U.S. banking conditions improved year over year through Q2 2026, but the improvement was uneven across institutions and balance-sheet dimensions. Stronger banks sit high in their historical range, while the weakest end remains much closer to normal; smaller institutions continue to face Asset Quality and CRE pressure even as capital and securities positions improve; and relative funding vulnerability remains concentrated in a small, rotating, asset-light cohort.

Five findings

Every tile summarizes a governed finding already published on p4-p12. No new calculation, threshold, ranking or interpretation originates on this page. Each tile carries its own basis and denominator: year-over-year change, point-in-time cross-section and historical rank appear side by side here, and the populations behind them differ. The weak composite tail and the run-vulnerability watch zone are separate definitions over separate denominators and are not comparable cohorts despite sharing a visual form.

No grade tile and no dashboard composite appear on this page. Grades are peer-relative and their interpretation is governed on p6; constructing a headline index whose only home is this page would add a number the report has not tested.

Exhibit 1

Grey mark is 2025Q2, teal mark is 2026Q2; the connecting band is the shift. Measured on the absolute composite score, not on grades — grades are peer-relative and their distribution is close to fixed by construction, so only the absolute score can show whether the underlying distribution moved. Matched population: 4,286 institutions reporting in both periods.

Exhibit 2

Institutions are classified using the report’s calibrated flat band — the 25th percentile of absolute historical year-over-year change, excluding the reporting period — recomputed at the level being classified (0.0583 for individual banks, from 78,253 bank-quarters). The cohort-level threshold (0.0204) governs cohort aggregates and is not applied to individual institutions: averaging a cohort shrinks dispersion, so using the aggregate threshold bank by bank would mark 90% of institutions as having moved.

Takeaway

Exhibit 3

Large figure is the year-over-year change in the category score; small figure is that category’s contribution to the cohort’s composite change. Breadth measures participation in a category movement, not claimability. Some categories — especially Concentration Risk — are mechanically coarse and therefore move for relatively few banks across all cohorts.

Exhibit 4

Bars step cumulatively from zero to the cohort composite. Muted, dashed bars are diagnostic-only contributors — shown rather than hidden, because withholding the largest measured driver without displaying it would let the remaining bars imply a complete explanation. △ marks narrow breadth.

Cohort takeaways

Exhibit 5

Denominator: 4,286 matched banks — institutions reporting in both periods, from a 4,477-bank prior population (95.7% matched). Cells show bank count and share of the row. Shaded diagonal = unchanged; green = improved; red = deteriorated. Largest corridors: C→B 282, B→C 265, D→C 201. Cross-zone movement was close to offsetting: A/B→C/D/E 416 against C/D/E→A/B 405.

Exhibit 6

Denominator: 4,296 full eligible population — a different denominator from Exhibit 5, which requires two observations per institution. The two are stated separately and never combined. This distribution describes where the population sits within its peer groups today; because the grade bands are defined on cohort percentile, it is not a measure of whether the banking system is historically stronger or weaker.

Takeaway

Exhibit 7

Cells show the year-over-year change in the absolute category score. Contribution to composite is deliberately not repeated here — it is the subject of Exhibit 1. Bands are reporting subdivisions nested inside the existing BankAtlas size cohorts; they do not alter peer-percentile construction, grade assignment or any production scoring cohort, and sub-band counts reconcile exactly to their parent cohort.

Exhibit 8

Vertical rule marks the calibrated materiality threshold (0.02039). Values close to the rule on either side should not be read as categorically different; the threshold is a calibrated reference, not a cliff.

Takeaways

Exhibit 9

Axes use existing BankAtlas production thresholds, not cutoffs selected for this report: rvi_pctile >= 90 and comp_pctile < 30. Both axes are PERCENTILE-defined positions in a contemporaneous peer distribution, not absolute economic cutoffs. Quadrant boundaries move with the population. Asset share is shown alongside institution count because the share of banks above the RVI P90 line is close to fixed by construction.

