A methods-note preprint: each forecast is sealed before its event window, SHA-256-hashed into a Bitcoin-anchored manifest, and scored with proper scoring…
Vijay Jyotish, Vijay Jyotish LLC. ORCID 0009-0009-0832-9998. Preprint — methods note. Version of 2026-06-29. Licensed CC-BY-4.0. Companion dataset: JYOTINT Sealed Forecast Corpus, DOI 10.5281/zenodo.20630257. Full text: /dataset/jyotint-methodology-preprint.md.
Abstract. Public forecasting records are, as a class, difficult to audit. Predictions are routinely edited, deleted, retro-dated, or selectively recalled after the fact, and the absence of a tamper-evident anteriority proof makes it impossible for a third party to distinguish a forecast that was genuinely committed before an event from one reconstructed afterward. This note documents a protocol that addresses the anteriority and falsifiability problem directly, and that separately scores the resulting record with standard proper scoring rules. Each forecast is published verbatim to public infrastructure before the event window, canonicalized as the string objectId|dateIssued|title|claim, hashed with SHA-256 into a manifest, and the manifest is committed to the Bitcoin blockchain via OpenTimestamps; anyone can independently recompute the hashes and verify the time-anchor with a zero-dependency verifier. A pre-committed grading rubric (HIT / NEAR / PARTIAL / MISS) is fixed at seal time, and missed calls are retained in the denominator. Over a corpus of 68 graded forecasts (58 HIT, 3 NEAR, 3 PARTIAL, 4 MISS) the self-assigned Brier score is 0.0717 (log-loss 0.322); the launch-timing subset (n = 23) scores 0.0360. We report calibration against the IPCC AR6 calibrated-likelihood bands, and an information-theoretic statistic — Information Yield — that measures bits of surprise-if-true per call (median ≈ 8.8 bits).
Limitations, stated first. The probabilities and grades are operator-assigned, not independently adjudicated; the sample is small and high-confidence-weighted; the generative method is a disclosed unconventional one (Vedic jyotish). The paper makes no claim about the validity of that mechanism. Its contribution is narrow and checkable: the protocol — sealing, falsifiability, retained misses, and reproducible scoring — not a claim that any particular method works.
Contents: 1. Introduction (the anteriority/skill split); 2. The sealing protocol (canonical string, SHA-256, OpenTimestamps, independent recomputation); 3. Grading (pre-committed rubric, misses retained, the frozen grading ledger); 4. Metrics (Brier, log-loss, IPCC-AR6 calibration bins, Information Yield, SITA); 5. Results (the 68-call corpus, headline metrics, the externally-adjudicated subset); 6. Limitations and threats to validity; 7. Reproducibility (the zero-dependency verifier, the open corpus, the API); 8. Conclusion; References. Deposit-ready for OSF / Zenodo. Download the full preprint.