idkmesh

Issue #13 scaling synthesis

Date: 2026-08-29 Issue: #13

Project-owner request

The project owner asked multiple agents to work professionally in parallel on distinct open issues, solve what can be supported, and integrate through the repository’s pull-request safeguards. This workstream was assigned issue #13.

Assistant interpretation

Before changing code, inspect live assignees, pull requests, branches, and the current main branch. Claim the issue publicly only if no other active owner is visible. Audit the merged R1/R2/R3/R4 and E-series artifacts against issue #13, then implement the highest-value bounded missing experiment or synthesis.

The work must not describe synthetic probabilities as real coding-agent evidence. Issue #13 may close only if every minimum configuration, deliverable, metric, and success criterion has actual support.

Finding

R1 already provided a one-size diversity experiment, a help/hurt assumption sweep, and a conservative adapter for future real ResultManifest and VerificationResult corpora. R2, R3, R4, and the E-series provide separate scheduling, orchestration, routing, and verification evidence.

The missing bounded layer was an explicit R1 curve over N with marginal success, resource deltas, repeated difficulty assumptions, and a durable audit of which issue #13 arms remain unmeasured. The adjacent experiments cannot honestly be combined as if they shared one held-out coding-task corpus.

Decision and implementation

Add a deterministic synthetic scaling runner over N = 1, 2, 5, and 10 for homogeneous, structurally diverse, and diverse-verifier families. Preserve raw seed trials, equal-attempt comparisons, uncertainty intervals, negative or saturated returns, cost deltas, and an explicit evidence-coverage ledger.

Publish a frozen 10-seed reference result and reproducibility command. Keep the evidence label at synthetic mechanism and use Advances #13 in the pull request.

Open gates

Community impact

The contribution makes the next work legible: contributors can see exactly which experiment cells exist, which are synthetic proxies, and which real observations are still needed. It avoids converting a large collection of unrelated green experiments into a false scaling-law claim.