Rule Q13 and Q15: types from huginn-mind, and mind.redb

The client depends on the organ's library for types only, pinned by a grep
that it never opens or admits. The mind file is named for the engine that
holds it, because the selector and Studio both key on the extension. The
engine's provenance is recorded beside the ruling so nobody asks again.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
Meta Vens 2026-09-16 16:39:16 +02:00
commit 5a71c4bb3e

View file

@ -2357,9 +2357,9 @@ Huginn: +about 2,100 lines (store 120, mind 260, receipt 220, refusal 90, admiss
### Operator questions
- **Q13. Where do the batch and outcome types live for the client?** `eureka-state` (Cut 13) must construct `PipelineAdmissionBatch` and read `PipelineAdmissionOutcome`/`MindRefusal`, and the map says it does not depend on `huginn-mind` ("would invite a second validator", Cut 13). Cut 10 puts the wire types in the `huginn-daemon` *binary* crate, which a client cannot import either. Options: **A.** `eureka-state` depends on `huginn-mind` for types, with a negative grep that it never calls `open`, `admit` or `admit_prepared`; the graph cost is one package beyond what the leaf already brings (`redb` comes with the leaf regardless). **B.** A fourth crate `huginn-wire` holding only the request/response/outcome types; doctrine's "a named external consumer with a hard dependency boundary" argument (D1) applies weakly, since the boundary is a discipline a grep can pin. **C.** Put them in `epiphany-pipeline`; refused here because D6 already rules Huginn's transport contracts are Huginn's. **Recommended: A.** Cut 8 makes the types `pub` and `JsonSchema` so A or B both work; the choice is Cut 10/13's, but it should be ruled before Cut 10 writes `wire.rs` into a binary crate.
- **Q13. Where do the batch and outcome types live for the client?** `eureka-state` (Cut 13) must construct `PipelineAdmissionBatch` and read `PipelineAdmissionOutcome`/`MindRefusal`, and the map says it does not depend on `huginn-mind` ("would invite a second validator", Cut 13). Cut 10 puts the wire types in the `huginn-daemon` *binary* crate, which a client cannot import either. Options: **A.** `eureka-state` depends on `huginn-mind` for types, with a negative grep that it never calls `open`, `admit` or `admit_prepared`; the graph cost is one package beyond what the leaf already brings (`redb` comes with the leaf regardless). **B.** A fourth crate `huginn-wire` holding only the request/response/outcome types; doctrine's "a named external consumer with a hard dependency boundary" argument (D1) applies weakly, since the boundary is a discipline a grep can pin. **C.** Put them in `epiphany-pipeline`; refused here because D6 already rules Huginn's transport contracts are Huginn's. **Recommended: A.** Cut 8 makes the types `pub` and `JsonSchema` so A or B both work; the choice is Cut 10/13's, but it should be ruled before Cut 10 writes `wire.rs` into a binary crate. **Ruled A, 2026-09-16** ("I'll take your recommendations"). Cut 13's "no dependency on `huginn-mind`" becomes "depends on it for types only, with a negative grep that it never calls `open`, `admit` or `admit_prepared`"; Cut 10's `wire.rs` moves out of the binary crate accordingly when Cut 10 is refreshed.
- **Q14. How does Huginn run the mutation harness?** **A.** Add `-Repo` (default: the harness's own parent) to `tools/eureka-mutations.ps1` in Epiphany, about four lines, and keep one harness; Huginn holds only its entries file and its verification names Epiphany's path. **B.** Copy the harness into Huginn: a second copy of a 19 KB script the operator will notice, one bug away from two behaviours. **C.** Move the harness to the Eureka skill repo (`GameCult/Eureka`), the owner of cross-repo Eureka tooling, and point both repos at `~/.claude/skills/eureka/tools/`: coherent, three repos touched, and Cut 15 ("Skill wiring") is its natural home. **Recommended: A now, C in Cut 15.** *Taken as a default by Self, 2026-09-16: not a product fork.*
- **Q15. `mind.redb`, not `mind.cc`.** D3 wrote `mind.cc`; `F:\Projects\CLAUDE.md` says "all state should be CultCache `.cc` files". A redb-backed CultCache store is CultCache state in a different container, and the extension is what Epiphany's backend selector and CultCache Studio key on. **Recommended: `mind.redb`**, and the doctrine sentence reads "CultCache state", which this is. If the operator wants `.cc` literally, the cost is Epiphany's selector convention and Studio misreading the file, both real.
- **Q15. `mind.redb`, not `mind.cc`.** D3 wrote `mind.cc`; `F:\Projects\CLAUDE.md` says "all state should be CultCache `.cc` files". A redb-backed CultCache store is CultCache state in a different container, and the extension is what Epiphany's backend selector and CultCache Studio key on. **Recommended: `mind.redb`**, and the doctrine sentence reads "CultCache state", which this is. If the operator wants `.cc` literally, the cost is Epiphany's selector convention and Studio misreading the file, both real. **Ruled `mind.redb`, 2026-09-16.** The operator asked how redb got in: it entered Epiphany's runtime spine as the keyed backing store on 2026-08-08 (`5d6a7316`) and CultLib with the return of the Rust CultCache engine on 2026-09-03 (`b672fb7`); it is CultCache's own embedded keyed backend at the shared pin, not a dependency this campaign added.
- **Q16 (carried, not new). Q6/ruling 18 stands:** identity is declared. This spec builds no credential path; `ForeignInstance` is collision and attribution control (D8). *Not a question: ruling 18 is in force and nothing here reopens it. Left as written so the reader sees it was checked.*
### Findings not assignable to a cut