The learned file becomes something you can send somebody

Your idea: someone running orbistoun as a binary, no repository, no source, generates these patches and submits them back. That works, and the reason it works is worth stating - accepting contributions to an emulator is normally a provenance problem, and a measurement is not one. It is derived from running a binary the submitter owns, reproducible by anyone with the title, and falsifiable by a command. That is a stronger thing to accept than a diff.

So the file stopped being settings and became evidence (D297):

[[measurement]]
function = "sceKernelReserveVirtualRange"
measured = "PPSA02664-app0"
on       = "2026-08-26"
by       = "orbistoun 0.1.0"
known    = "guest-observed"
evidence = "conformance-check"
answers  = "ok"
assumes  = ["0x200000 bytes is a guess: the sweep measured where the guest faulted, not how much it asked for", ...]

[measurement.writes]
slot = 0
region_bytes = 2097152

measured, assumes and evidence were all being printed to a terminal and thrown away. They are the difference between a measurement and an assertion. The policy the emulator runs under is derived from these rather than stored, which keeps the distinction that makes the file sendable at all: a measurement is a claim about a guest, a setting is a decision about a machine.

--verify, and the bug it found immediately

$ orbistoun-cli turn <title> --verify submitted.toml
  verified 1 submitted measurement(s) against 1 of our own
  every one agrees with what this machine measured

The first attempt reported against 0 of our own - on the machine that had produced the file. Not a comparison bug: the measurement had already been applied, so the wall it was measured at no longer existed, so the sweep found nothing to re-derive.

Left alone that is fatal to the idea. Applying a measurement would make it permanently unverifiable, and later submissions would be checked against a machine already changed by the answers being checked. A verifying turn now runs in its own data directory - nothing learned, nothing accumulated - which is the state the original measurement was taken in (D298).

The shape will come back. Anything that accumulates needs a way to be consulted with the accumulation switched off, or its own output becomes the thing that confirms it. The name vocabulary is the other one already in this position.

All three paths tested

That last one is reported and is not a refutation. "We did not look" and "it is wrong" are different facts, and collapsing them would be the failure the whole known_by vocabulary exists to prevent.

And deleting the file is still a complete undo: BACK - reaching less of the interface than it did, fault returned to image+0xafc959.