graff — a fast agentic coding harness in Zig: multi-provider, MCP, workflows, DGM evolution loop, TS/Python SDKs
Python
evals: name the PR that carries the measured build
justrachd1b901b45696
diff --git a/evals/graff-vs-codex-app-server/RESULTS.md b/evals/graff-vs-codex-app-server/RESULTS.mdindex 1b1312f2..ff6a8f35 100644--- a/evals/graff-vs-codex-app-server/RESULTS.md+++ b/evals/graff-vs-codex-app-server/RESULTS.md@@ -4,7 +4,7 @@ Measured 2026-10-01. The same model (`gpt-6.1-sol`), ChatGPT account and machine| Arm | What ran || --- | --- |-| graff | The ADR 0231 build (the `perf: drop round trips the model does not need` commit on top of v0.0.302.12), as a scripted `graff repl` over the Codex Responses WebSocket. |+| graff | The ADR 0231 build (#1449: the `perf: drop round trips the model does not need` commit on top of v0.0.302.12), as a scripted `graff repl` over the Codex Responses WebSocket. || Codex app server | `codex app-server` from codex-cli 0.159.0, through [`graff-evals/codex_app_server.py`](../../graff-evals/codex_app_server.py). || pi | pi 0.84.2 on the ChatGPT plan (`--provider openai-codex`). pi has no MCP client, so it ran the 11 coding tasks only. |diff --git a/evals/graff-vs-opencode-and-pi/RESULTS.md b/evals/graff-vs-opencode-and-pi/RESULTS.mdindex 36ac2cfd..18ef93e3 100644--- a/evals/graff-vs-opencode-and-pi/RESULTS.md+++ b/evals/graff-vs-opencode-and-pi/RESULTS.md@@ -4,7 +4,7 @@ Measured 2026-10-01. One model, served through the Codegraff gateway's chat-comp| Arm | What ran || --- | --- |-| graff | The ADR 0231 build as a scripted `graff repl` (the `graff-gateway-repl` harness). |+| graff | The ADR 0231 build (#1449) as a scripted `graff repl` (the `graff-gateway-repl` harness). || OpenCode 2 | `@opencode/cli` 2.0.21: `opencode run --standalone --auto --format json`, through [`graff-evals/opencode2_run.py`](../../graff-evals/opencode2_run.py). || pi | pi 0.84.2 (the `pi-codegraff` harness). pi has no MCP client, so it ran the 11 coding tasks only. |