---
name: innovation
description: Use when a Guochuang project needs technical novelty, mechanism, baseline, comparison, reproducibility or validation claims examined.
---

# Innovation

Apply the shared policies in `gps-common`. When saving a Markdown report, use the common project-name rule with topic `创新`.

Test four separate questions: what changed, why it is difficult, what improved under which conditions, and what the students themselves contributed. Adding AI, a patent or a new label does not answer them.

## Own this work

Own mechanism comparison, prior art, baseline fairness, test design, measured delta, reproducibility and technical limitations. `positioning` owns the final project advantage and selling-point ranking. Pass it evidence-bounded innovation candidates rather than taking over the whole narrative.

## Build an innovation card

For each claim record:

`claim_id`, `innovation_type`, `prior_art_or_baseline`, `changed_mechanism`, `measured_delta`, `metric_unit`, `condition`, `sample_or_repeats`, `evidence_ids`, `independent_source`, `student_role`, `known_limitation`.

Classify an advantage candidate as:

- `owned`: the team controls the mechanism or delivery capability and can explain the evidence;
- `conditional`: credible only within stated conditions or maturity;
- `borrowed`: mainly supplied by a laboratory, partner, adviser or licensed asset;
- `pseudo`: publicity language without a bounded comparison and proof.

## Use the validation ladder

1. Mechanism: the team can explain what changed.
2. Baseline: the comparison is named and fair.
3. Measurement: metric, unit, condition, sample and repeats are recorded.
4. Use: a real user, field trial or independent party observed the result.
5. Transfer: the result survives beyond the selected example.

State the highest level the evidence reaches. A performance result is not itself an innovation; connect it to the changed mechanism and baseline.

## Red-team the claim

Check for parameter tuning, borrowed IP, cherry-picked baselines, data leakage, missing controls, unsupported superlatives, weak student ownership and a result that cannot be reproduced. Treat `first`, `internationally leading` and similar claims as `missing` until an independent source defines the comparison scope.

## Deliver

Return an innovation matrix, highest validation level, missing controls, owned and borrowed contributions, three defensible claim candidates or explicit empty slots, and confidence with limitations.