Repository navigation
Phase 4: reach 100% on the L4 type and L5 class gates #77
Description
Why this phase
Phase 1 makes L4/L5 measurable; Phases 2 and 3 unblock the corpus. This phase turns "measurable" into "passing". The first sub-issue is nearly free — it is a mapping table — and it is worth doing first because it converts 153 corpus cases from invisible to tracked.
Baseline
M4 (61 mapped cases) and M5 (92 mapped cases) are both unmeasured. The corpus splits as:
| M4 code | Cases | M5 code | Cases |
|---|---|---|---|
TypesDoNotUnify |
41 | NoInstanceFound |
53 |
HoleInferredType |
10 | OverlappingInstances |
8 |
EscapedSkolem |
2 | InvalidInstanceHead |
7 |
CannotApplyExpressionOfTypeOnType |
2 | OrphanInstance |
7 |
ExpectedType |
2 | InvalidNewtypeInstance |
6 |
InfiniteType |
2 | ClassInstanceArityMismatch |
4 |
AmbiguousTypeVariables |
1 | CycleInTypeClassDeclaration |
2 |
IntOutOfRange |
1 | MissingClassMember |
2 |
PossiblyInfiniteInstance, DuplicateTypeClass, DuplicateInstance |
1 each |
Sub-issues
Tracked as children of this epic.
Gate
L4 and L5 reach 100% agreement on their mapped codes, with the scoreboard from Phase 1 reporting the per-code numbers.
References
docs/design/D-04-suite-roadmap.md— M4, M5, FE-08..FE-16crates/psrs-typecheck/src/typecheck/error.rs— the kinds to mapdocs/design/frontend/type-system/— the normative type-system designdocs/implementation/frontend/roles-and-coercions.md,docs/implementation/frontend/rank-n.md
Validation
cargo fmt --all --check
cargo test --workspace
cargo clippy --workspace --all-targets -- -D warnings
PSRS_ORACLE=annotations cargo test -p psrs-driver --test suite -- --ignored --nocapture
Activity
biuld commentedon Oct 22, 2026
The mapped Phase 4 gate is complete: L3 50/50, L4 65/65, L5 101/101, with 166 measured L4/L5 cases. This does not assert full frontend or runtime completion.
Verified by PR #164 (41a16e2): formatting, default Clippy, full workspace (2331 passed, zero failed), five separately executed frontend boards, all 22 official purs differential tests, and the complete serial runtime board (297/340; 25 FFI exclusions; every agreeing exit zero). Required Wasmtime CI execution check 119961920075 completed successfully. The three losses versus the older board reproduce on the unmodified starting HEAD; BigFunction remains an explicit incomplete compiler verdict.
Evidence: Phase 4 acceptance review.