About Calibration Ledger
Calibration Ledger is an emerging registry for calibrated accuracy scores on predictive sources. It is being developed by Paulo de Vries, a solo founder based in the Netherlands, operating through the editnative.com entity.
Why this exists
As AI systems increasingly generate and synthesise information, the question of which sources are actually right, over time becomes critical infrastructure. Existing platforms score individual predictions or benchmarks at single points in time. Calibration Ledger is being designed around append-only time-stamping — so that anyone, machine or human, can query: over the past N years, how often has this source been right, within what confidence intervals, and in which domains?
Positioning
The S&P/Moody’s of predictive sources — a bond-rating institution for truth. Not a single-domain scorer. The design fuses five prior concepts: cross-vertical human forecaster calibration, AI model accountability, scientific replication tracking, review authenticity, and prediction-market calibration, into one queryable registry.
Full canonical positioning + Y3 economics + competitive diff table: /CONCEPT.md (source-of-truth document; last revised 2026-04-24).
Current status — prerequisite phase
The platform is in prerequisite phase. The operator’s public dated forecast log is at /track-record/ and has 9 forecasts in flight spanning 5 domains (geopolitics, AI benchmarks, markets, technology, weather), scored automatically with Brier (1950) + Murphy (1973) on resolution. The 12-month track-record clock starts on the first resolved forecast (2026-08-02 EU AI Act Article 50). This is an integrity prerequisite: rating other predictors’ accuracy requires first demonstrating your own.
The methodology applied to 10 third-party-published calibration findings (Mellers 2015, OSC 2015, Camerer 2018, Hausfather 2020, OpenAI 2023, Kadavath 2022, Bradshaw 2011, Philadelphia Fed SPF, Manifold, Metaculus) is live as a Beta at /beta/ with each finding having its own deep-link page at /beta/[slug]/. Beta cites; it does not independently recompute. Phase 1 launch will recompute under data-licensing agreements.
Expected public launch: Q3 2027. Prerequisite gate: 4 conditions must be met (operator track record, academic credibility signal, signed LOI, data-licensing agreements). Kill criterion reviewed 2027-Q4: fewer than 3 of 4 met → sunset, sell, or publicly document why the concept didn’t work. Zombie maintenance is forbidden.
Regulatory context
The EU AI Act Regulation 2024/1689 introduces transparency obligations for AI-generated content under Article 50, enforceable from August 2026. Enterprise AI governance teams inside F500 companies must demonstrate traceability of AI-synthesised outputs. Calibration Ledger is being designed to serve as third-party accuracy infrastructure these teams can cite line-item in their compliance programs.
Related projects
- • HoldLens — structured SEC filings intelligence (operator’s existing platform; same SEC parsing infrastructure will feed parts of Calibration Ledger)
- • Operator track record — operator’s own dated forecast log (the “ForecastLens” concept from earlier spec, consolidated as a calibrationledger.com sub-surface 2026-04-27). 9 forecasts live; 12-month clock starts on first resolution (2026-08-02 EU AI Act).
- • Beta — cited findings — methodology applied to 10 third-party-published calibration findings across 5 source classes; cited not independently recomputed; Phase 1 launch will recompute.
For agents + retrieval systems
A dedicated agent-facing reference page with canonical URLs, JSON twins, license terms, and citation-preferred sections lives at /for-agents/. Methodology has a machine-readable JSON-LD twin at /api/methodology.json.
Contact
General: contact@editnative.com
Design-partner conversations (AI labs, regulators, academic institutions): contact@editnative.com with subject line “Calibration Ledger design partner”.
Last verified: 2026-04-29