Institution

The eight domains

Every domain is a collapsible panel, numbered in the canonical order. Only the first opens by default; the rest load when you open them, so a slow domain never holds up the page.

• 01 Capital — leverage, CET1, tier 1 and total risk-based capital, the conservation buffer, and the CBLR election where one is on file.
• 02 Lending capacity — concentration against capital, including the supervisory CRE and construction-and- development measures.
• 03 Deposit franchise — the funding base, its composition, and the uninsured share.
• 04 Liquidity structure — how the loan book is funded and what it leans on.
• 05 Securities and rate — the portfolio, unrealised positions carried with their sign, and the AOCI treatment.
• 06 Credit — asset quality, the allowance, and provisioning.
• 07 Earnings quality — returns, adjusted where a sub-S election makes the raw figure incomparable.
• 08 Efficiency and scale — cost structure relative to the group.

Each domain header states how many signals it holds, the quarter they were read at, whether a trend is drawn, and how many are in breach. Below the list, a provenance line names the source, the lane, the as-of quarter, the engine version, and whether the underlying store has been validated — with a date only when it actually has been.

The five states — and the one that is not a state

Every signal carries one word, in colour but never colour alone:

• Breach — the value is on the wrong side of a published supervisory level.
• Approaching — near a published level. There are no invented warning bands; if a regulator did not publish the band, the product does not draw one.
• Drifting — moving consistently in one direction over the window, independent of any level. Only some signals are structurally eligible for drift; the rest say so rather than implying stability.
• Passed — evaluated against a level, and clear of it.
• Not evaluated— off the severity scale altogether. It never means "passed". A CBLR elector, for example, shows not evaluable — CBLR election on the risk-based ratios the framework makes irrelevant.

Reading a signal row

A row gives you the name, the value, and — where the measure is thresholded and expressed in percent — a meter showing the value against the published level. Under it sits the basis line, verbatim: the rule, the level, and where the level comes from. A value derived rather than filed is tagged estimated.

Peer context sits on the same row. Where FFIEC publishes a peer figure it renders with the peer group, your percentile within it and the group size — labelled a trimmed average, theirs, because that is what FFIEC computes. Where you have an active group of your own, that line takes the row instead and reports an untrimmed median, ours plus your ordered position among observed values. See Find & Compare Banks.

Sparklines and the period control

The period control at the top of the shell sets the trend window: 4Q, 8Q, 12Q, 40Q. It is page-local and always opens at the shortest — it is a lens, not a saved preference.

A sparkline draws only from quarters that were actually filed and reported. Fewer than two drawable points and nothing is drawn at all rather than a line implying data that is not there. A direction arrow comes from the last quarter's move, and the line itself is muted: a trend makes no claim about good or bad. Where a merger completed inside the window, the quarter is marked — a step change is not organic growth.

Click a sparkline and it opens larger, in an overlay: the same series with a value axis, quarter labels, a hover readout and a table of every filed quarter behind it. The overlay carries its own window toggle — the same 4windows, scoped to that chart only, so widening one measure does not move the page. Escape or the backdrop closes it and focus returns to the sparkline you came from. Nothing changes about the honesty of the drawing: the line stays muted, no threshold band is added, an unreported quarter is a gap and never a zero, and the coverage line says how many of the window's quarters were published or filed.

The inspector

Select any signal and the inspector opens beside it. It is the whole audit trail for that one number, and it opens on the part of it a reader needs first:

• The status and the sentence explaining why that status was assigned.
• The trend window — first and last quarters, the values at each end, and how many of the window's quarters were actually filed.
• The drift trace, where the signal is drifting.
• The basis and its citation, companion measures, context that is explicitly never a threshold, and what the measure cannot see.
• The FFIEC peer group block, including how your percentile has moved and whether your peer group itself changed between quarters — and your own group's block, with its definition version and member count.

Under those sits Technical detail, closed until you open it: the formula, the derivation inputs as a table (every MDRM code, its description and its value, with "not reported" distinguished from a reported null), computed versus reported so you can see where our arithmetic and the filing disagree, the FFIEC published code, and any seams in the data. Nothing honest has been removed — the codes are one interaction away rather than in front of a reader who wanted to know whether a number was a problem.

Proctor: explain this

Every inspector carries a Proctor: explain this button. It is on request only — nothing is generated because you clicked a row — and what comes back is a short plain-language reading of that one measure, marked as generated and carrying its own provenance line. Institutional and above

Two guards run on every answer before you see it. The first refuses any figure that is not in the block of values the surface handed over, so the narration cannot contain a number the engine did not compute. The second refuses the vocabulary of advice, prediction, ratings and invented watch levels, and refuses to relabel the published peer figure — FFIEC computes a trimmed average and the narration has to call it one. If either guard trips, the explanation is withheld and the surface says so: the value, its basis and its citation above it are unaffected, because they come from the filing and not from the explanation.

