[Sad Catapella N4][Scientific identity] scope_groups accepts two biological repeats of one condition in the same paired group #42

Audit basis

Normalization Batch 4 at f784c95e3049483f87466150ffc5764375fa631d. Confirmed validation omission, with a direct source-path counterexample. No repository tests executed.

Evidence

• scope_groups() lines 115–125 checks only for missing pair keys and appends every row to the mapped group; _repeat_group() lines 129–135 accepts mapping values without validating per-group scientific uniqueness.
• _incomplete_repeat_groups() lines 150–157 reduces each group to a set of conditions, concealing duplicates within one condition.
• The owning upstream experimental-design contract explicitly requires one biological-repeat folder per condition per repeat group, enforced by preprocessing ExperimentDesign.conflicts. The standalone normalization API need not trust arbitrary mappings to have passed that check.
• Test verifies normal one-to-one cross-condition pairing, not many-to-one repeat conflicts.

Reproducer

Records for control/r1, control/r2, and treated/r3. Mapping {('control','r1'):'A',('control','r2'):'A',('treated','r3'):'A'}. The group A contains two independent control biological repeats and one treated repeat, but the incomplete-group warning is absent (both conditions exist). With pooling='conditions' each condition is weighted equally, yet the control side mixes two distinct biological replicates, violating the specified pairing unit.

Impact

Silent merging of independent biological replicates; normalization references can depend on accidental repeat allocation. Distinct from the FOV-pseudoreplication QC issue #21, which concerns how reference statistics weight FOVs.

Fix / acceptance

Validate the mapping for all present (condition, repeat) pairs before grouping: per resolved repeat-group label, at most one distinct biological-repeat identifier per condition. Refuse with a named actionable error listing all conflicting pairs and their target label. Confirm the group label is nonempty text and no unsafe normalization of label case/whitespace occurs silently; preexisting preprocessing-validated mappings must remain accepted. Do not automatically split or invent pairing. Tests for duplicate records from one repeat (legal), distinct repeats from one condition (error), missing mapping, single group, explicitly paired differently named repeats, and unused mapping entries (document policy). Numba not applicable.