Count the edges. A chain of n entries carries n-1 hash edges, because entry one has no predecessor to embed [1]. A verifier holding entries one through seven can recompute six of them [2][1]. Six records are named by a successor. One is not, and it is the newest [1].
That is all a verification pass returns: a walk over the edges present in the bytes it was given. It never had an input for entry eight, because entry eight was never delivered [5]. There is no gap to complain about. The numbering ends at seven, and seven is a legal end [6].
The reason this belongs in a design review rather than a footnote is where these records get produced. A signed scanner report attached to a release proves who signed those bytes and what they covered [10]. A scan that ran afterwards with worse findings sits outside that scope, and the dev.to piece makes the sharp point that the last scan before the ship decision is the only one that could still change the decision [10]. Evaluation sweeps have the same shape: stop publishing after the sweep that regressed and the visible sequence still holds no broken edge and no internal contradiction [9]. Incident timelines end where the story gets less flattering, and earlier edits leave scars while a missing final segment leaves none [11].
The first repair most teams reach for is publishing to two independent places, and the source is precise about what that buys, which is divergence detection: if record seven says one thing in one location and something else in the other, at least one copy is false or the feed split [14]. That property is real, and it is orthogonal to this failure. If publication stops, both mirrors hold the same prefix, and comparing them returns agreement [2]. Two silent tips instead of one.
For an absence to be observable, the object under verification has to commit to something that has not happened yet [3]. A declared interval, a promised slot for entry eight, anything that turns end-of-file into a missed deadline. Previous-hash linking gives you none of that, and the source is explicit that this is not a defect in the chain: it does what it claims for the prefix in hand, and separating "this is the end" from "this is where delivery stopped" was never in scope [15].
The failure already has a respectable name in another field. A voluntary benchmark submission can be signed, reproducible inside a stated harness and bound to a commit, and still say nothing about the abandoned runs that scored worse [12]. The source argues that publication bias is structurally the same property as tip truncation, because whoever sees the result before publication holds the write path [13]. If you read benchmark tables as claims about someone else's workload, the rows that were never published are the ones that would have told you whether the number transfers.
So the question to put to any log-integrity design is not whether the edges verify. It is what the verifier is holding that commits the publisher to a record it has not yet sent [3].
