- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper presents Innovation Prior, which reconstructs scientific discovery trace from multi-source materials to capture problem analysis and abandoned design choices, instead of only learning final methods from finished paper. Based on a dataset of 1,185 discovery units and 164 research evolution chains, the authors SFT Qwen3.5-4B and 9B then apply GRPO with 1,300 extra program optimization tasks. Experiments on four benchmarks show Innovation Prior improves RL model performance at both scales, with larger gains for the 4B model.
Critical strengths
1.The paper proposes a concrete, verifiable method for constructing data, using author accounts, the original paper, and review responses as supplemental data to show design backgrounds, alternative plans, and thought processes that the final paper can't display. 2.The paper designs a comprehensive experiments, and directly tests the combined effects of SFT and RL. The results found SFT alone doesn't consistently improve performance, but when combined with RL, the overall results are better than just RL, showing that learning the reconstruction of discovery trace can help subsequent RL.
Critical weaknesses
1.Whether the gains stem from the specific innovation prior, other fine-tuning effects, or RL stability differences remains questionable. The current ablation keeps the C++ data, which helps compare parts of the Innovation Prior, but it still can't completely isolate the source of gains during the whole fine-tuning phase. 2.The claim that reconstructed discovery traces reflect scientists’ reasoning processes requires stronger evidence. It remains to be validated that the resulting gains indicate improved scientific reasoning.
Questions for authors
1.Result stability: The work lacks multiple independent training replicates, task-wise variance and confidence intervals. It remains unclear whether the 4B direct RL degradation is persistent and whether the marginal gains at 9B exceed random fluctuations. 2.Gain attribution: The performance brought solely by C++ data and generic reasoning data with matched token budget remains unreported. It is unclear whether the ablations match the total number of training tokens and the distribution of example lengths. 3.Discovery traces authenticity: Independent human sampling audit is absent. The proportion of genuine author-tried paths versus post-hoc rationalization and reconstructed derivations within key steps is unknown.
5 comments
Response to Reviewer TLkY (2/3)
Result Stability
"...lacks multiple independent training replicates, task-wise variance and confidence intervals..."
"...whether the 4B direct RL degradation is persistent..."
Thank you for raising this; we agree that the persistence of the 4B degradation needed to be established, and we have now done so. We focused the reruns on the 4B Base + RL baseline. It trains normally up to step 15, as the table below shows, where scored counts rollouts with a nonzero score, Inside counts rollouts that end inside the reasoning before any answer, At limit counts rollouts that reach the length limit, Length is in tokens, and the last column gives the entropy of RL from the prior.
| Step | Reward | Scored | Inside | Length | At limit | Entropy | KL | Grad norm | Entropy, prior |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 6.1 | 23% | 0% | 15.5K | 5% | 0.74 | 0.24 | 0.86 | 0.75 |
| 2 | 8.8 | 31% | 0% | 16.0K | 4% | 0.73 | 0.25 | 0.66 | 0.74 |
| 3 | 7.5 | 23% | 0% | 19.7K | 12% | 0.74 | 0.22 | 0.37 | 0.76 |
| 4 | 8.9 | 30% | 0% | 19.2K | 18% | 0.65 | 0.21 | 0.43 | 0.63 |
| 5 | 8.2 | 27% | 1% | 18.9K | 21% | 0.62 | 0.19 | 0.33 | 0.61 |
| 6 | 6.3 | 25% | 5% | 22.5K | 48% | 0.43 | 0.13 | 0.46 | 0.56 |
| 7 | 3.9 | 16% | 4% | 25.4K | 61% | 0.36 | 0.13 | 0.39 | 0.50 |
| 8 | 5.7 | 16% | 7% | 25.7K | 70% | 0.24 | 0.08 | 0.31 | 0.41 |
| 9 | 6.0 | 22% | 4% | 22.4K | 49% | 0.34 | 0.12 | 0.25 | 0.50 |
| 10 | 9.3 | 26% | 3% | 20.4K | 44% | 0.35 | 0.14 | 0.24 | 0.46 |
| 11 | 6.4 | 26% | 2% | 20.0K | 40% | 0.35 | 0.13 | 0.21 | 0.43 |
| 12 | 5.9 | 24% | 1% | 19.3K | 37% | 0.35 | 0.14 | 0.24 | 0.50 |
| 13 | 8.5 | 32% | 1% | 17.0K | 29% | 0.35 | 0.16 | 0.21 | 0.37 |
| 14 | 9.7 | 35% | 1% | 16.2K | 28% | 0.32 | 0.17 | 0.26 | 0.48 |
| 15 | 8.4 | 32% | 2% | 16.5K | 30% | 0.30 | 0.17 | 0.27 | 0.46 |
From the step-15 checkpoint we ran steps 16 to 20 three times, and all three runs collapse in the same way, as the table below shows with the median rollout length in characters. The degradation is therefore persistent: under this recipe, RL from the 4B base cannot be trained stably to step 20. We have added this to Section 3.2 of the revised paper.
