docs: rescope the job-architecture extraction

[retr0h]

Task 3.2 is scoped wrong, and applying it is what showed why.

What it said

propose a capability from job-architecture.md's "Architecture Principles"; leave its remaining 550 descriptive lines in place

Those 550 lines are where the requirements actually are. The block named "Architecture Principles" is twelve lines of slogans — KV-First Storage, Stream Notifications. The rules that govern how an operation gets built are further down:

• Semantic routing rules (line 129) — whether an operation reaches jobs.query or jobs.modify
• Target types (line 153) — _all, _any, hostname, label selector
• Hierarchical label matching (line 163)
• Label limits (line 202) — five labels, with the consumer arithmetic that follows

Following 3.2 as written would have extracted the slogans and left the rules.

The routing rule has already drifted

The document classifies ten suffixes: .get, .query, .read, .status, .do → query; .update, .set, .create, .delete, .execute → modify.

The code uses twenty:

22 get   13 update   10 list   10 execute   10 delete   9 create
 1 each: stop start signal shutdown restart remove reboot
         password install enable disable sources query do

.list appears ten times and the rule does not classify it. Ten more verbs are unanticipated. .read and .set are documented and unused.

Nothing enforces any of it — the caller picks Query() or Modify() directly. The code outgrew the document, and because a description is checked against nothing, no one found out. A requirement with scenarios would have failed the moment .list appeared.

Changes

• 3.2 rescoped to a job-routing capability drawn from the whole document
• 3.2a added — establish the real routing rule before writing it down, since the documented one is a third accurate
• 3.2b added — the descriptive remainder stays: job states, append-only status, package layout, performance notes

design.md records the finding, including why renaming the three architecture documents should wait until after extraction rather than before.

No new capability here. The documentation capability already settles that a document mixing requirements with description gets split — this only corrects a task that misjudged which parts were which.

🤖 Generated with Claude Code

[github-actions]

Thank you for contributing to this project! 😊🕹️