on master help sign in
Name the core's face and join rulings in the milestone 2 brief

The finish brief's checker-core step now lists the rulings Montelot chose on 2026-11-03 from the syntax review, which the core must build in: a capital binds only the type under a face, joins keep a face once a position takes one, and the syntax thread's unface, S4 and S5. It also says the review-fix thread planned the same core, so the two don't start it twice. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01U4FzmSi6ruQLNHA1xAcXsN

commite208bcb2f19777b80628d81703e5dee465f91bbc authorClaude <noreply@anthropic.com> dateTue 3 Nov 2026, 03:59 (12 hours ago) parentf22d19d
1 file changed+2−1
docs/prompts/milestone-2-finish.md+2−1
docs/prompts/milestone-2-finish.md+2−1
@@ -13,7 +13,7 @@ You are finishing milestone 2 of Jibb, a UI language for Urbit: a Hoon dialect w
13133. [docs/spec/implementation-plan.md](../spec/implementation-plan.md), milestone 2, and the spec tabs it points at, closely Grammar's Tag resolution, Literals by context, Patterns, Modules and Imports.
14144. [docs/reviews/2026-11-02-questions.md](../reviews/2026-11-02-questions.md): the owner's answers, marked *(chosen)*. Q5 and Q6 set the direction below. Q16 to Q28 are proposals waiting for @iko, not spec.
15155. [docs/development.md](../development.md): setting up (`nix develop`), the tools, the test suites, and the checks before pushing (`tools/check` or `nix flake check`).
16−6. `git log origin/master`: @iko works in `jibb.hoon` in parallel, and another thread is implementing syntax-review proposals S1, S2, S4 to S7, S11 to S13 and S15 ([syntax review](../reviews/2026-11-02-syntax-review.md)). Rebase often and keep your changes out of theirs where you can.
16+6. `git log origin/master`: @iko works in `jibb.hoon` in parallel, and another thread is implementing syntax-review proposals S1, S2, S4 to S7, S11 to S13 and S15 ([syntax review](../reviews/2026-11-02-syntax-review.md)). Rebase often and keep your changes out of theirs where you can. The thread that fixes the 2026-11-02 review's findings planned this same core on 2026-11-03, in stages that keep every check green; ask in the project chat who is building it before you start step 2.
1717
1818## Ground rules
1919
@@ -38,6 +38,7 @@ The owner chose this on 2026-11-02: union-find bindings with levels, a queue of
3838- **Deferred constraints**: tags, `~`, numbers and cells, wing access, changes and `?-` on a type not yet known each wait in one queue, retried as variables are bound, as the Grammar's Tag resolution says. A tag-constrained variable does not generalize and stays open until something in the module fixes it. At the end of the module, each still-open constraint is an error at its source: "nothing fixes the union of %add; add a `^-`".
3939- **Order**: infer `++`s by the strongly connected components `839f6b0` already computes, a group of mutually recursive gates together; reject a non-gate value that depends on itself, naming the cycle.
4040- **Kinds by type**: check `main` as an ordinary expression. `(program init update view subs)`, `(page load view)` and `(store init update)` are calls whose unification fixes their components' types, messages included, so a tag in `view` is resolved by `main`. Then read the kind from `main`'s type, as the Grammar says, and run each kind's own checks after inference: gate-free island data and models, naming the field; for islands, a walk of the `(list (sub M))` that asks for a watchability proof only for `%watch`; for pages, a load that is any gate to `(ask D)`, `pure` and typed paths included.
41+- **Rulings it must build in**, chosen by Montelot on 2026-11-03 from the syntax review: a capital that meets a faced position binds only the type under the face, never the face, so `[A A]` fits `[x=@ y=@]`, `(hunt lt-a `1 `2)` checks, and `(roll ~[1 2] 0 add-f)` checks with `++ add-f |= [a=@ud b=@ud] (add a b)`; a join keeps a face once a position takes one, so reordering a list's items never changes whether it checks; and, as the syntax thread lands them, the explicit `unface` (S1), bare decimals fitting only `@ud`, `@u`, `@` and `*` (S4), and a bare-name pattern that spells a union tag being an error (S5). Take their final wording from `decisions.md`, and add each ruling's examples as golden cases in the stage that builds it.
4142- **Then delete** the program, page and store recovery chain, the deferred transactions, the prepared-tuple route and the other example-shaped slices as the core covers their tests (the earlier review counts about 1,180 lines of program recovery alone).
4243
4344When this is done, M2 to M5 are gone: every valid variant in the review's [probe corpus](../reviews/2026-11-03-milestone-2-probes/) checks, and each error in an island is reported where it is, as `button "Add" %submitt` giving "msg has no case `%submitt`" at the tag, with the union's site.