Reported only for $1-10B, the sole cohort where both the reporting coverage and the number of watch-zone institutions support a comparison. Suppressed: <$1B (0.3% cohort coverage, 2 zone banks reporting); $10-100B (99.2% cohort coverage, 9 zone banks reporting); >$100B (100.0% cohort coverage, 2 zone banks reporting). Among $1–10 billion institutions, where uninsured-deposit reporting is sufficiently complete, watch-zone banks show lower median uninsured-deposit reliance than other banks in the same size cohort. Coverage is insufficient to generalize this relationship to the broader watch-zone population. Liquidity, core and brokered deposits and unrealized losses are excluded from this comparison because they are inputs to the run-vulnerability index itself, and CRE because it is an input to the composite; reporting them here would restate the zone definition rather than characterize its members.

Exhibit 10

Matched population 4,286. Zone size 339 → 349 (+10 institutions). Entry and exit describe movement of institutions across a peer-relative boundary. They do not measure a change in the level of system-wide funding risk.

Takeaways

Exhibit 11

Change is the median of bank-level year-over-year changes and therefore need not equal the difference between the two cross-sectional level medians. Color is set by each measure’s frozen direction, so a rise in equity and a rise in CRE are colored oppositely despite both being increases.

Exhibit 12

Share of matched institutions moving favorably, approximately flat, or adversely, using a flat band calibrated per measure from its own historical bank-level year-over-year dispersion, excluding the reporting period.

Supplemental

Uninsured-deposit reliance rose among reporting institutions, adversely for roughly half of banks above $100 billion. This runs counter to the broader balance-sheet improvement and is shown separately because its reporting population is not the population used elsewhere on this page.

Takeaways

Exhibit 13

Each bank is scored on how many of the six core pressure families moved adversely year over year, using the same matched panel, the same measure-specific flat bands and the same direction contract as Exhibit 11. No new threshold is introduced and no stress cut-off is defined. The expected distribution is the Poisson-binomial implied by each family’s own observed adverse rate — what the count would look like if the families moved independently within institutions.

Exhibit 14

The benchmark is rebuilt within each cohort from that cohort’s own six adverse rates, so unequal rates across sizes cannot create or mask apparent dependence in the pooled figure. Every cohort fits its own benchmark: the largest chi-square is 6.5 against a 5% critical value of 12.59 at 6 degrees of freedom. Percentage-point cell deviations are scale-dependent and are reported for transparency only. In $10-100B, >$100B a single institution shifts a share by several points, so the chi-square, which carries the sample size, decides.

Takeaway

Exhibit 15

Tail defined by comp_pctile < 30, the production A–E banding cutpoint separating D/E — a position in a contemporaneous peer distribution, not an absolute economic threshold. Membership moved from 1,280 to 1,284 institutions. Entry and exit describe movement across a peer-relative boundary. A bank can cross because it changed, because its peers changed, or both. Grade composition (D 856, E 428) is descriptive metadata only: the percentile rule is the D/E boundary, so the mix is derived rather than evidence.

Exhibit 16

Compared within size cohort, never pooled, because the tail is 75% sub-$1B and several of these measures vary strongly with size. Suppressed cells name the gate that failed rather than appearing blank.

Takeaways

Exhibit 17

Higher score = stronger. Absolute composite score only; grade shares and aggregate run-vulnerability are excluded because both are defined against a contemporaneous peer distribution, so a historical rank of either would measure the construction rather than the banking system. Dashed line marks the reporting quarter.

Exhibit 18

Each bar is the share of the 93 prior quarters falling below the current reading; the vertical rule marks the historical midpoint. The scored population ranges from 4,296 to 8,445 institutions across the series and stands at 4,296 today — the bank population has roughly halved since 2003 through consolidation, so a historical comparison spans a shrinking and changing population, not a fixed panel of institutions.

Listed because they emerged from the series, not selected in advance: the six lowest median readings in 94 quarters all fall in 2008–2010. They are shown for scale only. Nothing here equates the current environment with those periods, and no future outcome is inferred from historical similarity.

Takeaways

Registry

Shared outcomes

Stating these once prevents a row from quietly defining persistence differently from its neighbour, and keeps a failure to detect movement from being recorded as evidence that movement stopped.

Contents

Controls

This issue

Exhibit numbers run in one sequence across the issue. Numbering is final.

Disclosure

Source: BankAtlas, FDIC Call Report data. Data through Q2 2026. BankAtlas grades and scores are proprietary analytical measures and are not credit ratings.