◎The Label Lens · one store, — lenses, — rooms below

Architecture · one design, — lenses · every figure read from this kit’s own run records

Your data is labelled.
Nobody has checked the labels in three years.

A data store carries a label that says how sensitive it is. Two other systems read that label to decide what to do — one masks the sensitive spans inside a document, the other decides who may open it at all. Neither of them ever checks whether the label is true. This is the whole of the system that does, drawn once as — boxes, and read — ways. Pick a lens. The boxes stay where they are.

Start here · 90 seconds

Welcome. Press Play and the page walks itself.

What this is

One worked question — “is this store labelled correctly?” — and the whole of the system that answers it. Every figure you will see is read from this kit’s own run records, not typed onto the page.

What you are about to watch

It moves stop by stop. The — boxes never move: each stop lights the ones it is about, redraws the band underneath to whatever that lens measures, and says in words what just happened.

Why — lenses

Because — things are worth saying and a flow chart has an axis for none of them: what the two labels are, what the evidence was, what was already free, which half of the guardrail the money actually moved, what the system refuses to do, and what a wrong label breaks downstream. That count is this use case’s number, not a fixed one.

What you should come away knowing

The paid call found — contradictions where a free rule that opens nothing found —. And it is not clean: one store was cut off mid-answer, counted wrong, and left in the denominator rather than re-run. That is published, not hidden.

Play runs the stops of the lens you are on · ← → step one at a time · pick any lens on the rack below to jump to it · — boxes open a second level — the sample pack and the prompt.

One store travelling from its recorded label through a sampled reading to a derived tier and a named contradiction

The quantity this lens measures

Look inside

— boxes open a second level — use a button, or double-click the box.

Every figure on this page is read from this kit’s own run records rather than typed here: the — graded category cells, the — citations, the — stores and the — measured bill are re-derived on every build from build/measured/runs/, the same records the kit board publishes from. The eight categories, the four tiers and the three verdicts are the schema the corpus carries. Every score is the re-checked column — the tier and the contradiction are derived in code after the call, so the first pass has nothing to score.

The masked columns in both rooms below are the output of the kit’s own masker, run over the real corpus and committed as data. Nothing on this page hand-writes an “after” column. No model is involved in masking at any point: it is pattern substitution, deterministic, $0.00, and identical on every run.

DS-CLASS-2026 is invented for this kit and is written as one organisation’s own internal data-classification standard. It is not a regulator’s rule, a statute, an industry specification or any real organisation’s register, and nothing in it may be quoted as governing any real store, label or disposal decision. The corpus is synthetic, from a fixed seed: every name, identifier, card number, address and care statement in it is fabricated and belongs to no real person.

Re-scoring any of this costs nothing, because every grader is code and no model judges anything anywhere in this kit.