#240's premise hook is inert: assume_max/assume_range have no non-test callers, so fits_below_bit can never fire

[avrabe]

OptimizationEnv::assume_max / assume_range — the value-range premise hook added for #240's "proof-carrying IR premises" half — have no non-test callers. The optimizer never produces a premise through them.

Evidence

Every call site is inside a test module:

loom-core/src/lib.rs:22523    env.assume_range(...)   # wsc_facts_factset_tests
loom-core/src/lib.rs:22538    env.assume_range(...)   # wsc_facts_factset_tests
loom-core/src/lib.rs:22551-2  env.assume_range(...)   # wsc_facts_factset_tests
loom-shared/src/lib.rs:6886   env.assume_max(...)     # tests
loom-shared/src/lib.rs:6904   env.assume_max(...)     # tests
loom-shared/src/lib.rs:2831   self.assume_range(...)  # inside assume_max itself

Surfaced while wiring the #231 fact source (#322): the agent planned to harvest facts from premises the mid-end had already derived, and found there were none to harvest. The consumer of the hook was written before any producer existed.

Why it matters

fits_below_bit — the gate that lets the mid-end drop a redundant mask — can only fire when something has asserted a bound. With no producers, that gate is permanently false, so the #240 premise-driven rewrites cannot trigger on real input. The rules are implemented and unit-tested against hand-asserted premises, and inert in the pipeline.

This is the same shape as three other findings in this repo: a trap gate with 19 passing tests and zero callers (#288), a pin defended only by a comment (#311/#318), and a proof that had never run. In each case the artifact exists, is tested, looks maintained — and is not connected to anything. Tests prove a component works; they do not prove anything calls it.

What to do

1. Decide whether Carry Verus value-range/no-alias facts as IR premises + algebraic mid-end — feed synth's beat-LLVM specialization #240's premise half is wanted. If yes, wire a producer: the obvious candidates are the same structural sources feat(#231): wire the wsc.facts value-range SOURCE (constants, masks, booleans) #322 uses (mask ⇒ [0,K], constant ⇒ [c,c]) plus anything the guard/branch analysis already knows. If no, delete the hook rather than leave a tested-but-unreachable API implying capability we do not have.
2. Either way, add the anti-regression test that would have caught this: assert the premise path is exercised end-to-end by a pass on a real module, not just by a test that calls assume_range itself. Same fix pattern as fix(#288): wire trap_gate onto the runtime path (additive trap backstop) #289 used for the trap gate — make reachability the asserted property, not just correctness.

Refs #240, #231, #322, #288