这个版本更新了什么#
This release tidies the code behind 0.9.2, with a few changes a player can notice. Mini-game review finds its proofs about four times faster, so the facts after a finished mini-game arrive sooner. The app hashes the engine's network file once per launch instead of once per engine, and decodes the bundled course once, so review, training and a network game start with less work. The opponent-strength picker uses the same names everywhere: "First Steps" and "Keep Custom Settings". A saved game's PGN tags no longer carry control characters. Nothing else is meant to differ. The pass shares one helper where modules had each written their own (hashing, JSON reading, private file writes, the rules engine's JavaScript calls, the training geometry tables), reuses rules instances instead of creating them, and gives the tests and the acceptance tooling the same treatment: one app-driver module behind the pointer gates, one scenario script behind four gates, one tree-hash comparison behind the runner and the release scripts. Automated verification: the full acceptance suite (24 gates, including every pointer and UI gate and the two-Mac network gate) passed on the test host for these sources before the version bump, in targeted runs during the work rather than one certified run, and the mini-game calibration was regenerated with an unchanged outcomes digest. The live voice check did not run for this release, which was packaged without acceptance evidence or a live-voice report (the release scripts now treat both as optional and warn when either is absent); the voice coach's changes (one speech-segment decoder, a single base64 decode, deferred setting changes) are verified by unit tests only. No person has reviewed the changes in play; the human-acceptance criteria noted for 0.9.0 (review grades, the beginner opponent, how the voice coach sounds in review) are unchanged and still open.
以上是作者写在 GitHub 上的英文更新说明原文，未经翻译。查看原页面