Seven sources and a real test contract—but no Go dependency planner

Research question

Where does research-to-program construction stop on the Go dependency-wave task?

Why this matters

Separating acquisition and test observation from candidate generation identifies which part of the construction path remains blocked.

What was tested

Research public references, recover examples and rejection constraints, then attempt bounded typed program induction and a separate relational-induction route.

Baseline

The unchanged Go dependency-wave acceptance contract; no task-specific planner implementation.

Comparison conditions

October 23 acquisition/build attempt; October 31 direct search; separate October 31 schema-gated diagnostic. Their operation budgets are not combined.

Test environment

Isolated Go task workspace; immutable acceptance tests; bounded candidate search.

Observed result

Seven sources were acquired in the earlier attempt and test examples were recovered, but no application was built. The later direct search stopped at 100,001 operations against a 100,000-step limit with no candidate. The separate shape-gated route generated zero candidates before type checking and did not exhaust its search budget. The oracle remained unchanged.

Outcome

Documented negative result; dependency-wave planner not demonstrated.

Measured results

Measurement conditions: Counts are from distinct dated attempts and diagnostics; do not add their separate search budgets together.

public Sources Acquired

7

direct Search Operations

100001

direct Search Limit

100000

direct Search Candidates

0

shape Gated Candidates

0

shape Gated Budget Exhausted

No

application Files Written

No

oracle Changed

No

Evaluation method

Run the immutable Go baseline, inspect research and test-observation receipts, then compare the direct typed search and the separately shape-gated relational search diagnostics.

External model usage

Zero external-model calls in the cited bounded diagnostics.

Verification and lifecycle

verified · failed · Measured 2026-10-31T01:11:19.307Z · Report LARI Full Report 2026-10-31T11:42:43.574Z

Limitations

One Go dependency-wave task and its frozen tests. Search counts are configuration-specific, not a speed benchmark. No holdout was added to the shape-gated diagnostic. A compiler sketch is not task completion; the planner remains unsolved.

Architecture component: execution →

Related experiments

• A bounded relational procedure passed held-out tests and failed after exact removal
• LARI built a small JavaScript app twice, starting from zero each time

Public provenance

Owner-reviewed summary of dated acquisition and current-runtime diagnostic receipts; source paths, hashes, and private artifacts are withheld.

Source Go dependency-wave acquisition and current-runtime diagnostics · 2026-10-23 to 2026-10-31 · Published 2026-10-31T20:03:40.716Z