Changelog

0.3.0

• Additive teammate-credit experiment: rejected; the original gameplay reward remains active.
• Three fresh cover/spawn scenarios and 16 fresh policy seeds; both original and additive rules tested, with fixed and frozen controls.
• Immutable v1 source/evidence archive; independent choice/feedback/commit verification and combat replay checks.
• human-fixed-v1 browser playtest: two warm-ups, two concealed 12-round conditions, movement changes, questionnaires, recovery/export, source fingerprints, and explicit QA exclusions.
• Reusable browser kit, npm tarball, source bundle, macOS arm64 C ABI bundle, standalone Rust/browser guard hosts, and timing harness.
• Public-facing playable demo, evidence explorer, limitations, and versioned downloads.

The contextual-bandit algorithm and snapshot schema are unchanged. No game-engine adapters or model-generated tactics were added. Human enjoyment and external-game effectiveness remain unmeasured.

0.2 prototype

Illustrated arena, two-wave scout missions, saved practice memory, learning inspector, guided movement exercise, adaptive/random comparison, typed Wasm client, and direct-observation adaptation benchmark.

0.1 prototype

Rust core, JSON contract, WebAssembly wrapper, native C ABI, deterministic sampling, bounded updates, observation gates, and snapshot replay.