Criterion C6 — Complementation Validity

Validity type	Construct
Pass condition	The construct’s labeled subdivisions are functionally distinct — not just different labels on one mechanism
Evidence family	Causal
Minimum reporting	Named subdivisions, test showing each performs a function the others do not
Common failure mode	Labeling subdivisions by observed behavior without testing functional independence

What this criterion requires

Complementation validity borrows from the genetics complementation test (Benzer, 1955). In genetics, two mutations are in the same gene if they fail to complement — if combining them does not restore function. The analogy: two labeled subdivisions of a circuit (e.g., “name movers” and “backup name movers”) are functionally distinct only if they complement — if one can rescue a function the other cannot.

Satisfied when:

1. The construct has labeled subdivisions. The claim names sub-components or sub-roles.
2. Each subdivision performs a function the others do not. Removing subdivision A breaks function X but not Y; removing B breaks Y but not X.
3. The subdivisions are not redundant copies. If removing either breaks the same function and restoring either recovers it, they are redundant — one construct, not two.

MI example

The IOI circuit names “name movers,” “backup name movers,” and “S-inhibition heads.” Complementation validity asks: do these subdivisions perform genuinely different functions, or are they different labels on overlapping sets of heads? The induction heads case study (token copying) is capped partly by C6 — it is unclear whether “induction heads” and “previous-token heads” are complementary subdivisions or overlapping descriptions of one mechanism.

Connection to the chain

Required for Validated tier. Without complementation validity, a circuit’s labeled parts may be one mechanism described at two granularities.