[sdubois777]

A creature's energy shield now refills at a fifth of its maximum per second, starting three seconds after it last took damage (#1562). Until now every creature's shield regeneration was set to zero, so a damaged shield never came back.

The rate is the design document's own figure: docs/Cataclysm_GDD_v2.md, the Energy Shield section, "It refills at a fifth of its own maximum per second ... Every source of maximum energy shield grants that rate alongside it", sourced there to Path of Exile's 20% per second. The Succubus section says its shield "behaves exactly like the player's" and "refills three seconds after the last damage it took".

What changed

• game/Source/Cataclysm/Character/CataclysmEnemyCharacter.h and .cpp: in ApplyStartingAttributes, a creature's energy shield regeneration is set to its maximum shield times EnergyShieldRegenShareOfMaximum (0.2). This covers both the Shielder enemy modifier's shield and a shield the creature's kind carries (as the Succubus's does). A creature with no shield still regenerates nothing; health and mana regeneration stay zero. The existing three-second wait is unchanged.
• game/Source/Cataclysm/Tests/CataclysmEnemyModifierTests.cpp: two tests, one for a floor creature that draws Shielder and one for a creature whose kind carries a shield.
• docs/DECISIONS.md: the entry and window w78's record.

Evidence: window w78, at df85590 on development 4faf66b

• Build: 72 actions, 62 files compiled.
• Each new test alone: 1 performed, 1 succeeded. Printed:
  • "Shielder: maximum 80.5, 40.2 after the hit, 40.2 at 2.5 s, 56.3 at 4 s" (one second of refill after the wait: 16.1 = a fifth of 80.5)
  • "the kind's own shield: maximum 500.0, 250.0 after the hit, 250.0 at 2.5 s, 350.0 at 4 s" (100 = a fifth of 500)
• Whole Unreal suite: 4351 performed, 4351 succeeded, 0 failed (registered 4351; 4351 declared in the tree, gap 0). Memory: peak 8,007 MB, final 7,686 MB.
• Python: "7024 passed, 8 skipped"; JUnit tests=7032, failures=0, errors=0, skipped=8. ruff: "All checks passed!".
• Proof P1 (the rate written as 0): PROVED. The Shielder test failed with the break in and passed restored.
• Proof P2 (the rate taken from Shielder's share only): PROVED. The kind's test failed with the break in and passed restored.

After the tested commit, only docs/DECISIONS.md changed (the record).

🤖 Generated with Claude Code