Standard: CRG v2.0 STRICT Current Grade: E Assessed: 2026-07-21 (demoted D → E) Assessor: Jonathan D.A. Jewell
The previous assessment (2026-04-12) recorded Grade D on the strength of
a parser, AST, recursive-descent implementation and a 57-test matrix,
all of which lived in KRLAdapter.jl. That repository no longer
exists — it was discarded, deliberately and not recoverably.
None of the D evidence can be checked. Under the CRG demotion table,
D → E applies when "`the scope narrows so far that the component
barely does anything`", which is precisely what happened: with the
adapter gone, nothing in this repository can parse or execute a KRL
program.
This is a correction to the record, not a regression in the work. The grade was restated rather than left standing on evidence nobody can inspect.
Grade E criterion: "`Does something slight … there is a kernel of value … at least one successful test case demonstrating the kernel of functionality, and documentation of known failures and limitations.`"
Every item below was executed on 2026-07-21, not inferred from documentation.
| Artefact | Check | Result |
|---|---|---|
|
114-line EBNF, v0.1.0 |
present |
|
4 example programs |
present |
|
lexical conformance of examples to the grammar |
20 checks, all pass |
|
|
3/3 pass |
|
|
produces |
|
|
4, all covered by 11 Zig exports |
|
SPDX, banned constructs, ABI/FFI correspondence |
4/4 pass |
|
full local pipeline |
4/4 pass, negative-controlled |
The kernel of value is the specification plus a set of examples that provably conform to it at the lexical level, over a C ABI that compiles and is tested.
-
No parser, and therefore no execution. Nothing in this repository can read a
.krlprogram and produce a result.grammar_smoke.shis lexical only and says so in its own header. -
The specification is contested.
spec/grammar.ebnfhere andquandledb/spec/grammar.ebnfboth claim to be KRL v0.1.0 and are disjoint on core vocabulary;|is bound to opposite meanings in the two. See README. -
No conformance suite. There is no executable artefact that an implementation can be tested against, so “conforms to the KRL spec” is not currently a checkable claim.
-
Proof obligations are unmet.
PROOF-STATUS.mdrecords 0 of 8 obligations proven, 2 partial. -
rust-ci.ymlgates nothing — it calls the shared Rust reusable, but this repository contains noCargo.toml.
Grade D requires a matrix of tested scenarios and at least one test per claimed capability. Concretely:
-
Reconcile the two grammars into one normative specification, resolving the
|collision. -
Write an executable conformance suite — programs plus expected results — so that spec conformance becomes testable rather than asserted.
-
Run that suite against
quandledb/server/krl/, the actual implementation (3,035 lines of Julia plus 1,732 lines of tests). One passing test per claimed capability is the D bar.
Until at least (1) and (2) exist, this repository specifies a language nobody can be shown to implement.
-
spec/grammar.ebnf(v0.1.0 EBNF) andspec/grammar-overview.md -
4
examples/programs -
tests/smoke/grammar_smoke.sh(16 lexical assertions)
Decision “Option B — Julia in KRLAdapter.jl”; lexer, AST,
recursive-descent parser and lowering implemented there, with 57
dedicated parser tests. The repository holding all of it has since been
discarded, so none of this is verifiable. Retained here as history, not
as evidence.
-
Grade restated against what is actually present and runnable in this tree.
-
Zig FFI shim repaired: it had never compiled (
opaquetype with fields). -
Three vacuous or false gates repaired (
aspect_tests.sh,e2e.sh). -
False
TangleIRlowering claims and deadKRLAdapter.jlreferences removed from the README.