New rule `setting` (C011): a bolt.bend setting that names no rule or group (BOLT-CFG-7)

[ngngardner]

SPEC.md's BOLT-CFG-7 says "A setting in a bolt.bend whose name is no rule or group is a finding." Nothing implements it today: bolt/config.bend reads every def <name>() and grading only ever looks names up, so an unknown name is silently ignored. This issue proposes the rule that makes CFG-7 true, so it can then be proved like the other CFG rows (#107).

Why

A typo in a bolt.bend fails open. def wrp() -> String: "off" leaves wrap at its group level, and nothing says so. CFG-3 already made an unknown level word fail closed ("a typo makes a rule stricter, never silently off"); an unknown name is the same mistake one token to the left. Retired names hit it too: def quantify() (L004) and def shadow() (C001) are now dead settings, and def law() is the trap bolt/README.md already warns about.

Proposal

slug	setting
code	C011, the next correctness code (C001 stays retired)
group	correctness, default error, like the rest of the group: a setting that does nothing is a config bug, and CFG-3 already chose fail-closed
message	`wrp` names no rule or group, so it sets nothing. plus, when one slug or group is one edit away, Did you mean `wrap`? (optional; drop it if it costs more than it earns)
position	the def line of the setting in the bolt.bend: bolt.bend:L:1

Known names are exactly Codes.table()'s slugs plus the five group names. A retired slug (quantify, shadow) is unknown, and gets the finding.

Where it runs. A bolt.bend is read, never linted (BOLT-SCOPE-5), so this cannot be a per-file rule over Src. It belongs where the config is read: the grader already finds the nearest bolt.bend per file, so the natural home is a check over each bolt.bend that grading reads, reported once per bolt.bend in the run, after the per-file findings and before coverage/unsafe (BOLT-OUT-3's order gains one slot; the row would say so). The LSP, which runs per-file rules only, would report it when the open document is a bolt.bend, which today gets no lint (BOLT-LSP-4's "an open bolt.bend gets no lint" would change to "gets only setting").

Its own level. setting is graded like any rule, by the bolt.bend it reports on, so def setting() -> String: "off" turns it off, and def correctness() sets it with its group.

Files it touches

• bolt/codes.bend: a setting/C011/correctness row (and the one-per-row proof arm OUT-1's laws in bolt/PROOF.bend need, test(output): prove the code table one-to-one and the finding format (BOLT-OUT-1, BOLT-OUT-2) #112).
• bolt/config.bend: a pure unknown(cfg: Config) -> List<(name, line)> over the parsed settings, which needs the settings to keep their line.
• bolt/lint.bend (and bolt/lsp/ for the open-bolt.bend case): report it.
• bolt/README.md: the group table, the code table, the rule's entry.
• SPEC.md: BOLT-CFG-7 cites the law; BOLT-OUT-3 and BOLT-LSP-4 wording as above; a BOLT-RULE-C011 row.

The law (CFG-7)

Over every list of settings: unknown(Config{sets}) holds exactly the settings whose name is neither a slug of Codes.table() nor one of the five groups, in file order, one each. Stated against a spec function in bolt/LAWS.bend, proved in bolt/PROOF.bend, next to the CFG laws.

Open questions

1. Correctness/error, or style/warn (S005) so existing projects with a stale def quantify() get a warning, not a failing gate, on upgrade?
2. Keep the "did you mean" suggestion, or leave it out of the first version?

---

Generated by Claude Code