# Terminal-Bench 2.0 Smoke24 Task List

This file documents the fixed 24-task Smoke24 corpus used for the model-card
summary results. It intentionally lists the task names and harness shape only;
it does not publish per-task outcomes, model-by-model traces, or aggregate CSV
details.

## Corpus

- Benchmark: Terminal-Bench 2.0
- Corpus ID: `tb20-coder-smoke24-fast-success-failure-20260616`
- Created: 2026-06-16
- Size: 24 tasks
- Selection policy: 12 shortest prior successes and 12 shortest prior failures
  from a recovery-corrected Qwopus3.6-27B-v2-GPTQ-Pro-v1 aggregate.

## Harness Shape

- Agent: `terminus-2`
- Concurrency: `1`
- Sandbox: 32 CPU / 48 GiB RAM
- Task timeout: 30 minutes
- Max output: 40k tokens
- Thinking token budget: 32768
- Sampling: temperature `1.0`, top-p `0.95`, top-k `20`

## Tasks

### Prior Success Group

- `git-leak-recovery`
- `prove-plus-comm`
- `fix-git`
- `modernize-scientific-stack`
- `kv-store-grpc`
- `openssl-selfsigned-cert`
- `headless-terminal`
- `multi-source-data-merger`
- `fix-code-vulnerability`
- `nginx-request-logging`
- `build-pmars`
- `code-from-image`

### Prior Failure Group

- `feal-differential-cryptanalysis`
- `raman-fitting`
- `log-summary-date-ranges`
- `video-processing`
- `sqlite-with-gcov`
- `sanitize-git-repo`
- `torch-pipeline-parallelism`
- `count-dataset-tokens`
- `configure-git-webserver`
- `cancel-async-tasks`
- `mteb-retrieve`
- `hf-model-inference`