P1.2.2 — κ DSL Parser — Verification

Step 5 of the 5-step executor chain. Records evidence that the implementation in commit 2250c863f3d73f86d6cb457df88c58a10e5bef3e satisfies the contract in docs/contracts/p1-2-2-parser-contract.md.

§1. Run record

Item	Value
Round	R84 (κ Wave 3)
Task	P1.2.2
β task ID	`f6482fe9-2ee8-41a8-a562-e44638cc671d`
Branch	`feature/p1-2-2-parser`
Worktree	`.worktrees/claude/p1-2-2-parser`
Base	`origin/main` @ `6345ba7aec8d2507337fa5161928c13d4a3b4d3e`
Implementation commit	`2250c863f3d73f86d6cb457df88c58a10e5bef3e`
Audit commit	`aee36c94…`
Contract commit	`ee303713…` (amended in implementation commit per change log §10)
Packet commit	`b539abf1…`

§2. Test gate (CLAUDE.md §5 — all three must be green)

Sequence run from the worktree:

cd .worktrees/claude/p1-2-2-parser
npm run build && npm run lint && npm test

§2.1. Build

npm run build (TypeScript tsc + postbuild migration copy) — green.

> colibri@0.0.1 build
> tsc

> colibri@0.0.1 postbuild
> node scripts/copy-migrations.mjs

copy-migrations: copied 6 migration(s) ...

§2.2. Lint

npm run lint (eslint src) — green with zero warnings.

> colibri@0.0.1 lint
> eslint src

(no output ⇒ no errors)

§2.3. Test

npm test (jest --coverage) — green, all 1467 tests passing across 31 suites.

Test Suites: 31 passed, 31 total
Tests:       1467 passed, 1467 total
Snapshots:   0 total
Time:        24.822 s

Pre-implementation baseline (post-merge of origin/main @ 6345ba7a): 1409 tests passing across 30 suites. Delta: +58 tests (parser.test.ts) and +1 suite. Zero pre-existing tests regressed.

§2.4. Coverage (informational)

The new parser file covers:

File	Stmts	Branch	Funcs	Lines
`src/domains/rules/parser.ts`	89.44%	73.17%	82.64%	89.31%
`src/domains/rules/lexer.ts`	100.00%	83.33%	100.00%	100.00%

Uncovered parser lines are mostly the recoveryValueFunc: () => undefined arrow callbacks (one per RULE — never invoked except during error recovery the test corpus doesn’t trigger) and a handful of rare-branch error paths inside OPTION actions that only fire when sub-SUBRULEs return undefined after recovery. These are defensive belt-and-braces null-checks, not load-bearing logic.

§3. Acceptance criteria — verification matrix

Maps the headline criteria from the dispatch packet → contract invariants → test cases that prove them.

AC	Description	Contract ref	Test	Status
AC1	Chevrotain `EmbeddedActionsParser` (not CstParser)	§1 / §5	`parser.ts` line 359 (`class KappaParser extends EmbeddedActionsParser`)	✓
AC2	All 11 AST node types exported with `type` discriminant	§2.1	F7.14 (visits AcceptCommitment AST and asserts every `type` is one of the known 11)	✓
AC3	Stratified operator precedence (not table-hack)	§4	F2.4 (`+` left-assoc), F2.5 (`*` over `+`), F2.6 (parens), F2.7 (`not`/`or`/`and`), F2.11 (`and` over `or`)	✓
AC4	4 rule kinds parse (Admission/StateTransition/Consequence/Promotion identified by name convention)	audit §6	The parser produces `RuleNode` regardless of name; classification is downstream (P1.2.4). Test F1 confirms parsing the AcceptCommitment rule, F7.5 parses three rules with arbitrary names, F7.7 parses a rule named `admissionRule`.	✓ (parser produces RuleNode; classification is downstream per audit §6 and PR description)
AC5	`recoveryEnabled: true`; first 5 errors reported	§3 / §I15	`parser.ts` line 902 (`recoveryEnabled: true`); F3.5 asserts `parseErrs.length <= MAX_PARSE_ERRORS`	✓
AC6	AST cap at 10000 nodes enforced at parse time	§2.3 / §3 step 4 / §I14	F4.2 (count > cap → error + omit), F4.3 (just under), F4.4 (just under upper bound)	✓
AC7	Round-trip property tested	§I17	F5.1 (AcceptCommitment), F5.2 (precedence fixture), F5.3 (interleaved bad-parse) — currently `parse(s) == parse(s)`; canonicalize TODO(P1.5.4)	✓
AC8	`npm run build && npm run lint && npm test` ALL THREE green	CLAUDE.md §5	§2.1–§2.3 above	✓
AC9	Zero new lint warnings	§I — corpus rule	§2.2 above	✓
AC10	Existing tests still pass (~1409)	regression	1467 - 58 (new parser tests) = 1409 baseline; zero regressions	✓

§4. Forbiddens — compliance

Re-checked against task §FORBIDDENS:

✗	Forbiddon	Status
1	AST nodes as classes with behavior	Pure data only — every AST type is an `interface` (no methods). Verified by reading `parser.ts` lines 86-202.
2	Semantic validation	The parser only enforces grammar + AST cap. Empty `guards { }` blocks parse cleanly (F7.6) — validator’s job (P1.2.3). The grammar accepts arbitrary `IntLiteral` values without overflow checks (P1.2.3 / P1.3.1 concern). String/Variable misplacement in expression position parses but downstream validation is the gatekeeper.
3	Operator precedence collapsed	Stratified — 8 distinct productions (`orExpr`, `andExpr`, `notExpr`, `comparison`, `additive`, `multiplicative`, `unary`, `primary`). Verified by reading `parser.ts` §5 and tested by F2.1–F2.12.
4	Edited main checkout	`git rev-parse --show-toplevel` of the worktree returned the worktree path. No commits on main.
5	Bumped Chevrotain version	`package.json` retains `chevrotain: 11.0.3` (pinned by P1.2.1 lockfile). No `package.json` / `package-lock.json` deltas in the implementation commit.

