Held-Out Test Set
A held-out test set is a portion of data deliberately excluded from a model's training or fine-tuning process and reserved purely for evaluation, so performance is measured on examples the model has genuinely never seen — the only way to get an honest read on how it generalizes rather than how well it memorized.
The core discipline is isolation: once a test set is designated "held out," it must never leak back into training, prompt examples, or few-shot demonstrations — even accidentally including a handful of held-out examples in a fine-tuning run inflates every score measured against that set afterward, silently.
Teams that fine-tune or iterate on prompts frequently need a rotation discipline too — if the same held-out set is used to guide every iteration decision, engineers start implicitly overfitting to it through repeated trial and error, even without ever training on it directly. Some teams keep a second, truly untouched set for final validation specifically to catch this.
For QA purposes, a held-out set built from real production edge cases (anonymized, with known-correct answers) tends to be far more diagnostic than a generic public one, since it reflects the actual distribution of inputs the system will face.
