# pd.3 — retire prelude.ail.json + carve-out cleanup — Implementation Plan > **Parent spec:** `docs/specs/0028-prelude-decouple.md` > > **For agentic workers:** REQUIRED SUB-SKILL: use `skills/implement` > to run this plan. Steps use `- [ ]` checkboxes for tracking. **Goal:** Close the prelude-decouple milestone by deleting `examples/prelude.ail.json` from the working tree and sweeping the remaining `include_str!` consumers + spec text + carve-out inventory that depended on it. After this iter the only on-disk source of the prelude is `examples/prelude.ail`, the form-a-default-authoring spec's §C4 (b) carve-out category is empty, and a new pin test guarantees the JSON file stays gone. **Architecture:** Pure deletion + adapt iter. Two `include_str!` of `prelude.ail.json` survive pd.2: (a) the migrate-bare-cross-module-refs defensive include in `crates/ail/src/main.rs:472-484` (a documented no-op — the inserted entry is unconditionally skipped by the loop's `if mod_name == "prelude" && path.as_os_str().is_empty() { continue; }` guard at lines 500-505); (b) the cross-form-identity preflight test in `crates/ailang-surface/tests/prelude_module_hash_pin.rs` (which the spec explicitly scoped as pd.2-only — its purpose was to ratify pd.3's load-bearing assumption, which it did at hash `3abe0d3fa3c11c99`). Both must go before the JSON file. Plus one in-mod test in `crates/ailang-core/src/workspace.rs` (~line 2654 area, pd.2-introduced) re-points from inline `serde_json::from_str(include_str!("prelude.ail.json"))` to `ailang_surface::parse_prelude()` via the existing core→surface dev-dep edge. **Tech Stack:** `ail` CLI (one block delete + lockstep skip-branch delete), `ailang-surface/tests/` (one test fn + supporting bytes delete + new carve-out pin), `ailang-core` (one in-mod test re-point), `ailang-core/tests/carve_out_inventory.rs` (list + comment update), `docs/specs/0025-form-a-default-authoring.md` (§C4 (b) status note), `examples/prelude.ail.json` (deleted). --- **Files this plan creates or modifies:** - Create: `crates/ailang-surface/tests/prelude_decouple_carve_out_pin.rs` — asserts `examples/prelude.ail.json` does NOT exist on disk. - Modify: `crates/ailang-surface/tests/prelude_module_hash_pin.rs` — delete the `prelude_ail_and_json_parse_to_identical_module` fn and its supporting `PRELUDE_JSON` const + `PRELUDE_AIL` orphan-import + the `const _: &str = PRELUDE_AIL;` shape-pin. Trim the module doc-comment to describe only the surviving long-term hash anchor. - Modify: `crates/ailang-core/src/workspace.rs` — re-point the in-mod test that uses `include_str!("../../../examples/prelude.ail.json")` (around line 2654 area; pd.2 introduced this as a replacement for the deleted `load_prelude()` call) to `ailang_surface::parse_prelude()` via the core→surface dev-dep edge. - Modify: `crates/ail/src/main.rs:472-505` — delete the defensive `include_str!` block (472-484) AND its lockstep skip-branch (500-505, the comment + the `if mod_name == "prelude" && path.as_os_str().is_empty() { continue; }` guard). The guard is dead-code without the synthetic insert. - Modify: `crates/ailang-core/tests/carve_out_inventory.rs` — drop the `"prelude.ail.json"` entry from the EXPECTED list (line 25); rewrite the doc-comment block (lines 6-11) to "Seven carve-outs" and "§C4 (a) subject-matter: 7 fixtures (canonical-form rejection / unbound / class-def rejection); §C4 (b) compile-time-embed: retired by prelude-decouple milestone, 2026-05-14"; update the header sentence at line 2 from "eight" to "seven". - Modify: `docs/specs/0025-form-a-default-authoring.md:270-311` — add a "Status: RETIRED 2026-05-14 by prelude-decouple milestone" marker at the top of §C4 (b) and a one-paragraph epitaph at the bottom; preserve the original §C4 (b) text below as historical context. Upstream callouts (lines 232-236, 462-465, 477-484) stay as historical references — their "as of 2026-05-13" temporal qualifiers preserve correctness. - Delete: `examples/prelude.ail.json` (the milestone's primary artefact retirement). --- ### Task 1: Re-point the in-mod test in `workspace.rs` from `include_str!