Skip to content

READINESS.adoc

Latest commit

 

History

History
186 lines (140 loc) · 7.27 KB

READINESS.adoc

File metadata and controls

186 lines (140 loc) · 7.27 KB
Raw

llm-grace Component Readiness Assessment

Standard: Component Readiness Grades (CRG) v2.0 STRICT
Assessed: 2026-05-18
Assessor: Jonathan Jewell
Previous assessment: <date or "none">

Current Grade: X

This line is parsed by just crg-grade / just crg-badge. The grade above is the worst-graded in-scope component — the project’s weakest link sets its grade. See the per-component table in §3.

Note

Honest grading. Per CRG v2.0 Principle 4: grade as-is today, not aspirationally. Long alpha is discipline, not shame. Demote immediately if evidence doesn’t support the claim.

CRG v2.0 Grade Reference

Grade Name Release Stage Stability Posture Shorthand

X

Untested

—

—

—

F

Harmful / Wasteful

—

—

reject/delegate

E

Minimal / Salvageable

Pre-alpha

Unstable

pre-alpha

D

Partial / Inconsistent

Alpha

Unstable

alpha-unstable

C

Self-Validated

Alpha

Stable in home context

alpha-stable

B

Broadly Validated

Beta

Stable for broad trial

beta-stable

A

Field-Proven

Stable

Stable

stable

Evidence gates (v2.0 is stricter than v1.0):

  • D: RSR-compliant + per-capability tests + documented scope. Test matrix must cite counts and live in .machine_readable/6a2/STATE.a2ml or CI.

  • C: All of D, plus active dogfooding in home context with no known home-context failures over an evidence window, plus deep per-file and per-directory annotation (purpose, boundaries, invariants, execution/test/proof surfaces, per-directory orientation READMEs). STATE.a2ml [dogfooding-status] populated with concrete "done — <evidence>" entries.

  • B: All of C, plus six genuinely diverse external targets with feedback fed back. STATE.a2ml [external-targets] holds target identities + dates + outcomes; [issues-fed-back] holds the closed-issue trail. Internal-capability items do not count.

  • A: All of B, plus multi-source real-world external feedback confirming value, no harm in the wild. STATE.a2ml [field-signal] populated.

Publication gate: non-abstract implementation claims require B+. Below B, any publication must be explicitly abstract or provisional.

Headline Evidence (as of 2026-05-18)

Metric Value Source

Test count

<N>

.machine_readable/6a2/STATE.a2ml or CI

Formal-verification posture

<N believe_me / sorry / Admitted — irreducible only>

grep of proof dirs

Dangerous patterns (assert_total, unsafeCoerce, Obj.magic)

<N>

grep 2026-05-18

Per-unit README coverage

<M/N (percentage)>

filesystem scan of unit dirs

CI status

<workflow name + last green run>

.github/workflows/

RSR mandatory workflows

<N/17>

.github/workflows/

Third-party badges (if any)

<e.g. Glama AAA; Scorecard N/10>

external

LIVE deployment (if any)

<URL + last health check>

production

Component Assessment

All components graded as-is today, per CRG v2.0 Principle 4. Stability posture reflects the component’s state within its home context only unless noted.

Component Grade Posture Evidence Summary Promotion blocker Last Assessed

<component 1>

X/F/E/D/C/B/A

<shorthand>

<what was checked + counts>

<specific, actionable gap to next grade>

2026-05-18

<component 2>

…

…

…

…

2026-05-18

<component 3>

…

…

…

…

2026-05-18

Rules of thumb for populating this table:

  • One row per independently-gradable unit. If a subsystem can ship or break independently, it gets its own row.

  • Evidence Summary must cite numbers (test counts, LOC, file counts) or paths (e.g. src/abi/Module.idr), never vague claims.

  • Promotion blocker must be actionable — "add external consumer in home context, then annotate" beats "needs more validation".

  • Re-assess every release cycle; date-stamp each row.

What’s Needed for D → C

CRG v2.0 requires two new pieces of evidence on top of D:

  1. Active dogfooding in home context with no known home-context failures.

    • Start date: <date of first dogfood-log entry>.

    • Home context: <what "home" means for this project>.

    • "No known failures" is a moving claim — must hold continuously across the evidence window (recommended: 4 weeks, daily use).

    • Populate STATE.a2ml [dogfooding-status] with one entry per capability: capability = "done — <concrete evidence>".

  2. Deep code and folder annotation.

    • Per-directory orientation READMEs in every non-trivial source subtree.

    • Per-file header comments: purpose, boundaries, invariants, execution/test/proof surface.

    • Per-unit (cartridge/panel/plugin/module) README covering: purpose, tools, architecture-at-a-glance, build steps. Overview-level alone is not sufficient — depth is required where a reviewer would otherwise have to read source to orient.

Minimum first-ring targets for C promotion: list the trunk components here. If these aren’t C, nothing else can be.

What’s Needed for C → B

Six genuinely diverse external targets with feedback fed back into the component. Candidate diversity axes:

  • Different language runtime (not just variants of the same one).

  • Different OS family (at least one must not be Linux).

  • Different hardware class (cloud / server / desktop / embedded / mobile).

  • Different auth posture (unauthenticated, API-key, vault-brokered, mTLS).

  • Different topology (star, mesh, peer-to-peer).

  • Different trust model (same-org, federated, adversarial).

Populate STATE.a2ml:

  • [external-targets] — target-id → "<date> — <outcome> — <contact/URL>".

  • [issues-fed-back] — issue-id → "<date> — <external reporter> — <resolution>".

Six variations of the same use case do not count. Re-classify any items currently under [grade-b-status] (legacy) and move external-eligible ones into [external-targets].

What’s Needed for B → A

Real-world external feedback confirming value. Populate STATE.a2ml [field-signal] with multi-source entries:

  • External users beyond the six B-targets.

  • Third-party writeups, talks, papers, or testimonials.

  • No unresolved safety / correctness incidents in the last 90 days.

Summary (2026-05-18)

  • Project grade: <grade>. Why: <1-2 line justification>.

  • Delta since last assessment: <what changed; none if first assessment>.

  • Next milestone: <specific promotion target + timeline>.

  • Machine-readable grade line present (§ header) for just crg-grade / just crg-badge.

Companion Artefacts

  • docs/governance/CRG-CRITERIA.adoc — grade definitions (boilerplate).

  • docs/governance/CRG-AUDIT-<DATE>.adoc — formal audit (populate from rsr-template-repo/docs/governance/CRG-AUDIT-TEMPLATE.adoc).

  • docs/practice/DOGFOOD-LOG.adoc — dated dogfood evidence (required for C).

  • .machine_readable/6a2/STATE.a2ml — authoritative state (canonical keys [dogfooding-status], [external-targets], [issues-fed-back], [field-signal]).

Run just crg-badge to generate the shields.io badge for your README.