# Residuals — what could not be verified, and the confidence that attaches

Card t_592b3c4a, verification wave, 2026-10-16. Ten of the 70 checked items could not be re-verified at
primary source this run. Each is listed with the reason and the confidence that should attach to the
claim as it stands. Nothing here is asserted; where a claim rests only on a residual, say so.

## 1. SSNAP audit point values carried from the vault (highest consequence residual)

- **Claims**: SLT assessment within 72 h 46.8%; within 24 h 24.8%; no assessment for organisational
  reasons 7.6%; patients receiving the NICE 128 min/day total therapy dose 35.4% (2024/25).
- **Why unverified**: the SSNAP annual results portfolio (XLSX) and the APT synthesis data file were not
  downloaded and parsed. The values are carried from the vault, whose own definitions were confirmed
  against the SSNAP SoN 2025 [1] and the 2023 guideline [19], but the point values were not re-derived.
- **Confidence**: medium. The definitions and denominators are corroborated; the exact percentages are
  not. L2 already marks these [UNVERIFIED at primary source]. Anyone using the 46.8 / 24.8 / 7.6 / 35.4
  numbers in a published artefact should download the SSNAP results portfolio first.

## 2. Paywalled / abstract-only studies (cannot be checked by anyone without a subscription)

- Monnelly 2024 ICAP systematic review — 21 studies, 13 ICAPs, N=485 (L1 source [9]).
- Monnelly 2026 provisional theory of change for ICAPs (L1 source [10]).
- Raymer & Roitsch 2023 CILT appraisal (L1 source [24]).
- Pierce et al. 2024 CIAT-Plus / M-MAT low-moderate intensity, and the COMPARE dose (L1 source [25]).
- Schwen Blackett et al. 2025 semantic-vs-phonological (L1 source [26]).
- **Confidence**: low-to-medium. Each was used at abstract level only, and the lanes say so. The COMPARE
  15 h/week x 2 wk vs 6 h/week x 5 wk arithmetic is internally consistent (both = 30 h) but the dose
  figures themselves are not verified.

## 3. Documents whose text layer could not be read

- **StepByStep computer-therapy manual** (L1 source [33], a PDF): page 19 needs OCR. The "about 30 hours
  over 2 weeks" CILT dose and the CIAT summarisation were read from search-retrieved text, not from the
  manual itself. Confidence: low for the exact dose; the 30 h figure is consistent with the published
  ICAP definition (3 h/day x 5 days x 2 weeks = 30 h).
- **RCSLT aphasia-plus-intro PDF** (L1 source [31]): 430 KB PDF not parsed; used as metadata-level only.
- **Big CACTUS TIDieR "20-30 min/day over six months"** dose: not present in the retrieved abstract and
  full-text extract. Confidence: medium (the trial is real and UK; the exact daily dose is plausible but
  unconfirmed here).

## 4. Figures fetched at metadata level only

- Stroke Association "434 support groups / around 13,000 people / interpreting support for 480 people"
  — these sit in the Charity Commission filing description, not on a fetched page. L2 flags them.
  Confidence: low-to-medium. Do not present as independent.
- Northcott 2016 UK SLT survey — n=124; 99% social participation; 50% psychological difficulty. Fetched
  at abstract level in the parent lane; not re-fetched this run. Confidence: medium.
- Hilari et al. UK telehealth survey — >80% used telehealth in COVID; >90% planned to continue. Not
  re-fetched this run. Confidence: medium.
- Aphasia Alliance — 13 member organisations. Not re-fetched this run. Confidence: medium.
- Care-home resident provision — the audit section 4.5 exists but the figures were not extracted by the
  lane and were not extracted here. Confidence: n/a (no value claimed).

## 5. Registry detail not present in the fields retrieved

- **NCT07386652 "4-week, 100-hour ICAP"**: the ClinicalTrials.gov fields retrieved (arm descriptions,
  primary outcomes, enrolment) do not state a 4-week / 100-hour dose. Enrolment (334), status
  (NOT_YET_RECRUITING) and completion (2030-03-31) are confirmed. Confidence: medium for the 4-week /
  100-hour dose.

## 6. Scope limits on otherwise-confirmed claims

- **D3 (no mapped strategy has UK effectiveness evidence)** is CONFIRMED only as an internal count over
  the lane's own sources (validated_uk = 0). It does not assert that no UK aphasia implementation
  strategy has ever been evaluated anywhere. Read it as "none of the strategies this map selected carries
  effectiveness evidence in the sources these lanes used".
- **B25 (six-month review discrepancy)** is a real source anomaly: the Stroke Association 10-Year-Plan
  response states 39% of survivors received a six-month review in 2024-25, while SSNAP [1] reports 35.1%
  for 2024/25 (and 39% for 2023/24). The two cannot both be true for the same year. The lane flags the
  contradiction and does not average it; verification agrees with that treatment. Confidence in the SSNAP
  figure: high (national audit). Confidence in the charity figure: low for this measure.

## What the lanes got right

The three parent lanes are substantially right on their load-bearing numbers. Of 70 items checked, 55
were confirmed verbatim at source, including every figure that carries the central argument: the
provision gap (median 6.3 h/3 months [3]; mean 128 min/week [18]), the guideline standard (3 h/day x 5
days [19]), the SSNAP pathway and review figures [1], the RCSLT workforce figures [2], the waiting-time
figures [9], the charity-reported reach and the CHSS unmet-need survey [10]. The single KILLED item
(Palmer n=114) had already been caught and flagged by L1 itself as a vault error. The corrections are
narrow: one unit (V2), one inferred TIDieR cell (V3) and one over-strong labelling claim about the CFIR
vocabulary (V4). None of the corrections changes a conclusion.