Start with the two commitments that are not optional, because the framework requires footprints agreed around natural communities and plans for integrated neighbourhood teams focused on high priority cohorts. Confirm the data-sharing arrangements in the same cycle, since without them patients cannot be identified and results cannot be evaluated, and the framework treats this as a minimum basic requirement. Then fix the outcome before the money moves. The reason is in the record: the Vanguard programme, which covered around 9 per cent of the population, slowed the rise in emergency admissions in care home sites but produced no significant reduction in bed-days, and Sibbald's review of 119 studies found that shifting care reduced outpatient activity while risking quality and losing economies of scale. So protect the community nursing establishment first, state plainly in the plan that a fall in hospital activity may not follow, and name the measure on which the system will be judged in 2029 before the board signs it off.

For the pooled budget, replace the intention with the sum, the services it buys and the years it covers. Name the pooling vehicle, the ICB contribution and the local authority contribution, and state that the fund will finance urgent community response and reablement capacity for the frailty cohort, since bed days for multimorbid patients rose by 4.9% a year where pooling was not matched by redesigned services. For data sharing, the plan cannot say in principle. It must name the datasets, the controller, the lawful basis, the retention period and the accountable officer who signs the arrangement, because the English pilots showed that a necessary data-sharing agreement can simply never arrive. Ownership should name one organisation and one post, not a committee, for each deliverable, and publish the 10% reduction in non-elective admissions as the measure the board accepts.

I would not adopt any single existing boundary by default. I would build the footprint from patient flows first, because the Greater London work shows that even a data-defined network holds only a median of 70.1 per cent of its patients, so any boundary will leak and the question is how much it leaks. I would then test each candidate against four criteria: the share of patients whose care stays inside it, the distribution of deprivation and ethnicity across it, the travel time for residents at its edge, and whether it follows the local authority boundary where that alignment is real rather than assumed. I would size the team to the frailty cohort, 3 to 5 per cent of the population, and not to the whole list, because that is the caseload a named team can hold. Finally I would expect the footprint to miss the inclusion health groups Humphrey and colleagues describe, and I would fund outreach registration rather than assume the boundary catches them.

I would constitute the single neighbourhood provider, but only on two conditions. First, one named person accountable for the cohort, holding a delegated budget that spans health and social care, because the Evercare evaluation, the Medicare trials and the Pioneer evaluation all show that arrangements without that single point change activity without changing responsibility. Second, a shared writable record and protected shared time, since co-ordination happens between visits rather than in them. What I would refuse is a promise of a 10% fall in non-elective admissions in year one. The trials suggest the early gains are quality of life, experience and fewer emergency department visits, with admission effects confined to well-targeted high-risk groups and appearing late. I would publish those measures by March 2029 and report the community waiting position honestly.

Before the tool touched a caseload I would ask four questions and publish the answers. First, coverage: who is in the training data, and whose ethnicity, housing or social care record is missing, since a field absent from most emergency records cannot carry an equity claim. Second, the label: does the score predict cost or clinical need, because a cost target converts unequal access into apparent good health. Third, the threshold: which groups fall just outside the cut, and what the service offers them, since most flagged people will still not be admitted. Fourth, subgroup accuracy: does discrimination hold for the most deprived fifth, for older people with frailty, and for people with severe mental illness, whose relative risk of death is 2.22 and who rarely appear as acute activity first. I would publish the audit, the threshold and the plan for those not selected. I would not switch the tool on if the supplier refused the audit or the data sharing it needs.

I would keep a universal offer and intensify it, because concentrating only on the most deprived cohort forfeits the population effect: most preventable events occur in people at moderate risk, which is the prevention paradox. The funding argument points the same way, since the benefits arrive over years while the service is paid for now, so I would want the offer to reach the people least likely to opt in rather than to advertise widely. The stop smoking services show the design that works: reach further into disadvantage, accept a lower quit rate, and the overall effect is still equity positive. I would cite the bowel screening gradient, where uptake was 35% in the most deprived areas against 61% in the least deprived, and the widening vaccination gap, where the second MMR dose moved from -9.6% to -13.4%. I would measure uptake and outcome separately for the most and least deprived fifths and publish both, so that a rise in the average could not conceal a widening gap. I would not judge the offer on total reach alone.

Choose the hub, and say so on the evidence. Remote monitoring works on a defined risk in a known patient: the Cochrane review found COPD related readmission probably reduced, and the Whole Systems Demonstrator found lower admissions and lower mortality, but BEAT-HF found no readmission difference at 180 days in heart failure. It manages an episode. It does not reach the people the neighbourhood service exists for. The exclusion data decide it: 41 per cent of patients booked online but only 29 per cent of those aged 75 to 84 did so, and 8.5 million adults lack the most basic digital skills. The hub model works when it is recipient centred, in community spaces, with devices and support, so it must be funded as service infrastructure rather than as a project. Before either goes live the board needs a data sharing agreement naming the lawful basis and purposes for every party, a common record and identifier set, and a named person accountable for information governance, because identifying and evaluating a whole population cannot rest on individual consent. The single patient record promised for 2028 will make that unavoidable.

The panel should refuse to read the specification as a compliance list, since it introduces no compliance mechanism and uses must only where law already requires it. Four decisions matter. First, ordering: is the clinical brief settled before the site was chosen, and which named service change does the building enable? Second, revenue: the funding criteria require a credible revenue case covering unitary charges, occupancy and utilisation, so the panel should ask what the annual charge is, what it represents as a share of the responsible providers' turnover, and what happens in year one before activity shifts. Third, access: opening for 12 hours a day only helps the patients who can reach the building, so the panel should ask for the distance evidence on the catchment and a plan for those who will not travel. Fourth, integration: co-location does not equal integration, so require shared records, a named accountable lead, and a commitment to measure in the third year whether the rooms deliver the service model rather than merely fill.

I would run the first year as a block contract with a published development plan, and move to outcome-based payment only once two conditions are met. The first condition is a payment cycle that does not require a small provider to fund a year of work before it is paid, because the evidence is that the transaction costs of contracting fall hardest on small organisations and that two thirds of government contract income already goes to providers with incomes above 10 million pounds. The second is an outcome set that counts work preventing admission as well as activity delivered, because community services currently cannot evidence their contribution to hospital avoidance. I would promise the small providers a staged payment, a place on the 50,000 population contract at their own scale rather than through a prime contractor, and a published evaluation. I would not promise savings in the first year, because the audit office found none from the pooled fund.

The first question is accountability: which named post holds the high priority cohort and the budget that goes with it? The next steps letter asks plans to confirm organisational ownership of planned deliverables, which an organisation can satisfy while no person is answerable, and the National Audit Office found it unclear who monitors whether integration is working. An acceptable answer names a person, a cohort and a budget. The second question is money in year two: which budget line pays in 2028 to 2029, and is it recurring? Neighbourhood health is to be funded by rebalancing existing resources, local priorities have no protected budget, and MacInnes and colleagues found that a lack of ongoing resources ended the spread of the Vanguard models. An acceptable answer is a recurring line and a statement of what stopped being funded. The third question is measurement: which measure would fall if the service stopped working well for people? If every measure counts activity, as the 18 week community access measure does, nothing measures the person. An acceptable answer names at least one outcome for people and the data sharing arrangement that will capture it.