Public: anyone can read this; the crew can change it.

A new forge is proven once it is up: the spawn answer no longer says "Running, not reachable" for a forge that answers a second later #13801

Merged opened this

Check · Review · Land · Live done — Landed and live in production.

Merged

spawnDoor started the new forge's cell behind the answer (void bootHostedChild) and proved its public address at once with a 3 s deadline (firstProof, PROBEMS). A cell takes about 4 s to answer, so the first proof always ran out. Measured on production (0.7.613), 2026-11-16, 20 probe forges (scale-1 ...

Notes

shipped production is serving it deploy 59903fa

opened this

What

spawnDoor started the new forge's cell behind the answer (void bootHostedChild) and proved its public address at once with a 3 s deadline (firstProof, PROBE_MS). A cell takes about 4 s to answer, so the first proof always ran out.

Measured on production (0.7.613), 2026-11-16, 20 probe forges (scale-1 ... scale-20, size small): every spawn answer read

Running, not reachable: home no answer (The operation timed out.). Still checking; the owner is alarmed if it does not come up in 3 minutes.

in 17 of 20 answers, and each of those forges answered its home 0.2 to 2.4 s after the answer (spawn call 8.2 to 11.4 s; home answering 8.5 to 13.0 s from the call, at host load 65 to 190). The other 3 were made at the awake ceiling, where the proof read "home answered 429, not by the forge" 5.7 s in, for forges that answered at 9.1 to 9.7 s. An agent or a person reading the answer is told a healthy forge is broken (the workshop pilot reported the same three times).

Change

  • src/spawn/door.ts: the answer waits for the forge to come up before it is proven, for as long as a visitor's request is held for a waking forge (wakeLimits.waitMs, 20 s). A cell that takes longer is proven as before and settled behind the answer, so the answer is never held past that. A page's spawn (page: true) does not wait, as before.
  • src/spawn/reach.ts: proves(app), so nothing waits where proving is off (the suite).
  • test/spawn-reach-door.test.ts: a public address that answers only while the forge is awake; the home proof is ok in the spawn answer. The test fails on main (Expected: true, Received: false).

Cost: the spawn call returns about 1 s later (when the forge answers) instead of 3 s after the record is made, and says the truth. No page changes.

Checked

  • bun x tsc --noEmit: exit 0. bun scripts/gate.ts: exit 0 (comments rewrapped with --fix).
  • Linux (release image, Bun 1.4.4): test/spawn-reach-door.test.ts, test/spawn-reach.test.ts, test/release-journey.test.ts, test/first-five-minutes-journey.test.ts, test/client-list.test.ts: 49 pass, 0 fail.

Try: call the spawn tool with dryRun false for a new name and read reach.says in the answer: Every public address answered.

From the scale measurement in scratchpad/scale-hosting.md (the mitosis repository).

🤖 Generated with Claude Code

Approved

The spawn answer now waits for the forge to come up before it proves it from outside. The wait is bounded by wakeLimits.waitMs, which is the same bound a visitor request gets for a waking cell. Where the proof is off (tests), it is skipped. The change is small and the one new test would fail without it. It adds no input, write path or dependency, so the security list is untouched. It does not touch src/web or the terminal. I did not open the preview or the screenshots, because no page changes. Two things I would do next: (1) The create call can now take up to 20 s on a slow boot, and a caller on a phone sees only a spinner. Return the wait in the answer, or show "starting" while it waits, and add a test where boot never finishes to pin the bound. (2) Put the wait in reach.ts beside firstProof, so door.ts holds only the call.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs