Run hypergraph-native region optimization after every solve

[seveibar]

Summary

• pin tiny-hypergraph#166 at 8573307
• run its post-solve region-cost optimizer after every shared tiny-hypergraph pathing solve
• always consume the optimizer's solved hypergraph as the port-point pathing output
• preserve serialized fixed routes using existing hypergraph connection metadata
• use the dependency's final path-relinking pass to remove speculative mutations that are not needed to hold the optimized peak

There is no optimizer toggle, per-pipeline branch, dataset classifier, density gate, or target threshold.

Scope boundary

This is the focused replacement for #2100, which remains open only as an implementation and benchmark reference.

This PR intentionally does not:

• pass SimpleRouteJson into hypergraph pathing
• infer obstacle connectivity or reconstruct pad/obstacle geometry
• add pipeline-specific optimizer behavior
• change DRC evaluation, high-density repair, stitching, or trace simplification
• add thresholds, sample/dataset checks, alternate portfolios, or conditional stage execution

Why these changes are necessary

The tiny-hypergraph section pipeline owns the post-solve stage, but the autorouter wrapper previously returned the preceding section solver's graph. This integration supplies only hypergraph-native options and selects the optimizer stage's output. An optimizer failure is surfaced as an error rather than hidden behind a fallback.

The dependency now follows single-route descent before opening pair ejection, preserving the full neighborhood while avoiding repeated quadratic work. It then reverse-relinks the accepted mutation history under the achieved peak as a hard constraint. This retains temporary plateau moves when they unlock a better peak, but removes unrelated lane displacement and superseded route replacements before detailed routing.

Pipeline7 evidence

• Tiny-hypergraph ./benchmark.sh: all 8 committed Pipeline7 hypergraphs completed and improved; P50 maximum region cost reduced 59.1%, P95 reduced 51.8%.
• Paired SRJ18 same-machine run: relaxed-DRC success increased from 43.8% to 50.0% (+6.2 pp) with 1 improved outcome and 0 regressions. Completion held at 81.3%, P80 time fell 22.3%, and average vias fell 1.5%.
• In that paired run, dense sample 2 changed from solved with relaxed-DRC errors to a relaxed-DRC pass while falling from 349.4s to 271.4s.
• SRJ18 sample 3: completes in 10.4s and passes relaxed DRC; the prior optimizer revision's unrelated same-net via regression is gone.
• SRJ18 sample 13: keeps a 33.8% peak-cost reduction (4.277 to 2.831). Trace simplification finishes in 12.3s instead of remaining in same-net via merging past 60s, and the full sample completes in 79.2s rather than timing out. Its existing dense-layout DRC failures remain.

The paired benchmark used runtime commit 45cc81c; subsequent commits change SVG snapshots only.

Validation

• tiny-hypergraph: 144 tests passed, 0 failed
• tiny-hypergraph: bunx tsc --noEmit, format check, and git diff --check
• tiny-hypergraph: ./benchmark.sh — P50 59.1%, P95 51.8%
• autorouter universal-stage integration test: 1 passed, 0 failed
• autorouter focused affected route/snapshot tests: 12 passed, 0 failed
• autorouter CI: all 8 test shards, build, type-check, format-check, and added-code check passed
• autorouter: git diff --check

The autorouter diff contains no new SimpleRouteJson, geometric-connectivity, DRC-repair, high-density-repair, threshold, or sample-specific plumbing.

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
capacity-node-autorouter	Ready	Preview	Aug 15, 2026 7:05am

[tscircuitbot]

Benchmark This PR

Run benchmarks by commenting on this PR:

/benchmark [benchmark.sh args...]
/benchmark-long [benchmark.sh args...]
/benchmark-all
/benchmark --same-machine
/benchmark-all --same-machine

Comment /benchmark to run the default dataset, or append any arguments accepted by ./benchmark.sh.
Comment /benchmark-long for an 8-vCPU run that defaults to 8 workers and has an eight-hour timeout.
Comment exactly /benchmark-all to start separate workflow runs and result comments for the default dataset plus srj18, srj19, srj20, srj21, and srj23.
Append --same-machine to /benchmark or /benchmark-all to compare current main and the PR head sequentially on the same 8-vCPU Blacksmith runner.

Everything after /benchmark or /benchmark-long is safely forwarded to ./benchmark.sh, except --profile-solvers, which enables profile comparison tables, and --same-machine, which enables the paired runner.
Examples: /benchmark --dataset 18 --sample-timeout 2000s, /benchmark --pipeline 7 --scenario-limit 20, /benchmark all 20 --concurrency 8, /benchmark --same-machine, /benchmark-all --same-machine, and /benchmark-long --dataset 18.

Use /update-snapshots (or /us) to run BUN_UPDATE_SNAPSHOTS=1 bun test --timeout 120_000 on the PR branch and auto-commit snapshot updates.
Use /usf to read recent failed test files, update and verify their exact CI test shards, and auto-commit only their snapshots. It uses the configured fast benchmark runner by default; use /usf --ubuntu-latest for GitHub-hosted x64 CI parity.

Any PR whose title contains [BENCHMARK TEST] will automatically run one default-dataset benchmark on PR updates; it does not post a PR result comment.

[seveibar]

/benchmark --same-machine --dataset 18

[tscircuitbot]

Same Machine Benchmark Results

Both revisions ran sequentially in one Blacksmith job on blacksmith-01m019et9y79cs6m3kdqdvfnbc-8vcpu.

Dataset: srj18 · Scenarios: 16
Main: 35f0116 · PR: 1627a35

Solver	Metric	Main	PR	Delta
Pipeline7	Completion	81.3% (🕒18.8%)	75.0% (🕒25.0%)	-6.3 pp
Pipeline7	Relaxed DRC pass	43.8% (🕒18.8%)	37.5% (🕒25.0%)	-6.3 pp
Pipeline7	Timeouts	3	4	+1
Pipeline7	P50 time	142.4s	151.9s	+6.7%
Pipeline7	P60 time	204.8s	216.0s	+5.5%
Pipeline7	P70 time	240.5s	269.9s	+12.2%
Pipeline7	P80 time	352.9s	360.0s	+2.0%
Pipeline7	P90 time	360.0s	360.0s	0.0%
Pipeline7	P95 time	360.0s	360.0s	0.0%
Pipeline7	Average vias	210.15	205.83	-2.1%

Outcome changes: 1 improved, 3 regressed. Timing percentiles include solved and timed-out samples; negative timing deltas are faster.

Changed outcomes (4)

Solver	Sample	Main	PR	Main time	PR time	Delta
Pipeline7	2	Solved (DRC failed)	DRC passed	352.9s	259.6s	Improved
Pipeline7	3	DRC passed	Solved (DRC failed)	40.8s	37.3s	Regressed
Pipeline7	5	DRC passed	Solved (DRC failed)	33.1s	36.1s	Regressed
Pipeline7	13	Solved (DRC failed)	Timeout	239.4s	360.0s	Regressed

Workflow: View run
Artifact: Download raw reports

[seveibar]

/benchmark --dataset 18 --same-machine

[tscircuitbot]

Same Machine Benchmark Results

Both revisions ran sequentially in one Blacksmith job on blacksmith-01m01djnsvb2a815a1ye9npz9b-8vcpu.

Dataset: srj18 · Scenarios: 16
Main: 35f0116 · PR: 74c2905

Solver	Metric	Main	PR	Delta
Pipeline7	Completion	75.0% (🕒25.0%)	75.0% (🕒25.0%)	0.0 pp
Pipeline7	Relaxed DRC pass	43.8% (🕒25.0%)	43.8% (🕒25.0%)	0.0 pp
Pipeline7	Timeouts	4	4	0
Pipeline7	P50 time	147.5s	154.0s	+4.4%
Pipeline7	P60 time	208.2s	208.1s	-0.1%
Pipeline7	P70 time	241.9s	268.7s	+11.1%
Pipeline7	P80 time	360.0s	360.0s	0.0%
Pipeline7	P90 time	360.0s	360.0s	0.0%
Pipeline7	P95 time	360.0s	360.0s	0.0%
Pipeline7	Average vias	181.83	205.67	+13.1%

Outcome changes: 1 improved, 2 regressed. Timing percentiles include solved and timed-out samples; negative timing deltas are faster.

