> Data revision: baf7c2f4307d3a989a8fb9363df8000fda3d9310
> Source files:
- [FINDINGS.md](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/FINDINGS.md)
- [hypotheses/f08-language-spec.md](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/hypotheses/f08-language-spec.md)

# F08 — Language and executable spec

> Source narrative: values not linked to an eligible claim ledger row are source-reported and not independently promoted here.

Cross-round review: [Rust vs Elixir: what our runs showed](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/research/RUST-VS-ELIXIR-EVIDENCE.md).

**Hypothesis:** H91–H99 ask whether language or executable specifications change runtime, parity, or task outcomes. The records inform a bounded product choice but do not establish a broad language effect.

**What ran:** The stack one-shot conformance round compares the planned Rust, Go, and TypeScript builds at their first recorded post-build conformance snapshots. It is **DESCRIPTIVE** because suite versions, spec revisions, hosts, and gate timing differ. The Sol-medium language replication adds 27 new cells and 31 reused cells; its task-equal rates are a post-hoc sensitivity analysis ([CL-SOL-LANG-REPLICATION-POSTHOC-EQUAL-TASK](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/results/claim-ledger.jsonl), **STATUS_ONLY**). Other R70 original, fixed-skeleton rerun, extension, compile-time, and task-eight smoke records are distinct cohorts.

**Result:** Source-reported, not reproducible from public data: one-shot snapshots show Rust at 76/196 cases (38.8%), Go at 61/236 on the historical macOS v1.2 suite and 76/278 on Linux v1.3, and TypeScript at 90/244 recorded case rows before its partial run stopped. The later Rust INT6 and final v1.3 results are separate from the one-shot comparison. In the related Sol-medium replication, raw totals were Rust 18/20, Go 18/20, and TS-Bun 15/18. Its post-hoc equal-task mean is Rust 90.3%, Go 88.9%, and TS-Bun 83.3%; [CL-SOL-LANG-REPLICATION-POSTHOC-EQUAL-TASK](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/results/claim-ledger.jsonl) is **STATUS_ONLY**. Rust conformed most by pass fraction in the one-shot snapshots and is top or tied-top across the recorded measures; the evidence does not support a matched causal language effect. Kogen continues in Rust.

**What it does not show:** The records do not establish that Rust is broadly superior to the other tested languages or that executable specifications improve agent completion. Different suite sizes and spec revisions, the partial TypeScript run, Go/TypeScript access to Rust reference code, different hosts, and different gate timing limit the stack comparison. The Sol-medium result has a separate status and limits; its pooled language comparison is not justified.

**Evidence:** [Stack one-shot conformance](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/rounds/stack-oneshot-conformance/README.md), [Sol-medium language replication](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/rounds/lang-sol-replication/README.md), [post-hoc equal-task claim](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/results/claim-ledger.jsonl), [F08 hypothesis page](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/hypotheses/f08-language-spec.md), [R70 original language cohort](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/rounds/r70-rve/README.md), [R70 rerun](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/rounds/r70-rve-rerun/README.md), [R70 extension](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/rounds/r70-rve-ext/README.md), [R70 compile-time record](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/rounds/r70-compile/README.md), and [R70 task-eight smoke](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/rounds/r70-task8/README.md). Spec map: [acceptance tests and stack adapters](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/EVIDENCE-MAP.md#24-acceptance-tests-and-stack-adapters) and [sources and precedence](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/EVIDENCE-MAP.md#sources-and-precedence).

See the [language round summary](/languages/) for the current comparison and final-round status.


## Related records

### Related hypotheses

- [H91](/hypotheses/h91/)
- [H92](/hypotheses/h92/)
- [H93](/hypotheses/h93/)
- [H94](/hypotheses/h94/)
- [H95](/hypotheses/h95/)
- [H96](/hypotheses/h96/)
- [H97](/hypotheses/h97/)
- [H98](/hypotheses/h98/)
- [H99](/hypotheses/h99/)

### Related experiments

- [lang-sol-replication](/rounds/lang-sol-replication/)

### Related claims

- [CL-SOL-LANG-REPLICATION-POSTHOC-EQUAL-TASK](/claims/cl-sol-lang-replication-posthoc-equal-task/)
