Skip to content

acceptance.md

Latest commit

 

History

History
69 lines (48 loc) · 3.86 KB

acceptance.md

File metadata and controls

69 lines (48 loc) · 3.86 KB
Raw

Acceptance results

Verified October 14, 2026 against https://api.iteracompute.com/v1 with qwen/qwen3.8-27b and a user-supplied test key. All four examples returned non-empty replies and exited successfully. Tests ran on Windows with Python 3.11.4, Node.js 25.2.1, Python OpenAI 3.30.0, TypeScript OpenAI 7.32.0, tsx 4.23.15 and OpenCode 1.19.0.

Machine-readable evidence contains actual stdout, coding tool outcomes, code changes and the independent test rerun. Credentials, local machine paths and session IDs are omitted.

Python and TypeScript quickstarts

Ran the documented Python command and npm run quickstart sequentially. Each printed Hello! and exited with code 0. TypeScript also printed npm's script banner. Both explicitly passed the IteraCompute key and used the exact API model ID.

Hello!

Summarizer

Ran examples/summarizer/summarize.py examples/summarizer/meeting.txt --output summary.txt from the root using the virtual environment. Actual stdout:

• Customer feedback dashboard launch decision: go live on October 21, but the launch is blocked until CSV export tests pass and the support guide is reviewed.
• Maya owns the release checklist and must finish it by October 18; Leo will add CSV export before launch.
• Mobile layout is deferred to the next release.

Saved: summary.txt

Exit code 0; the saved file was non-empty and matched the printed summary. Reviewed all three bullets against the source: launch date, owners, checklist deadline, CSV export, deferred mobile work and blockers were preserved. Wording will vary on another run.

OpenCode

Installed OpenCode from the included lockfile, copied the starter and configuration into workspace, and ran the documented Build-agent command.

Check Actual result
Starter before repair 4 tests, 1 pass, 3 fail; exit code 1.
Model tool execution Read the function and tests, wrote the repaired function, then ran node --test slugify.test.mjs successfully.
Test integrity slugify.test.mjs remained byte-for-byte identical to the starter. Its SHA-256 is in the evidence file.
Agent completion Non-empty final reply reported all four tests passing.
Independent rerun 4 tests, 4 pass, 0 fail; exit code 0.

Actual model-written function:

export function slugify(text) {
  return text
    .toLowerCase()
    .replace(/[^a-z0-9]+/g, "-")
    .replace(/^-+|-+$/g, "");
}

The configuration explicitly sets free Qwen's prices to zero; the final run's OpenCode cost estimate was zero. Client estimates are not billing evidence; check allowance and charges in the IteraCompute Console.

This verifies a real code-edit and test workflow on Windows. macOS/Linux, larger repositories, hosted deployments and browser applications were not tested.

Offline checks

Verified October 14, 2026 on Windows with Python 3.11.4 and Node.js 25.2.1:

Check Actual result
Python dependency install openai==3.30.0 installed in a fresh virtual environment.
TypeScript dependency install openai@7.32.0, tsx@4.23.15 installed from the package manifest; lockfile included.
python -m unittest discover -s tests -v 7 tests passed, including missing/blank/placeholder keys, explicit IteraCompute authentication, exact API model ID, empty replies, HTTP 401/429 and summary file writes.
npm run code -- --version in examples/opencode 1.19.0
npm run code -- models iteracompute with a fake key The configured iteracompute/qwen/qwen3.8-27b selector was recognized. No chat request was made.
node --test slugify.test.mjs in a fresh starter workspace 4 tests, 1 pass, 3 fail, exit code 1. This reproduces the coding task before a model repairs it.

These offline SDK checks use fake credentials and mocked HTTP transport. They complement the real API and coding workflow results above.