UK aphasia provision: evidence base and learning site

UK aphasia treatment and service provision · compiled 16 October 2026

Current UK aphasia treatment and provision: what is delivered, what blocks it, and how we know

A review of what is actually provided to people with aphasia in the UK, and the barriers to providing it. Ask the assistant below, or read the findings.

Ask the assistant anything in this evidence base

How much therapy people actually receive. Which treatments have no description. How many mapped strategies have UK evidence. It answers only from this work, and refuses anything outside it.

Open the assistant

The one page here that uses a server. No account, nothing stored, no medical advice.

12Treatments extracted

64Barriers mapped

0 / 64Strategies with UK evidence

76Distinct works cited

What is on this site

The findings, with their sources and their verification status. The evidence base itself, as it was reviewed. The method behind it, including why a barrier map is not an evidence base. Every asset the work produced, downloadable. And a process record at the end, for anyone who wants it.

01 What this is, and what it is not

This is an internal working evidence base, not a published report and not clinical guidance. It describes what the sources say. It gives no medical advice, and it is not a route to advice about an individual's treatment.

Two disciplines were imposed throughout, and they are why the numbers carry labels rather than standing alone:

Every claim carries a source, and the access level is recorded: full text, abstract only, or metadata only. A claim shown to exist is not a claim shown to say what is attributed to it.
Every mapped strategy carries an evidence standing. Without that column, a barrier-to-strategy table reads as a plan; with it, the reader can see what is evidenced and what is reasoning.

The one-sentence version

The effective treatments largely exist and are described in the literature. What is missing in the UK is delivery at anything like an effective dose, and a pathway that holds after discharge. Separately: not one of the 64 implementation strategies mapped to the barriers has UK effectiveness evidence, so the fix list is a hypothesis, not a plan.

02 The brief

Establish, for the UK: current treatment (what is delivered, at what dose, in what format, by whom, on what evidence); current service provision and contexts (the pathway from acute care to community, who provides what, and where it breaks); the key barriers, expressed as implementation determinants using CFIR for the determinant side and ERIC for the strategies; and TIDieR extraction where relevant, meaning a structured description of named interventions rather than a summary.

Scope decisions: adults with aphasia after stroke; UK services, with non-UK evidence labelled and never presented as UK provision; every figure carrying its date, and anything older than three years flagged. Excluded: clinical recommendations, cost-effectiveness modelling, primary data collection.

03 What we found

Routine provision sits far below any effective dose

Measure	Value	Source and standing
Median therapy received by community-dwelling people with aphasia	6.3 hours over three months, about one 60-minute session per fortnight	Palmer, Witts and Chater 2018; 278 patients across 21 UK SLT departments
Mean weekly provision reported by UK services	128 minutes (SD 86.5)	Monnelly et al. 2023, UK SLT survey
NICE standard	At least 3 hours a day, 5 days a week	NICE NG236, full text
ESO high-dose recommendation	At least 20 hours	ESO guideline 2025

Read those two delivery figures carefully

They are not like for like: one is what patients received (a median), the other what services reported providing (a mean with a wide spread). Read as a pair they point hard at under-delivery. They should not be averaged, and neither should be quoted as though it were the other.

Provision drops away after discharge

In 2024/25, 35.1% of applicable patients had a six-month review, down from 39%. The two tabulations disagree; the disagreement was recorded rather than averaged. Separately, around one fifth of hospitals were reported as not offering early supported discharge (a Stroke Association claim on a 2020 basis).

Twelve treatments, and the description problem

The list matters less than the description quality: 1 treatment had a strictly complete 12-item TIDieR description, 8 were usable, and 4 had no retrievable intervention description at all, including the Stroke Association Communication Support Service and generic semantic and phonological therapy as standalone packages. A treatment that cannot be described cannot be replicated, commissioned or audited.

Covered: impairment-focused therapy; constraint-induced and intensive language action approaches; comprehensive and high-dose programmes (ICAPs, CHAT); conversation and communication-partner training; technology and telepractice (Big CACTUS, Remote LUNA); adjuncts such as befriending (SUPERB).

Workforce and access context

23,590Adults waiting for SLT in England (Nov 2024)

14%SLT posts vacant (RCSLT 2025)

16.5%Services able to deliver an intensive programme

04 The barrier map, and why it is a map

Sixty-four barriers were mapped: 56 documented in a source and 8 inferred, kept apart and never merged. By CFIR domain: Outer Setting 15, Inner Setting 16, Individuals 12, Implementation Process 13, Intervention characteristics 8.

