alunduil / alunduil/blog.alunduil.com
A decision format nobody else reads decays fastest
- 主要語言
- Astro
- 星號
- 1
- 分支
- 1
- 平均合併
- 6 小時 34 分鐘
- 30 天內合併 PR
- 118
描述
## Spark
Stuart Rimell, *Mental Models: All the Way Down*, highlighted 2026-08-03:
> Visualisations frequently form key elements of mental models. If folks share
> a common understanding of a system through a shared visualisation, then
> coherence will increase. Ensure static visualisations (diagrams) are updated
> to reflect changes in mental models.
The same week, genshin.dungeon.studio filed issues asking for exactly those
artefacts — a data flow diagram of client write paths, a record of which layer
owns writes — and one asking to abandon its bespoke DSGEP decision format for
standard ADRs.
## Why it could be interesting
The quote's second sentence is where the cost lives, and it's the sentence
everyone skips. A diagram that isn't maintained is worse than no diagram: it
still transmits a mental model, just the wrong one, with full confidence.
The DSGEP-to-ADR switch is the worked instance of a sharper claim: a bespoke
documentation format decays faster than a standard one, not because it's worse
designed but because nothing outside the project reinforces it. Standard
formats come with tooling, with skills, with reviewers who already know the
shape, and with an answer to "where does this go" that doesn't need relitigating.
The maintenance burden the quote names is paid down by conventionality, and
that's an argument for choosing boring formats that has nothing to do with the
format's merits.
## Open questions
- [ ] Is the claim about diagrams, about formats, or about the general property
that shared artefacts decay unless something reinforces them?
- [ ] What actually went wrong with DSGEP — did it drift, or was the cost just
anticipated? The post needs the real reason, not the tidy one.
- [ ] Does the ADR skill count as the reinforcement mechanism? If so, the claim
is narrower and more interesting: standard formats are cheap because the
tooling is already written.
- [ ] Which diagram would actually get maintained? Naming one that has survived
a change is stronger evidence than filing an issue for a new one.
- [ ] Relationship to #176 (ADRs co-created with an agent) — that's about
production cost, this is about decay. Sibling or one post?
## Source material
- Highlight — Stuart Rimell, *Mental Models: All the Way Down*: https://read.readwise.io/read/01kz4aseq90435wj4bx57h4xn7
- dungeon-studio/genshin.dungeon.studio#1132 — client write paths and state copies in a data flow diagram
- dungeon-studio/genshin.dungeon.studio#1133 — DSGEP-006 records which layer owns writes
- dungeon-studio/genshin.dungeon.studio#1136 — architecture decisions use a standard ADR format instead of DSGEP
- dungeon-studio/genshin.dungeon.studio#1167 — DSGEP-003 implementation notes describe the schema layout that shipped
- Brittany Ellich, *How GitHub engineers learn new codebases*: https://read.readwise.io/read/01jp840wf5e2qpjab1evkyqpw8
- Adjacent existing ideas: #176, #251, #328, #67
貢獻指南
這個儲存庫沒有索引到貢獻指南
評估
這個 Issue 還沒有評估資料。