The constitution of the machine
The laws.
The standing rules this machine runs under, published in full. Each law is enforced in the
schema or the pipeline, not in a policy nobody reads. Most were paid for — with a
demonstrated failure, or a refused temptation. A law changes only by an explicit ruling,
dated and recorded. What this machine refuses is why it can be trusted.
I
The record
Law 01
Receipt or it does not exist.
Every row in the spine carries the address it was fetched from, the moment it was fetched, and a hash of what was fetched. Every published figure wears its year and a receipt any reader can replay. A claim without a receipt is not published, however plausible it looks.
Why The entire product is trust. One unreceipted sentence makes every other sentence deniable.
Law 04
No model touches the harvest.
Harvesters fetch, parse deterministically, and store verbatim. No language model writes, rewords, or summarizes a record on its way in.
Why A model in the ingest path can silently rewrite the record. Verbatim-in is what keeps every downstream claim auditable.
Law 05
The model has four jobs, each checked.
Parse documents into structure, quoted verbatim. Compile a reader’s question into register joins. Explain nominated residue in plain language, a second pass checking the first. Investigate nominations, every sentence held against a receipt. Nothing else.
Why The system reduces; the model reads the residue; a second pass checks. A model that free-answers from the pile is a horoscope with a database.
Law 16
A read has three states.
Rows are the record. Empty means the harvest holds nothing, said plainly. A failed read says READ FAILED and the surface stays silent about the value — never a zero, never a stale figure.
Why A catch that returns empty is a false all-clear. A zero standing in for a failure is the same lie with better manners.
Law 27
The intake test.
Before any dataset ships, four questions. Is it official and receipted. Is the subject an institution or an aggregate, never a person. Is the metric defined, with known coverage. Does it join the spine on place, subject and time. Four yeses and it ships. Any no, and it waits.
Law 28
The gate.
No second country and no next dataset until the layer below is green on Denmark. Coverage is stated on the first screen, at full size, and never implied beyond what the ledgers support.
Why The vision is a map, not a backlog. The first reader who catches this machine implying coverage it lacks has broken the whole thesis.
II
Persons
Law 02
Institutions, never persons.
The marks of this system are places, institutions and documents. No person is scored, ranked, flagged, tracked or profiled. Where a public register prints a person’s name, the name renders exactly as the register published it — no portrait, no profile, no judgment attached, and no join that could build one.
Why This is the founding refusal. The person-scoring version of this machine already exists, and the Nordic states refused it. The person line is not a handicap; it is the license to exist.
Law 22
Aggregates, not individuals.
Official aggregate statistics and institutional performance pass the intake test. Tracking an individual does not, ever. Finding people is the police’s lawful job, not this machine’s.
Law 24
A tip is a search instruction.
A tip is never a receipt. Nothing publishes unless the receipts stand entirely without it. Nothing identifying is held, by architecture. This is not a legal whistleblower channel, and it says so, pointing to the official ones.
Law 26
The duty lands with the authority.
Counterparty integrity renders at company level — companies are legal persons — and as the institution’s duty: the buyer is legally obliged to verify exclusion grounds, and the record shows whether verification happened. Person-level checks land with the authority that lawfully owns them.
III
Verdicts
Law 03
The verdict happens in the reader.
The system renders, bands and receipts; it does not conclude on contested ground. No composite score of a place, ever — a composite is a verdict wearing a number.
Why Any verdict printed here becomes our politics. Rendered positions let every reader keep their own.
Law 06
Peer bands before flags.
No per-capita figure may flag outside a band of its true peers, and the band is named in the sentence that flags.
Why An early detector confidently named the worst-administered communes in the country. Every one was a small island. An outlier detector without peers only ever discovers smallness.
Law 07
No number travels alone.
Every figure renders beside its sibling measures, its definition, its baseline and its deflator. Money renders in real terms by default, deflated by the official index, receipted.
Why Nobody fabricates official numbers; everybody selects the flattering frame. Selection dies when all the frames share one wall.
Law 08
The mirror, not the pillory.
