- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper investigates ontic–epistemic reasoning across video revisits, distinguishing physical state transitions from conclusions justified by incomplete visual evidence. The authors construct structured VisitPacks and RevisitPacks containing evidence-complete, evidence-withheld, and nuisance-modified observations. Building on the existing Behavior Pack Optimization (BPO) framework, they propose Bottleneck Behavior Pack Optimization (BBPO), which reallocates anchor-relative advantages toward unsatisfied behavioral relations.
Critical strengths
- The distinction between what physically happened and what can be inferred from available visual evidence is meaningful and offers an interesting perspective on persistent video reasoning. -The proposed VisitPacks and RevisitPacks provide a systematic way to examine transition recognition, evidence sufficiency, and invariance to irrelevant visual changes.
- The paper compares BPO and BBPO using matched training settings and reports results across multiple seeds.
Critical weaknesses
- The central optimization framework is directly inherited from Behavior Pack Optimization for Video MLLM Post-Training (NeurIPS 2026), including behavior packs, counterfactual interventions, behavioral contracts, and anchor-relative advantages. The proposed BBPO primarily introduces normalized failure-aware weighting of existing BPO advantages. This modification closely resembles established hard-example mining and difficulty-aware loss reweighting strategies. Although the ontic–epistemic revisit formulation is interesting, the paper does not sufficiently establish that its optimization mechanism constitutes a substantial methodological advance over BPO or standard reweighting approaches.
- The evaluation focuses on seven predefined binary object-state properties and structured multiple-choice questions. This restricted setting may allow models to exploit recurring question templates, object-state priors, or intervention-specific cues instead of performing genuine ontic–epistemic reasoning.
- The experiments use only 270 simulated scenes, with 216/27/27 scenes allocated to training, validation, and testing.
- The experiments use only Qwen3.5-2B and Qwen3.5-4B, which belong to the same model family and are relatively small backbones.
- The main comparisons focus on SFT and standard BPO, without evaluating whether the proposed gain can be reproduced using simpler failure-aware weighting or established post-training objectives.
- The propositions primarily establish straightforward properties of normalized weighting, including mean preservation and greater weights for failed relations. They do not provide guarantees concerning optimization convergence, generalization, or improved ontic–epistemic reasoning.
Questions for authors
- Given that BPO already introduces behavior packs, intervention-based rewards, and anchor-relative optimization, what is the fundamental algorithmic novelty of BBPO beyond failure-aware advantage reweighting?
- Why are established post-training alternatives such as GRPO, difficulty-weighted policy optimization, and standard supervised objectives not included under matched training conditions?
- Could the authors extend the evaluation to larger and architecturally different VLMs, such as 7B/8B or larger backbones, and independent real-world video datasets?
2 comments
Response 2 to Reviewer qH5j
3. Seven binary states and shortcuts
Our method supports seven binary property types, not seven object identities. The M3R-5 experiments actually span four property types, 13 object categories, and 152 persistent scene-object instances. We agree this is bounded, structured multiple-choice reasoning, not arbitrary open-world state understanding. However, the proposed shortcut explanation is testable: all intervention views within a pack share exactly the same question and option order but require different correct answers. Under deterministic inference a text-only/template-only constant answer cannot complete such a pack.
To investigate whether the model can solve the task using question templates or answer priors alone, we introduce a blank-video control that preserves the question, answer-option order, and number of input frames while replacing all visual frames with zeros. Under this intervention, BBPO's Overall AV decreases from 66.3% to 0%, indicating that the evaluated model cannot satisfy the complete behavioral contracts using the retained nonvisual information alone. Together with the natural temporal-withholding ablation (Table 7), the experiment addresses two important shortcut mechanisms.
| Settings | Overall AV | Revisit AV | Visit AV | Revisit ID | Visit ID | Transition | Nuisance | Missing |
|---|---|---|---|---|---|---|---|---|
| Normal BBPO | 66.3±0.5 | 57.5±3.3 | 70.7±1.4 | 89.7±2.8 | 82.2±0.8 | 85.6±1.6 | 88.5±0.8 | 91.4±1.9 |
| Masking cue (Table 7) | 41.4±2.5 | 12.6±3.3 | 55.7±3.5 | 71.3±1.6 | 82.2±0.8 | 75.9±2.4 | 71.3±3.3 | 55.7±5.9 |
| Blank video | 0.0±0.0 | 0.0±0.0 | 0.0±0.0 | 0.0±0.0 | 3.4±2.4 | 0.0±0.0 | 0.0±0.0 | 50.0±0.0 |
| Normal − blank | +66.3 pp | +57.5 pp | +70.7 pp | +89.7 pp | +78.7 pp | +85.6 pp | +88.5 pp | +41.4 pp |
4. Scope of propositions, failure-density analysis
We agree that mean preservation and relative weighting do not prove convergence or generalization. Instead, these propositions characterize how BBPO redistributes optimization credit among intervention relations within a behavior pack. In particular, Proposition A.2 describes how the relative emphasis on failed relations depends on their prevalence within the pack, yielding a testable prediction about the relationship between failure density and bottleneck-aware optimization.
