Skip to content
mediaite-ghostink

ADR-008: Observational limits and causal identification (M-22)

Status

Section titled “Status”

Accepted — methodology documentation (2026-04-26).

Context

Section titled “Context”

The pipeline compares stylometric and embedding time series before and after a calendar split date. It does not embed a randomized treatment, an instrument, or a difference-in-differences design with parallel trends.

Decision

Section titled “Decision”

Findings are observational. Outlet-wide CMS changes, beat reassignment, house style, syndication, and staff-mix shifts are alternative explanations that cannot be ruled out by the current feature battery alone.

Consequences

Section titled “Consequences”
  • External control corpora (see M-08 / config.external_controls.example.toml) and human spot-checks remain necessary for stronger claims.
  • Optional DiD or natural-experiment scaffolding may be added later as separate ADRs; it is not implied by the existing stage contract.
Section titled “Links”
  • Punch list: M-22 (prompts/punch-list/current.md)
  • Guardrails — serial correlation caveat (M-16): docs/GUARDRAILS.md