`simp` abstracts too many meta-variables

Prerequisites

Please put an X between the brackets as you perform the following steps:

• Check that your issue is not already filed:
  https://github.com/leanprover/lean4/issues
• Reduce the issue to a minimal, self-contained, reproducible test case.
  Avoid dependencies to Mathlib or Batteries.
• Test your test case against the latest nightly release, for example on
  https://live.lean-lang.org/#project=lean-nightly
  (You can also use the settings there to switch to “Lean nightly”)

Description

simp abstracts meta-variables in proof terms, even if they existed before the elaborating the term. This results in it falsely generalizing

Context

I'm writing my own simp and I noticed a kernel error after switching to abstractMVars (what simp uses) which lead me to finding out about this issue.

Steps to Reproduce

inductive SomeThing.{u} : Prop where
  | mk (_ : PUnit.{u})

set_option pp.universes true in
def testMe (a : PUnit) : PLift SomeThing := by
  /-
  a : PUnit.{?u.10}
  ⊢ PLift.{0} SomeThing.{?u.14}
  -/
  constructor
  simp only [SomeThing.mk a] -- SomeThing.mk.{?u.10} a

Expected behavior: simp should fail since the universe meta-variables are different and simp shouldn't assign them.

Actual behavior: simp abstracts the universe meta-variable ?u.10 even though it relies on it specifically being ?u.10, causing a kernel application mismatch. A similar issue occur for regular meta-variables.

Versions

Lean 4.23.0-nightly-2025-07-09

Impact

Add 👍 to issues you consider important. If others are impacted by this issue, please ask them to add 👍 to it.

Start by reproducing the minimal Lean example and inspect the interaction between simp and abstractMVars, including the analogous regular meta-variable case. The fix is done when simp no longer abstracts pre-existing meta-variables it relies on, and the example fails as expected without a kernel application mismatch.