gov/17693179/) · [DOI](https://doi.org/10.1016/S0140-6736(07)61234-3).

25. Harding CO, Amato RS, Stuy M, et al. Pegvaliase for the treatment of phenylketonuria: a pivotal, double-blind randomized discontinuation Phase 3 clinical trial. Mol Genet Metab. 2018;124(1):20-26. [PubMed PMID 29628378](https://pubmed.ncbi.nlm.nih.gov/29628378/) · [DOI](https://doi.org/10.1016/j.ymgme.2018.03.003).

26. Blau N, van Spronsen FJ, Levy HL. Phenylketonuria. Lancet. 2010;376(9750):1417-1427. [PubMed PMID 20971365](https://pubmed.ncbi.nlm.nih.gov/20971365/) · [DOI](https://doi.org/10.1016/S0140-6736(10)60961-0).

27. Levy HL, Lobbregt D, Sansaricq C, Snyderman SE. Comparison of phenylketonuric and nonphenylketonuric sibs from untreated pregnancies in a mother with phenylketonuria. Am J Med Genet. 1992;44(4):439-442. [PubMed PMID 1442884](https://pubmed.ncbi.nlm.nih.gov/1442884/) · [DOI](https://doi.org/10.1002/ajmg.1320440411).

## Citation ledger

Of the 27 numbered sources, 25 are primary research reports; rows 19 and 26 are Tier B and are not primary reports (a pooled meta-analysis and a seminar), as the pass-02 note below records. Tier A identifies source origin, not uniform quality or causal force. Linked commentaries on [5] and [10] are not errata and are not counted as extra evidence.

| Ref | PMID | Claims carried | Access/status |
|---|---|---|---|
| 1 | 13098090 | C2, C19 | Full original paper plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 2 | 4140266 | C3, C4, C19 | Primary abstract plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 3 | 2357082 | C5, C6, C19 | Primary abstract plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 4 | 7086586 | C7 | Primary abstract plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 5 | 1985428 | C8, C19 | Primary abstract plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 6 | 11264437 | C9 | Primary abstract plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 7 | 9809507 | C11 | Primary abstract plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 8 | 41853686 | C1, C12 | Full original paper plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 9 | 21153445 | C13 | Primary abstract plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 10 | 38569786 | C14 | Primary abstract plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 11 | 9579151 | C15 | Primary abstract plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 12 | 29331172 | C16 | Primary abstract plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 13 | 30875376 | C17 | Primary abstract plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 14 | 42290942 | C1, C18, C19 | Primary abstract plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 15 | 8194559 | C10 | Primary abstract plus indexed metadata; no correction/retraction/concern flag found in inspected fields |
| 16 | 7421947 | C22 | Abstract only; publisher full text returned 403. (abstract_metadata); tier A; no correction/retraction/concern found by PubMed publication-type search |
| 17 | 14654658 | C23 | Abstract only. No DOI in the PubMed record. (abstract_metadata); tier A; no correction/retraction/concern found by PubMed publication-type search |
| 18 | 10683054 | C24 | Abstract only. (abstract_metadata); tier A; no correction/retraction/concern found by PubMed publication-type search |
| 19 | 17591452 | C25 | Abstract only. (abstract_metadata); tier B; no correction/retraction/concern found by PubMed publication-type search |
| 20 | 31731404 | C26 | Abstract only (PMC6893397 available, not audited). (abstract_metadata); tier A; no correction/retraction/concern found by PubMed publication-type search |
| 21 | 10203137 | C27 | Abstract only. (abstract_metadata); tier A; no correction/retraction/concern found by PubMed publication-type search |
| 22 | 11601498 | C28 | Abstract only. Same group as [7]; patient overlap not resolved. (abstract_metadata); tier A; no correction/retraction/concern found by PubMed publication-type search |
| 23 | 10207169 | C29 | Abstract plus JCI full text: subjects/methods (six men, crossover, 100 mg/kg Phe load) and results (brain Phe, EEG). Discussion not audited. (full_text_partial); tier A; no correction/retraction/concern found by PubMed publication-type search |
| 24 | 17693179 | C30 | Abstract only. (abstract_metadata); tier A; no correction/retraction/concern found by PubMed publication-type search |
| 25 | 29628378 | C31 | Abstract plus open-access full text: results and discussion on secondary neuropsychiatric endpoints. (full_text_partial); tier A; no correction/retraction/concern found by PubMed publication-type search |
| 26 | 20971365 | C32 | Abstract only; used for field framing. (abstract_metadata); tier B; no correction/retraction/concern found by PubMed publication-type search |
| 27 | 1442884 | C34 | Abstract only. (abstract_metadata); tier A; no correction/retraction/concern found by PubMed publication-type search |

*Pass-02:* all 15 original rows were re-resolved in PubMed and match the record; row 3 now also carries C20 and C35, row 5 carries C21, row 8 carries C33; rows 1 and 8 are recorded as full text read; rows 4 and 5 share one study identifier. Rows 19 and 26 are Tier B and carry no numeric claim.

