Skip to content
This repository was archived by the owner on Oct 20, 2026. It is now read-only.

spec(rfc-0003): ratify Tranche A3 — status only #23

Merged
s0fractal merged 3 commits into from
Aug 26, 2026
Merged

spec(rfc-0003): ratify Tranche A3 — status only #23

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
6079f2e
spec(rfc-0003): Tranche A3 is RATIFIED — and that moves exactly one o…
s0fractal Aug 26, 2026
2dd453b
spec(rfc-0003): the tree said RATIFIED and candidate at the same time
s0fractal Aug 26, 2026
94c33d5
docs(rfc-0003): §17 misdescribed the Qwen verdict's closing status
s0fractal Aug 26, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
67 changes: 32 additions & 35 deletions conformance/cnp-0-jcs-v0/CONTRACT.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
<!-- Verbatim copy of contracts/CANONICAL_ENCODING.v0.1.md at sha256:f5d6f23426b85ea42ef4f957c6d91d009f3de2ab4b426ce9c4014da07d2d24c1.
<!-- Verbatim copy of contracts/CANONICAL_ENCODING.v0.1.md at sha256:847f56a376078c42f8e546014873093bcb4bd68a7596a5828fb82366b2f6781e.
§5.1.3 requires the contract inside the kit: a kit that cited a document it
did not carry would not be usable by someone who has only the kit, which is
the whole point. tools/build_kit.py --check fails on a one-byte drift.
Expand All @@ -9,28 +9,27 @@
type: "ContractDescriptor"
version: "0.1"
title: "CNP-0-JCS Canonical Encoding"
status: "draft"
status: "ratified"
---

# CNP-0-JCS Canonical Encoding — candidate
# CNP-0-JCS Canonical Encoding — ratified

> **Status: candidate / unratified / partial implementation.** This document
> restates the encoding RFC-0003 Part 01 §5.1 selects, in the form an
> implementer needs. It does **not** ratify Tranche A3, does not lift the
> federation blocker, and is not evidence that any substrate has adopted it.
> Part 01 §5.1.3 requires `CANONICAL_ENCODING.v0.1`, the corpus, a reference
> encoder, a verifier-only path sharing no code with it, a standalone
> conformance kit, and steward ratification before A3 is closed. Substrate
> adoption (`adoption-evidenced`) and independent maintenance
> (`interop-confirmed`) are separate states above it and do not block it. See
> "What is still missing" below.
> **Status: RATIFIED 2026-08-26.** This document restates the encoding RFC-0003
> Part 01 §5.1 selects, in the form an implementer needs. Tranche A3 is ratified
> and the specification-side federation blocker is lifted.
>
> **It is not evidence that any substrate has adopted the encoding, and not
> evidence of independent interoperability.** Those are two separate states,
> both false: `adoption-evidenced: false`, `interop-confirmed: false`. A3 being
> closed moves neither.

- **Normative source:**
[RFC-0003 Part 01 §5.1.1–§5.1.3](../docs/rfc/0003-heterogeneous-state-protocol/01-canonical-identity-and-encoding.md).
Where this document and Part 01 disagree, **Part 01 governs**; this file is a
restatement, not an amendment, and adds no normative rule.
- **Executable candidate:** [`probes/cnp-0-seed-v0/`](../probes/cnp-0-seed-v0/)
— reference encoder, verifier-only rejection path, corpus, negative controls.
- **Reference implementation:**
[`probes/cnp-0-seed-v0/`](../probes/cnp-0-seed-v0/) — reference encoder,
verifier-only rejection path, corpus, negative controls.
- **Conformance kit:**
[`conformance/cnp-0-jcs-v0/`](../conformance/cnp-0-jcs-v0/) — implement §5.1
and score yourself, without running or trusting anything of ours.
Expand Down Expand Up @@ -84,9 +83,9 @@ separate because only the first is shared with prior art:

§5.1.3 requires a verifier-only path that "rejects non-canonical ratios and
malformed raw input", and the corpus pins "a stable rejection class" per
negative case. The class names below are this candidate's vocabulary. **The
names are an implementation choice; the rejections they stand for are the
clauses above.**
negative case. The class names below are this contract's vocabulary. **The names
are an implementation choice; the rejections they stand for are the clauses
above.**

`invalid-utf8`, `syntax`, `trailing-bytes`, `duplicate-member-name`,
`malformed-escape`, `unpaired-surrogate`, `number-not-cnp0-integer`,
Expand Down Expand Up @@ -131,7 +130,7 @@ normative rule here.** They are open questions for the steward.

