Screener

Mode one — criteria and measures

Start with the identity criteria: states, charter classes, an asset range, active institutions only. That alone is a screen. On top of it you can add up to five measure clauses — a measure, a direction, and a value — which all have to hold together.

The thresholds are entirely yours. There is no placeholder and no suggested value, because a suggested threshold is a threshold the product asserted without a citation.

Two things the results always state. First, the result set is capped, and the cap binds on the identity match before the measures are read — so the surface tells you the measure screen ran over the largest N, not the whole population. Second, an institution that does not report a measure you filtered on is excluded and counted, never treated as failing. CBLR electors not reporting a risk-based ratio is the common case, and the exclusion line names it.

Saving a screen: live group or snapshot

What you get when you save depends on what you screened on, and the two outcomes are labelled differently because they behave differently.

• Criteria only saves as a live criteria group. Its membership re-evaluates as institutions move in and out of the criteria.
• Any measure clause saves as a snapshot — a manual group frozen to the institutions that passed at that quarter. Group membership evaluates on identity, not on measures, so a measure screen cannot be a live rule. The full screen definition is stored with the group so you know what produced it.

Opening a result, and getting back to the list

A completed screen is written into the address bar — the definition, never the rows. So clicking a result to open that institution and then pressing Back returns you to the same screen, re-run: the results you come back to are read fresh rather than restored from a stale bookmark, and the screen re-runs exactly once on arrival. Re-running a screen replaces the current address rather than stacking another history entry, so Back stays one step from where you were.

A definition the surface cannot parse is refused outright rather than half-restored. Results that do not match the definition on screen would be worse than losing the run.

Mode two — complementary screening

Complementary screening starts from your filed position and finds institutions that sit on the other side of it. It needs a subject institution and refuses to run without one, because there is no complement to nothing.

The pairings are declared, not inferred. Before you run anything, each one shows you the axis it works on, which side counts as the constrained side, why the two sides are complementary, and — in its own labelled block — what that pairing cannot see. You switch pairings on and off; the product does not weight them.

The only aggregate on a result row is a count of the pairings that matched. There is no score and no index. The "complements on at least N pairings" control is described on screen for what it is: a display filter over stated facts, not a threshold the product asserts.

The diligence read

From any complementary result you can open a diligence read on that institution. Its order is deliberate and worth knowing before you use it:

1. What public data cannot see, first and in full. The blind spots come before the findings, so nothing below is read as complete.
2. Questions their filings generate — each with the filed figure that produced it, why it fired, what to ask for, and a citation.
3. Further documents, each tied to the gap that produced it.
4. Regulatory approval considerations, from filed facts only — including the case where the enforcement register cannot see that institution's supervisor, which is deliberately not skim-readable as a clean record.
5. Coverage — which of the eight domains could actually be read.

Zero questions is explicitly not a clean read, and if the read fails nothing partial renders: a shortened blind-spot register is worse than none at all.