The strategy side: 40 distinct ERIC strategies across all 9 published clusters, every one of them within the published set of 73. And their evidence standing:

Standing	Meaning	Count
Validated in the UK	Effectiveness evidence in a UK aphasia or closely comparable context	0
Validated elsewhere	Effectiveness evidence exists, but not in the UK	8
Mapped only	From the published matching set, no effectiveness evidence found	47
Untested	No evidence located; the mapping is plausible reasoning	9

That distribution is the published pattern in this literature, not an accident of this review. Matching-tool studies are front-heavy and back-light: 96% formulate strategies from the tool, 51% adjust during implementation, 16% validate the mapping, and 8% evaluate whether the matched strategies worked. The value of this map is that it says so on its face.

04a The 64 barriers, specified

Maddy asked for the barriers by name. Below are all 64, grouped by CFIR domain. Each row carries the construct, the barrier as it appears in the source (or "INFERRED" where the mapping is reasoning rather than evidence), the source it came from, and the standing of the matched ERIC strategy. The full machine-readable form is in cfir_eric_map.csv.

Mobile-friendly: open a domain to expand its barriers.

How to read a row

Construct: the CFIR category within the domain.
Barrier: what the source says, in its own words, or marked INFERRED if it is reasoning.
Evidence: the source the barrier comes from.
Standing of the matched ERIC strategy: validated UK, validated elsewhere, mapped only, or untested. 0 rows in this map have UK effectiveness evidence.

Innovation Characteristics (8, 6 documented, 2 inferred)

01. Relative Advantage DOC

CPT / conversation-focused therapy is not recognised in workload or as direct intervention; 'gold standard but not essential', deprioritised against dysphagia.

Evidence: APT convergent barrier 1; concept-cpt-identity-problem (vault)
Matched ERIC strategy: Inform local opinion leaders (Develop stakeholder interrelationships) mapped only

02. Relative Advantage DOC

Current community provision is below any effective dose (~6.3 h per 3 months; mean 128 min/week against a 45 min/day x 5 day guideline), so the innovation's advantage is not demonstrable in the status quo.

Evidence: Palmer 2018; Monnelly 2023
Matched ERIC strategy: Promote adaptability (Adapt and tailor to context) mapped only

03. Complexity DOC

No standardised approach: 'different SLTs mean different things by CPT'; no structured manual; conceptual confusion about what CPT is.

Evidence: APT convergent barrier 2; WP2 conceptual clarity
Matched ERIC strategy: Develop a formal implementation blueprint (Use evaluative and iterative strategies) mapped only

04. Complexity DOC

ICAP/comprehensive therapy is operationally difficult to configure within mainstream services; only 16.5% of services felt able to deliver an ICAP.

Evidence: Monnelly 2023
Matched ERIC strategy: Promote adaptability (Adapt and tailor to context) mapped only

05. Cost DOC

Cost-effectiveness of intensive/comprehensive aphasia therapy is not established; a 100-hour multidisciplinary programme is resource-intensive.

Evidence: Kellar 2026 comprehensive rehab argument (open counter-argument)
Matched ERIC strategy: Fund and contract for the clinical innovation (Utilize financial strategies) untested

06. Evidence Strength and Quality DOC

Uncertainty about the effectiveness of comprehensive models weakens the case for investment (first definitive ICAP RCT reports 2030).

Evidence: Kellar 2026 comprehensive rehab argument
Matched ERIC strategy: Capture and share local knowledge (Develop stakeholder interrelationships) mapped only

57. Cost INF

INFERRED: the true economic case is likely understated because SSNAP's health-economics methodology was not built around communication outcomes.

Evidence: Pattern across SSNAP health-economics programme (vault entity-ssnap) - no study found on aphasia-specific costing.
Matched ERIC strategy: Fund and contract for the clinical innovation (Utilize financial strategies) untested

64. Design Quality and Packaging INF

INFERRED: the paucity of aphasia-accessible trial materials/consent is itself a barrier to the evidence base that would justify provision.

Evidence: Shiggins 2022 exclusion (vault UK-gap paper); inference from participation gap.
Matched ERIC strategy: Develop educational materials (Train and educate stakeholders) validated elsewhere

Outer Setting (15, 13 documented, 2 inferred)

07. Policy and Incentives DOC

Lack of national and/or local funding due to ICB deficits is the top reported challenge to delivering stroke rehabilitation.

Evidence: Stroke Association 2025 (Aug-2024 survey, 300+ professionals)
Matched ERIC strategy: Access new funding (Utilize financial strategies) untested

08. Policy and Incentives DOC

Only one third of ICSs prioritised stroke in their Joint Forward plans; ISDNs are not consistently funded by their ICB.

