CALIBRATION / v9.0
CAFE Calibration Ledger
A concise ledger of what CAFE is now calibrating, what remains provisional, and which boundaries are constitutional.
Commission state
CAFE v9.0
Attention Field: active
Axes: 12 independent domains
RABBIT: v0.4 all-axis atlas
MSC / FI: archived
Current calibration target
Calibration now focuses on event clustering, source-family diversity, topic classification, semantic lineage, receiver health, and source-role disclosure — not global estimates of how internet users spend their time.
Event geometry
Related headlines are clustered using lexical overlap. A cluster receives greater attention weight when independent receiver families converge on it. Outlet repetition inside one family adds only modest additional weight.
Edge damping
Edge receivers carry a disclosed attention-authority factor below Core Press to prevent a deliberately dense ideological panel from overpowering the field by construction. This is a sampling correction, not a truth judgment.
What remains provisional
Lexical axis assignment, cluster thresholds, family definitions, edge attention-authority weights, and RABBIT semantic vectors remain calibration decisions. They should be revised from observed failure modes and documented rather than silently tuned.
What is categorical
Attention cannot upgrade evidence. EDGE cannot become a RABBIT source through popularity. Editorial affinity cannot rescue a path that fails semantic/evidence admission.
Archive note
The v8 MSC/FI calibration ledger is preserved by the installer backup and the repository history. v9 intentionally stops investing in population-equivalent Mode Spectrum calibration.