("prelude.ail.json")` to `ailang_surface::parse_prelude()` **Files:** - Modify: `crates/ailang-core/src/workspace.rs` — the in-mod test that inlined `serde_json::from_str(include_str!("prelude.ail.json"))` as a pd.2 replacement for the deleted `load_prelude()` call. Per recon, the reference is at lines 2654 / 2656 (inside `#[cfg(test)] mod tests`). The dev-dep edge `ailang-core → ailang-surface` (declared in `crates/ailang-core/Cargo.toml`'s `[dev-dependencies]`) is reachable from in-mod tests when surface is consumed via fully-qualified path in the test body, but the lib-test-target compilation cycle blocks top-level `use ailang_surface::*;` lines. Use the fully-qualified `ailang_surface::parse_prelude()` at the call site to avoid the import issue (per pd.2's experience). - [ ] **Step 1: Read the existing in-mod test to confirm shape** Open `crates/ailang-core/src/workspace.rs`, navigate to lines 2640-2670. Identify the test fn that contains the `include_str!("../../../examples/prelude.ail.json")` and `serde_json::from_str` pair. Per the pd.2 journal this is `build_workspace_accepts_assembled_modules_and_runs_validation` (and possibly one sibling using the same pattern; check for both occurrences at lines 2654 and 2656). If the line numbers shifted post-pd.2 (very possible given pd.2's deletions), use: ```bash git grep -n 'include_str!.*prelude\.ail\.json' crates/ailang-core/src/workspace.rs ``` - [ ] **Step 2: Verify `ailang_surface` is reachable from in-mod tests** Read `crates/ailang-core/Cargo.toml` `[dev-dependencies]`. Confirm `ailang-surface = { path = "../ailang-surface" }` is present (per brainstorm-recon: yes). This is sanity-checking; if absent, **STOP** — pd.2 was wrong about the edge existing and the migration target needs revisiting. - [ ] **Step 3: Replace the inlined `serde_json::from_str` with `ailang_surface::parse_prelude()`** In each affected test fn body, replace the pattern: Old (per pd.2 journal): ```rust let prelude: Module = serde_json::from_str(include_str!( "../../../examples/prelude.ail.json" )).expect("embedded prelude must parse"); ``` New: ```rust let prelude = ailang_surface::parse_prelude(); ``` The `Module` type annotation can drop because `parse_prelude` returns `Module` directly. The import line for `Module` (if newly orphan) either stays or is removed depending on other test bodies in the same mod. - [ ] **Step 4: Run the affected tests to verify they still pass** Run: `cargo test -p ailang-core build_workspace_accepts_assembled_modules_and_runs_validation` (And the second test name if recon surfaces one — Step 1 will have named it.) Expected: PASS. The semantic content is unchanged (same prelude Module, just constructed via the canonical post-pd.2 path). - [ ] **Step 5: Run the full core test suite** Run: `cargo test -p ailang-core` Expected: all green. --- ### Task 2: Delete the cross-form-identity preflight + supporting bytes **Files:** - Modify: `crates/ailang-surface/tests/prelude_module_hash_pin.rs` — delete the `prelude_ail_and_json_parse_to_identical_module` test fn (lines 22-43 per recon), the `PRELUDE_JSON` const (line 20), the `PRELUDE_AIL` orphan import (line 18 — keep `parse_prelude` only), the `const _: &str = PRELUDE_AIL;` shape-pin (line 60), and trim the module doc-comment (lines 1-15) to describe only the surviving long-term hash anchor. - [ ] **Step 1: Delete the cross-form-identity test fn + its supporting bytes** Edit `crates/ailang-surface/tests/prelude_module_hash_pin.rs`. Delete: - Lines 1-15: the module doc-comment block (will be replaced in Step 2). - Line 18: the import `use ailang_surface::{parse_prelude, PRELUDE_AIL};` — replace with `use ailang_surface::parse_prelude;` (PRELUDE_AIL no longer consumed). - Line 20: `const PRELUDE_JSON: &str = include_str!