Construct & measurement-method paper · OSF
A deterministic index of challenge intensity in an AI cognitive contradictor — the Δ-CSI (Cognitive Sovereignty Index, delta)
| Author | Francesco Saverio Canepa |
|---|---|
| Type | Construct & measurement-method paper — version 1.0 (23 July 2026). No empirical results. |
| Status | Public OSF project · osf.io/ru3qk · DOI 10.17605/OSF.IO/RU3QK · CC BY 4.0 |
| Specification | Δ-CSI Specification v1.0 — Appendix A of the paper, published as an open standard |
| Companion | Cognitive Sovereignty (the construct) · Calibrated Dissent (the protocol, DOI 10.17605/OSF.IO/5QC8T) |
If a cognitive contradictor exists to challenge a decision rather than validate it, an immediate question remains open: how hard did it challenge? Without a reproducible answer, «the AI pushed back on me» stays an impression — and an impression cannot be compared over time or across sessions. This paper defines that quantity and specifies its operational definition.
The Δ-CSI (Cognitive Sovereignty Index, delta) is a deterministic index in [0, 100], computed from five weighted components: implicit-assumption count modulated by mean severity, falsification-test count, burden-of-proof question count, cited-source count, and thesis-versus-counter-scenario divergence. Counts are capped, which precludes gaming by verbosity; weights are constrained to sum to one hundred; the interpretation bands are fixed and the versioning scheme is explicit.
The paper contributes three things. (i) The construct, delimited by four negations: challenge intensity is not decision quality, not the probability of success, not a verdict on the decision-maker, and not whether the decision was subsequently changed. (ii) The specification, together with its reproducibility guarantee, its fail-closed defenses against sycophantic output and a register of its declared weaknesses, two of which are residual limits the author states openly: the self-reported severity channel is gameable by construction, and the count components register quantity, not argument quality. (iii) The comparability and calibration discipline: cohorts keyed by computation version, divergence method and weights, with statistics held to descriptive-only below five recorded outcomes.
What the paper does NOT claim
No claim of empirical validity or effectiveness. The paper does not assert that a high Δ-CSI corresponds to better decisions — nor, conversely, to weaker ones — and it does not assert that the index predicts outcomes: it demonstrates construct definition and measurement method only. Empirical validation is stated as a deferred programme, structured as a pre-registered evidence-accumulation roadmap. This honesty is a design requirement, not a caveat: an index of challenge intensity that overstated its own validation would be inconsistent with the very stance it exists to measure.
Declaration of interest. The Δ-CSI is the metric computed by CounterBrain, the reference implementation developed by the author. The specification is published as an open standard precisely so that the index does not depend on the implementation that produced it: anyone can recompute it, contest it or replace it.
Read the full paper
Full version (PDF), OSF project and reference implementation.
