Skip to content

Release notes: 2026-05-25 — tier2-physics-li-builds-wp-t2

Status: Ready for review
Repo: li-langverse/lic
PR: feat/fix-tier2-li-builds
PH / REQ: Phase 5b benchmarks (WP-T2 tier-2 physics Li harness)
Author: agent

Summary (one sentence)

Tier-2 physics Li wrappers for cloth_swing, combustion_passive, euler_fluid_2d, rigid_body_stack, and wind_field_bc now satisfy extern contract + raises IO gates and emit checksums via li_rt_volatile_sink_f64 so bench.py --tier 2 produces CSV rows instead of compile failures.

Agent continuation (required)

  1. Read: benchmarks/tier2_physics/*/li/main.li (five kernels above); benchmarks/harness/bench.py build_li / verify_benchmark_results.
  2. Run: ./scripts/build.sh then python3 benchmarks/harness/bench.py --tier 2 --only cloth_swing,combustion_passive,euler_fluid_2d,rigid_body_stack,wind_field_bc --runs 3.
  3. Then: ingest benchmarks/results/latest.csv to benchmarks dashboard if publishing perf rows; extend same pattern to remaining tier-2 stubs (orbit_two_body, etc.) if CI still reports unknown.
  4. Blocked on: human merge; no self-merge.

Changed (specific)

Area What Evidence
Harness extern proc requires/ensures, main raises IO, checksum sink bench.py --tier 2 --only five ids → verify ok + updated latest.csv
Paths benchmarks/tier2_physics/{cloth_swing,combustion_passive,euler_fluid_2d,rigid_body_stack,wind_field_bc}/li/main.li local run 2026-05-25

Not changed (scope fence)

  • Shared C oracles (common/*_core.c), bench.py spec table, compiler contract policy, li-tests corpus, Lean/trusted.lean, tier-0/1 benches, dashboard ingest scripts.

Breaking / Security / Performance

Category Status
Breaking N/A — benchmark wrapper-only
Security N/A — no trusted surface
Performance N/A — harness timing only; no codegen change

Downstream

N/A — benchmarks repo ingest optional after merge.