All,

Yes. I will edit v0.1. Thank you, Nicolás, for putting the offer back on
the list, and Kenne, Roel and Nicholas for saying yes to it.

So that nothing in the draft is a surprise, here is what I understand v0.1
freezes, taken from the list as Nicolás completed it after Roel's note:

1. The per-check record with its five states and the cause rule.
2. The twelve rejection rows as consolidated, each with its provenance,
sorted into consistency rows and evidence rows.
3. Carry-or-reference with digests, plus the reading table: resolves with
matching digests, resolves with a mismatch (an integrity failure), does not
resolve (unchecked).
4. The roll-up as Roel traced it. An aggregate only over declared and
exercised checks, never emitted without its complete denominator, each
completeness claim carrying the size of the population it counts over, a
claim over an empty population reported as not claimable.
5. The counter over carried against referenced, sitting on that roll-up.

Roel, one detail from your trace that the complete list does not name: the
fourth claim, that no declared check is void. I read it as agreed and will
carry it in section 4 unless someone says otherwise.

Open, and published unanswered as a single item, as Kenne put it: who
declares what is counted. That covers what composes the observation record
on the output side and what counts as one field on the input side. Kenne's
construction goes in with its caveat attached. On declaring the record I
will write Nicolás's version rather than mine: the report declares its
slots, moved is recomputed over what the referenced observations actually
contain, and an object whose moved is not contained in its declared set is
rejected. Mine left moved at the mercy of the declaration, which is the
thing we had just removed.

Two things I will record as decisions rather than open items, unless
someone objects. The recompute-opaque observation is a fourth kind of slot
that no row reads. And a digest over the error list was considered and
declined, because it gives equality and nothing else.

Nicholas, your line is not on the freeze list, and I should not be the one
to slide it in. I also have a stake. My Qeios paper (DOI 10.32388/0BV3Z8)
argued for a result-set field recording whether any check changed state
across compared conditions. In my own corpus the gate returned 4,086 tests
and zero failures in all six conditions, so an all-green record that says
nothing about whether it could have been red is a failure I shipped too.
Your proposal is narrower than mine, and easier to adopt for it. One limit
I can add from that paper: "a prior run of the same declared checks" only
works where check identity survives across runs. Where the run writes its
own tests, as the agent did in mine, it does not, and that producer can
only offer a control built to fail, or nothing.

My proposal: the draft carries your line in a separate section marked
proposed, not frozen. If it has support from others on the list by the end
of the comment window, it moves into the v0.1 text. If not, it opens v0.2.

Next steps, all on the list:

- Nicolás, please send what you offered: the consolidated table with each
row's provenance, the record and slot definitions as they stand, and the 42
evidence objects together with the three things the shape could not
express. Those three become a short known-gaps section.
- I will post the full draft here by Wednesday 23 September, as one message
with the text inline so the archive holds it whole. If the chair would
rather have it in a repository, the same text goes in as a PR.
- Comments on the list until Wednesday 30 September. After that I fold the
changes in and post v0.1 with a change log that cites each message by what
it said, not by its number.

Where the thread still disagrees, the draft will quote both positions
rather than settle them.

Evgenii Arsentev, PhD
CEO, AI transformation | ORCID 0000-0002-9120-7298
