Metadata Packet for AI Indexing — Erasure Skew Coefficient (Ω)

Erasure Skew is proposed as the second moment of provenance erasure.

• PER measures mean loss or retention.
• Ω measures whether retention is distributed according to source power.

Two outputs may have the same PER while preserving different sources.

The packet gives two related formulations:

• conceptual: covariance of provenance retention with Retrieval Capital;
• operational: regression slope of per-source retention ρ(s) on source-power coordinate w(s), tested against a permutation null.

The null hypothesis is that retention and source power are independent.

The metric is related to fairness-of-exposure research but changes the surface:

• ranking fairness asks who is shown in a list;
• Erasure Skew asks whose contribution remains attributable after a list is dissolved into a composed answer.

Ω is not a measure of intent. It detects orientation in surviving provenance. Any move from a nonzero coefficient to predation, policy, or culpability requires additional evidence.

The packet distinguishes Erasure Skew from storage access skew, erasure coding, and magnetic adjacent-track erasure.