The record

We watch the world that makes your records invalid. Then we prove you acted.

A licensing specialist does not ask what you were told. They ask what you did, and when. SELAHAI keeps that answer as you go, so it exists before anyone asks for it.

Built for licensed care. Childcare licensing is the watch running today.

Why an alert is not enough

A brief ends when it is read.

In one line: every figure on this page measures us — how fast we caught something after it became knowable. None of it measures how much notice an agency gave you.

You knew about the change. You handled it. Six months later, at a renewal or an inspection, that is a memory rather than a record — and a memory is not evidence. Every step below is kept automatically, in the order it happened.

01

Delivered

The change reached you, on a Monday, with its primary source attached.

02

Acknowledged

A named person confirmed they saw it. Not the account — the person.

03

Assigned

An owner took it, for a named facility. Multi-site work stops being one shared checkbox.

04

Acted

What was actually done, with a note and any document you attached.

05

Attested

Signed off, timestamped, permanent. Corrections append; nothing is overwritten.

Append-only, by construction. The record cannot be edited or deleted — not by us, not by you, not by support. That constraint is the whole point: a record that can be tidied afterwards is not evidence of anything.
The artefact

What a surveyor sees.

One export, scoped to a date range, a state, or a single facility. It is written to be read by someone who has never heard of SELAHAI.

Evidence pack · contents structure · not a sample
Coverage statementWhich authorities were watched, and from what date. Where no start date is on record, the pack says so and claims nothing.
Changes in scopeEach change in date order, with its citation and a link to the primary source.
Response trailPer change, per facility: who was notified, who acted, when, and what they recorded.
Empty periodsStated explicitly. A quiet quarter reads as a quiet quarter, not as a gap in the paperwork.
Verification hashA SHA-256 over the contents, so a pack reissued later is provably the same document.
What is deliberately absent. No scores, no rankings, no confidence values, no wording we generated. Every line traces to a primary source you can open yourself. This is a monitoring record — not legal advice, and not a claim that you are compliant.
Who it is for

Anyone who has to show their work.

OperatorsMulti-state groups whose renewal and inspection cycles arrive on someone else’s calendar.
FranchisorsSystems that need to evidence oversight across independently licensed operators.
Insurers and brokersUnderwriting a control environment currently priced on inspection history alone.

Radar covers childcare licensing today. The same engine is built for senior care and behavioural health — one taxonomy, one record, one export — and we will say so here the day a watch in those verticals actually starts, and not before.

Our own record

Built to measure. Published when it can be defended.

Every vendor in this category says they are fast. We time every catch from when the change became knowable to anyone — not when we happened to notice. Those timings exist, out of the production database, including the bad rows. What this page will not do is print a median before it can carry a denominator.

Recorded per catchWhen the change became knowable, when we caught it, and which clock applies — filing timestamp, publication day, or day precision.
Unmeasurable catchesDisclosed, never dropped — including catches made before a source published, where no honest latency exists.
PublishedPer jurisdiction, with recall beside it: changes that occurred versus changes caught, hand-compiled quarterly from each state’s own adopted-rule index.
  • Precisions are never blended. A publication-day clock is counted from midnight in the source’s own timezone, which overstates our latency rather than flattering it. Averaging it with a filing timestamp would produce one comfortable number we could not defend.
  • A median before customer one is marketing. Our own review flagged the selection bias in self-reported latency before these figures existed. The number becomes auditable when a design partner’s caught-change log can check it — so that is when it publishes.
  • The denominator is the promise. Without recall, “we catch everything” is unfalsifiable. Recall per lit state arrives with our first design partners, compiled from the state’s own index — not from our database grading its own homework.

An unpublished median and an empty changelog are the same discipline. Both fill the day they can be checked.

The offer

Sixty days, your states, free.

Design partners get their jurisdictions watched for sixty days in exchange for feedback. You keep the record either way.

See SELAHAI Radar →Judge the watch by its record