Civic Kernel a thin protocol for public decisions

When the system fails you, what could a protocol anyone can check actually do?

Three real UK cases the Good Law Project has fought, run against the Civic Kernel, a thin protocol for public decisions that nobody can rig and anybody can check. For each: what happened, what a checkable record would, and wouldn’t, change, and what helps today. No jargon; every link goes deeper.

Records

Decisions taken where no one can read them

During the pandemic, government business ran on private, auto-deleting messaging. The UK Covid-19 Inquiry found records were “not retained… some deliberately.” One health minister spelled out the tactic in a group chat:

“I’m deleting the messages in this group. They can be captured in an FOI…”

, evidence to the UK Covid-19 Inquiry
What a checkable record could do

A transparency log is a record no one can quietly rewrite: change yesterday and the figures stop matching every copy of today, and any phone can tell.

Where it stops

It can guarantee a record isn’t altered. It cannot force a decision to be written down in the first place. Here the real decision never entered any record; it was destroyed at source. A tamper-proof log around an empty space is still empty.

What you can do

Use Freedom of Information, and treat “no information held” as a finding, not a dead end: it is evidence that business was kept off the record. Back records-retention duties, and the Inquiry’s own recommendation that government run on approved, retained channels.

See this case checked against the protocol, step by step →

Public money

Every contract was lawful. The lane was not.

In the pandemic, firms referred by ministers, MPs and officials were routed to a “high-priority lane.” They were about ten times more likely to win a PPE contract, roughly one in ten, against one in a hundred, and £3.8 billion of contracts went through it. The High Court ruled the lane unlawful, then found the firms would “likely have been awarded contracts anyway,” cancelled none, and declined to publish the money wasted. Working out the pattern took the National Audit Office about two years.

What a checkable record could do

Every award can be logged with who referred it and by which route. A privileged lane stops being a two-year investigation and becomes a signed number anyone can check, while it is happening, not years later.

Where it stops

A number is not a remedy. Even after a court called the lane unlawful, no contract was cancelled and no money came back. And whether a price was fair is a judgement a record can’t make, only the pattern testifies.

What you can do

Ask for contracts to be published as they are signed, who won, by what route, for how much. The wrong is rarely in any single contract; it lives in the pattern, so push for the pattern to be visible in real time.

See this case checked against the protocol, step by step →

The vote

A document check that turns real people away

Photo ID at polling stations turned away about 14,000 people at the May 2023 elections, and many more went uncounted, because the law barred staff from recording who was refused. It fell hardest on disabled, unemployed, younger and poorer voters. A minister later admitted it was a “clever scheme” that “came back to bite” his own party.

What a checkable protocol could do

Prove you are allowed to vote without showing a document at the booth, eligibility proven by maths, not a passport. The identity check happens once, when you enrol, and never again at the ballot box.

Where it stops

It moves the gate; it does not remove it. And it can’t decide who is eligible, that is a political choice a protocol must carry out, not overrule. A rule designed to exclude can still exclude, but now it must be argued in the open, not slipped in as routine admin.

What you can do (Great Britain, today)

If you don’t hold accepted photo ID, apply for a free Voter Authority Certificate; it exists for exactly this. Check the accepted-ID list before an election, and help someone who might otherwise be turned away.

See this case checked against the protocol, step by step →

For the advocate

If you fight cases like these for a living, in a law centre, a campaign, a chambers, the corpus behind this page is built to be used, not just read: fifty-two cases, 140 cited precedents, every one machine-validated. Four things worth knowing.

How to read a scenario as a worked remedy

Each scenario is a case file: the documented situation, then a step-by-step walkthrough in which every act of power must produce a record conforming to two small public schemas, who acted, under which cited rule, when, witnessed by whom. The walkthrough is the remedy made concrete: it shows which record, required at which moment, would have changed the case, and, just as precisely, which harms no record reaches.

The evidentiary spine: 140 cited precedents

A scenario cannot pass validation without citing documented real-world precedents, inquiry findings, judgments, audit and ombudsman reports. Across the fifty-two cases there are 140 such citations. Nothing rests on a hypothetical: every claimed failure happened, on the record, to real people, and each card names its sources.

How verdicts are reached, and why most are not victories

Every case ends in one of three verdicts. Holds: the protection fully works (one of the sixty-two). Strains: it helps, and a hard limit is named (fifty-three). Breaks: it honestly fails (eight). The limits are findings, not caveats, a corpus that only reported wins would be advertising. The whole argument in one page: necessary, and not enough.

Propose the next case

Cases are chosen by a triage that shows its working: process-integrity failures are in scope; cases that would need the protocol to judge substance, who is right, what is true, are refused, with the refusal recorded. If a case you know has the shape of these, a record quietly absent, altered, or ignored, propose it, and the triage will answer in the open.

Part of something larger

These three are drawn from a wider set: real situations walked through the protocol step by step, each machine-checked against the same rules, each honest about where it holds, strains, or breaks.