Changed outcomes (3)

Solver	Sample	Main	PR	Main time	PR time	Delta
Pipeline7	2	Timeout	DRC passed	360.0s	265.5s	Improved
Pipeline7	3	DRC passed	Solved (DRC failed)	45.0s	38.6s	Regressed
Pipeline7	13	Solved (DRC failed)	Timeout	234.2s	360.0s	Regressed

Workflow: View run
Artifact: Download raw reports

[seveibar]

/benchmark --dataset 18 --same-machine

[tscircuitbot]

Same Machine Benchmark Results

Both revisions ran sequentially in one Blacksmith job on blacksmith-01m020zsvt0byrhptv6w65x9f5-8vcpu.

Dataset: srj18 · Scenarios: 16
Main: 35f0116 · PR: 45cc81c

Solver	Metric	Main	PR	Delta
Pipeline7	Completion	81.3% (🕒18.8%)	81.3% (🕒18.8%)	0.0 pp
Pipeline7	Relaxed DRC pass	43.8% (🕒18.8%)	50.0% (🕒18.8%)	+6.2 pp
Pipeline7	Timeouts	3	3	0
Pipeline7	P50 time	140.2s	150.1s	+7.1%
Pipeline7	P60 time	199.0s	202.2s	+1.6%
Pipeline7	P70 time	242.6s	241.5s	-0.5%
Pipeline7	P80 time	349.4s	271.4s	-22.3%
Pipeline7	P90 time	360.0s	360.0s	0.0%
Pipeline7	P95 time	360.0s	360.0s	0.0%
Pipeline7	Average vias	210.15	206.92	-1.5%

Outcome changes: 1 improved, 0 regressed. Timing percentiles include solved and timed-out samples; negative timing deltas are faster.

Changed outcomes (1)

Solver	Sample	Main	PR	Main time	PR time	Delta
Pipeline7	2	Solved (DRC failed)	DRC passed	349.4s	271.4s	Improved

Workflow: View run
Artifact: Download raw reports

[seveibar]

/benchmark --dataset 18 --same-machine

[tscircuitbot]

Same Machine Benchmark Results

Both revisions ran sequentially in one Blacksmith job on blacksmith-01m02xd0yjrdpqxa5m60amg9he-8vcpu.

Dataset: srj18 · Scenarios: 16
Main: 35f0116 · PR: efe68be

Solver	Metric	Main	PR	Delta
Pipeline7	Completion	81.3% (🕒18.8%)	81.3% (🕒18.8%)	0.0 pp
Pipeline7	Relaxed DRC pass	43.8% (🕒18.8%)	50.0% (🕒18.8%)	+6.2 pp
Pipeline7	Timeouts	3	3	0
Pipeline7	P50 time	137.1s	152.7s	+11.3%
Pipeline7	P60 time	199.6s	204.6s	+2.5%
Pipeline7	P70 time	231.9s	251.4s	+8.4%
Pipeline7	P80 time	346.3s	274.8s	-20.6%
Pipeline7	P90 time	360.0s	360.0s	0.0%
Pipeline7	P95 time	360.0s	360.0s	0.0%
Pipeline7	Average vias	210.15	206.92	-1.5%

Outcome changes: 1 improved, 0 regressed. Timing percentiles include solved and timed-out samples; negative timing deltas are faster.

Changed outcomes (1)

Solver	Sample	Main	PR	Main time	PR time	Delta
Pipeline7	2	Solved (DRC failed)	DRC passed	346.3s	271.1s	Improved

Workflow: View run
Artifact: Download raw reports

[github-actions]

This PR has been automatically marked as stale because it has had no recent activity. It will be closed if no further activity occurs.

[github-actions]

This PR was closed because it has been inactive for 1 day since being marked as stale.