Evidence: Stroke Association 2025
Matched ERIC strategy: Involve executive boards (Develop stakeholder interrelationships) mapped only

09. Policy and Incentives DOC

CPT is invisible in the national stroke audit (SSNAP does not distinguish CPT from impairment-based therapy or dysphagia), so there is no national measurement or feedback loop.

Evidence: SSNAP/APT synthesis (vault); SSNAP 2024/25
Matched ERIC strategy: Develop and organize quality monitoring systems (Use evaluative and iterative strategies) mapped only

10. Policy and Incentives DOC

Aphasia is undercounted nationally: SSNAP records presence/absence on NIHSS item 9 (~72% sensitivity) with no severity, type or standardised screener.

Evidence: Kellar 2026 UK evidence gap (vault)
Matched ERIC strategy: Develop and implement tools for quality monitoring (Use evaluative and iterative strategies) mapped only

11. Policy and Incentives DOC

The six-month cliff-edge: only 35% of stroke survivors received a six-month review in 2024/25 (39% in 2023/24), leaving ~60,000/yr with no formal review.

Evidence: SSNAP/APT synthesis; concept-six-month-cliff-edge (vault)
Matched ERIC strategy: Change record systems (Change infrastructure) mapped only

12. External Pressure DOC

Resources are skewed towards acute provision, leaving community rehabilitation under-resourced; there is variation in access and some survivors pay privately.

Evidence: Stroke Association 2025
Matched ERIC strategy: Alter incentive/allowance structures (Utilize financial strategies) untested

13. Local Conditions DOC

Geographical variation in therapy intensity (c.60-180 min/week by region) and in service capacity raises equity concerns.

Evidence: Monnelly 2023
Matched ERIC strategy: Tailor strategies (Adapt and tailor to context) validated elsewhere

14. Local Conditions DOC

Transport and geography (rural/remote access) restrict face-to-face attendance; pre-pandemic UK data show ~1 h 23 min/week post-discharge in the North West.

Evidence: Bowen 2012 via Monnelly 2023; telepractice rationale (vault)
Matched ERIC strategy: Change service sites (Change infrastructure) untested

15. Local Conditions DOC

Digital exclusion: telehealth access depends on a carer/relative being present and capable, excluding people with aphasia who live alone.

Evidence: UK SLT telehealth survey, doi:10.1111/1460-6984.12996
Matched ERIC strategy: Prepare patients/consumers to be active participants (Engage consumers) mapped only

16. Patient Needs and Resources DOC

Half of people with aphasia (50%) report support they needed but could not access, the top unmet need being more speech and language therapy (24%).

Evidence: CHSS '1 in 5 Aphasia' (1,886 responses)
Matched ERIC strategy: Conduct local needs assessment (Use evaluative and iterative strategies) mapped only

17. Patient Needs and Resources DOC

Two thirds (67%) of people with aphasia report some difficulty accessing services.

Evidence: CHSS '1 in 5 Aphasia'
Matched ERIC strategy: Involve patients/consumers and family members (Engage consumers) mapped only

18. Local Attitudes DOC

Cultural and linguistic barriers: only 23% of the national workforce feels confident adapting CPT for non-English speakers; materials are absent in other languages and interpreters are unreliable.

Evidence: concept-cultural-linguistic-barriers (vault, WP1/WP2)
Matched ERIC strategy: Tailor strategies (Adapt and tailor to context) validated elsewhere

19. Local Attitudes DOC

Societal attitudes and stigma: 48% of people with aphasia report loneliness (vs 34% of all stroke survivors); friends disappear.

Evidence: CHSS '1 in 5 Aphasia'
Matched ERIC strategy: Use mass media (Engage consumers) untested

58. Local Conditions INF

INFERRED: outer-setting funding and inner-setting capacity interact non-linearly (a funded service still fails without staff), a pattern CFIR's interaction deletion cannot represent.

Evidence: Nilsen 2026 (interaction/dynamics critique); concept-cfir deletion diagnostic (vault).
Matched ERIC strategy: Tailor strategies (Adapt and tailor to context) validated elsewhere

59. External Pressure INF

INFERRED: commissioning cycles and contract durations may be misaligned with intervention timeframes, sustaining a revolving-door pattern of short programmes.

Evidence: Pattern across APT WP2 7-month re-referral wait and 6-week programme; no source states a causal link.
Matched ERIC strategy: Alter incentive/allowance structures (Utilize financial strategies) untested

Inner Setting (16, 15 documented, 1 inferred)

20. Structural Characteristics DOC

Fewer than a third of community-based rehabilitation teams meet recommended staffing levels; less than 50% of inpatient services meet any core-discipline level.

Evidence: SSNAP Post-acute Organisational Audit 2021
Matched ERIC strategy: Revise professional roles (Support clinicians) mapped only

21. Structural Characteristics DOC

Only 17% of community-based services have appropriate access to all core disciplines (PT, OT, SLT).

Evidence: Stroke Association 2025 citing SSNAP 2021 audit
Matched ERIC strategy: Recruit, designate, and train for leadership (Develop stakeholder interrelationships) mapped only

22. Structural Characteristics DOC

SLT workforce shortage: 14% overall vacancy, 14% in NHS adult services, 21% in London; 23% in the independent sector.

Evidence: RCSLT 2025 vacancy survey
Matched ERIC strategy: Access new funding (Utilize financial strategies) untested

23. Structural Characteristics DOC

Under-represented disciplines (psychology, dietetics, social work, family/carer support) and a lack of band-4/assistant staff in stroke rehabilitation teams.

Evidence: SSNAP PAOA 2021
Matched ERIC strategy: Create new clinical teams (Support clinicians) mapped only

24. Networks and Communications DOC

No standardised referral pathway for CPT; inconsistent referral/recruitment (ImPACT recruited 41 of 504 pooled dyads).

Evidence: APT WP2; Wielaert 2018 (ImPACT)
Matched ERIC strategy: Build a coalition (Develop stakeholder interrelationships) mapped only

25. Networks and Communications DOC

Interpreters are not always available for the critical first session and lack aphasia training.

Evidence: concept-cultural-linguistic-barriers (vault, WP2)
Matched ERIC strategy: Develop resource sharing agreements (Support clinicians) mapped only

26. Culture DOC

Competing priorities and deprioritisation: CPT is not recognised in workload and is squeezed by dysphagia and impairment targets.

Evidence: APT convergent barrier 1; cpt-deprioritisation (vault)
Matched ERIC strategy: Conduct local consensus discussions (Develop stakeholder interrelationships) mapped only

27. Culture DOC

Staff moral injury: clinicians experience distress and overwhelm at knowing the standard but being unable to deliver it.

Evidence: APT WP2 (vault); concept-staff-moral-injury
Matched ERIC strategy: Provide ongoing consultation (Train and educate stakeholders) mapped only

28. Implementation Climate DOC

Time and caseload pressures are the most frequent barrier to delivering psychosocial/family work (72% of surveyed UK SLTs).

Evidence: UK SLT psychosocial survey (n=124)
Matched ERIC strategy: Facilitation (Provide interactive assistance) validated elsewhere

29. Implementation Climate DOC

Staff turnover and loss of a local champion derailed implementation in some teams; change fatigue reported.

Evidence: Scobbie 2020 (G-AP); CPT implementation trial, doi:10.1080/09638288.2025.2534525
Matched ERIC strategy: Identify and prepare champions (Develop stakeholder interrelationships) mapped only

30. Implementation Climate DOC

Poor 'fit' between the innovation and the team's organisational structure; simultaneous delivery of other systems (e.g. parallel goal-setting tools) impeded adoption.

Evidence: Scobbie 2020 (G-AP)
Matched ERIC strategy: Model and simulate change (Develop stakeholder interrelationships) mapped only

31. Readiness for Implementation DOC

Limited leadership engagement from team leaders/managers is a documented barrier to implementation.

Evidence: Scobbie 2020 (G-AP)
Matched ERIC strategy: Recruit, designate, and train for leadership (Develop stakeholder interrelationships) mapped only

32. Readiness for Implementation DOC

Practice changes were not sustained: short-term CPT provision improved but gains were not maintained due to organisational barriers.

Evidence: CPT implementation trial, doi:10.1080/09638288.2025.2534525
Matched ERIC strategy: Purposefully reexamine the implementation (Use evaluative and iterative strategies) mapped only

33. Resources (Access to Knowledge and Information) DOC

Only 23% of community-based teams offered stroke-specific training for all disciplines.

Evidence: SSNAP PAOA 2021
Matched ERIC strategy: Conduct ongoing training (Train and educate stakeholders) mapped only

34. Resources DOC

Funding, printing and materials constraints: most SLTs lack access to resources; no time to prepare materials.

Evidence: APT convergent barrier 10
Matched ERIC strategy: Develop resource sharing agreements (Support clinicians) mapped only

60. Implementation Climate INF

INFERRED: deprioritisation and time pressure compound, so the constraint is not any single barrier but their conjunction ('deal breakers' interact).

