Re: Re: Conformance cases for mid-task halts

Hi Abisheak,

One correction on the proposed lab run. #210 and #212 are candidate tests
for Lee's RTTP/IQA addressing draft, not NTI. #206 contains candidate cases
based on the NTI discussion, and #216 is a separate NTI submission that I
still need to review.

Publishing candidate tests does not establish an implementation result.
NTI's authors can run the reviewed cases and submit a pinned record with
reproducible commands and per-case output. Their run would be labeled
author-produced.

The lab records what was tested, by whom, and what the evidence
establishes. It does not certify implementations or standards.

Best,
Tymofii

On Tue, Nov 03, 2026 06:19 PM, Abisheak P <abisheakp197@gmail.com> wrote:

> Hi Tymofii, Steven, Lee, and the CG,
>
> This is exactly how a neutral standard should operate. We score frozen
> bytes, not reputations—and Steven, your public correction and byte-level
> diff is the definition of cryptographic integrity.
>
> The Decision on Versioning:
> We go with Option A. The canonical source of truth for v0.1.4 is the
> original W3C email attachment (83d88a70). GitHub mirrors are just
> convenience—they are not the canon. Steven, please align the GitHub mirror
> back to the exact att-0108 text so the hashes match end-to-end. Any
> formatting cleanup can be deferred to a future v0.1.5 release with a proper
> diff summary.
>
> On the Lab Execution:
> Lee, congratulations on PRs #210 and #212 being merged into the lab
> mainline. The families are officially registered.
>
> Tymofii, you noted that "the lab has not run an NTI-1 implementation."
> That is the missing piece. We have the specs, we have the vectors, and we
> have the independent candidate tests.
>
> I am formally requesting the LF lab to run the NTI-1 reference
> implementation against the newly merged PRs (#210, #212) and Steven's
> candidate vectors (unsigned token -> Deny; unmet quorum -> Blocked). The
> NTI Rust core is live, the PyO3 bindings are ready, and the CI is green.
>
> We do not self-certify. The lab needs to execute the run and publish the
> sealed report. If it breaks, publish it. If it passes, we lock the standard.
>
> Larry, we await the rescheduled workshop details.
>
> Let's keep the momentum.
>
> Best,
> Abisheak
>
> On Wed, 4 Nov, 2026, 4:58 am Steven Zhao, <zhaoxinghua07@126.com> wrote:
>
>> Hi Tymofii, Abisheak, and the CG,
>>
>> Tymofii — thank you, and you are right. A correction from my side.
>>
>> ## 1. What I verified just now (with the actual bytes)
>>
>> I downloaded the original attachment from my 30 October message and
>> recomputed
>> its hash:
>>
>>   https://lists.w3.org/Archives/Public/public-agentprotocol/
>> 2026Oct/att-0108/nti-transition-contract-spec-v0.1.4.md
>>   27477 bytes, sha256
>>   83d88a70237aedb6bffdb446b8bd8e3e4df41291d2c178f33367a6e10d8fc763
>>
>> The lab's pin is exactly reproducible. My earlier statement that 83d88a70
>> "does not match any published bytes" was wrong — I had only tested the
>> GitHub
>> mirror and byte-level transforms of it, never the list attachment itself.
>> I
>> withdraw that statement with apologies; my clarification post of 4
>> November
>> (reply_hash_clarification) should be read as superseded by this message.
>>
>> ## 2. The real difference between the two versions (full diff, not
>> guesses)
>>
>> Comparing the att-0108 text (83d88a70, 27477 bytes, hard-wrapped) with the
>> GitHub mirror at commit 679b6e4
>> (https://github.com/zhaoxinghua09-cell/nti-lgd-transition-contract — file
>> nti-transition-contract-spec-v0.1.4.md; 25532 bytes, sha256
>> a0d40175cc025213bcd356babd74c56171ee452ba362681ff8094ca249d8bab4),
>> whitespace-insensitive normalization still shows a 1,297-character gap.
>> Nine
>> difference blocks total; most are markdown formatting (angle-bracket
>> links,
>> lint comments). Three are substantive deletions in the mirror:
>>
>>   a) Section 2.4: the mirror drops the sentence "Proposed on the
>>      agent-protocol thread (2026-10-30) ... **Not agreed**: it carries no
>>      fingerprint until circulated, and §0.3 applies to it like the rest of
>>      this text."
>>   b) OPEN-1 row: the mirror shortens "LGD (Core Spec gate decision record)
>>      + NTI, joint table | Section 1.1 is untestable; the conformance
>> family
>>      cannot start" to "LGD+NTI, joint table | Section 1.1 untestable" —
>>      dropping the blocking statement.
>>   c) Section 7: the mirror drops the title qualifier "(direct links, kept
>>      adjacent to the citation)" and the sentence "A citation whose link
>> does
>>      not resolve is not a citation."
>>
>> ## 3. What this means under the document's own freeze rule
>>
>> Section 0.2 says any character change to a released version raises the
>> minor
>> version and carries a diff summary. The mirror changed characters but kept
>> "v0.1.4". Two honest ways to fix the chain:
>>
>>   Option A (preferred): align the GitHub mirror back to the att-0108 text
>> so
>>   that 83d88a70 is the single frozen text both the lab record and the
>> mirror
>>   reproduce. Formatting-only cleanup, if wanted, becomes v0.1.5 with a
>> diff
>>   summary.
>>
>>   Option B: keep the mirror text but re-release it as v0.1.5 with the diff
>>   summary above, and let the lab record stay pinned to the att-0108 text
>> as
>>   the v0.1.4 release.
>>
>> I will not touch the mirror until you and Abisheak say which option, since
>> the lab's published record currently rests on the att-0108 bytes.
>>
>> ## 4. Candidate submission (unaffected)
>>
>> The two vectors (unsigned capability token -> Deny; unmet quorum ->
>> Blocked)
>> are packaged per CONTRIBUTING.md / existing candidate PRs (CANDIDATE.md,
>> vectors.json, check.py, mutate.py) and pin whichever text the CG settles
>> on;
>> the checker and mutation test are pinned to behavior, not to a byte hash,
>> and
>> currently print 8/8 agree and 6/6 mutants killed.
>>
>> Best,
>> Xinghua Zhao (Steven Zhao)
>> ORCID 0009-0001-0512-1237
>>
>>
>>

Received on Wednesday, 4 November 2026 02:40:34 UTC