docs: AFFIRMATION.adoc (MUST/INTEND/WISH state) + next-Claude handoff prompt (#95)

## What

Two handoff/attestation docs capturing where WokeLang stands after this
session's proof + bytecode-VM work (PRs #90–#94).

### `AFFIRMATION.adoc` (repo root, timestamped 2026-06-19)
A point-in-time **affirmation** of current state, organised as the
project's normative **MUST / INTEND / WISH** standing (using the
`Intentfile` status vocabulary):
- **MUST** — the normative core (working language, *enforced* consent,
machine-checked + CI-gated type safety, sound non-rotting proofs,
ethical scaffolding, RSR compliance): **met** for the scoped core.
- **INTEND** — committed/in-progress (VM as a real compilation target =
the live frontier; broaden the verified core; reconcile docs;
echo-types; eval↔spec correspondence).
- **WISH** — near/mid/far horizon items.
- A "where we are" synthesis + provenance (the merged PRs).

### `docs/NEXT-CLAUDE-PROMPT.md`
A copy-paste prompt to hand a fresh Claude Code session: what to read
first, current state, best next steps (parity harness, `.wbc`
serialization, consent grant-sources, workers, by-reference/multi-level
closures, float-arith proofs, doc reconciliation), and the CI/convention
gotchas (preflight `fmt` + `clippy`; proof gates).

## Notes
- Follows the RSR `.adoc` attestation conventions visible in this repo
(cf. `RSR_COMPLIANCE.adoc`, MPL-2.0 headers). The canonical
`AFFIRMATION.adoc` template in `hyperpolymath/standards` was **not
reachable from this session** (repo-management tooling unavailable) —
flagged in both the doc and here; **reconcile against that template when
available**.
- Docs-only; no code change.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

https://claude.ai/code/session_013wg3Mtq2QFhYi4XVw1Z6z7

---
_Generated by [Claude
Code](https://claude.ai/code/session_013wg3Mtq2QFhYi4XVw1Z6z7)_