Evidence: Scobbie 2020 explicitly argues barriers act collectively as deal breakers; extrapolated to aphasia provision.
Matched ERIC strategy: Conduct cyclical small tests of change (Use evaluative and iterative strategies) mapped only

Individuals (12, 11 documented, 1 inferred)

35. Deliverers: Knowledge and Beliefs DOC

Most SLTs had received no CPT training at all; knowledge gaps about the intervention.

Evidence: APT convergent barrier 3; Shrubsole 2023 ('Skills')
Matched ERIC strategy: Conduct educational meetings (Train and educate stakeholders) validated elsewhere

36. Deliverers: Knowledge and Beliefs DOC

The video-feedback component is rated essential by almost all SLTs but very few use it: the largest single evidence-practice gap.

Evidence: APT convergent barrier 4; concept-video-feedback-in-cpt
Matched ERIC strategy: Make training dynamic (Train and educate stakeholders) mapped only

37. Deliverers: Self-efficacy and Confidence DOC

Only 42% of surveyed UK SLTs felt confident addressing clients' psychological needs.

Evidence: UK SLT psychosocial survey (n=124)
Matched ERIC strategy: Provide clinical supervision (Provide interactive assistance) mapped only

38. Deliverers: Self-efficacy and Confidence DOC

Just over half of SLTs feel they have the skills to manage emotional/relationship issues; fewer than half are confident using video.

Evidence: APT convergent barrier 3 (WP1)
Matched ERIC strategy: Provide ongoing consultation (Train and educate stakeholders) mapped only

39. Deliverers: Capability (COM-B) / Knowledge DOC

77% of the workforce does not feel equipped to adapt CPT for non-English speakers.

Evidence: concept-cultural-linguistic-barriers (vault, WP1)
Matched ERIC strategy: Conduct ongoing training (Train and educate stakeholders) mapped only

40. Deliverers: Capability / Skills DOC

SLT scope is heavily oriented to impairment-level work (naming, repetition, comprehension) with little family involvement, psychosocial support, employment or participation work.

Evidence: Monnelly 2023; Kellar 2026 comprehensive rehab argument
Matched ERIC strategy: Work with educational institutions (Train and educate stakeholders) mapped only

41. Deliverers: Motivation / Personal Attributes DOC

Existing clinician knowledge shapes how training is 'heard'; operationalisation drift means the same strategy is enacted differently by different clinicians.

Evidence: APT WP2; concept-operationalisation-drift (vault); Nilsen 2026
Matched ERIC strategy: Conduct local consensus discussions (Develop stakeholder interrelationships) mapped only

42. Recipients (Patients/Clients): Characteristics DOC

Many people with aphasia have no consistent communication partner; family attendance at therapy is often low.

Evidence: Shrubsole 2023 (Environmental context / Social influences); APT convergent barrier 6
Matched ERIC strategy: Involve patients/consumers and family members (Engage consumers) mapped only

43. Recipients: Motivation / Beliefs about Consequences DOC

Family and client preference for direct therapy; CPT can be read as 'giving up' or 'no hope for improvement'.

Evidence: Shrubsole 2023 (Beliefs about consequences); APT convergent barrier 7
Matched ERIC strategy: Intervene with patients/consumers to enhance uptake and adherence (Engage consumers) mapped only

44. Recipients (Families): Characteristics DOC

Carer burden and unmet carer need: 81% of people with aphasia are cared for by a family member or friend; communication partners are not assessed at all.

Evidence: CHSS '1 in 5 Aphasia'; SSNAP/APT synthesis
Matched ERIC strategy: Prepare patients/consumers to be active participants (Engage consumers) mapped only

45. Recipients: Opportunity (COM-B) DOC

Equity gap: people without an advocate 'fall by the wayside'; access is shaped by financial and cognitive capacity to push for services.

Evidence: concept-cultural-linguistic-barriers (vault, WP2); APT WP2
Matched ERIC strategy: Increase demand (Engage consumers) mapped only

61. Recipients: Motivation INF

INFERRED: patient and carer readiness is dynamically shaped by relationship and timing, not a fixed trait, so a single pre-implementation assessment will misclassify dyads.

Evidence: APT convergent barrier 7 (readiness shaped by relationships); Nilsen 2026 (temporal dynamics flattened).
Matched ERIC strategy: Assess for readiness and identify barriers and facilitators (Use evaluative and iterative strategies) mapped only

Implementation Process (13, 11 documented, 2 inferred)

46. Planning DOC

Training is not coupled to implementation planning: staff were prepared to deliver the innovation but not to plan, monitor and tailor its local implementation.

