The Fidelity Metrics That Catch a Programme Drifting as It Scales

A study published in Health & Place this month tracked twelve categories of third place — coffee shops, libraries, museums, recreation centres, restaurants — across the United States from 2010 to 2021. Every category lost ground between 2019 and 2021. The losses were not spread evenly. They concentrated in census tracts with higher social vulnerability and in rural areas.

If you are scaling a connection programme from one or two pilot sites into a wider rollout, this is not background noise. It is a direct warning about the assumption most scale-up plans quietly make: that the thing which worked in the pilot neighbourhood — a walkable café, a functioning library branch, a community centre with a room to spare — exists in the same form wherever you replicate.

The replication assumption

Call it the replication assumption: if the model worked here, it will work there, provided you copy the referral pathway, the group format, and the training. Programme design literature going back to Ray Oldenburg’s account of the third place, and more recently to Eric Klinenberg’s argument that libraries and parks function as social infrastructure, treats physical venues as a precondition for informal contact, not a detail. A model that depends on a good venue is not portable to a place without one. Scaling documents rarely say this out loud, because venue quality is not the thing anyone measures. Referral counts and attendance are.

That is the gap the Health & Place data exposes. Two sites can have identical referral volume and near-identical staffing, and still produce different outcomes, because one sits in a tract that lost its library branch and its diner between 2019 and 2021, and the other did not.

What to measure before you approve a new site

Most programme scale-ups measure fidelity to the intervention — did staff follow the script, run the right number of sessions, use the same materials. That is necessary but not sufficient. Add a site-readiness check before committing budget to a new location:

1. Third-place density within a reasonable travel radius of the referral point. Count venues in the same categories the pilot relied on — not any café, the kind of low-cost, low-commitment space the model actually used.
2. Public transport or walking access to those venues, not just their existence. A library three miles away on a route with no bus is not the same asset as one five minutes on foot.
3. Staff-to-participant ratio adjusted for local caseload, not copied from the pilot. Rural sites in the study’s data lost amenities disproportionately; if your model relies on staff filling the gap left by a missing venue, that costs more staff time per participant, not less.
4. Baseline social infrastructure trend for the tract, where available, rather than a single snapshot. A site in decline needs a different plan than a stable one, even if today’s numbers look similar.

None of this is exotic. It is the kind of check a commissioner should ask for before signing off on expansion, and the kind that almost never appears in scale-up business cases because it was never a problem at the pilot site — the pilot site was chosen, consciously or not, partly because it had the infrastructure the model needed.

What this means in practice: before opening a new site, run the venue-density and access check as a formal go/no-go gate, not a soft consideration. If the check fails, budget for either a different format (staff-led, no external venue dependency) or a genuinely higher staffing ratio — do not assume the pilot’s numbers will hold.

What the evidence actually supports

Claim	Evidence status
Third-place closures 2019–2021 concentrated in socially vulnerable and rural tracts	Directly evidenced, single study
Physical venues shape rates of informal social contact	Well-established in the social infrastructure literature, though largely descriptive rather than causal
Structured, purposeful group activity works better than unstructured contact alone	Supported by qualitative synthesis of social prescribing programmes
A site-readiness check of the kind proposed here improves scale-up outcomes	Not tested; this is an inference from the venue-access finding, not a validated protocol
Social prescribing for older adults reduces isolation at scale	Still unclear — a 2025 systematic review protocol notes only one peer-reviewed randomised controlled trial exists in this area

The last two rows matter for anyone writing a funding case. The venue-density argument is a reasonable extrapolation from a real finding, not itself a tested intervention. Say that plainly to a funder rather than presenting it as settled.

Why this differs from ordinary fidelity monitoring

Standard fidelity monitoring in social prescribing and community programmes tracks whether the intervention was delivered as designed — session count, group size, facilitator training. That is a check on your organisation’s behaviour. The check proposed here is a check on the environment you are placing the programme into, something outside your control that the pilot happened to have and a new site might not. Cross-country data on social connection backs this distinction: the OECD’s 2025 review of social connections found that people are meeting in person less frequently than in the past across member countries, and that declines were sharpest for groups — younger adults and men — not previously flagged as high risk. A model built around one demographic’s typical meeting patterns will not transfer cleanly to a site with different patterns, even before the venue question comes up.

What this does not solve

None of this fixes the underlying scarcity. If a rural tract has lost its library and its diner and there is no comparable venue within reach, no amount of site-readiness checking manufactures one. The honest response in that case is either a higher-cost, staff-intensive model that does not depend on informal venues, or an acknowledgment that the programme should not expand there in its current form. Scale-up plans that promise uniform delivery across uneven ground are promising something the physical geography does not support — and the data on where third places have actually disappeared says that unevenness is getting worse, not better.