[Sad Catapella N7][Scientific QC] Entirely nonfinite channels produce no warning and are labelled Histogram_Exact=true #57

Baseline and classification

Normalization Batch 7 at f784c95e3049483f87466150ffc5764375fa631d. Confirmed existing deliberate blank-statistics policy, but an unassessed-measurement diagnostic gap requiring a scientific contract decision rather than claiming an accidental regression. Distinct from #22 (zero-MAD reference skipped) and #46 (malformed nonfinite tail moments). No tests executed.

Evidence

• _summary() lines 181–192 catches NoValidPixelsError for a channel, sets median/low/high/saturated fraction to None, but sets exact=True, a vacuous exactness declaration for an unmeasured channel.
• intensity_checks() and checks do not emit a distinct missing-intensity diagnostic. _image_outliers lines 194–199 excludes missing values, _saturated_images lines 278–280 skips None, and _no_spread lines 303–306 skips when p_low is None.
• Contract §7.1–7.2 explicitly documents blank channels are not checked. Test verifies blank fields but does not verify a conspicuous warning; test_checks checks clean full-finite data.

Scientific impact

An entirely nonfinite channel (all NaN/±inf) can yield a summary row with missing intensities and Histogram_Exact_Ci=True, and if other metrics are in range, intensity_checks returns no warning. A user consuming only warnings cannot distinguish 'healthy' from 'channel unassessable'. The absence of valid pixels is not a biological zero signal.

Proposed policy/acceptance

Add a distinct channel-specific diagnostic such as no_finite_intensity_pixels / intensity_unassessable, with affected image and channel, and ensure calling bricks present it clearly. Define exactness for missing channels: prefer None / not-applicable if schema evolution permits, or keep True with an explicit paired availability field and prominent warning; do not silently infer a valid zero. Compare with fitting's planes_without_finite_pixels warning to avoid duplicated notifications. Add all-NaN, all-infinite, one missing channel, partly missing Z/T, no-spread-but-finite, and clean image tests. This is scientific communication/QC, not a Numba task.