feat(toolchain-readiness-grades): add TRG v1.0 spec + 25 reference audits

Toolchain Readiness Grades v1.0 — strict superset of CRG for the core
toolchain components of a programming-language project (grammar, lexer,
parser, AST, semantic analyser, type checker, IR, codegen, linker,
runtime, stdlib, REPL, formatter, LSP, diagnostics, debugger, package
manager, proof dispatcher, ABI/FFI surface, backend registry).

Grade-compatible with CRG (X|F|E|D|C|B|A) but with significantly higher
evidence bars because toolchain components are critical infrastructure.
Promotion cascade: banned constructs ZERO at E, panic-attack zero
Critical/High at E, A2ML+manifest+Immaculate Guide at E, RSR-FULL at D,
fuzz harness at D, full Testing-Taxonomy 16x14 in Six-Sigma range at C
(alpha gate), multi-prover (>=2 of Idris2/Lean4/Rocq/Agda) at B,
multi-prover (>=3 with one kernel-soundness-proven) at A, type-theory
upstream contribution at A.

A-grade primary release gate is Six Sigma vs trailing-12-month IPO
cohort + 5M external processor-hours (<=10% internal CI) + 21
fully-proven programs cross-validated by >=3 independent provers + full
hyperpolymath standards conformance (RSR 90d-green, launch-scaffolder,
PanLL, Hypatia zero, panic-attack zero, VeriSimDB, Stapeln, Groove,
SLSA L3, reproducible builds) + unanimous 6+6 (humans + LLMs) signoff.

Demotion is immediate on loss of any gate (no grace period). Per-language
extension hook at \u00a71.1 for tightening profiles (Eclexia, AffineScript
etc.). Canonical Mathematical/Scientific/Engineering Proof Suite at \u00a76.6
as the gate to leave grade E.

References: 25 component-audit MK2 documents from 007 as canonical
exemplars of the four-tier MoSCoW audit format (Must -> E, Should -> D,
Could -> C, Aspirational -> B).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: hyperpolymath