Co-authored-by: Claude <noreply@anthropic.com>

Author: hyperpolymath

AFFIRMATION.adoc

@@ -0,0 +1,138 @@
+// SPDX-License-Identifier: MPL-2.0
+// SPDX-FileCopyrightText: 2026 Jonathan D.A. Jewell <j.d.a.jewell@open.ac.uk>
+= WokeLang — Affirmation of Current State
+:toc:
+:sectnums:
+:affirmed: 2026-06-19
+:revdate: 2026-06-19
+
+[abstract]
+A point-in-time *affirmation* of where WokeLang stands: a timestamped
+attestation of what is true now, organised as the project's normative
+*MUST / INTEND / WISH* standing. It is a report, not a gate — the hard gates
+live in `.machine_readable/contractiles/` (the `must` contractile).
+
+[NOTE]
+====
+*Provenance & method.* Produced in an AI-assisted (Claude Code) working
+session, reflecting the state after the proof and bytecode-VM work merged on
+2026-06-18/19 (PRs #90–#94). It follows the RSR conventions visible in this
+repository; the canonical `AFFIRMATION.adoc` template in
+`hyperpolymath/standards` was *not reachable from this session*, so reconcile
+this document against that template when available.
+====
+
+== Affirmation
+
+As of *{affirmed}*, we affirm that WokeLang is a *working, consent-driven,
+human-centred programming language* whose *type safety is machine-checked and
+CI-gated* for a deliberately-scoped expression/statement core. Its normative
+MUSTs hold. The live development frontier is making the bytecode VM a true
+second compilation target and reconciling documentation to reality.
+
+== MUST — the normative core (what WokeLang must be to *be* itself)
+
+[cols="3,1,4",options="header"]
+|===
+| MUST | Status | Where we are
+
+| A working language: lex → parse → typecheck → run `.woke`
+| ✅ met
+| Rust toolchain (parser, typechecker, interpreter, bytecode VM, LSP, DAP, formatter, linter, REPL); 188 lib tests; OCaml reference core
+
+| Consent-driven safety actually *enforced* (the defining feature)
+| ✅ met
+| Interpreter always; the *VM now enforces* under `#care` (deny-by-default, PR #92); consent lives in the type system of the Idris calculus
+
+| Machine-checked type safety (progress + preservation)
+| ✅ met (core)
+| Lean + Coq (expression core + arrays + statement typing) and the Idris consent calculus. A *deliberate subset* of the surface language
+
+| Proofs sound and non-rotting (CI-gated, `sorry`/axiom-free)
+| ✅ met
+| `lean-proofs.yml` + `coq-proofs.yml`; axiom-free bar Coq's unavoidable reals axioms
+
+| Human-centred + ethical scaffolding (consent, gratitude, emotes, Result)
+| ✅ met
+| In the interpreter and now the VM
+
+| RSR repository compliance (license/governance/contractiles)
+| ✅ mostly
+| Template `Mustfile` checks pass; commit-signing `must` is the soft spot
+|===
+
+== INTEND — committed next-actions (we WILL do this)
+
+[cols="3,1,4",options="header"]
+|===
+| INTENT | Status | Where we are
+
+| Make the bytecode VM a real compilation target
+| in_progress
+| Major strides 2026-06-18/19: machine opcodes, full expr/stmt coverage, consent enforcement, Result constructors, closures + upvalue capture. Remainder → wishes
+
+| Broaden the verified core toward the Rust type system
+| in_progress
+| binop parity + arrays + statement typing landed; float arithmetic still open
+
+| Reconcile documentation to reality
+| declared
+| README/EXPLAINME stale; `TOOLCHAIN-WISHLIST` lists *built* tools (LSP/DAP/WASM) as "missing"; completion figure contradicts (`STATE.a2ml` 35% vs wishlist 85%)
+
+| Integrate echo-types into the type checker
+| declared
+| `STATE.a2ml` frontier; not started
+
+| Resolve eval ↔ spec correspondence (PROOF-NEEDS #2)
+| declared
+| Needs a maintainer decision (treat the spec as normative, or build a second model)
+|===
+
+== WISH — horizon aspirations (we WISH this were true)
+
+[cols="4,1,1",options="header"]
+|===
+| WISH | Horizon | Status
+
+| Interpreter↔VM parity harness (conformance through both) | near | declared
+| `.wbc` bytecode serialization | near | declared
+| Consent grant-sources (`superpower` declarations) on the VM | near | declared
+| Worker / concurrency statements on the VM | near | declared
+| By-reference + multi-level closure capture | near | declared
+| Full surface-language type safety (statement dynamics, units, pattern matching) | mid | declared
+| WASM codegen correctness | mid | declared
+| Remaining dev tooling (debugger UI, profiler, package manager, doc-gen) | mid | declared
+| Ordinal/echo Bachmann–Howard consumer-evidence; formal VM↔spec compiler-correctness | far | declared
+| Establishment (paper / Zenodo) + production-grade toolchain | far | declared
+|===
+
+== Where we are (synthesis)
+
+WokeLang's MUSTs are met for a deliberately-scoped core: a real, running,
+consent-enforcing language whose type safety is machine-checked and CI-gated.
+The frontier is the INTEND layer — turning the bytecode VM into a genuine
+second compilation target (which moved a great deal this session: it went from
+crashing on most programs to running the core language *with consent enforced
+and closures that capture state*) and reconciling the stale documentation and
+scope numbers. Everything past that is the WISH horizon.
+
+The one honest asterisk: `STATE.a2ml` pegs whole-vision completion at *~35%* —
+that figure counts the entire surface language plus full tooling plus ecosystem
+(the large WISH/INTEND surface), so it coexists with "the core MUSTs hold." The
+gap between "MUSTs met" and "35%" *is* the roadmap above.
+
+== Provenance
+
+Work merged to `main` that this affirmation reflects:
+
+* PR #90 — Lean array typing (mirror of Coq `T_Array` / `T_Lit_Array`)
+* PR #91 — statement typing judgment + metatheory (Lean + Coq)
+* PR #92 — VM: full expression coverage + Result/record/index execution + consent enforcement under `#care`
+* PR #93 — VM: Result constructors from surface syntax + closures
+* PR #94 — VM: upvalue capture for closures
+
+Deeper detail: `docs/proofs/verification/AUDIT.md` (proofs) and the per-PR
+descriptions. Next-session handoff: `docs/NEXT-CLAUDE-PROMPT.md`.
+
+// Affirmed 2026-06-19 (UTC date) in an AI-assisted Claude Code session, on
+// behalf of the maintainer. Reconcile against hyperpolymath/standards.