Hover an acronym

Regulatory prose is acronyms end to end. Every term the surface actually renders — 124 of them today, including the call-report and published-UBPR code prefixes — carries a definition on hover and on focus, wherever it appears in a measure name, a basis line, a status reason, a peer line or a provenance footer.

A definition is one factual line. It never says whether a number is good, never names a level to aim for, and never tells you what to do: a hover is the least-reviewed text in any product and therefore the worst possible place to keep an opinion. The supervisory levels stay in the basis line, with their citations, where they can be argued with. Copy that has no matching term passes through untouched.

The four panels below the domains

Under the eight domains sit four more panels. They load on open, like the domains. Institutional and above

• 09 Observed deposit beta — how deposit cost actually responded across each observed rate cycle, with the arithmetic shown in full against the published benchmark path. It is evidence, never a parameter you set.
• 10 What changed — the quarter-over-quarter diff in three tabs: values, statuses, and structure. Structure covers the things that are not a number moving — identity, form type, merger seams, filing vintage, a signal appearing or vanishing.
• 11 Obligations— this institution's slice of the ledger, with its own coverage register saying which signals feed it and why the rest do not. It is one of the two entrances to the Work ledger; see Dashboard.
• 12 Cohort bands — where this bank sits within its group, as bands whose width is the number of institutions in them, with the observed values at each edge. No status chips appear here: position is not a verdict.

Your own bank reads as a position

There is no second reading surface for your own institution. Open your own bank and the view you already know grows additional bases; open anybody else's and it is exactly what it always was. A separate readiness page would be a second place for the same number to live, and two places eventually disagree. Your Dashboard is the home for this reading and links straight into it. Institutional and above

Ownership is not something a link can claim. The view attaches your record only when your account, your entitlement and your organization's recorded relation to that certificate all say it is yours. A demonstration bank is not ownership, and nothing folds in on one. What is added, and nothing else:

• The rule basis, on the measure's own row. Where a rule in the call-report rule book reads a measure, its verdict is folded in underneath as a continuation of the row — the rule's name, the state it reached, the basis sentence the engine itself renders, the citation's honesty note, and a link into the ledger filtered to that measure where work can genuinely exist. A rule the validation programme will not let assert a verdict appears as context, never as a finding. A measure no rule reads carries no rule basis and is not badged.
• Coverage, per domain, at the top of the panel. How far the rule book reaches these measures, split by what those rules may claim; which posture area is bound to the domain and how much of it has an answer on record; and what your own figures would add. One domain can be fully rule-covered while the next carries only published figures, and that difference is the point. Counts of rules that could fire are never phrased as counts of rules that did.
• Your recorded posture, beside the numbers. A posture row has no value, line or meter, so it does not borrow the shape: a statement taken verbatim from the specification, a recorded answer — recorded yes, recorded no, or nothing recorded — its citation, and a link that opens the field itself. Only the two posture areas unambiguously about one domain are bound here; everything else is deliberately unbound and says so. The full record is in Prepare.
• The run panel, above the domains. The provenance for everything folded in below it: when findings were recorded, how many, which quarter was read, which feed produced them — or the honest statement that a run which finds nothing writes nothing, so an empty record cannot be told apart from an engine that was never run here. You can run the engine from the panel, and that is the only write on the page.

A run records findings; it does not persist a row for every rule that passed. So a rule with nothing on it is one of three genuinely different things — a finding is on record, no finding was recorded for this rule in the run on record, or nothing is on record at all — and the surface never collapses them. There is no branch in which a silence renders as passed.

On your own bank the measures inside each domain are ranked: status first, then rule-based above peer-relative, then proximity to the line, then rate of travel. The eight panels themselves are not ranked — domain order is fixed, Capital first, and the surface says which of the two orderings is active rather than leaving you to infer it.

The public profile of the same bank

Every institution also has a public profile — the same artifact, run in a reduced mode, indexable, and reachable without an account. It shows all eight domains at the current quarter only, with full inspectors, meters, basis lines and the FFIEC peer line, because all of that is public data.

What it withholds, and says it is withholding:

• History. No trend series is fetched at all; the period control renders as a greyed label.
• Cohorts. It never loads a group, and the rows where a group reading would land carry a label saying so.
• The four panels below the domains — deposit beta, what changed, obligations and cohort bands.

It is useful in its own right: it is the link to send someone who does not have an account when you want them looking at the same figures you are.

A first pass, in order

1. Open the institution view from the rail. The identity line confirms which bank, which form type and which quarter you are reading.
2. Scan the domain headers for breach counts before opening anything — that is the whole picture in eight lines.
3. Open a domain and read the rows: value, meter, basis, status.
4. Select the row that surprised you and read the inspector top to bottom. If the number differs from your own, the derivation inputs table is where you will find out why.
5. Widen the period control to 8Q or 12Q and look at the same row again. A number that is fine and moving is a different fact from a number that is fine and flat.