epic: Game/index.ts remediation (types #96 + decomposition #132 + coverage #134)

[nodots]

Epic: Game/index.ts remediation

src/Game/index.ts has accumulated two related forms of tech debt that both need to be paid down in the same file. This epic tracks them together so the work is sequenced coherently rather than colliding.

The three parts

• chore: Eliminate as any casts in Game/index.ts with proper type definitions #96 — Type safety: eliminate the as any casts (~46 on feat/4.6.4-RC), including the untyped undo stack and cube offeredThisTurnBy. Requires additions to @nodots/backgammon-types.
• refactor: split Game/index.ts god file into cohesive modules #132 — Structure: split the 2491-line god file (~40 static methods, six concerns) into cohesive modules (shared, guards, undo, cube, lifecycle, turnFlow, robot) behind a facade Game class, with no public-API change.
• test: expand core coverage, starting with Game/index.ts extraction targets #134 — Test coverage: raise coverage on the extraction targets before they move (characterization tests), then lift core coverage more broadly + add a threshold gate. Core is at ~57% line / ~41% branch; Game/index.ts at ~54% stmt / ~46% branch, lowest exactly on the cube/undo/resign/restore methods refactor: split Game/index.ts god file into cohesive modules #132 moves.

Why group them

All three edit or protect the same file and touch the same weak spots — the undo stack and cube fields are simultaneously untyped (#96), scattered across concerns (#132), and untested (#134). Doing them blind to each other risks merge churn and rework; the coverage gap makes the refactor riskier precisely where it's largest.

Sequencing

1. chore: Eliminate as any casts in Game/index.ts with proper type definitions #96 first (or its type additions). Promote the undo/cube/stateVersion types in @nodots/backgammon-types, then remove the casts. This lets the decomposition land into already-typed code.
2. refactor: split Game/index.ts god file into cohesive modules #132 second. Phase 2 of the split consolidates undo-stack access into a single undo.ts helper, then deletes the casts once the type is fixed.

Type changes are in-scope (authorized). The undo casts are removable outright: promote undo?: { frames: BackgammonGameMoving[] } from BackgammonPlayMoving to BasePlay in packages/types/src/play.ts, using a type-only import type for BackgammonGameMoving (defined in the same package at game.ts:193). Contrary to an earlier note, this is not a types→core dependency — it's only a play.ts ↔ game.ts file cycle inside the types package, which is harmless for an erased type-only import.

Either order works as long as the two are coordinated and not run in parallel on overlapping lines.

3. test: expand core coverage, starting with Game/index.ts extraction targets #134 interleaves ahead of refactor: split Game/index.ts god file into cohesive modules #132. For each extraction phase in refactor: split Game/index.ts god file into cohesive modules #132 that moves an under-covered method (cube, undo, resign, restore, switchDice), the matching characterization tests from test: expand core coverage, starting with Game/index.ts extraction targets #134 land first, so a green suite after extraction proves behavior was preserved. The broader coverage lift + threshold gate in test: expand core coverage, starting with Game/index.ts extraction targets #134 can follow independently.

Done when

• Zero unexplained as any in src/Game/index.ts (chore: Eliminate as any casts in Game/index.ts with proper type definitions #96)
• index.ts is a thin facade delegating to per-concern modules (refactor: split Game/index.ts god file into cohesive modules #132)
• Extraction targets have characterization coverage; core coverage lifted with a threshold gate (test: expand core coverage, starting with Game/index.ts extraction targets #134)
• core typechecks, the Game/__tests__ suite is green, and api + ai build unchanged

[nodots]

Decisions (2026-07-15): (1) one PR per phase across #132 and #134 for reviewability; (2) coverage gate at 75% (stmts/branch/funcs/lines), added to jest.config.js once coverage clears 75% — ratcheted up as coverage rises rather than one jump at the end.