Evidence: Scobbie 2020 (G-AP)
Matched ERIC strategy: Develop a formal implementation blueprint (Use evaluative and iterative strategies) mapped only

47. Planning / Assessing Context DOC

No formal context/readiness assessment before implementation: services adopt innovations without evaluating inner/outer setting readiness.

Evidence: Scobbie 2020 (G-AP); CFIR 2.0 'Assessing Context' (vault source-damschroder-2022)
Matched ERIC strategy: Assess for readiness and identify barriers and facilitators (Use evaluative and iterative strategies) mapped only

48. Planning DOC

Delays between training and implementation (staged referral, waiting lists) let momentum dissipate before delivery begins.

Evidence: Scobbie 2020 (G-AP)
Matched ERIC strategy: Stage implementation scale up (Use evaluative and iterative strategies) mapped only

49. Engaging DOC

No co-design or coproduction with people with aphasia and families in service design; recipients are consulted late or not at all.

Evidence: BMJ Open 2021;11:e047398; APT WP2 reframing
Matched ERIC strategy: Involve patients/consumers and family members (Engage consumers) mapped only

50. Engaging DOC

Information and education gaps: aphasia education is delivered but the purpose and value of CPT are not understood by communication partners.

Evidence: APT convergent barrier 12
Matched ERIC strategy: Develop educational materials (Train and educate stakeholders) validated elsewhere

51. Engaging DOC

Goal-setting difficulties: goal tools 'don't get to the heart of it'; families want to be told what to do; goals change between sessions.

Evidence: APT convergent barrier 9
Matched ERIC strategy: Conduct local consensus discussions (Develop stakeholder interrelationships) mapped only

52. Executing DOC

SLT assistants are underutilised: 88% of SLTs deliver CPT alone; Band 3/4 staff lack CPT-specific training and supervision.

Evidence: vault concept-slta-role (WP1/WP2)
Matched ERIC strategy: Revise professional roles (Support clinicians) mapped only

53. Executing DOC

Follow-up and continuity are not executed: no long-term follow-up pathway exists after the six-week block; ~3 h total therapy over 6 weeks.

Evidence: APT WP2 (vault); concept-six-week-cliff
Matched ERIC strategy: Change service sites (Change infrastructure) untested

54. Reflecting and Evaluating DOC

Lack of ongoing local monitoring, evaluation and tailoring of implementation; no site-specific performance feedback.

Evidence: Scobbie 2020 (G-AP recommendations)
Matched ERIC strategy: Develop and organize quality monitoring systems (Use evaluative and iterative strategies) mapped only

55. Reflecting and Evaluating DOC

The CFIR-ERIC cascade: across 53 studies using the matching tool, 96.23% formulated strategies but only 15.69% validated them and 7.84% evaluated effectiveness.

Evidence: Meng et al. 2026 scoping review
Matched ERIC strategy: Purposefully reexamine the implementation (Use evaluative and iterative strategies) mapped only

56. Reflecting and Evaluating DOC

Matched strategies are selected from a taxonomy without mechanism specification, so failure is undiagnosable and nothing specific is validated.

Evidence: Meng 2026; Waltz 2019 (1-20 strategy recommendations per barrier); Nilsen 2026
Matched ERIC strategy: Audit and provide feedback (Use evaluative and iterative strategies) validated elsewhere

62. Planning INF

INFERRED: because determinants are measured at baseline only, teams cannot tell which barrier was consequential when, and so target the wrong strategy.

Evidence: concept-cfir temporality deletion (vault); Nilsen 2026 challenge 5.
Matched ERIC strategy: Conduct local needs assessment (Use evaluative and iterative strategies) mapped only

63. Engaging INF

INFERRED: equity of reach is not modelled by any implementation plan, so improvement in aggregate uptake can mask worsening inequality of access.

Evidence: Palmer 2018 and Monnelly 2023 both note equity was not stratified; inference from that gap.
Matched ERIC strategy: Capture and share local knowledge (Develop stakeholder interrelationships) mapped only

05 The evidence base, and how far it can be trusted

Three research lanes ran in parallel so they could not read each other's conclusions: treatments, services and contexts, barriers. A fourth lane then tried to falsify their load-bearing numbers, and a fifth consolidated the result. The verification lane's brief was to break the work, not summarise it: a figure it kills with evidence is worth more than one it carries over.

81Numbered citations

76Distinct works

55 / 70Claims confirmed at source

10Residuals carried

Of the load-bearing claims re-fetched at source: 55 confirmed, 4 corrected (3 distinct), 1 killed, 10 unverifiable and carried as residuals. The killed claim was ours: a note recorded the Palmer 2018 sample as 114; the paper reports 278. Two further corrections: a community SLT establishment figure of "0.4 WTE per unit" is 0.4 WTE per 100 patients, and a TIDieR cell naming a trained assistant as alternate deliverer should be a local collaborator SLT.