| Run | Step | Inside reasoning | Length | Zero reward | Mean reward |
|---|---|---|---|---|---|
| 1 | 16 / 17 / 18 / 19 / 20 | 2% / 5% / 30% / 58% / 86% | 35.7K / 33.5K / 26.7K / 18.6K / 9.8K | 66% / 62% / 60% / 67% / 81% | 6.9 / 8.9 / 11.2 / 8.3 / 3.4 |
| 2 | 16 / 17 / 18 / 19 / 20 | 2% / 5% / 31% / 68% / 93% | 35.8K / 34.1K / 29.2K / 18.7K / 9.0K | 67% / 60% / 56% / 73% / 90% | 7.5 / 10.8 / 13.1 / 7.7 / 2.4 |
| 3 | 16 / 17 / 18 / 19 / 20 | 2% / 7% / 51% / 62% / 87% | 38.1K / 33.0K / 27.1K / 17.5K / 9.8K | 65% / 62% / 71% / 70% / 83% | 8.2 / 11.9 / 7.5 / 7.5 / 3.8 |
"...whether the marginal gains at 9B exceed random fluctuations..."
We would like to note that the 9B comparison is not a single noisy number. Each FrontierCS and ALE-Bench score already pools five samples per problem, and the 9B rows of the table in our previous comment show RL from the prior above RL from the base on every metric of all four benchmarks, which is not the pattern of a random fluctuation. Figure 6 further shows that it keeps the lead over many rounds of test-time feedback. Section 3.2 of the revised paper now states this consistency directly.
Response to Reviewer TLkY (3/3)
Faithfulness of the Traces and Scientific Reasoning
"...The claim that reconstructed discovery traces reflect scientists' reasoning processes requires stronger evidence..."
"...Independent human sampling audit is absent. The proportion of genuine author-tried paths versus post-hoc rationalization...is unknown..."
Thank you for raising this important point. The faithfulness of the traces is central to our data, and we agree that it deserves a direct answer. We control it during construction and audit it afterward, unit by unit.
During construction, a fallback enters a trace only if the records show that the authors considered or tried it (Lines 209–210), a trace may not report the outcome of an experiment the narrator claims to run, and an adversarial verifier checks the key step of every trace against the sources (Appendix D). The agent also records in notes which sources it collected and which steps each of them supports.
After construction, a separate audit checks every trace against these notes and the source files rather than trusting the construction agent: a source counts as used only if the auditor can point to the step that depends on it, and labels of low confidence get a second reading. The table below gives, for each kind of source, the number of discoveries whose trace uses it under this audit, and we have added this procedure to Section 2.1 of the revised paper.
| Kind of source | Ancestor works | Code | Third-party explanations | Self-accounts |
|---|---|---|---|---|
| Discoveries whose trace uses it, of 1,185 | 871 | 546 | 340 | 221 |
The data then went through several more rounds in which agents re-audited the units under the same rules and improved those that fell short, with an adversarial verifier or an automatic check confirming each change. All discoveries and ladders went through these rounds. The authors also inspected samples of the labels and audit results by hand.
