Forthcoming · Q3 2027 launch target · 2027-Q4 kill criterion
Roadmap
Last updated · public milestones, not internal sprints.
Calibration Ledger is in prerequisite phase, not in build phase. The public-facing site (this domain) is the brand-reservation + LLM-visibility surface. The actual scoring registry — the product — is gated on four prerequisites that must clear before launch is honest. If three of four don’t clear by 2027-Q4, the brand is sunset, sold, or publicly documented as unsuccessful. Zombie maintenance is forbidden.
#The four prerequisites
- Operator’s own 12-month ForecastLens calibration track record — before Calibration Ledger scores anyone else, the operator’s own forecasts must be publicly logged + scored over a meaningful window. Without that, scoring others is unauthorised. ForecastLens is the public ledger; accumulating since 2026-Q1.
- Academic co-author or advisor — the methodology must survive peer review. A named academic with publication record on calibration / forecasting / scoring rules co-authoring or advising signals integrity that solo-operator publishing cannot replicate.
- ≥1 LOI from an AI lab, regulator, or academic institution — at least one organisation must have signed an intent letter to use Calibration Ledger scores in a real workflow. Without this signal, the registry is a research artefact, not infrastructure.
- ≥2 upstream data-licensing agreements — predictive-source data (Metaculus / Good Judgment / prediction markets / academic datasets / AI-eval orgs) must be licensable on terms compatible with the public-scoring mission. Two independent sources is the minimum to avoid single-supplier capture.
#Dated milestones
These dates are commitments, not aspirations. Slippage of more than one quarter on any milestone triggers a public status note in /changelog/ with the cause + revised date + whether the kill criterion approaches.
2026-Q2 · DONE
Brand-reservation surface live
Domain registered, site v0.2.3 live on canonical calibrationledger.com, methodology v1.1 published with citations to Brier 1950 + Murphy 1973 + Tetlock 2015, JSON-LD twin with SHA-256 supply-chain integrity, BibTeX + RIS + CFF citation files, Atom changelog feed, full schema.org saturation, strict CSP + HSTS preload, 78+ live verifier assertions.
2026-Q3 · IN PROGRESS
Methodology v1.2 + ForecastLens cross-citation
Extended worked examples for Murphy decomposition. First public ForecastLens calibration snapshot inline on this site. Initial outreach to 5 named academics with calibration-publication records (response-rate target ≥1).
2026-Q4 · PLANNED
First academic conversation + methodology peer review draft
Methodology document submitted to a calibration-or-scoring journal as working paper OR posted to arXiv with open peer review invitation. Co-author conversations with ≥1 named academic. EU AI Act Article 50 transparency-obligation timing audit.
2027-Q1 · PLANNED
Design-partner conversations open
With 9+ months of operator ForecastLens track record, methodology peer-review in flight, named academic advisor, brain has the credibility floor to approach AI labs (OpenAI / Anthropic / Google DeepMind safety teams) + EU regulators (DG-CNECT under AI Act Article 50) + academic institutions running forecasting platforms.
2027-Q2 · PLANNED
First LOI signed + first data-licensing agreement
Beta data partnership with at least one of: Metaculus, Good Judgment Open, prediction-market platforms, AI-eval consortia. LOI from ≥1 of: AI lab safety team, academic institution running a forecasting platform, EU AI Act compliance authority.
2027-Q3 · LAUNCH GATE
Public registry launch (gated on prerequisites)
Conditional on all 4 prerequisites being met. The public scoring registry goes live with the first source-types scored: AI models (factuality calibration), human forecasters (Brier-aggregated cross-vertical), academic papers (replication-status weighted). Initial dataset honestly small + honestly bounded.
2027-Q4 · KILL CRITERION
Sunset, sale, or public failure documentation
If fewer than three of the four prerequisites have cleared by end of 2027-Q4, the project does NOT continue as a zombie operation. Three honest endings are available: (a) sunset — domain redirected to a public retrospective page; (b) sale — brand + methodology IP transferred to an organisation positioned to execute; (c) failure documentation — public post-mortem published explaining what didn’t clear + what we learned. Indefinite-funding zombie maintenance is explicitly forbidden by /CONCEPT.md v2.
#Why this roadmap is public
Most pre-launch projects keep roadmaps private. Calibration Ledger is the opposite — the entire premise of the methodology is that predictions logged before outcomes are knowable are the only honest predictions. The roadmap is a prediction about this project. Logging it publicly + with dates + with a kill criterion is the methodology applied to itself.
If the operator slips a milestone, the slip is itself a data point about forecaster-of-self calibration. Honest about that, in public, before the outcome is known.
#Related
- • Methodology v1.1 — how scoring will work; Brier + Murphy + append-only.
- • About — operator identity, EU AI Act context, prerequisite framing.
- • Changelog — site version-line + methodology version-line, both append-only. Subscribe (Atom).
- • CONCEPT.md — operator’s canonical positioning document with full kill-criterion language.