BenchEWS · Studio 3.0 / ECHO

BenchEWS Studio 3.0 — SCF Development Plan

Public project plan and preliminary requirements · October 13, 2026

BenchEWS analyses warning signals in observed systems. SCF adds a prior question: are observations and missing feedback suitable for interpreting those signals reliably?

PLANNED / SPECIFICATION — not implemented, verified, or released

Scientific basis

The SCF manuscript “When Does Suppressed Feedback Impair Self-Correction?” studies selective reporting, receiver models and early-warning artefacts in synthetic dynamical systems. It is not peer-reviewed and does not study real organisations. Published October 13, 2026 as a Zenodo preprint (v1, DOI: 10.5281/zenodo.23345829). The publicly displayed PDF has 36 pages; its version still needs reconciliation with the previously reviewed 38-page manuscript.

Planned requirements

Observation Integrity

Expose missing observations, provenance, possible state dependence and data-quality limits. Reporting mechanisms are often not identifiable from reports alone.

Receiver Awareness

Compare naive and mechanism-aware decoding under explicit assumptions. Known thresholds and mechanisms are best-case assumptions.

EWS Robustness

Test variance, AR(1) and other warning indicators against selective missingness, level shifts, truncation and threshold errors; include level-only negative controls.

Evidence & Uncertainty

Separate information loss from decision harm; show assumptions, alternatives, uncertainty, null states and reproducible evidence.

Project stages and quality gates

  1. Map requirements to the existing Studio 3.0 specification and OBJECT workspace
  2. Specify data contracts, model assumptions, null states and evidence UI
  3. Reproduce synthetic SCF benchmarks and negative controls
  4. Implement only after gate authorization; then test, independently review and authorize release separately

Scientific boundaries

No automated diagnosis of intentional suppression, global risk score, causal claim about organisations, certain tipping-point prediction or unsupported recommendation. Studio 2.0 remains unchanged.