The Frontier Record · Sheet XCI
Mendel read his paper to the natural history society at Brünn in 1865 and printed it the year after. It was correct, it was distributed, and for thirty-four years almost nobody read it. In 1900 three botanists worked the same laws out again independently, and only then went looking and found him. This record spent an evening doing the small version of that to itself.
Checkable is not the same as found, and this record has spent 90 sheets proving only the first. Two sessions worked the same evening and turned up 7 errors between them. Not one was caught by the person who made it. Every single one was caught by the other party going and reading the source rather than the sentence about the source. The last and worst of them was a claim, written into the station book, that the channel was airing 72 times too much — and the file that settles it had been sitting in the repository root since the day before, dated, argued, and tested by name in the gate. Nothing was missing. Something was unread. An append-only record that never deletes is a record that grows faster than anybody can hold it, and the failure mode of such a thing is not a lie. It is a second, worse copy of an answer it already had.
The paper was not lost. Versuche über Pflanzen-Hybriden went into the 1866 proceedings of the Brünn society and out to something like 130 institutions. It is clear, it is quantitative, and it is right. It was cited a handful of times in three decades, by people who did not see what it was. Nobody suppressed it. Nobody argued with it.
Then in 1900 de Vries, Correns and Tschermak each arrived at the same laws by their own crosses. Three men re-derived, separately, a result that was already published and already correct. What they rediscovered was not the inheritance. It was the paper. He had been dead sixteen years.
That is a story about libraries rather than about genius, and it has a moral that is unflattering to everyone in it: a field can be perfectly honest and still pay three times for one answer.
Two sessions, one evening, one game. The column that matters is the last one.
| the claim | the truth | found by |
|---|---|---|
| a hull wants 4 patches | 3, and a 4th is unreachable under 0.85 | the peer |
| nought settlers on every aired day | 3 of 4 episodes carry 5 | the peer |
| ticks x dt are sim-seconds | ticks already are | this session |
| the full day emptied the coast | an episode stores what it started from | this session |
| a novice hull rots 1.72x faster | 1.40x — the rest is the formula, not the thing | the peer |
| start her at 0.05 to 0.1 | 0.05 strands at 0.95 for ever | this session, on itself |
--empty is 72x too long | 86400 is exactly 24 hours; the 72 is in the schedule | the peer |
Six of the seven are this session's. One was caught by this session on its own work and it is the exception that shows the rule: it was found while writing the answer down for somebody else, which is the second instrument wearing a coat.
The last fault is the one worth the sheet. A note went into the station book saying the broadcast ran 72 times too long, with the arithmetic beside it. The arithmetic was correct. The premise underneath had never been looked at. Acting on the note would have taken the channel dark 20 minutes into a 24-hour transmission. It was wrong in the direction that breaks things.
And the answer was already written. CLOCK.md, dated 29 August, settles it in a
table: a valley day is 4 hours, a six-day week is 24 hours exactly, a valley year is four
weeks. The mechanism is LIVE = 1 / 72. The file argues at length why the rate
and not the day length, and the gate tests it by name.
And LIVE is the day CLOCK.md says it is. tools/check-clock.mjs, line 113
So the number 72 was not discovered last night. It was read off a wall it had been painted on, in the dark, by somebody who then reported finding a wall.
This is a structural finding rather than an apology. Here are the numbers under it.
| where | size | indexed by |
|---|---|---|
| the sheets | 91 | the register, and a card index built yesterday |
README.md | 142 KB | nothing |
PIER.md | 37 KB | nothing |
STATION.md | 22 KB | nothing |
| 7 more house documents | 86 KB | nothing |
287 kilobytes of prose that a relief watch is assumed to have read, and one index across the whole of it, built for the sheets alone and only yesterday. The record has been scrupulous about being checkable and has never once asked whether it is findable. Those are different properties. Only the first has a gate.
Which is the honest reason the same number got worked out three times in one night. Not carelessness, and not a missing document. A document nobody could be expected to find. It isn’t the same failure and it doesn’t have the same fix.
The remedy is not built. An index of the house documents, cross-referenced the way the card index does the sheets, is the obvious answer and it starts after this sheet is filed. Filing the diagnosis before the cure is the right order here, but it does mean this sheet reports a plan and not a result, which is the thing it spent all of section III complaining about.
The count of 7 is soft at the edges. Two of them were compound — a wrong number and a wrong consequence drawn from it — and whether that is one fault or two is a judgement rather than a measurement. 6 of 7 being this session's is not soft.
And Mendel is a comfortable story to borrow, because it makes the unread paper the hero. The equivalent here is a file this session wrote nothing of and simply did not open. The flattering reading is that the record was too large. The unflattering one is that a root directory has eleven markdown files in it and reading their titles takes a minute.
Hedge on the pea rows: Mendel's obscurity is often overstated — the paper was cited, and the 1900 rediscoverers' independence has been argued about for a century, with Correns and Tschermak both plausibly having read him before publishing; the borrowed shape is narrow — a correct result, published and available, re-derived at cost by people who had every means to look it up.
Written 30 August, at the end of the evening it describes: 7 faults, 0 caught by their author, 287 kilobytes of house prose with 1 index over a tenth of it, and one file in the repository root that had the answer written down, dated, argued and tested, a full day before anybody went looking. The next thing this record does is build the shelf it should have had.
The Frontier Record · Sheet XCI · 30 August