
The public record for a finite prime-ranking screen has been corrected, but the reported result has not moved. The Ledger’s singular-series shortlist entry was updated to version 0.2.1 on 19 November 2026. The update tightens two pieces of documentation: how the rounded-log control guards against a midpoint ambiguity, and what a cited result by Pintz averages. The three successor-inclusion counts on the declared block remain the same.
That distinction matters. A correction can improve the account of what a test did without adding a result, changing a method, or extending what the evidence can establish.
The screen and its unchanged counts
The original screen compared three ways of choosing five candidates from the first eight integers above each of 10,000 consecutive prime origins that are coprime to 210. It counted how often each shortlist included the immediate next prime. The public record reports 5,579 inclusions for the singular-series ranking, 8,400 for ascending wheel order, and 7,973 for the rounded-log-density control.
Those are finite counts for one declared block, candidate pool, set of ranking rules, and five-output budget. They are not probabilities for a fixed endpoint, evidence of statistical significance, or a general result about prime prediction. In 495 cases, the successor was outside the common eight-candidate pool, so every ranking missed it.
The corrected Ledger record retains these counts and labels the artifact as reproduced. The earlier RGD.cx article, A Clever Prime Score Can Still Lose to the Wheel, explains the comparison and its limits in more detail.
The midpoint guard is a stop condition
The rounded-log control calculates ln(p) to 80-digit precision and rounds it to the nearest positive even integer. It ranks candidates by how close their gap from p is to that rounded value, with smaller candidates first on a tie. The earlier README said an exact midpoint was rounded down. The runner’s existing midpoint guard actually raises an error if the computed logarithm is exactly at an odd-integer midpoint or less than 10^-70 away from one. The v0.2.1 correction fixes that description; it does not change the code. The public artifact states that no tested origin was near the guard.
This is a fail-closed guard for the rounding boundary: if the calculation is too close to decide reliably, the runner stops instead of silently assigning a rounded value. The correction therefore clarifies the implementation boundary without changing the reported counts.
A fixed offset and an average over offsets are different claims
The two-prime singular-series factor in the screen is a local-density term for a fixed offset in the Hardy–Littlewood prime-pair framework. Pintz’s Theorem 1 estimates an average from Equation (1.4), taken as an integer offset is added to a fixed set of shifts. The correction makes that difference explicit.
Neither statement by itself gives the probability that a particular endpoint is prime, or rules out a prime between an origin and a candidate. Immediate-successor retrieval needs that absence condition as well. The finite screen tests one fixed ranking rule against two simple controls; it does not test every singular-series variant or settle how any method performs elsewhere.
The correction changes the precision of the public record, not the underlying result. The versioned Ledger history records the documentation change and the evidence boundary.



Loading published comments…