Grouped multiverse report v0.1

Multiverse results can contain estimates with incompatible units. A subtractive delta-F estimate and a divisive delta-F/F estimate may both be defensible, but their magnitudes, medians, and direction fractions must not be pooled.

MultiverseReportGroup makes that boundary executable:

from fiberphotometry import MultiverseReportGroup

groups = (
    MultiverseReportGroup.from_choice(
        result,
        name="Divisive normalization",
        units="ΔF/F",
        node="normalization_window",
        alternatives=("divide_standard", "divide_early", "divide_displaced_baseline"),
    ),
    MultiverseReportGroup.from_choice(
        result,
        name="Subtractive normalization",
        units="acquired fluorescence",
        node="normalization_window",
        alternatives=(
            "subtract_standard",
            "subtract_early",
            "subtract_displaced_baseline",
        ),
    ),
)
result.write_grouped_html("robustness.html", groups)

The renderer refuses to proceed unless every compatible universe belongs to exactly one group. Overlaps, omissions, unknown IDs, and attempts to assign an incompatible universe are errors. Each evidence lane calculates its own estimate range and success/failure/blocked counts. No pooled multiverse summary is displayed.

Failed lanes receive an explicit empty state and amber limitation edge. Declared incompatibilities remain in a separate ledger, while the complete compatible ledger retains estimates, intervals, choices, failures, and blocking reasons.

The report is self-contained, accessible, printable, and follows the persistent interface system.

The deterministic synthetic report used to evaluate whether scientists understand this contract is generated by examples/grouped_multiverse_report.py. The frozen formative usability study separates participant comprehension from implementation assumptions and retains failed review outcomes.

Each group may also declare a unit-local smallest_effect and direction. The machine-readable robustness-summary.json reports lane membership, counts, estimate range, median, direction fractions, threshold, and practical-effect stability. Failed and blocked compatible workflows remain in fraction denominators, so execution loss cannot make robustness appear stronger. The artifact follows the normative multiverse-lane-summary-v1 schema.