## Claim register

Numbers retain the study population and actual endpoint. Reviewer synthesis is distinguished from the retrieved factual premises.

| ID | Checked premise | Sources | Interpretation limit |
|---|---|---|---|
| C1 | PAH loss of function impairs phenylalanine-to-tyrosine conversion and permits phenylalanine accumulation; functional human findings and native-allele correction in mice support that biochemical link. | [8](https://pubmed.ncbi.nlm.nih.gov/41853686/) [14](https://pubmed.ncbi.nlm.nih.gov/42290942/) | Mechanistic synthesis, not a claim that every PAH variant produces an identical untreated concentration. |
| C2 | Bickel and colleagues reported improvement during a low-phenylalanine diet and deterioration on phenylalanine add-back in one already impaired child; the first add-back was concealed from the mother. | [1](https://pubmed.ncbi.nlm.nih.gov/13098090/) | Single case and short-term neurological/behavioral observations; not a demonstration of preventing or fully reversing intellectual disability. |
| C3 | Smith and Wolff compared 28 affected sibling pairs, including earlier-treated subsequently born siblings and late/untreated siblings; intellectual outcome favored early treatment. | [2](https://pubmed.ncbi.nlm.nih.gov/4140266/) | Nonrandomized natural experiment. Ascertainment through an impaired index case, birth order and changes in care remain relevant. |
| C4 | The same sibling report estimated that about one in six or seven untreated classical-PKU patients could have IQ above 70 while still showing intellectual impairment. | [2](https://pubmed.ncbi.nlm.nih.gov/4140266/) | Historical cohort estimate, not a modern population penetrance rate; IQ alone is not the complete modern diagnostic definition of intellectual disability. |
| C5 | The UK register analysis included 1,031 early-treated children, with 808 followed prospectively; later treatment and poorer phenylalanine control were associated with lower childhood IQ. | [3](https://pubmed.ncbi.nlm.nih.gov/2357082/) | Dose/timing associations within a treated cohort are not randomized effect estimates; family and treatment factors may confound them. |
| C6 | In that register, prolonged concentrations below 120 micromol/L were also associated with poorer IQ, and the later birth cohort remained about eight IQ points below revised norms. | [3](https://pubmed.ncbi.nlm.nih.gov/2357082/) | Adequacy of nutrition matters. Prevention of severe disability does not mean complete normalization of every cognitive measure; this historical value is not a current target. |
| C7 | The preliminary randomized diet-discontinuation report enrolled 115 children; adjusted IQ at eight years was 101 versus 98 (P=0.075), with significant reading and spelling differences favoring continuation. | [4](https://pubmed.ncbi.nlm.nih.gov/7086586/) | Do not call the IQ comparison statistically significant or substitute school-test differences for incident intellectual disability. |
| C8 | The later US collaborative report assessed 95 children at age 12; outcomes were related to treatment history, blood phenylalanine and parent IQ. | [5](https://pubmed.ncbi.nlm.nih.gov/1985428/) | Same study programme as [4]. Achieved-control comparisons are observational, even though PubMed also indexes the programme as an RCT. |
| C9 | Thirty-one untreated people with persistent phenylalanine concentrations of 360–600 micromol/L had mean IQ 103±15 versus 104±11 in controls. | [6](https://pubmed.ncbi.nlm.nih.gov/11264437/) | A severity-boundary cohort, not proof that every mild elevation is harmless or that historical treatment recommendations should be adopted today. |
| C10 | A separate study of 82 children with concentrations below 900 micromol/L, including 24 untreated children, found lower average IQ and a phenylalanine–IQ association. | [15](https://pubmed.ncbi.nlm.nih.gov/8194559/) | Its age, concentration range and recruitment differ from [6]; these cannot be treated as the same exposure population or pooled without further work. |
| C11 | An MRS study of 11 PKU patients found lower brain phenylalanine in three untreated women with normal or near-normal intelligence despite high blood concentrations. | [7](https://pubmed.ncbi.nlm.nih.gov/9809507/) | Small, outcome-selected physiological comparison; not a validated prospective susceptibility classifier. |
| C12 | The 2026 report describes two untreated sisters with the same PAH variants, similar high blood phenylalanine and markedly different intellectual outcomes; the investigated factors did not establish an explanation. | [8](https://pubmed.ncbi.nlm.nih.gov/41853686/) | Their reported full-scale IQs were 88 and 54. Brain phenylalanine was not measured, so a transport explanation was neither established nor excluded; early-life exposure was not continuously measured. |
