The number nobody keeps

On the fifteenth of January 2026, 718,838 children and young people in England held an education, health and care plan.

A year earlier the number was smaller by an eighth.

Each plan is a list of promises. Speech therapy, so many hours a week. A teaching assistant, so many sessions. Transport, every school day. The law is plain about who owes them: the council must secure what the plan specifies, and it must specify it closely enough to be counted.

So a great deal is counted. The number of requests: 162,702 in a single year. The number refused: 43,300. The number of new plans: 110,708. The share of those issued inside the twenty weeks the regulations allow: 46.1 percent, down a little from the year before.

Fewer than half.

When a family believes the plan is wrong, it may appeal. In the year to July 2025, 25,000 appeals were registered, more than ever before. Of the appeals the Tribunal decided, 99 percent went at least partly the family's way.

That figure is worth reading twice. It does not describe a contest. It describes a record that could not be defended once someone independent looked at it.

The Ombudsman found fault in 94 percent of the complaints about special educational provision it investigated. A committee of the House of Commons recorded its concern at the frequent failure to deliver the support set out in a plan once the plan has been issued.

And here the counting stops.

There is no national figure for whether the provision in the plans was delivered. Not for the speech therapy, not for the sessions, not for the transport. The regulations count the plans. Nobody counts the promises inside them.

Where a record of delivery exists at all, it is usually kept by the body that owes the delivery. A week in which nothing was written down looks, in that record, exactly like a week in which everything happened.

The money moves anyway. £12.2 billion of high-needs funding in 2026-27. Deficits that the Office for Budget Responsibility expects to reach £14 billion by 2028-29, after a one-off transfer of £6 billion to clear most of what came before. Funding per plan, in real terms, a third lower than it was.

Every pound of it is spent against promises whose keeping is not measured.

The change asked for is small to say. Keep the count somewhere the council cannot write. Let the setting, the therapist or the family record each delivery. Let the clock mark a missed deadline by itself, so that silence reads as what it is. Let only an authority other than the council change what is owed, and keep the old version when it does. And let the record be sealed, so that nobody, later, can make a missed month read as a delivered one.

The families already win almost every appeal that reaches a judge.

What they do not have is the record that would have made the appeal unnecessary.

The numbers, in one table

Every figure is cited in the sources: block above, with its date.

The arithmetic — a case for £1 million that can be checked

A case for spending public money can prove two things before the money is spent: the thresholds — how small an effect pays for it — and the instrument — that what is built measures the effect honestly. It cannot prove the effect in advance. Anyone who claims otherwise is selling. What follows proves the first two and says plainly what the third will take.

1. What the million buys — a proposed allocation, an assumption to be quoted, not a measured cost:

build what is armed red          the clock, the Tribunal's
                                 amend rung, fair routing,
                                 access control on case pages
the sealed record                duty verdicts anchored on a
                                 public chain, signed by the
                                 certifier, not the council
a pilot                          three councils, twelve months
an independent evaluation        someone else measures it

2. The pilot's size. Three councils out of 153 is 1.96%. If the pilot councils hold an average share of plans (an assumption — councils vary widely):

plans in the pilot  ≈ 718,838 × 3/153  ≈ 14,100
cost per plan-year  ≈ £1,000,000 / 14,100  ≈ £71

3. Against what is spent on the same plans.

high-needs funding per plan  ≈ £12.2bn / 718,838  ≈ £16,970
the pilot, per plan          ≈ £71  ≈ 0.42% of that

So the break-even, in the pilot year, is:

the record must redirect or recover
≥ 0.42% of the high-needs money spent on each pilot plan
— ≈ £71 per plan — that would otherwise buy nothing delivered

4. At national scale, after the build. The build is paid once. If the whole million is set against one year of national high-needs funding:

break-even share  =  £1,000,000 / £12,200,000,000  ≈ 0.0082%

That is one pound in every twelve thousand two hundred. Against the charity estimate of tribunal spend alone (not official), the same million is 1.7% of one year's £59.8m.

5. What is already proven about the instrument — each a test that runs against the code (text/ehc-routing.md, text/duty.md):

• the body that owes a duty cannot record its own delivery; the database refuses it;
• a deadline that passes with the count short is a breach, without anyone noticing;
• no routing weight, of any size, can change a duty's verdict — 5,000 random duties, and the test goes red when a weight is planted;
• the adaptive part excludes a provider that keeps failing whatever its size, and every decision it makes can be recomputed from the log.

6. What is not proven, and must not be claimed.

• That delivery improves. Nobody knows today's delivery rate, because nobody counts it. The pilot's first result is the baseline itself.
• That tribunal costs fall. The only cost figure is a charity estimate.
• That the chain is needed for the arithmetic. It is not. The duty rules run without it. The chain is bought for one property only: a verdict that neither the council nor anyone else can quietly rewrite, which a family or the Tribunal can check without trusting the council's own server.
• That it runs anywhere today. The duty rules and the proofs are built and tested. The clock, the amend rung, fair routing and the sealed record are armed red. No council uses it. The on-chain contracts run on a test network only.

The provable case, in one line. For about 0.42% of what is already spent on each pilot plan, England would have, for the first time, a count of whether the promises in a plan were kept — a count the body owing them cannot write — and would then know, rather than argue, whether it pays for itself.

Where the metaphor stops being true

The story says "sealed". In this repo that is a Promise or duty object on Sui (pay/contracts/generated/sources/promise.move), generated, on testnet only, and no duty is anchored on any chain today. The story says "the clock marks a missed deadline by itself"; duty:breach exists and nothing calls it (duty.md row 2). "Only an authority other than the council" is refused today with blocked_on: 'tribunal-rung' (duty.md row 3). The locked verbs appear here only in skin 10: deliver, breach, de-list (metaphors.md).

The family this story generates