Every detector that can find the worst in a band must surface the best. A flag never renders without its peer-proof — the place of the same shape that already solved this, at a documented cost. Improvement is a first-class public event.
Why The peer-proof converts an accusation into a referral. This exists to do actual good, not to point fingers.
Law 09
Solutions age.
Every peer-proof wears its date and its what-changed-since receipts — statutes, prices, technology. Right then, wrong now is the reader’s conclusion from rendered differences, never this machine’s sentence.
Law 10
Comparison only where definitions align.
The Nordic statistical offices define differently. A cross-border comparison renders only where the definitions genuinely align, both of them receipted; where they do not, the coverage ledger says so.
Why A naive cross-country comparison is lying with numbers — the kind of lie this machine exists to catch.
Law 21
Contested ground.
Official indicator series render with their published definitions, whoever they flatter or embarrass. The machine must be equally ready to embarrass every narrative. A causal arrow onto contested science prints only under the three-arrow law, and a question that is a verdict wearing a math costume is refused.
Law 23
The institution, not the operation.
Defense renders as an institution — procurement, budgets, public audits, commitments against spend. Operational tracking is refused.
Law 25
Simulations are branded.
A simulation always renders marked as a simulation, runs on deterministic scoring, and is never presented as evidence about real people.
IV
Money
Law 17
The same page.
The subject of a record reads exactly what everyone reads. No secret dossiers, no analyst-only views, no early access to findings. Institutions pay for access — API seats, briefs, integrations, service levels — never for exclusive truth.
Law 18
Findings are un-buyable.
Paying never touches a flag. Only an actual repair does, measured from the registers. The pattern computation is blind to the customer table by construction.
Why Without this, the whole thing reads as a protection racket — and deserves to.
Law 19
Never an invoice.
Outreach is a proposal wearing its price, never an unsolicited invoice. Pitch and price never travel apart.
Law 20
No reply channel on the record.
Officials’ public statements are harvested as receipts. Corrections flow through the public corrections ledger, in the open. Help flows out — briefs, benchmarking, open exports of an institution’s own rows. Verdicts never flow in.
V
The forward field
Law 11
The three arrows.
A causal chain renders only when every link is receipted and every arrow declares its source: the actor’s own stated reason; legal or mechanical necessity; or bare temporal sequence, drawn dotted and labelled — sequence, not cause. A statistically inferred arrow is never printed as fact. The record does not explain this is itself a valid finding.
Law 12
Never an invented forecast.
The forward field has four lawful fills: the institutions’ own published projections; their forecast track record; labelled arithmetic on the record’s own numbers; and peer history as the future’s receipt — the place that already hit this wall, its costs documented in its own audits. Nothing else is drawn past now.
Law 13
Deadlines are theirs, or arithmetic.
A before-a-given-year sentence renders only when the deadline is a statute date, the institution’s own projection, or a labelled arithmetic crossing of a red line the law itself defines.
Why Borrowed thresholds make the law the judge, not this machine.
VI
The learning loop
Law 14
The three ledgers.
Coverage: every number wears its coverage, and the system states its own blind spots in the interface. Corrections: this machine’s errors, found and fixed, in public. Track record: every claim written in falsifiable form with a check date, rechecked against fresh harvest, precision published with the misses included.
Why The misses are what make the hits believable — the exact number the industry never publishes.
Law 15
Thresholds refit on our own data.
Never another country’s constants. Bands re-derive on cadence from the harvested record; rules are versioned like law and change only through the same door.
Law 29
The learning loop.
Detector versions are first-class: every claim registers the version that made it, and each version’s curve is public. Proposed refinements are proposals, never rewrites; a new version runs in shadow beside the old on the same matured claims and is promoted by receipted comparison, never by taste. Every matured miss is classified and joins the detector’s golden set. Claims score on boldness against their registered base rate, never on raw accuracy.
Why A raw hit-rate target breeds cowardice, or moved goalposts. A public curve that starts low and climbs is a better trust story than a flat number, because it is visibly real.