extending the record without rewriting it

An archive had stopped at a fixed date, and a study needed it to continue past that date. The dangerous move is the obvious one: rebuild the whole series from new inputs and accept whatever comes out, because then every earlier result silently changes underneath the reports that already cite it.

So the extension ran as a parity exercise. Before any new span was accepted, every previously committed row had to come back byte-identical from the rebuild. The earliest span — the one used to fit — had to stay exactly invariant, and the frozen summary derived from it had to hash the same as before. A truncation audit compared cuts across the series. Only then were the new rows appended.

The outcome was not clean, and the record says so. There is a genuine hole: several months where the inputs carry nothing usable, written down as a hole rather than filled. A handful of series gained nothing at all, for two separate reasons — some had been legitimately cut far earlier and stay cut, and others lose the post-hole stretch to the existing quality bound because the gap itself exceeds it. No values were synthesised to close either case. A second archive family was deliberately left unadvanced, because nothing consumes it yet.

The cost is that any report citing this study has to carry the hole with it. The gain is a series that grew without a single earlier number moving.

What it proves: history here is append-only, gaps included.