(...);` - Lines 22-43: the entire `#[test] fn prelude_ail_and_json_parse_to_identical_module` body. - Lines 59-60: the shape-pin comment + `const _: &str = PRELUDE_AIL;`. The surviving content is the `prelude_parse_yields_canonical_hash` test fn (post-edit lines will be ~3-15). - [ ] **Step 2: Replace the module doc-comment with the surviving-only version** At the top of the file, insert: ```rust //! pd.2/pd.3 (`prelude-decouple` milestone): pin the canonical //! identity of the parsed prelude as a literal hex hash. Updates //! intentionally on prelude-content changes (record the why in the //! per-iter journal); accidental drift trips immediately. //! //! pd.2 also carried a `prelude_ail_and_json_parse_to_identical_module` //! cross-form-identity preflight that compared `parse(prelude.ail)` //! against `serde_json::from_str(prelude.ail.json)`. The preflight //! discharged its purpose at pd.2 close (PASSED at hash //! `3abe0d3fa3c11c99`) and was retired in pd.3 alongside the //! `prelude.ail.json` file. ``` - [ ] **Step 3: Run the surviving test** Run: `cargo test -p ailang-surface --test prelude_module_hash_pin` Expected: `prelude_parse_yields_canonical_hash` PASSES (still pinning hash `3abe0d3fa3c11c99`); the deleted `prelude_ail_and_json_parse_to_identical_module` no longer appears in the suite. If the hash assertion fails, the prelude content has changed since pd.2 close — investigate. With no source-edit between pd.2 commit and pd.3 task 2, the hash MUST be unchanged. - [ ] **Step 4: Verify orphan imports are clean** Run: `cargo build -p ailang-surface --tests 2>&1 | grep -E '(warning|error)' | head -10` Expected: no `unused import` warnings on the edited file. If `Module` or any other type was the only consumer of an import line that now has no consumer, drop the line. --- ### Task 3: Delete the defensive include in `crates/ail/src/main.rs` + its lockstep skip-branch **Files:** - Modify: `crates/ail/src/main.rs` — two adjacent code blocks inside the `Cmd::MigrateBareCrossModuleRefs` arm: the defensive include + insert at lines 472-484, and the skip-branch at lines 500-505. - [ ] **Step 1: Delete the defensive include block** Edit `crates/ail/src/main.rs`. Delete lines 472-484 (verify with `git grep -n 'Include the embedded prelude' crates/ail/src/main.rs`): ```rust // Include the embedded prelude (so the implicit fallback // matches the loader's behaviour). let prelude_json: &str = include_str!( "../../../examples/prelude.ail.json" ); let prelude: Module = serde_json::from_str(prelude_json) .expect("embedded prelude must parse"); if !modules.contains_key("prelude") { modules.insert( "prelude".to_string(), (PathBuf::new(), prelude), ); } ``` Block-locals `prelude_json` and `prelude` are never read after their binding scope ends; deletion is safe. - [ ] **Step 2: Delete the lockstep skip-branch** Same file, locate the loop body. Delete the comment + guard at lines 500-505: ```rust // Pass 2: rewrite each module's bare cross-module refs. // Skip the synthetic prelude entry (no path to write to). let mut rewritten = 0usize; for (mod_name, (path, m)) in modules.iter_mut() { if mod_name == "prelude" && path.as_os_str().is_empty() { continue; } ``` becomes: ```rust // Pass 2: rewrite each module's bare cross-module refs. let mut rewritten = 0usize; for (mod_name, (path, m)) in modules.iter_mut() { ``` The `mod_name` binding is no longer read inside the loop after the guard's removal (verify by reading the next ~15 lines). If the unused binding generates a warning, change to `for (_, (path, m)) in modules.iter_mut() {`. - [ ] **Step 3: Verify the subcommand still compiles** Run: `cargo check -p ail 2>&1 | head -20` Expected: clean. No new warnings on `main.rs`. - [ ] **Step 4: Spot-test the migrate subcommand (optional sanity check)** Run: `cargo run -p ail -- migrate-bare-cross-module-refs --help 2>&1 | head -10` Expected: prints help text successfully (the binary builds and the subcommand enumerates). The subcommand's substantive behaviour (rewriting bare cross-module refs across a directory of `.ail.json` files) hasn't been exercised since the canonical-type-names milestone; no in-tree fixtures are known to need rewriting today, so a deeper test is out of scope. --- ### Task 4: Update `carve_out_inventory.rs` — drop §C4 (b) **Files:** - Modify: `crates/ailang-core/tests/carve_out_inventory.rs` — three edits: header sentence count (line 2), comment block (lines 6-11), EXPECTED list (line 24-25 area). - [ ] **Step 1: Update the header sentence** Edit `crates/ailang-core/tests/carve_out_inventory.rs`. Change line 2 from: ``` //! spec. Asserts that exactly the eight named carve-out files exist ``` to: ``` //! spec. Asserts that exactly the seven named carve-out files exist ``` - [ ] **Step 2: Rewrite the carve-out comment block (lines 6-11)** Old: ``` //! Eight carve-outs at iter form-a.1 close: //! - §C4 (a) subject-matter: 7 fixtures (canonical-form rejection / unbound / class-def rejection). //! - §C4 (b) compile-time-embed: 1 fixture (prelude.ail.json). //! //! When the prelude-embed-refactor milestone retires §C4 (b), this //! test's pass-count drops to seven and `EXPECTED` is re-published. ``` New: ``` //! Seven carve-outs post prelude-decouple (2026-05-14): //! - §C4 (a) subject-matter: 7 fixtures (canonical-form rejection / unbound / class-def rejection). //! - §C4 (b) compile-time-embed: retired 2026-05-14 by milestone //! prelude-decouple; the prelude is now embedded as `prelude.ail` //! in `ailang-surface` and parsed at compile time via //! `ailang_surface::parse_prelude`. ``` - [ ] **Step 3: Drop the `"prelude.ail.json"` entry from EXPECTED** Edit lines around 24-25. Delete: ```rust // §C4 (b) — compile-time-embed "prelude.ail.json", ``` If the comment "// §C4 (b) — compile-time-embed" is on its own line above the entry, delete the comment too (it has no other content beneath it after this edit). - [ ] **Step 4: Run the test — should now PASS against the post-pd.3 corpus** Run: `cargo test -p ailang-core --test carve_out_inventory` Expected: PASS — but note this test STILL READS `examples/prelude.ail.json` because the file hasn't been deleted yet (Task 7). The EXPECTED list no longer contains it, so the assertion `matches_carve_outs` will FAIL saying "unexpected file: prelude.ail.json". That's expected at this task — it goes green only after Task 7 deletes the file. If you want intermediate green between this task and Task 7, swap the task order (Task 7 before Task 4). The plan's order keeps the test edits adjacent and defers the file-system mutation to last; either order is correct. --- ### Task 5: Update `docs/specs/0025-form-a-default-authoring.md` §C4 (b) **Files:** - Modify: `docs/specs/0025-form-a-default-authoring.md:270-311` — add a "Status: RETIRED" marker at top of §C4 (b) and a one-paragraph epitaph at the bottom; preserve the original text as historical context. The spec is a historical document about the closed form-a-default-authoring milestone; the convention is to retire sections in place rather than delete them, so future readers tracing back the prelude-embed-refactor milestone's prerequisites can find the original framing. Per the recon, upstream callouts at lines 232-236, 462-465, 477-484 stay as historical references — their "as of 2026-05-13" temporal qualifiers preserve correctness. - [ ] **Step 1: Insert a status marker at the top of §C4 (b)** Edit `docs/specs/0025-form-a-default-authoring.md`. Find the `**(b) Compile-time-embed carve-outs (one file as of 2026-05-13):**` heading (line 270 area). Insert immediately above: ```markdown > **Status: RETIRED 2026-05-14 by milestone prelude-decouple.