docs/NEXT-CLAUDE-PROMPT.md

@@ -0,0 +1,72 @@
+<!--
+SPDX-License-Identifier: MPL-2.0
+SPDX-FileCopyrightText: 2026 Jonathan D.A. Jewell <j.d.a.jewell@open.ac.uk>
+-->
+# WokeLang — handoff prompt for the next Claude
+
+Paste the block below into a fresh Claude Code session working on
+`hyperpolymath/wokelang`. (State as of 2026-06-19 — see `AFFIRMATION.adoc`.)
+
+---
+
+You are continuing work on **WokeLang**, a consent-driven, human-centred
+programming language. Primary implementation is **Rust** (`src/`), with an
+**OCaml** reference core (`core/`) and formal proofs in **Lean/Coq**
+(`docs/proofs/verification/`) and **Idris 2** (`src/abi/`). Match the
+interpreter (`src/interpreter/mod.rs`) as the reference semantics.
+
+**Read first, in order:**
+1. `AFFIRMATION.adoc` — the current MUST / INTEND / WISH standing (timestamped).
+2. `.machine_readable/6a2/STATE.a2ml` — project state checkpoint.
+3. `docs/proofs/verification/AUDIT.md` — proof state (Lean / Coq / Idris).
+4. `PROOF-NEEDS.md`, `TEST-NEEDS.md`, `TOOLCHAIN-WISHLIST.md` — needs and wishes.
+   (Note: `TOOLCHAIN-WISHLIST.md` is **stale** — LSP, the DAP debugger, and the
+   WASM backend already exist; and it says 85% complete while `STATE.a2ml` says
+   35%. Reconciling these is itself an open task.)
+
+**Where things stand (2026-06-19):**
+- **Proofs.** Lean + Coq at parity (expression-core type safety + arrays +
+  statement typing) and an Idris consent calculus (typing with a handler
+  context + small-step with an audit log + `progress`). CI-gated
+  (`lean-proofs.yml`, `coq-proofs.yml`), `sorry`/axiom-free.
+- **Bytecode VM** (`src/vm/`). Now runs the core language: Result types,
+  records, indexing, control flow, **consent enforcement under `#care`**
+  (deny-by-default), and **closures with upvalue capture**. 188 lib tests green.
+  It went from crashing on most programs to running the core language this
+  session (PRs #92–#94).
+
+**Best next steps (near-horizon, mostly independent — pick one):**
+- *VM → real compilation target:*
+  - **Interpreter↔VM parity harness:** run each `conformance/*.woke` and
+    `examples/*.woke` through both the interpreter and `vm::run_vm`, assert
+    equal results, CI-gate it. (Highest-value correctness mechanism.)
+  - **`.wbc` bytecode serialization:** `woke compile -o x.wbc` cannot write a
+    loadable file yet (`src/main.rs` says "not yet implemented"); add serde
+    round-trip so `woke run-vm x.wbc` loads it.
+  - **Consent grant-sources:** seed the VM's `granted` set (in `src/vm/machine.rs`)
+    from `superpower` declarations so `#care` programs can actually grant
+    capabilities (currently deny-by-default).
+  - **Worker statements on the VM** (the 5 concurrency forms — needs a VM
+    channel/scheduler model).
+  - **By-reference + multi-level closure capture** (current capture is by-value,
+    single-level).
+- *Proofs:* **float arithmetic** variants; **open-term expression preservation**
+  (the prerequisite for statement *dynamic* safety); decide the **eval ↔ spec
+  correspondence** question (PROOF-NEEDS #2) with the maintainer.
+- *Docs:* **reconcile** `README.adoc` / `EXPLAINME.adoc` / `TOOLCHAIN-WISHLIST.md`
+  to reality and fix the 35%-vs-85% completion contradiction.
+
+**Conventions & CI gates:**
+- Develop on the branch you are told to; push and open a **draft PR**.
+- Gates include **Build Check** (`cargo fmt --check` + `cargo clippy --lib --bins
+  -- -D warnings`), **Build + E2E (Rust + OCaml)**, **lean-proofs**, **coq-proofs**.
+  **Run `cargo fmt` + `cargo clippy --lib --bins -- -D warnings` locally before
+  pushing** — Build Check gates on both, and that has bitten past sessions.
+- Language policy (`.claude/CLAUDE.md`): Rust / OCaml / Lean / Coq / Idris as
+  present; no Go / Python / TypeScript; `Containerfile`, not `Dockerfile`.
+- Keep proofs `sorry`/axiom-free; pin each headline theorem the way `AUDIT.md`
+  describes.
+
+Tell me which item to take, or propose one and go.
+
+---