A relation fails unless the anchor answer, intervention answer, and required change/invariance relation are all correct. We report: (i) All exposures: mean number of failed non-anchor relations over every behavior pack encountered during training, including packs whose identity/anchor answer was wrong. (ii) Anchor-gated: the same mean, but only for training exposures where the anchor answer was correct and the BPO/BBPO policy loss was therefore active.
Online Training table summarizes failed non-anchor relations in online training exposures.
| Method | All exposures ↓ | Anchor-gated ↓ |
|---|---|---|
| BPO, 2B | 1.596±0.009 | 0.970±0.026 |
| BBPO, 2B | 1.597±0.039 | 0.945±0.059 |
| BPO, 9B-v2 | 1.469±0.087 | 1.153±0.085 |
| BBPO, 9B-v2 | 1.447±0.074 | 1.117±0.075 |
The failure-density analyses below provide additional insight into where BBPO improves Overall AV correctness on the 87 evaluation packs. We report: (i) Pack-seed pairs: one behavior pack evaluated by one trained seed. With 87 packs and 3 seeds, there are pack-seed outcomes. The same physical pack appears once per seed. (ii) BPO AV: the percentage of those outcomes where BPO answers every view in the pack correctly. For a row with one or more BPO failed relations, this is necessarily 0%.
For the 2B backbone, BBPO repairs 12/42 one-failure outcomes and 3/53 outcomes with two or more failures; it loses 8 previously complete outcomes.
| BPO failed relations | Pack-seed pairs | BPO AV | BBPO AV | AV delta | Mean failed-relation reduction |
|---|---|---|---|---|---|
| 0 | 166 | 100.0% | 95.2% | -4.8 pp | -0.102 |
| 1 | 42 | 0.0% | 28.6% | +28.6 pp | +0.262 |
| 2 | 9 | 0.0% | 11.1% | +11.1 pp | +0.222 |
| 3+ | 44 | 0.0% | 4.5% | +4.5 pp | +0.318 |
For the 9B backbone, BBPO repairs 4/36 one-failure outcomes and 2/43 outcomes with two or more failures; it loses 1 previously complete outcome. Again, results using 9B backbone are not tuned by the validation set. For the group containing three or more initial failures, BBPO exhibits an average reduction of 0.5 failed relations per pack-seed outcome.
| BPO failed relations | Pack-seed pairs | BPO AV | BBPO AV | AV delta | Mean failed-relation reduction (per pack) |
|---|---|---|---|---|---|
| 0 | 182 | 100.0% | 99.5% | -0.5 pp | -0.005 |
| 1 | 36 | 0.0% | 11.1% | +11.1 pp | -0.194 |
| 2 | 13 | 0.0% | 7.7% | +7.7 pp | +0.231 |
| 3+ | 30 | 0.0% | 3.3% | +3.3 pp | +0.500 |
Response 1 to Reviewer qH5j
Thank you for emphasizing the scientific distinction between physical state changes and what visual evidence can justify.
1. What is new beyond BPO and generic hard-example mining?
We explicitly build on BPO's single-video behavior packs, behavioral contracts, and anchor-relative advantage. We jointly introduce (1) an ontic–epistemic paired-revisit task that holds the world's physical transition fixed while varying the availability of premises across Visit 1 and Visit 2, (2) five-view Visit and eight-view Revisit Packs built from natural temporal withholding and nuisance controls, with a query-conditioned target-state representation for two visits (Appendix B), and (3) intervention-relation-defined, pack-local bottleneck weighting on top of adapted BPO. Unlike an ordinary answer-level difficulty indicator, BBPO defines its success signal using the anchor, the intervened answer, and the required cross-view relation.
2. Matched alternatives, more models
We introduce three additional comparisons. First, we implement difficulty-weighted BPO, which scales anchor-relative advantages according to individual answer correctness and normalizes the weights over the active global batch: A~iv=∣Bopen∣1∑(j,u)∈Bopen[1+α(1−cju)]1+α(1−civ)Aiv,α=1, where civ∈{0,1} is answer correctness and Bopen contains all active, anchor-gated view-rollouts in the distributed global batch. This preserves mean weight one globally. Second, we use direct supervised post-training using answer-token cross-entropy (CE) on the ground-truth response for every intervention view. Third, we compare GRPO and BBPO with four rollouts per view, using matched training packs, initialization, and optimizer steps.
The additional baselines demonstrate that BBPO outperforms the tested answer-error-weighted BPO objective and achieves higher Overall AV accuracy than GRPO under the matched four-rollout setting. Direct CE supervision is also highly competitive. The principal difference lies in the optimization formulation: direct CE explicitly trains the model toward the ground-truth answer for each intervention view, whereas BBPO updates sampled responses using verifier-derived rewards and pack-local weighting. Nevertheless, BBPO provides a reward-based mechanism for prioritizing poorly satisfied behaviors without directly teacher-forcing the corresponding answer tokens, which could be feasible when correct intervention labels are unavailable but a verifier can score behavior.
Furthermore, we evaluate with larger models as below. Please see more details in Response 1 to Reviewer wi5Q (part 1) and Response 1 to Reviewer VBVy (part 1).