** The > §C4 (b) compile-time-embed carve-out is empty post that milestone; > the prelude is now authored exclusively as `examples/prelude.ail` > and parsed at compile time via `ailang_surface::parse_prelude`. The > original §C4 (b) framing below is preserved as historical context; > it documents the doctrine inconsistency that prelude-decouple > resolved. See `docs/specs/0028-prelude-decouple.md` and the > per-iter journals 2026-05-14-iter-pd.{1,2,3}.md for the resolution. ``` - [ ] **Step 2: Verify the spec still reads coherently** Read the §C4 (b) section top-to-bottom (lines ~268-315 post-insert). Confirm the new status marker reads as a forward-pointer to the resolution and the original text reads as the historical motivation. The intent: a future reader seeing "the prelude is the only file authored as JSON" understands that statement as past-tense and finds the resolution one click away. No other edits in this task. Upstream callouts (lines 232-236, 462-465, 477-484) stay verbatim because their qualifiers ("as of 2026-05-13", "one file as of 2026-05-13") remain truthful as historical statements. --- ### Task 6: Create the carve-out retirement pin **Files:** - Create: `crates/ailang-surface/tests/prelude_decouple_carve_out_pin.rs` — single `#[test]` fn asserting `examples/prelude.ail.json` does NOT exist on disk. - [ ] **Step 1: Write the test** Create `crates/ailang-surface/tests/prelude_decouple_carve_out_pin.rs`: ```rust //! pd.3 (`prelude-decouple` milestone): pin that //! `examples/prelude.ail.json` does NOT exist on disk. Surfaces //! immediately if the file is recreated by accident or a future //! refactor reverts the embed-source swap. //! //! Companion to the `examples_ail_json_inventory_matches_carve_outs` //! test in `crates/ailang-core/tests/carve_out_inventory.rs`: that //! one asserts the carve-out list is exactly seven; this one //! asserts the eighth (retired) file specifically does not exist. //! Either test detects an accidental re-introduction; this one //! pinpoints the file by name in the error message. use std::path::Path; #[test] fn prelude_ail_json_does_not_exist_on_disk() { let path = Path::new("../../examples/prelude.ail.json"); assert!( !path.exists(), "examples/prelude.ail.json must NOT exist after the prelude-\ decouple milestone (closed 2026-05-14). The prelude is now \ embedded as prelude.ail in ailang-surface. If you intentionally \ re-introduced this file, you must also: (a) remove this pin, \ (b) restore the §C4 (b) entry in carve_out_inventory.rs, \ (c) revert the §C4 (b) status marker in \ docs/specs/0025-form-a-default-authoring.md, and \ (d) record the rationale in a per-iter journal." ); } ``` - [ ] **Step 2: Run the test — must FAIL today** Run: `cargo test -p ailang-surface --test prelude_decouple_carve_out_pin` Expected: FAIL because `examples/prelude.ail.json` still exists. The test is RED-on-purpose at this point; Task 7 turns it GREEN by deleting the file. If the test PASSES at this step, the file is already gone from the working tree — investigate (was it deleted in an earlier task by mistake? Recon should not have permitted that). --- ### Task 7: Delete `examples/prelude.ail.json` **Files:** - Delete: `examples/prelude.ail.json` — the milestone's primary artefact retirement. - [ ] **Step 1: Verify zero remaining production / test consumers** Run: ```bash git grep -n 'include_str!.*prelude\.ail\.json' -- '*.rs' ``` Expected: NO matches. (Tasks 1-3 cleared all three.) ```bash git grep -n 'prelude\.ail\.json' -- '*.rs' | grep -vE '(//|//!|/\*)' ``` Expected: NO matches outside comments / doc-comments / string-literals in test fixtures. The non-`include_str!