grammar artifacts: .k9 has no grammar or contract at all, and deed.abnf states both “sole normative” and “pending owner ruling”

[arena-ai-coding-agent]

Two grammar-artifact findings, from auditing what actually exists after #856.

---

Finding 1 — K9 has no grammar artifact, and may not need the one A2ML has

Every A2ML family ships an ABNF:

1-formats/a2ml/agentic/spec/agentic.abnf
1-formats/a2ml/anchor/abnf/anchor.abnf
1-formats/a2ml/ecosystem/spec/ecosystem.abnf
1-formats/a2ml/ecosystem/spec/family/scm-core.abnf
1-formats/a2ml/meta/spec/abnf/meta.abnf
1-formats/a2ml/meta/spec/shared/s-expression.abnf
1-formats/a2ml/state/spec/abnf/state.abnf

K9 ships none. 1-formats/k9/ carries prose only — SPEC.adoc, GUIDE.adoc,
IANA-MEDIA-TYPE-APPLICATION.adoc, TESTING.adoc, K9-ADOPTION-EXPANSION.adoc — and there is no
spec/abnf/ anywhere in it.

The reason this is not simply "write the missing ABNF"

A2ML has its own syntax (s-expressions, ; comments, the =-is-not-a-deed rule), so a grammar
file is the right artifact for it. K9 does not have its own syntax. A K9 file is a
*.k9.ncl — a Nickel file — distinguished only by a magic first line:

K9!
# SPDX-License-Identifier: MPL-2.0
{ pedigree = { schema_version = "1.0.0", security = { leash = 'Hunt, ... }, ... } }

SPEC.adoc states this directly: the format "operates via the must-just-nickel triad:
environment detection (must), task orchestration (Just), and typed validation (Nickel)".

So an ABNF that covers the whole file would be re-specifying Nickel. The grammar that is
actually missing is only the envelope, and the validation that is actually missing is a
normative contract, not a grammar:

Artifact	Exists?	Should exist?
Nickel's own grammar	upstream, in Rust	not ours to write
K9! magic envelope (K9! = \x4B\x39\x21) + SPDX header + top-level shape	prose in SPEC §Status	yes — small, ABNF-able
normative Nickel contract for pedigree (schema_version, component_type, security, metadata, warnings, side_effects)	no — only prose and the template-*.k9.ncl examples	yes — this is the real gap
normative leash taxonomy	prose (Hunt / Yard / Kennel, with `template-hunt	kennel
media type	drafted (IANA-MEDIA-TYPE-APPLICATION.adoc, application/vnd.k9)	in progress

Recommendation: rule on which of these K9 needs, rather than assuming the A2ML shape
transfers. My read is: envelope ABNF (small) + a normative Nickel contract (the load-bearing
one) + the leash set as a closed enum. Validation today appears to run through
.githooks/validate-k9.sh, .github/workflows/k9-contractile.yml and the estate's
k9_typecheck tests — worth confirming whether those check a specified contract or each
consumer's local assumption, because that difference is the whole point of a spec.

Naming hazard to carry into any K9 spec work

SPEC.adoc warns, in its own IMPORTANT block:

"K9's execution triad is must / just / nickel. It is not the contractiles family
must / trust / dust / intend, which is a separate estate concept. The word must appears in
both and means different things; never conflate them."

Any K9 contract should namespace or otherwise disambiguate must, since DEED conversion is
simultaneously pulling the contractiles family (which owns the other must) into the same tree.

---

Finding 2 — deed.abnf v1.0.0 contains a self-contradicting ruling note

1-formats/deed/spec/abnf/deed.abnf is now the sole normative grammar (#856 landed this; hazards
1–2 of #837 are cleared — thank you). Its header comment still reads, in sequence:

; NOTICE (standards#837): OWNER RULING 2026-09-19 — the grammar files have
; come together: THIS is the sole normative grammar (v1.0.0), ...
; The v0.1.0 DRAFT archive lives at
; archive/deed.abnf_v0.1.0-draft under its true version. Its only extra rule
; (version-field) is superseded: v1.0.0 folds :schema-version into `field`
; with the "exactly once" side condition.
; pending owner ruling. Grammar below is unchanged.

"OWNER RULING 2026-09-19 … THIS is the sole normative grammar" and "pending owner ruling" are
in the same block. A reader parsing this file as a normative artifact cannot tell whether the
ruling landed. The trailing fragment looks like a leftover from the pre-ruling revision.

Ask: delete or complete the pending owner ruling fragment so the normative grammar states
its own status unambiguously. Anyone writing a validator from this file will otherwise hand-encode
whichever reading they happened to land on — which is precisely hazard 4 of #837
("the filename-dispatch ordering side condition must be hand-encoded in every validator").

While there: the archive entry is named deed.abnf_v0.1.0-draft but R1 in the #837 ruling
register proposed deed.abnf_v0.1. Confirm the intended canonical name, since the register and
the tree currently disagree.

---

Acceptance

• Rule whether K9 needs an envelope ABNF, a normative Nickel contract, or both — do not
  assume A2ML's shape transfers.
• If a contract is ruled in: state whether today's validate-k9.sh / k9-contractile.yml
  implement it or their own local assumptions.
• Record the leash taxonomy (Hunt / Yard / Kennel) as a closed set with its
  capability differences.
• Resolve the must naming collision between K9 and the contractiles family before the
  contractiles translate to DEED.
• Clean the pending owner ruling fragment from deed.abnf; confirm the archive filename.
• Negative controls for whatever is specified — the estate has shipped gates that could never
  fire (rsr-antipattern.yml: BUILTIN_GLOBS bash block stranded outside Python heredoc (exit 127) #49, Hypatia dogfooding job red estate-wide — unresolvable setup-beam pins (companion to hypatia-side fix) #64), and a grammar with no rejecting validator is the same class of artefact.

Refs: standards#837, #856, #752; 1-formats/k9/SPEC.adoc; 1-formats/deed/spec/abnf/deed.abnf.