Four label families are preserved throughout and never merged, because merging them is how a review overstates: documented versus inferred; the four evidence standings; [UNVERIFIED] markers; and full text versus abstract versus metadata-only access. The access labels sit on the claims that depend on them, not on the source list.

The artefact passed a house style check (no dashes, no external assets, citation parity 81 of 81), an accessibility check with no hard findings, a tag balance check, and a layout probe at three widths.

Open the evidence base Ask the assistant about it

06 The method

The approach comes from the autoresearch family: long-horizon work that proposes, evaluates and refines, with a gradeable artefact at every step. Five rules govern a loop: fix the artefact and the grader before the first candidate exists; keep one episode bounded; leave a ledger line for every episode; keep selection separate from production; report the delta with its denominator.

Why the ledger matters more than it looks. An audit of autonomous research pipelines detected methodological faults at 53.7% accuracy from the finished manuscript alone, barely better than chance, but 81.5% when the full execution record was available. Polished prose hides exactly the intermediate decisions that carry the faults. A clean document with no trail is unverified, not verified.

The grader, for research work. Research tasks rarely have a test suite, so the grader is usually a fixture: a small set of items with known answers, scored field by field, with negative controls that must reject wrong and empty input. Reusing a small selection set overstates performance by 13 to 20 points on a 16-item set.

Selection integrity. Split the fixture before iterating: rows used to choose and rows used to report are different sets. Gate admissions with a named change, a cause and the test that passed. Treat confidence as a gate rather than a decider.

The idea-basin problem. Long runs plateau in different local optima because the running context anchors the search. Two documented escapes differ in what survives the reset: fork into fresh-context branches carrying the workspace, or discard the session entirely and keep only the best candidate. Both were measured on tasks with executable metrics, and neither has been shown to make bibliographic research better. That boundary is worth saying out loud.

07 Assets and next steps

Everything the work produced is below. The evidence artefacts are the reviewed versions. Two of the items are reusable machinery rather than findings.

Evidence base (HTML)The reviewed artefact: findings, treatments with TIDieR rows, services, the CFIR to ERIC map, limits and the source ledger.
Merged evidence base (JSON)The machine-readable merged base behind the page: cards, sources, labels, verification, corrections, treatments, TIDieR gaps.
Treatments table (JSON)Twelve treatments with provider, format, setting, dose, evidence standing and the 12 TIDieR fields each.
TIDieR gaps (JSON)The interventions with no retrievable description, and which of the 12 items are missing across the set, with counts.
Services table (JSON)Thirteen service types with access route, setting, coverage evidence and dated figures.
Provision gaps (JSON)The breakpoints in the pathway, each with its evidence, documented and inferred kept apart.
CFIR to ERIC map (CSV)Sixty-four rows: CFIR domain and construct, the barrier, its evidence, the ERIC strategy and cluster, and the evidence standing.
Map coverage (JSON)Counts per CFIR domain, per ERIC cluster, and the four evidence-standing counts.
Verification log (MD)Seventy load-bearing claims, the source re-fetched, what it says now, and the verdict.
Kill list (JSON)Every corrected and killed item with its replacement figure and source.
NHS and non-NHS pathway after stroke with aphasia (HTML)Working brief, 18 October 2026. Stage-by-stage route in England, Scotland, Wales and Northern Ireland, with verification status per nation and the open gaps.
Residuals (MD)What could not be verified and why, with the confidence each remaining claim deserves.
SSNAP Open Explorer (HTML, unofficial)Standalone explorer over SSNAP open aggregate outputs: national trends, ISDN comparator, league and hospital profiles, casemix, coverage.
SSNAP SLT figures from the store (HTML)National speech and language therapy measures read from the SSNAP data store, with the sheet and cell for each value.
Wave audit script (PY)Reusable machinery: the check that reads the board and flags an incomplete handoff before the next card consumes it.
Handoff contract (MD)Reusable machinery: metadata, the completion file, attachments, and the order to do them in.

Next steps, honestly

The barrier map needs a version decision. It is a CFIR 1.0 and 2.0 hybrid; tidying it to one version is a real piece of work, not a formatting pass.
The four undescribed treatments are the commissioning lead. A description gap is the most actionable thing in the evidence base.
The zero stands. Until a mapped strategy is evaluated in a UK context, the strategy list should be presented as a hypothesis to be tested rather than a plan to be implemented.