` references should all be either documentation, comments, or path-string constructions in tests (which fail gracefully on missing files). If any reference appears as a file-system access (read/open/etc), STOP and migrate it before proceeding. - [ ] **Step 2: Delete the file** Run: ```bash rm examples/prelude.ail.json ``` (Plain shell delete; not `git rm`. The file will appear as a deletion in `git status`. The Boss's commit step stages the deletion via `git add` of the path or `git add -u`.) - [ ] **Step 3: Run the carve-out retirement pin — must now PASS** Run: `cargo test -p ailang-surface --test prelude_decouple_carve_out_pin` Expected: PASS — `prelude.ail.json` no longer exists. - [ ] **Step 4: Run the carve-out inventory test — must now PASS** Run: `cargo test -p ailang-core --test carve_out_inventory` Expected: PASS — `EXPECTED` is exactly seven names, and exactly seven files exist under `examples/*.ail.json`. - [ ] **Step 5: Run the full workspace test suite** Run: `cargo test --workspace` Expected: all tests green. --- ### Task 8: Bench regressions + final acceptance **Files:** none directly. - [ ] **Step 1: Run regression baselines** Run: `python bench/check.py && python bench/compile_check.py && python bench/cross_lang.py` Expected: all three exit 0. pd.3 changes are file-system + spec text + test deletion + dead-code cleanup; no codegen, no runtime, no typecheck-result deltas. If `check.py` fires the same `bench_list_sum.bump_s` envelope-noise as in pd.1 / pd.2 / earlier audits, treat as established noise (per the prior 13-audit pattern) — pd.3 cannot be the source. - [ ] **Step 2: Confirm milestone-level acceptance** Per the parent spec (§Acceptance criteria): 1. `examples/prelude.ail.json` is removed from the working tree — verified by Task 7 Step 3. 2. `cargo build --workspace` succeeds with no `include_str!` of `prelude.ail.json` — verified by Task 7 Step 1. 3. `git grep -l '"prelude"' crates/ailang-core/src/` returns zero matches — already true post-pd.2; spot-check now to confirm pd.3 didn't regress. 4. `cargo test --workspace` is green — verified Task 7 Step 5 + Task 8 Step 1. 5. Baselines green at unchanged values — verified Task 8 Step 1. 6. `docs/specs/0025-form-a-default-authoring.md` §C4 (b) is updated — verified by Task 5. 7. CLAUDE.md and `docs/DESIGN.md` reviewed — spot-check now. For item 7: ```bash git grep -n 'prelude\.ail\.json' CLAUDE.md docs/DESIGN.md ``` Expected: NO production-doctrine assertions; only historical references acceptable. If a CLAUDE.md sentence still asserts that core embeds the prelude, edit it to reflect the post-milestone state. ```bash git grep -n 'authors write `\.ail`' CLAUDE.md ``` Expected: this sentence is consistent with the post-milestone state (authors write `.ail` for ALL files now, including prelude — no exception). If the wording still names a prelude exception, edit it to drop the exception. - [ ] **Step 3: Confirm zero literal-`"prelude"` strings in core/src/** Run: `git grep '"prelude"' crates/ailang-core/src/` Expected: NO matches (already true post-pd.2; pd.3 is purely file-system + downstream test/spec adjustments, no core/src/ source edits). If any match appears in production code, STOP — pd.3 should not have modified core's src/ source. --- ## Out of scope for pd.3 The following are NOT part of this milestone and MUST NOT be touched: - The `WorkspaceLoadError::ReservedModuleName` variant declaration in `crates/ailang-core/src/workspace.rs` — stays per spec (generic shape, no literal `"prelude"`). - The `crates/ail/src/main.rs:1220-1229` `W::ReservedModuleName` match-arm in `workspace_error_to_diagnostic` — variant tag preserved, no edit. - The `ailang-check` and `ailang-codegen` implicit-prelude resolution paths (`crates/ailang-check/src/lib.rs:143` area, codegen iter 23.2.4 behaviour) — these consume the prelude module from a populated workspace, do not embed prelude bytes, and were explicitly listed as out-of-scope in the parent spec. - The 3 cargo doc warnings on dangling `[`load_workspace`]` references (in `Workspace` and `Registry` rustdoc) noted as known debt in the pd.2 journal — pre-existing baseline; unrelated to pd.3 scope. - Audit (Boss-dispatched after pd.3 closes — separate skill invocation per `skills/audit/SKILL.md`).