Skip to content

RFC: bound DDR routing attempts and checkpoint validated copper - #4949

Draft
seveibar wants to merge 1 commit into
mainfrom
codex/propose-ddr-bounded-routing
Draft

seveibar wants to merge 1 commit into
mainfrom
codex/propose-ddr-bounded-routing

Conversation

@seveibar

@seveibar seveibar commented Oct 27, 2026 •

Copy link
Copy Markdown
Contributor

Status: this RFC is an isolated supervisor experiment, not a production routing implementation. The concrete solver implementation and fresh 47/47 DDR routing proof are now tracked in bus-lanes-solver #102. That PR adds bounded default group routing but does not yet provide persistent checkpoints or establish whole-board completion.

A late DDR routing failure should return usable validated progress instead of discarding it after an unbounded search. This proposes an external supervisor with request/attempt deadlines, isolated candidates, independent validation, content-addressed checkpoints, and finite retries.

Includes an executable Bun prototype, a separate six-layer AM3352 fixture with RAM at (0, -29) mm, native validation reports, and an integration proposal for core's worker runtime and routing phases. Production routing defaults are unchanged.

Independent evidence

  • No escape-placement planner from RFC: plan DDR escapes with movable decouplers #4948 is applied to this layout.
  • The complete-DDR monolithic control timed out at 120 seconds in shared-layer routing.
  • The staged supervisor validated and checkpointed byte 0 (11 connections), tried byte 1 twice, then returned incomplete after 11.724 seconds with the accepted copper preserved.
  • A real-router fault-injection trial killed a deliberately 1-second attempt after 1.015 seconds, validated its retry, and retained the same copper when byte 1 failed.
  • Resume revalidated the checkpoint and retried only byte 1 in 3.293 seconds. All 11 native checks passed on the retained byte lane.
  • Neither the complete 47-connection DDR bus nor the whole board routed successfully. These results establish bounded recovery and retained progress, not faster whole-board completion.

Validation and limits

Two supervisor tests pass: non-cooperative worker termination/fallback/resume, and preservation after rejected output/corruption rejection. The targeted strict TypeScript check passes. Tests using synthetic output are separate from native board validation.

The fixture is a different placement of the earlier populated AM3352 circuit. Both proposal experiments share a pinned solver plus a disclosed common investigation patch; this is not a clean-upstream speed benchmark. The Bun prototype is outside the production build and still needs process-tree/browser lifecycle support, read-only checkpoint isolation, cancellation/concurrency handling, and whole-board stages.

This remains draft because the requested independent complete-board proof is outstanding.

See proposal and reproduction instructions and machine-readable evidence.

@vercel

vercel Bot commented Oct 27, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
tscircuit-core-benchmarks Ready Ready Preview Oct 27, 2026 6:36am UTC

Request Review

This branch was successfully deployed

1 active deployment
Preview — 2d3de97a Deployed Oct 27, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant