Singularity's Bounty e/CC
@catblanketflower.yuwakisa.com
3.5K followers 1.8K following 35.4K posts
language is a construction set for hacking the human mind Large Language Models are a cornucopia for the curious I do computer stuff but that doesn't define me posts are not financial advice stuff at yuwakisa.com he him
![It compares the six-document format with a single SPEC.md and concludes that, for a game, the split was worth it, mainly because of three habits: tagging each item as your decision ([D]), my proposal ([P]) or tunable ([T]); recording the alternatives behind every decision; and writing criteria as input/expected tables. ONTOLOGY effectively became the spec, CRITERIA gave the best return because it turned almost directly into the test file, RATIONALE caught the soft-armor bugs and kept a tuning history, DYNAMICS mattered mostly at design time, and TELOS was rarely used. The costs were numbers repeated in up to five places, more files to keep in sync, and an /implement skill that only knows SPEC.md. It names specific failures: a confidently stated balance prediction that turned out wrong, the missing commands-and-events section, checks with no tests, tuning and browser tools left in a scratch folder, and an architecture log written after the fact. Its main recommendation is to change the order of work: write rules and their tests together, build the engine without a UI, tune with self-play before making balance claims, then build the UI. A follow-up section records the fixes now done: commands and events added to ONTOLOGY, the missing tests plus a check that every criterion has a test, and the tuning tool and browser test kept in the repo. The first comparison with the new tuning tool showed my claim about hearing range was wrong. Still open are keeping each number in one place, folding TELOS into DYNAMICS, and teaching /implement the format.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:cwa5qtro5bhfz25opigbe6qi/bafkreial2hwg6l2vkczubvjjh3jciqjor4ccdpjoaxawjtdmv3lyngvieq)
