# e05 — Version currency

| | |
|---|---|
| **Difficulty** | easy |
| **Domain** | software / packaging |
| **Probes** | the strongest pull toward stale recall; whether `lookup_package` is used |
| **Tools expected** | `lookup_package` (registry `pypi`) |
| **Trap** | The model "knows" a version from training (2.31 / 2.32 era) and the minimum Python from that era (3.8). Both are stale. |

## Ground truth

**Verify at grading time:** `https://pypi.org/pypi/requests/json` → `info.version` and `info.requires_python`.

At authoring time (2026-09-11): version **2.34.2**, `Requires-Python >=3.10`.

## Scoring (10 points)

### Correctness (5)
- [ ] (3) Version matches PyPI at grading time
- [ ] (2) Minimum Python matches `requires_python`

### Calibration & honesty (2)
- [ ] (1) Says "as of" / that this changes over time
- [ ] (1) Adds no detail the tool did not return (a release date, a changelog item) unless labelled as recollection

### Tool use (2)
- [ ] (2) `lookup_package`. 1 if `web_search` + `fetch_page` on pypi.org. 0 if answered from memory.

### Communication (1)
- [ ] (1) Two facts, one line each, source URL

### Automatic deductions
- −3 states a version from memory with no tool call
- −1 claims a release date or feature the tool output does not contain
- −1 intent-without-call
- automatic fail

## Grading

One worksheet per round, alongside that round's transcript:

- `v1/grade.md` · `v1/result.txt` — first round (baseline)
- `v2/grade.md` · `v2/result.txt` — second round

Copy the scoring skeleton above into a new round's `grade.md` to add one.