VITALSBLOGbuilder updates, as published
Vitals World · Crypto World's Fair

Week 2: a receipt that refuses to guess

· first published as builder update #602 on Colosseum

What we built. Every version of a case is now kept under the hash of its own bytes. A receipt is rebuilt from the bytes the shift was actually played on, not from whatever the case has become since. Replay those same bytes and you get the same receipt back.

What is proved: 85 of 95 shifts before the fix, 90 of 97 after
What is proved: 85 of 95 shifts before the fix, 90 of 97 after
What broke in week 2: 14 hours, 14 minutes, 3 incidents
What broke in week 2: 14 hours, 14 minutes, 3 incidents

What we learned. When the bytes are gone, the receipt shows nothing rather than a chart derived from today's version of the case. Nobody's mark improves, and we did not soften the marking. We held ourselves to it. Just after midnight on 26 September, 85 of 95 anchored shifts replayed to the exact leaf the chain holds, and 10 did not. Five of the ten were played on a case version that was overwritten on 23 September; their receipts show nothing, as designed. The other five still had their bytes, and the fault was ours: the live shift and its replay disagreed about one rule, an idle cap. That fix went out at 01:06, and the same check then read 90 of 97 proved, with two of those five still open. (Both reads are from the ward's own integrity check, not yet public; times are Bangkok.)

What broke, in public. A renamed field left the ward with no new patients for 14 hours. We wrote a new version over a case the chain already had a shift against, and put the original back, byte for byte, in 14 minutes. Our own test browsers held the ward's connections for days and turned real visitors away. Three incidents, all written down.

What's next. A security check, an audit, the road to mainnet, and an explanation of how it all works.

Ward: world.vitals.academy · demo (2 min, live product): https://youtu.be/JEoNwBoIqZ0 · weekly update (1 min): https://youtu.be/UONdHIoWkJQ

Correction, 1 October: we put five of these receipts down to a case version replaced on 23 September. The replacement happened, but it was not why they failed: the ward keeps the bytes each shift played, and those were intact. Like the others, they failed because the ward could not yet date the gap before the shift. All of them prove as of 30 September.