toolchain-readiness-grades/README.adoc

@@ -0,0 +1,122 @@
+// SPDX-License-Identifier: PMPL-1.0-or-later
+// Copyright (c) 2026 Jonathan D.A. Jewell (hyperpolymath) <j.d.a.jewell@open.ac.uk>
+
+= Toolchain Readiness Grades (TRG)
+:toc:
+
+*Status:* Active, v1.0, 2026-04-12 +
+*License:* PMPL-1.0-or-later (MPL-2.0 automatic legal fallback) +
+*Companion to:* Component Readiness Grades (CRG) v2.2 +
+*Part of:* Rhodium Standard Repositories (RSR)
+
+== What this is
+
+A *strict superset* of CRG specifically for the *core toolchain components*
+of a programming-language project: grammar, lexer, parser, AST, semantic
+analyser, type checker, IR, codegen, linker, runtime, stdlib, REPL,
+formatter, LSP, diagnostics, debugger, package manager, proof dispatcher,
+ABI/FFI surface, backend registry.
+
+TRG uses the same letter grades as CRG (`X | F | E | D | C | B | A`) but
+the evidence required at each grade is significantly higher, because
+toolchain components are *critical infrastructure* — every program
+ever compiled by the toolchain inherits its bugs.
+
+== Why it exists
+
+Three reasons:
+
+. CRG is a general-purpose component scheme. Toolchains need a tighter
+  rubric.
+. The 007 language project already runs MK2-style four-tier MoSCoW audits
+  (`audits/audit-<component>.md`). TRG promotes that pattern to an
+  estate-wide standard, with grade-thresholds attached to each tier.
+. Hyperpolymath's standing priority order (dependability > security >
+  interop > usability > performance > versatility > functional-extension)
+  needs an enforcement vehicle for the parts of the estate that sit upstream
+  of everything else.
+
+== Files in this directory
+
+[cols="2,5",options="header"]
+|===
+| File | Purpose
+| `TOOLCHAIN-READINESS-GRADES.adoc` | The normative spec. Read this first.
+| `README.adoc`                     | This file.
+| `references/007/`                 | The 24 component audit documents
+                                     copied verbatim from the 007 language
+                                     project, plus the seed lexer-MK2 audit
+                                     that originated the four-tier format.
+                                     These are the canonical exemplars of
+                                     the per-component audit format.
+| `templates/`                      | (Forthcoming) `AUDIT-TEMPLATE.adoc`,
+                                     `CANONICAL-PROOF-SUITE.adoc`,
+                                     `QUALIFYING-PROVERS.adoc`,
+                                     `FUZZING-CORPUS-FLOOR.adoc`,
+                                     `A-GRADE-LLM-PANEL.adoc`. Skeletons
+                                     for the templates referenced in the
+                                     spec.
+|===
+
+== Quick mental model
+
+[horizontal]
+X:: absent / not working / any Must row failed
+F:: tested + harmful / wasteful (delegate or remove)
+E:: pre-alpha. All Must passed. Banned constructs ZERO. panic-attack
+    zero Critical/High. Six A2ML files + 0-AI-MANIFEST present. Immaculate
+    Guide passing.
+D:: pre-alpha+. All Must + Should + the *Canonical Proof Suite* (this is
+    the foundational gate). RSR-FULL. panic-attack zero across all
+    severities. Fuzzing harness exists.
+C:: *alpha*. All Must + Should + Could. Full Testing-Taxonomy 16×14
+    coverage in Six Sigma range. Continuous fuzzing 14d. 5 fully-proven
+    programs. 3-nines on top-4 priority axes.
+B:: *beta*. All four tiers passed. Multi-prover cross-validation (≥2 of
+    {Idris2, Lean 4, Rocq, Agda}). Single external audit. Continuous
+    fuzzing 30d. 100 distinct external users. 5-nines top-4, 4-nines rest.
+A:: *stable / release*. Six Sigma vs trailing-12-month IPO cohort (bottom
+    decile). 5,000,000 external processor-hours. 21 fully-proven programs
+    cross-validated by ≥3 independent provers. Type-theory upstream
+    contribution in last 12 months. 1000 distinct external users.
+    Unanimous 6+6 (humans + LLMs) signoff. *Plus* full hyperpolymath
+    standards conformance (RSR, CI/CD 90d green, launch-scaffolder, PanLL,
+    Hypatia zero, panic-attack zero, VeriSimDB, Stapeln, Groove, SLSA L3,
+    reproducible builds, Coordination.k9, UX Manifesto).
+
+== Demotion is immediate
+
+Loss of any A-grade gate (canonical proof regresses, six-sigma slips,
+processor-hours window drops, signoff rescinded, standards drift)
+immediately demotes the component. There is no grace period. See §5.8.
+
+== How strict is this really?
+
+See §11 of the spec. Honest answer: TRG-A is, to the author's knowledge,
+*not matched by any standard currently in use anywhere*. The author expects
+approximately *zero* hyperpolymath toolchain components to hit A in the
+first year, and is comfortable with the possibility that *no toolchain
+component anywhere reaches A for many years*. That is the correct order of
+magnitude. If A becomes easy to hit, the bar is wrong and the standard MUST
+be raised.
+
+== Per-language profiles
+
+TRG is the *baseline*. Each hyperpolymath language is expected to publish
+a `spec/TRG-PROFILE.adoc` in its own repo that *tightens* TRG with
+language-specific obligations (e.g. Eclexia tropical-semiring soundness,
+AffineScript region/affinity discipline, Idris2 totality discipline). A
+profile may only tighten, never loosen. See §1.1.
+
+== Interaction with CRG
+
+A project that ships a language toolchain MUST publish *both*: a CRG grade
+for the project as a whole, and a TRG grade for the toolchain. Where the
+two rubrics overlap, the *stricter* grade governs. TRG is always at least
+as strict as CRG.
+
+== Open invitation
+
+If you think TRG is still too lax (or too strict in a way that lets bad
+work hide), file an issue with concrete evidence and a proposed
+tightening. TRG earns trust by surviving critique.