§5. Drift items handled

§5.1. ADR-006-dsl-grammar still missing

Documented in docs/audits/p1-2-2-parser-audit.md §3 — same drift surfaced by the P1.2.1 lexer audit. docs/architecture/decisions/ADR-006-dsl-grammar.md does NOT exist at base 6345ba7a. The actual ADR-006 is ADR-006-executable-meaning.md. The κ DSL grammar is not yet ratified by ADR.

Scope: out of scope for P1.2.2. Re-noted for the next docs-hygiene round.

§5.2. Test layout — colocated vs `src/tests/`

The task prompt suggests src/domains/rules/__tests__/parser.test.ts. The Phase 0/1 corpus convention is src/__tests__/domains/<name>/.... This task follows the corpus convention (matching the P1.2.1 lexer test placement). Documented in audit §1.3.

§5.3. Contract amendment: `ArgList` shared by `funcCall` and `effectCall`

During implementation, the test fixture revealed that the contract’s §4.2 phrasing (“A bare STRING is allowed only here [effectCall]”) was inconsistent with extraction §1’s EBNF, which defines ArgList = Arg { "," Arg } with Arg = Expression | STRING and uses ArgList for both EffectCall and FuncCall. The contract was amended in the same implementation commit (per the change-log discipline in §10) to clarify that FuncCall.args accepts strings via Arg = Expression | STRING. The implementation reuses a single effectArg SUBRULE for both call shapes. This is consistent with the AST shape — FuncCall.args is Expression[] and StringLiteral is part of the Expression union.

§5.4. Grammar correction: `not not` is a parse error

Test F2.10 originally asserted that not not $a parses to Not(Not($a)). Reading extraction §1 carefully reveals NotExpr = [ "not" ] Comparison — square brackets denote optional, not iteration. The grammar permits at most one not per boolean negation level; stacked negations require parens (not (not $a)). The test was updated to assert this correctly: F2.10 now expects a parse error for not not $a, and a new F2.12 confirms not (not $a) parses successfully.

§6. AST verification — pretty-print of the F2.1 precedence fixture

For visual confirmation that operator precedence is correctly stratified, here is the AST produced by parse(wrapExpr('$a + $b * $c == 10 and not $d')):

RuleNode { name: 'R', guards: [
  GuardClause {
    condition: LogicalOp {
      op: 'and',
      operands: [
        // left: $a + $b * $c == 10
        BinaryOp {
          op: '==',
          left: BinaryOp {
            op: '+',
            left: VarRef { path: ['a'] },
            right: BinaryOp {
              op: '*',
              left: VarRef { path: ['b'] },
              right: VarRef { path: ['c'] },
            },
          },
          right: IntLiteral { value: 10n },
        },
        // right: not $d
        LogicalOp {
          op: 'not',
          operands: [ VarRef { path: ['d'] } ],
        },
      ],
    },
    action: 'admit',
    reason: null,
  },
], effects: [] }

This shape is exactly what the spec demands. Tests F2.1–F2.3 walk this AST and assert each level.

§7. Surface diff (this PR only)

 docs/audits/p1-2-2-parser-audit.md           | new (228 lines)
 docs/contracts/p1-2-2-parser-contract.md     | new (337 lines, amended once for §5.3)
 docs/packets/p1-2-2-parser-packet.md         | new (582 lines)
 docs/verification/p1-2-2-parser-verification.md | new (this file)
 src/domains/rules/parser.ts                  | new (1113 lines)
 src/__tests__/domains/rules/parser.test.ts   | new (751 lines)

Total: 6 new files. Zero edits to existing files outside the chain docs.

§8. Writeback metadata

The β-pipeline writeback for this task lands in this PR’s body and in the thought_record MCP call. Concrete values for the writeback:

task_id: f6482fe9-2ee8-41a8-a562-e44638cc671d
branch:  feature/p1-2-2-parser
worktree: .worktrees/claude/p1-2-2-parser
commit:  2250c863f3d73f86d6cb457df88c58a10e5bef3e   # implementation
final_commit_in_pr: <set after writeback commit lands>
tests:   npm run build && npm run lint && npm test (1467/1467 passing across 31 suites)
summary: |
  Shipped Chevrotain EmbeddedActionsParser producing typed AST matching
  extraction §2 (11 node types). Grammar matches EBNF superset from
  extraction §1; operator precedence stratified across 8 productions.
  Error recovery with 5-error cap. AST cap at 10000 nodes enforced at
  parse time via post-parse recursive walker. Round-trip property tested
  (canonicalize is P1.5.4 — TODO marker left in code). 58 new tests
  across 8 fixture groups; zero existing regressions.
blockers: |
  None — P1.2.3 / P1.2.4 / P1.3.1 / P1.5.4 now unblocked.
followups:
  - ADR-006-dsl-grammar.md still missing (third drift report; out-of-scope
    docs hygiene candidate for the next round).
  - Canonical serialization (P1.5.4) replaces the parse-twice round-trip
    proxy in F5; TODO(P1.5.4) marker left in parser.test.ts §F5.

§9. Sign-off

All five steps of the executor chain have completed. The implementation satisfies every acceptance criterion in §3 and complies with every forbiddon in §4. The drift items in §5 are either documented for follow-up (§5.1) or have been resolved in-scope (§5.2, §5.3, §5.4). Three gates green; ready for PR + writeback.