The table below counts the problems found in the first version of the data and in the final version. A trace leaks hindsight when it mentions "the paper", "the authors" or an arXiv id, since the scientist reasoning at the time could not refer to the finished paper. Ladders chain several methods, so their steps are counted separately from the discoveries.
| Problem | Where | First version | Final version |
|---|---|---|---|
| Leaks hindsight from the finished paper | Discoveries | 82 of 1,184 | 2 |
| Leaks hindsight from the finished paper | Reasoning of ladder steps | 48 of 641 | 0 |
| Leaks hindsight from the finished paper | Answers of ladder steps | 65 of 641 | 0 |
| Answer code is not the original implementation | Discoveries | 595 of 1,172 | 6 |
| Non-English text in the reasoning | Discoveries | 9 of 1,184 | 0 |
The paper also gives examples that can be checked against the sources: Appendix B quotes, for each kind of source, records that hold reasoning absent from the paper, and Figure 5 and Appendix C tie the steps of full traces to the sources they come from. Finally, as an indirect test, rebuilding each trace from the paper alone, where post-hoc rationalization is most likely, lowers every metric, as the ablation table in our first comment shows.
"...It remains to be validated that the resulting gains indicate improved scientific reasoning..."
Thank you for this point. We agree that scores alone cannot show this, so Section 4 tests it in three further ways. The model judges research better although it was never trained to judge: on four tasks from three third-party sources, such as picking the equation the authors wrote or the paper later cited more, RL from the prior is ahead at 4B (Section 4.1). It weighs more methods before committing to one and sets most of them aside (Section 4.2). And human experts find that it reaches simple methods from first principles rather than by retuning known ones (Section 4.3). We have revised the opening of Section 4 to state this purpose more clearly.
Thank you again for your time and effort. We hope our response has addressed your concerns. If the issues have been resolved, we would appreciate it if you could reflect this in your evaluation. We welcome any further comments.
Response to Reviewer TLkY (1/3)
Thank you for your thoughtful review and for recognizing our concrete, verifiable construction method and the comprehensive experiments that test SFT and RL together. We address your concerns below.
Gain Attribution
Thank you for raising this important point. We ran a new experiment, Maintain-only. It fine-tunes the model on only the 338 C++ problems that Innovation Prior includes to maintain coding ability, without any discovery traces, and then runs the same RL; the SFT recipe, merge weight, RL recipe and evaluation are all unchanged. Maintain-only thus shares the fine-tuning step and the C++ data with InnoPrior and differs from it only in the discovery traces.
Before RL, Maintain-only is not consistently stronger than Base, so the C++ problems add little by themselves. After RL, InnoPrior + RL is above Maintain-only + RL on every metric at 4B and on all but one at 9B. Since Maintain-only + RL also trains stably, the gain comes from the discovery traces rather than from other fine-tuning effects or RL stability. We have added Maintain-only to Table 1 and Sections 3.1 and 3.2 of the revised paper.
Thank you for this observation. We agree, and Maintain-only closes exactly this gap: it keeps the C++ data and the whole fine-tuning phase and removes all the traces, with the results shown above.
Thank you for this question. The performance of the C++ data alone is Maintain-only above. For generic reasoning data, our ablations are the matched comparison: they reason about the same methods in the same form, with the sources beyond the paper or the exploration removed, and the table below shows that removing either lowers every metric.
Thank you for the careful question about matching. Each ablation rebuilds the same units, so it has the same number of training examples as the full data, and both ablations are trained for the same number of optimizer steps. We have stated this explicitly in Section 3.3 of the revised paper.