Process notes, for the record

Not part of the findings. Six defects surfaced while the work was being done, five of them in the machinery rather than the research, and they are kept here because they are the transferable part. Open if you want them.

Six defects, and what changed because of them

1. The worker handoff arrived empty

Symptom: Four of four workers stored an empty metadata object on completion and re-posted the real metadata as a comment. Two attached nothing, so their only deliverable sat in a workspace that is pruned afterwards. Anything reading metadata alone would have silently dropped every lane.
Cause: The completion call passed no payload, and nothing checked. The platform's stop-gate verifies only that a terminal board tool was called, not that its arguments were complete.
Fix: A pinned skill carrying the rule into every worker's prompt: pass the schema as metadata, write the same object to completion.json, attach every deliverable, attach first and complete second, and re-post if the payload drops. Plus a wave audit that reads the database and flags empty metadata, missing attachments and missing paths before the downstream card runs.

2. The worker operating contract was never loaded

Symptom: None of the rules written into a worker profile's AGENTS.md had any effect.
Cause: The configuration key meant to inject it appears nowhere in the product code, and project-context discovery reads the working directory chain, which a task workspace does not have. Every operating contract written that way had been inert from the day it was written.
How it was caught: By asking rather than assuming: a worker was asked whether a distinctive phrase from the freshly edited contract was in its loaded context, and replied NOT IN MY CONTEXT. The same rule, delivered as a pinned skill, was then quoted back verbatim.
Fix: Rules a worker must obey go in the card body, in a pinned skill (which the dispatcher preloads), or in the system-prompt identity file which is always included.

3. Remembered framework vocabulary

Symptom: The barrier map was labelled as using CFIR 2.0 constructs. Roughly 32 of its 64 rows were not CFIR 2.0 names: a hybrid of CFIR 1.0, 2.0 and COM-B labels.
Cause: The lane was told to use the canonical list and did not gate itself against it. The ERIC side of the same map came back clean, 40 of 40 strategies and 9 of 9 clusters, precisely because that list is finite and checkable.
Fix: A framework name in a brief is a gate, not a gloss: the lane must run its own validity count against the canonical list and report it before completing. No count, no completion.

4. A count typed during writing

Symptom: The synthesis introduction claimed "8 documented, 5 inferred" breakpoints. Its own table read 11 and 4; the merged data read 9 documented-only, 2 inferred-only, 2 both. No partition of the 13 produced 8 and 5.
Cause: A number typed while writing prose, inherited from a lane's completion sentence that already contradicted that lane's own data file.
Fix: Numbers in prose come from the data file, never from the writing: generate the sentence with a script, or assert the prose count against the JSON. The reviewer caught it; the cheap fix is upstream.

5. A wrong figure living in the notes

Symptom: A prior note recorded the Palmer 2018 sample as 114 community-dwelling stroke survivors. The paper reports 278 patients across 21 UK SLT departments.
Cause: Later notes cited the note rather than the paper, so a wrong denominator propagated without being re-checked.
Fix: Figures held in notes are claims, not facts, and are verified at the primary source before reuse. A discrepancy is reported as a finding. The correction itself is staged for the owner of the note rather than edited in behind them.

6. The attach rule was written in the wrong place

Symptom: Two lanes declared their deliverables as absolute paths in metadata, which do not survive workspace pruning.
Cause: "Declare deliverable files" is not the instruction "attach deliverable files", and only the second survives.
Fix: The contract names the attachment as the deliverable's home, and the audit treats an unattached artefact as a defect even when its path is recorded.

The pattern across all six

Check the handoff, not the handoff's description of itself. Read the database, the attached bytes and the served file, rather than the summary line that says everything is fine.

Sources

The evidence base carries 81 numbered citations resolving to 76 distinct works. The full generated list with its URLs is in the evidence base and in the assistant's corpus.

Palmer R, Witts H, Chater T (2018). What speech and language therapy do community dwelling stroke survivors with aphasia receive in the UK? PLoS ONE 13(7): e0200096.
Monnelly K, et al. (2023). UK SLT survey of intensive and comprehensive aphasia therapy. International Journal of Language and Communication Disorders.
NICE (2023). NG236 Stroke rehabilitation in adults.
Brady MC, et al. (2025). ESO guideline on aphasia rehabilitation.
Hoffmann TC, et al. (2014). Better reporting of interventions: TIDieR. BMJ 348:g1687.
Waltz TJ, Powell BJ, et al. (2019). Choosing implementation strategies to address contextual barriers (CFIR-ERIC matching).
RCSLT (2025). A profession under pressure: SLT retention and waiting times.
HQIP / SSNAP (2025). State of the Nation and post-acute organisational audit reports.