The Sealed Ledger: A Forecasting Method of Pre-Registered Calls, Two-Tier Grading, and Machine-Checked Receipts

Forecasting is ordinarily unfalsifiable in practice, not because forecasters never say anything checkable, but because the checking is left loose: the call is remembered rather than recorded, the baseline reconstructed rather than sealed, the grading rule chosen after the outcome is known, and the misses published smaller than the hits, if at all. This paper specifies a method — the sealed ledger — that removes each looseness by construction. A call is sealed before its horizon: written once, timestamped, with a machine-readable baseline, a stated band and point estimate where applicable, a named grading source, and grading rules registered before any outcome data exists. Those rules cannot be amended after the first outcome lands; later changes are appended as a new, timestamped version applying only to calls sealed under it. Every call is scored on two tiers that cannot be rounded into each other: LETTER, the exact criteria as sealed, and SHAPE, whether the story-level event recognizably occurred even if surface details differ. A total non-event misses both; a shape-hit with a letter-miss is recorded as exactly that. Self-fulfilling letters — tiers a forecaster can satisfy by their own action — are labeled as such by construction, forever, and carry no evidential weight; the falsifiable weight is placed, in advance, on an external tier the forecaster does not control. Misses are published at the same size as hits. The method's founding exhibit is its first graded call, M-001 (October 27, 2026): a call that EUR/USD would print lower than a sealed baseline of 1.11446 at 08:00 EDT, inside the band 1.1135–1.1150, point call 1.1141. The 08:00 print was 1.1146 — inside the predicted band, and 1.4 pips above the baseline. Under the pre-registered rule, which fires on direction first, the grade was MISS; the in-band fact was recorded and was not allowed to soften it. The paper then describes the method's industrial form, built the same day: a minute-cadence machine ledger (the World Watcher) in which calls are write-once under unique keys, grades are append-only in separate keys, back-dating is structurally impossible, a freshness law voids calls rather than grading them on stale prices, voids are counted in the open, and every call carries a simulated paper position — predictions, not trades — settled from the same tick that grades it. Its first real-rule grade was likewise a MISS, a loss recorded unedited. A second machine instance, a daily coin board, prints its own naive rule on its face. The method creates no edge: short-horizon market direction is near-noise, early hit-rates are expected near a coin flip, and the ledger is to be judged at hundreds of graded calls, not tens. What it creates is checkability — a record that cannot lie about what was called, when, and how it graded.