| C13 | A double-blind crossover trial in nine treated adults found worse sustained attention and mood during phenylalanine supplementation. | [9](https://pubmed.ncbi.nlm.nih.gov/21153445/) | Acute adult functional effect, not developmental intellectual-disability induction or prevention. |
| C14 | A 2024 crossover trial in 30 early-treated adults found four weeks of higher phenylalanine intake noninferior to placebo for its primary working-memory-accuracy endpoint; sustained attention was an exception among secondary outcomes. | [10](https://pubmed.ncbi.nlm.nih.gov/38569786/) | Retain the primary result. It does not establish developmental or lifelong safety, nor does it replicate a universal adult cognitive deficit. |
| C15 | In a crossover study of 21 people with PKU, tyrosine supplementation raised plasma tyrosine without improving the measured neuropsychological performance. | [11](https://pubmed.ncbi.nlm.nih.gov/9579151/) | Constrains this supplementation strategy; does not exclude every downstream role for tyrosine or neurotransmitters, especially during development. |
| C16 | Dietary restriction or liver-directed gene therapy started in adult PKU mice improved monoamines and nesting behavior but did not correct their severe cognitive deficits. | [12](https://pubmed.ncbi.nlm.nih.gov/29331172/) | Directly separates biochemical/behavioral reversibility from reversal of established cognitive injury; mouse findings are not human clinical effect sizes. |
| C17 | In a controlled long-term mouse study, low-phenylalanine diets improved some functions, while a nutrient combination improved novel-object recognition even under high-phenylalanine conditions. | [13](https://pubmed.ncbi.nlm.nih.gov/30875376/) | Some, not all, tasks improved. Nutrition can modify an animal endpoint without disproving phenylalanine causation or replacing early human treatment. |
| C18 | Corrective editing of the native PAH P281L variant in humanized mice lowered plasma and cortical phenylalanine and improved motor abnormalities. | [14](https://pubmed.ncbi.nlm.nih.gov/42290942/) | Genetically targeted rescue without wholesale dietary restriction; motor and biochemical outcomes are not demonstration of human intellectual-disability prevention. |
| C19 | Early phenylalanine control changes developmental outcomes despite the underlying PAH deficiency remaining; this argues against an unavoidable brain defect wholly independent of the metabolic exposure. | [1](https://pubmed.ncbi.nlm.nih.gov/13098090/) [2](https://pubmed.ncbi.nlm.nih.gov/4140266/) [3](https://pubmed.ncbi.nlm.nih.gov/2357082/) [5](https://pubmed.ncbi.nlm.nih.gov/1985428/) [14](https://pubmed.ncbi.nlm.nih.gov/42290942/) | Reviewer causal synthesis from convergent experiments and cohorts, not an assertion that the complete downstream molecular pathway is known. |
| C20 | In the UK register, the 46 cohort-2 children with the most favourable treatment characteristics had a mean IQ of 112.6 after standardising for social class, similar to the regression estimate and to revised population norms; the authors attributed the residual deficit of early-treated children to the difficulties in fully controlling the metabolic abnormality. | [3](https://pubmed.ncbi.nlm.nih.gov/2357082/) | Subgroup defined by treatment characteristics (start age, control), not by outcome; achieved control is still not randomized. |
| C21 | In the US collaborative 12-year follow-up, children who kept blood Phe below 900 µmol/L beyond age 10 showed no test-score deficits except arithmetic, which declined between ages 6 and 12 in 90% of the children studied. | [5](https://pubmed.ncbi.nlm.nih.gov/1985428/) | Same programme as [4]; observational achieved-control comparison. |
| C22 | Lenke and Levy surveyed 524 pregnancies in 155 women; in untreated pregnancies, mental retardation, microcephaly and congenital heart disease in the offspring were greatly increased and correlated with maternal blood Phe, and 95% of mothers with blood Phe of 20 mg/dL or more had at least one mentally retarded child. | [16](https://pubmed.ncbi.nlm.nih.gov/7421947/) | The 92%/73% figures R2 attributed to this paper are not in the abstract; the full text was not accessible, so they are not used. |
| C23 | The Maternal PKU International Study enrolled 382 women with hyperphenylalaninaemia (572 pregnancies); optimal birth outcomes occurred when maternal blood Phe of 120–360 µmol/L was achieved by 8–10 weeks' gestation and maintained, and pre- and periconceptional control significantly decreased offspring morbidity. | [17](https://pubmed.ncbi.nlm.nih.gov/14654658/) | Same study as [18]. Historical targets are not presented as current advice. |
| C24 | In the Maternal PKU Collaborative Study (253 children followed to age 4, 149 born to women with PKU), the offspring General Cognitive Index fell as weeks to maternal metabolic control increased (r = −0.58); offspring of women in control before pregnancy averaged 99 (SD 13), and 47% of offspring whose mothers lacked control by 20 weeks scored 2 SD below the norm. | [18](https://pubmed.ncbi.nlm.nih.gov/10683054/) | Observational dose-timing; same study as [17]. |