1. ~~**How a tagged form is recognized.**~~ **Resolved normatively on
2026-08-25.** §5.1.2.1 now reserves the member name `cnp0` and makes
recognition a property of the bytes. The earlier candidate read any map with
recognition a property of the bytes. An earlier draft read any map with
`kind` in `{bytes, ratio, fixed}` as a tagged form, which would have reserved
three values of this document's universal discriminator — 39 values across at
least twelve declared types. See Part 07 §14 and
Expand All @@ -146,7 +145,7 @@ normative rule here.** They are open questions for the steward.
### Open

3. **The `-0` literal.** §5.1.2(2) bars IEEE `-0.0` from the object model. `-0`
is nevertheless an integer token in the JSON grammar, so this candidate
is nevertheless an integer token in the JSON grammar, so this contract
rejects it as `signed-zero`. That is a derivation, not a quotation.
4. **What "canonical" means for a verifier.** The encoder canonicalizes member
order and whitespace; the verifier is asked whether _these_ bytes are
Expand All @@ -155,13 +154,13 @@ normative rule here.** They are open questions for the steward.
literally. Both expectations are pinned per case in the corpus.
5. **The `circle256` LUT shape.** §5.1.2.4 requires a LUT to be fixed-point,
content-addressed, and bound into the domain descriptor. Its member layout is
this candidate's.
this contract's.
6. **Quantizing signed zero.** `-0.0` maps to the integer `0`, following
§5.1.2(2)'s "an exact integral result MAY be converted to the integer `0`
before that boundary".
7. **How a full content digest is spelled.** §5.1.2.2 says `unit_ref` is "either
`null` or a full content digest" without fixing the string form. This
candidate requires 64 lowercase hexadecimal characters and rejects an opaque
contract requires 64 lowercase hexadecimal characters and rejects an opaque
handle, a truncated digest, or an uppercase one — fail-closed, and consistent
with §5.1's refusal elsewhere to accept a truncated handle as a reference. A
typed reference object would be the alternative, and choosing between them is
Expand Down Expand Up @@ -196,16 +195,16 @@ independent encoders** for A3 itself, which conflated a technical gate this
project can meet with an organisational precondition it cannot create — an
outside maintainer — and so blocked the whole RFC on someone else's decision.

### A3 ratification — what this project must produce
### A3 ratification — what this project produced, and did

| Artifact | Status |
| ------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| `CANONICAL_ENCODING.v0.1` | this file, **candidate** |
| The CNP-0 corpus | `probes/cnp-0-seed-v0/corpus/`, candidate, all eight §5.1.3 categories |
| A reference encoder | present as a candidate (`probes/cnp-0-seed-v0/ts/`) |
| A verifier-only rejection path sharing no code with it | present as a candidate; `ts/reject.ts` imports nothing from the encoder, which is the property that matters — it cannot repair what it judges |
| A reproducible conformance kit | **present as a candidate**, `conformance/cnp-0-jcs-v0/`. Normative extract quoted from Part 01 by byte range, 63 required cases with expected bytes and digests, a stdlib-only Python runner that scores an outside implementation, and `MANIFEST.sha256` pinning every file so an edited corpus is refused rather than quietly scored. It ships **no implementation**: scoring by agreement with our encoder would be asking an implementer to trust our encoder. |
| Steward ratification | **absent.** Nothing here is accepted, signed, or ratified. |
| Artifact | Status |
| ------------------------------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `CANONICAL_ENCODING.v0.1` | this file, **ratified** |
| The CNP-0 corpus | `probes/cnp-0-seed-v0/corpus/`, all eight §5.1.3 categories |
| A reference encoder | present (`probes/cnp-0-seed-v0/ts/`) |
| A verifier-only rejection path sharing no code with it | present; `ts/reject.ts` imports nothing from the encoder, which is the property that matters — it cannot repair what it judges |
| A reproducible conformance kit | **present**, `conformance/cnp-0-jcs-v0/`. Normative extract quoted from Part 01 by byte range, 63 required cases with expected bytes and digests, a stdlib-only Python runner that scores an outside implementation, and `MANIFEST.sha256` pinning every file so an edited corpus is refused rather than quietly scored. It ships **no implementation**: scoring by agreement with our encoder would be asking an implementer to trust our encoder. |
| Steward ratification | **present.** Ratified by `s0fractal` as steward on 2026-08-26; the statement is recorded verbatim in Part 07 §17. |

