[joelteply]

Placement switching occupied one 1,775-line file. Split it by responsibility: persona owns the adapter, pure decisions and fixtures; Grid owns fleet orchestration and opportunity placement. This removes the duplicate upward dependencies introduced by the initial split without raising the layer baseline.

The orchestration Rust API moves from persona::placement_switch::follow_the_fleet to modules::grid::placement::follow_the_fleet; its sole repository caller is migrated. Function bodies and placement behavior are preserved. Native-model eligibility and fallback behavior repairs remain separate work.

Validation: git diff --check passes. Whole-file source measurement is 154,999 excess lines, down 775 from current canary. The largest touched core source file is below 1,000 lines. Final repaired-head CI passes, including Linux tests, Windows compilation/custody and hygiene; independent exact-head source review approves. The earlier upward-import hygiene failure was corrected by moving orchestration to its actual Grid owner. No serving-host compilation or deployment was performed.

[joelteply]

Fable review at de04c44: APPROVE (source) as a pure split.

The claim I verified: no behaviour moved. I sorted the whitespace-normalized lines of placement_switch.rs at the merge-base and of the six resulting files, then compared the two sets.

• Tests: 16 before, 16 after.
• Lines only in the old file: mod tests {, one closing brace, fn of(...) (now pub(super) fn of), and SWITCHES.insert(switch.persona_id, switch). That last one is now switch.persona_id(), a getter, because the field crosses a module boundary.
• Lines only in the new files: module declarations, use super::*, re-exports (pub use adapter::PlacementSwitch, pub(crate) use decision::{…}, follow_the_fleet, follow_the_allocation), module docs, and #![cfg(test)].

Every file is ≤494 lines, so the split passes the oversized-edit gate on #5221.

Not checked: CI, and whether the old public path for each moved item still resolves outside persona/. The re-exports suggest it does; compile will prove it.

🤖 Generated with Claude Code

[joelteply]

Independent source review by astra_architecture_review, exact c86e096: APPROVE. Reviewer reports no authorship or prior involvement.

Compared against merge-base9c2e7b8d: moved function/impl bodies and all16 tests preserved after indentation/visibility normalization; register uses equivalent persona_id getter. Sole fleet caller migrated to modules/grid/placement; no old forwarding facade or duplicate fleet implementation. No new Persona-to-Grid import. Existing switch/decision public paths preserved; orchestration path migration is explicit. Excess-line ratchet falls775. Read-only review, no runtime performance claim.

Parent separately verified all exact-head CI green, including Linux tests, Windows compilation/custody, binding drift and hygiene. Integration is not deployment or proof of learning.