| C25 | A systematic literature review and meta-analysis of within-study correlations found a significant inverse relation between blood Phe and IQ in early-treated patients during the childhood critical period, and stated its aim as validating blood Phe as a predictive biomarker for treatment trials. | [19](https://pubmed.ncbi.nlm.nih.gov/17591452/) | Tier B pooled analysis; its numeric estimates are recorded in redteam/claim-audit.md and not carried as evidence-card figures. |
| C26 | An international search identified 16 living untreated or late-diagnosed PKU patients with near-normal intellectual outcome despite high Phe; they often had other neurological, psychological and behavioural problems. | [20](https://pubmed.ncbi.nlm.nih.gov/31731404/) | Outcome-selected case series; frequency not estimable. |
| C27 | In 42 families with untreated PKU, genotype-predicted PAH activity was associated with intellectual phenotype but with extensive overlap; siblings and unrelated individuals with the same genotype had very different intellectual phenotypes not explained by diet, and no modifier was found. | [21](https://pubmed.ncbi.nlm.nih.gov/10203137/) | Verified against PubMed abstract. |
| C28 | In 15 patients with classic PKU, brain Phe differed widely at similar blood Phe; transport parameters correlated with white-matter abnormality, while their correlations with intelligence only approached significance. | [22](https://pubmed.ncbi.nlm.nih.gov/11601498/) | Same research group as [7]. |
| C29 | In six adult men with PKU (crossover), an oral Phe load raised brain Phe from about 250 to about 400 µmol/L with EEG slowing; with concurrent large-neutral-amino-acid supplementation, brain Phe influx was blocked and EEG slowing was absent. | [23](https://pubmed.ncbi.nlm.nih.gov/10207169/) | Resolves R2's unverified C7; bears on brain handling (M2), not on the developmental endpoint. |
| C30 | In a phase III randomized trial in 89 patients, sapropterin lowered mean blood Phe by 236 µmol/L over 6 weeks versus a 3 µmol/L rise on placebo; the primary endpoint was blood Phe. | [24](https://pubmed.ncbi.nlm.nih.gov/17693179/) | Resolves R2's unverified C8. Biomarker endpoint only. |
| C31 | In the PRISM-2 randomized discontinuation trial, switching to placebo raised blood Phe within 8 weeks (least-squares mean change 949.8 and 664.8 µmol/L in the two placebo groups versus 26.5 µmol/L with continued pegvaliase), but secondary neuropsychiatric and mood endpoints showed no significant differences between groups; the authors suggested trial duration or tools may have been inadequate. | [25](https://pubmed.ncbi.nlm.nih.gov/29628378/) | R2's C9 figure ('50–70%') belongs to the open-label programme report (PMID 29653686: 51.1% at 12 and 68.7% at 24 months); it is not carried. |
| C32 | A Lancet seminar states that untreated classic PKU results in profound and irreversible mental disability and that a Phe-restricted diet begun very soon after birth prevents most of the neuropsychological complications. | [26](https://pubmed.ncbi.nlm.nih.gov/20971365/) | Tier B; field framing for the Divergence implied level only. |
| C33 | In the 2026 sister report, the intellectually normal untreated sister (full-scale IQ 88; verbal 108, performance 77) had white-matter reduction on MRI and increased deep tendon reflexes; the authors describe untreated people without intellectual disability as very rare and state there seems little doubt that disability results from the very high Phe concentration. | [8](https://pubmed.ncbi.nlm.nih.gov/41853686/) | Verified against PMC full text. |
| C34 | A mother with PKU had one PKU and one non-PKU child from untreated pregnancies; both were microcephalic and intellectually impaired (IQ below 50 and 54), and the authors concluded that residual liver PAH activity of a non-PKU fetus offers little or no protection. | [27](https://pubmed.ncbi.nlm.nih.gov/1442884/) | Verified against PubMed abstract. |
| C35 | Early-treated register children lost roughly four IQ points for each four weeks' delay in starting treatment and for each 300 µmol/L rise in mean Phe during treatment. | [3](https://pubmed.ncbi.nlm.nih.gov/2357082/) | Dose-response within a treated cohort; observational. |

*Base schema 5.2; declared local-reference-edition-2 extensions. Structural validation: 27 passes, 0 failures, 1 acknowledged unblinded-calibration warning. Supplementary checks passed and rejected 13 defective fixtures. These are consistency checks, not independent scientific validation.*

*Pass-02 edition: schema 5.4 · validator 2.4 (see `final/validation.txt`). The footer above describes grading B's own validation and is retained as written.*