The same-author caveat that used to disqualify the second code path still
applies to what it actually bears on — it is not evidence of independent
Expand Down Expand Up @@ -237,7 +236,5 @@ file may not be cited as showing implementation diversity.
do not contain the `cnp-0` profile members, ratios, fixed-point domains, or
the rejection corpus.

The honest status remains what Part 01 already records: **A3 ratification gate
defined; contract, corpus, reference encoder, verifier-only path and conformance
kit present as candidates; steward ratification pending; adoption not evidenced;
independent interoperability unconfirmed.**
The honest status remains what Part 01 already records: **A3: RATIFIED.
adoption-evidenced: false. interop-confirmed: false.**
8 changes: 4 additions & 4 deletions conformance/cnp-0-jcs-v0/MANIFEST.sha256
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@
88cf41c026fd0d04cdab555987dd4e81d0b660d68eea3bd3a10d96c6ee75988a CONTRACT.md
bfdc30247c4c399f9b9ab6ccd04c9de56785d5850f881b48064a1a69a8c6aa3d CONTRACT.md
b040b943c3cd4b13e1e18fd776b8ceef98c43272ef264344b7d2cb715ad7c172 INTERFACE.md
f85b040ba44c07c69b231ecb63a527604240dcf94dc7f9b67c51ffdffbb5f205 README.md
013d8af4e8e554d8fc852d6c14ff482fac9485bad5a7887cadfe47b827311e96 SPEC-EXTRACT.md
b19a92a73cbec859a853e4ed12c11dde7ab9c77269e311e16a4d901d9ba12e52 README.md
4f289ee10fcb3e5698b8823af6241291b34b326406354a40d79708371b39ab0d SPEC-EXTRACT.md
17aaf7ed6d92040b0d1c4b88f4157912c95029368b6d28e56f2144c103082e5a corpus/extended.ndjson
dde58a183fa746900cc82de3f1935a09177483297ab16a650c2cba25c07e9e19 corpus/required.ndjson
7c7fee476b75768345e882c27bc24066a7f3896472de3b3023750c8221248041 run_conformance.py
e7fc89b00d260bbf9a145d0e599b4d14d150b70b87eae8988f5261c92de49317 selftest.py
4b1f5f6b08940f18dd5494e6ff13899e133ce19762af9883170029fef346c2a8 tools/build_kit.py
e85d72a6e33ea26053a98f375ebbdb07fa8e7b93877ff1ae1a56b078faaa6198 tools/build_kit.py
2e909d228c1aea437544581737471cbcf127504fdc2052cdaacd30588d09f06b ts/kit_test.ts
11 changes: 9 additions & 2 deletions conformance/cnp-0-jcs-v0/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -154,5 +154,12 @@ extract is quoted from Part 01 by byte range. `tools/build_kit.py --check`
re-derives both and fails on a one-byte drift in either direction, so this kit
cannot silently diverge from the specification it claims to carry.

Status: **candidate**. This kit is itself an A3 ratification requirement and is
not yet ratified.
Status: **ratified**. This kit is one of the artifacts Tranche A3 required, and
the steward ratified A3 on 2026-08-26.

That settles the specification. It does not settle the world:
`adoption-evidenced` is **false** — no substrate computes references under
`hsp-jcs@v0` today — and `interop-confirmed` is **false** — no independently
maintained implementation exists. If you build one and it passes, you move the
second of those, and that is the evidence this project cannot produce for
itself.
Loading
Loading