[ucgmsim]

Closes #232

What was wrong

• src-rust/arias_intensity.rs: the module doc and the ARIAS_CONSTANT
  doc comment gave the formula as $\frac{\pi}{2g}$, but ARIAS_CONSTANT = G * PI / 2.0, i.e. $\frac{\pi g}{2}$. The code is correct — it expects
  acceleration in units of $g$ — but the comments described the formula for
  acceleration in physical units, which is off by a factor of $g^2$. The
  test_arias_physical_constant comment likewise claimed the input was
  "1.0 cm/s^2" with an expected result of PI / (2.0 * 981.0), when the
  input is actually 1 g and the expected result is 9.81 * PI / 2.0.
• src-rust/cav.rs: the module doc omitted the G factor the code applies,
  and the test_cav_constant comment claimed the input was "1.0 cm/s^2"
  with a result of "G (9.81 cm/s)", when the input is 1 g and the result is
  in m/s.
• src-rust/constants.rs: GAL was dead code — never referenced anywhere
  in the Rust crate. Removed it rather than wiring up a shared definition
  with Python's separate G = 981 in IM/ims.py, since the two live in
  different languages with no existing constant-sharing mechanism, and
  removing unused code is the smaller, lower-risk fix.
• IM/konno_ohmachi.py: the comment above the cache pop-and-reinsert said
  it "shifts the key to the front of the insertion order," but
  dict.pop + re-assignment actually moves the key to the end (most
  recently inserted position). The eviction loop a few lines down does walk
  from the front (oldest first), so the cache logic itself is correct — only
  the comment had the direction backwards.

No numerical or behavioral change; comments and one unused constant only.

How the new test shows it

test_arias_physical_constant previously asserted the result against
ARIAS_CONSTANT itself, which is circular: it would pass regardless of
whether the constant matched the documented formula. Changed the assertion
to compare against the literal 9.81 * PI / 2.0 from the corrected
comment. If ARIAS_CONSTANT were changed to match the old, wrong comment's
formula (PI / (2.0 * 981.0) ≈ 0.0016 vs. the correct ≈ 15.41), this test
would now fail, whereas it previously would not have caught that.

What to check closely

• The corrected formulas/comments in arias_intensity.rs and cav.rs
  match the actual runtime behavior of arias_intensity/cav (inputs in
  units of $g$).
• Confirm GAL truly has no other consumers (checked via grep across the
  Rust sources; only constants.rs defined it).

Cost $0.43, 27 turns.

🤖 Generated with Claude Code