arxiv: 2605.14063 · v1 · submitted 2026-05-13 · 💻 cs.LG

Recognition: no theorem link

Reliability-Gated Source Anchoring for Continual Test-Time Adaptation

Vikash Singh , Debargha Ganguly , Weicong Chen , Sabyasachi Sahoo , Sreehari Sankar , Biyao Zhang , Mohsen Harir , Shouren Wang

show 3 more authors

Osama Zafar Christian Gagn\'e Vipin Chaudhary

Authors on Pith no claims yet

Pith reviewed 2026-05-15 05:25 UTC · model grok-4.3

classification 💻 cs.LG

keywords continual test-time adaptationsource anchoringreliability gatingpredictive entropydomain shiftimage classificationreset-based adaptation

0 comments

The pith

RMemSafe uses normalized predictive entropy to gate source anchoring in continual test-time adaptation, disabling unreliable anchors when the source posterior flattens.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that continual test-time adaptation benefits from a dynamic gate that weakens or removes explicit source anchoring once the frozen source model begins to produce high-entropy predictions. This addresses blind anchoring, where fixed-strength anchors continue to pull the model toward a now-unreliable checkpoint even after the source accuracy has collapsed. By attenuating the anchor and agreement filter, the objective falls back to source-agnostic losses plus marginal calibration. Experiments on continual corruption streams demonstrate lower error than fixed-anchoring baselines and a shallower performance drop as source quality degrades.

Core claim

The central claim is that an entropy-driven reliability gate applied to ROID's source-coupled terms produces graceful degradation: when the frozen source posterior approaches uniformity the anchor and filter are removed, leaving only the base losses plus calibration, which yields lower error on eight of nine matched-split continual-corruption cells and a 1.13 times shallower harm slope than ROID plus ASR under controlled source degradation.

What carries the argument

RMemSafe's entropy gate, which scales down the source anchor and agreement filter strength in proportion to the normalized predictive entropy of the frozen source model.

If this is right

On CCC-Hard the gated method plus ASR records the lowest error on eight of nine cells and the best result among all reset-based methods on every cell.
A source-degradation sweep produces a harm slope 1.13 times shallower than the un-gated ROID plus ASR baseline.
When source entropy rises the objective automatically reduces to the source-agnostic fallback of ROID base losses plus marginal calibration.
The gate is shown to detect high-entropy collapse rather than low-entropy confident errors, with that limitation stated and evaluated.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same entropy gate could be ported to other source-anchored CTTA objectives beyond ROID to obtain similar graceful fallback behavior.
Long-running adaptation streams may need multiple complementary reliability signals, since entropy alone misses confident errors.
The approach implies that monitoring source posterior flatness can serve as a lightweight proxy for deciding when to switch from anchored to fully source-free adaptation.

Load-bearing premise

Normalized predictive entropy from the frozen source is a sufficient signal to decide when anchoring should be attenuated, without missing low-entropy but systematically wrong predictions.

What would settle it

A controlled run on a stream where the source model outputs low-entropy yet consistently incorrect labels, measuring whether error rises faster under RMemSafe than under an un-gated baseline.

Figures

Figures reproduced from arXiv: 2605.14063 by Biyao Zhang, Christian Gagn\'e, Debargha Ganguly, Mohsen Harir, Osama Zafar, Sabyasachi Sahoo, Shouren Wang, Sreehari Sankar, Vikash Singh, Vipin Chaudhary, Weicong Chen.

**Figure 1.** Figure 1: Continual test-time adaptation under collapsing source reliability. CTTA anchors the adapter to its frozen source, presuming the source stays a meaningful reference. Blue: on CCC, frozen RN-50 source top-1 collapses from 76% (clean ImageNet) to 1.3% (CCC-Hard). Red dashed: prior reset-based methods (ROID/ETA/EATA+ASR, ROID+RDumb) hold λ=2 across all severities, pulling the adapter toward near-noise output.… view at source ↗

**Figure 2.** Figure 2: Overview of RMEMSAFE. The reliability engine derives the source-reliability gate Rsrc = 1 − Hsrc from the frozen source’s entropy. Rsrc gates all explicit source-coupled uses: the Dynamic Anchor, the agreement filter (interpolating between source agreement and pass-through), and the source-divergence scaling inside the anchor, while leaving the base ROID losses and marginal calibration ungated. At inferenc… view at source ↗

**Figure 3.** Figure 3: Left: paired per-split CCC comparison (n=54); RMEMSAFE is below y=x on 51 splits, on the diagonal (|∆| ≤ 0.02 pp) on 2, and 0.17 pp above on 1 (a CCC-Hard ViT split where both methods are at ∼98% error). Center, right: controlled source-degradation on CIN-C, varying source clean-test accuracy S via Gaussian weight noise (Appendix L). Error bars: ±1 std over 3 seeds; x-axis reversed. The RMEMSAFE+ASR − ROID… view at source ↗

**Figure 4.** Figure 4: Component ablation on CCC ResNet-50 (mean over 27 splits). Two of the five ablated components (anchor, source-expert agreement) are multiplied by Rsrc ≈0.26 at CCC-Hard runtime (App. K); the three ungated components (marg. calibration, confidence-scaled LR, decoupled flip) are not. The decoupled flip is the only contribution with a large leave-one-out effect (+0.79 pp). Gated components interpretation [PI… view at source ↗

**Figure 5.** Figure 5: sweeps each of the five RMEMSAFE hyperparameters independently while holding the other four at their paper values. Each point is the mean error over 9 CCC-Hard ResNet-50 splits (50,000 samples each); CCC-Hard is chosen because it is our most variance-heavy cell and therefore the toughest test of robustness. Every sweep is flat to within 0.21 pp across the full range including 16× changes in λ and α and 100… view at source ↗

**Figure 6.** Figure 6: Source reliability Rsrc (blue, left axis) and gated Jensen–Shannon divergence Rsrc DJS (red, right axis) over a single CCC-Hard ResNet-50 split (split 3, 3,128 test batches). Traces are smoothed using a 25-batch running mean. The reliability stays near a low floor of 0.26 throughout the run rather than collapsing to zero, so the anchor term is scaled down by roughly 3.8× but is not deactivated. This predic… view at source ↗

read the original abstract

Continual test-time adaptation (CTTA) updates a pretrained model online on an unlabeled, non-stationary stream while anchoring it to a frozen source checkpoint. This anchor is useful only when the source remains reliable. On CCC-Hard, however, a ResNet-50 source falls to approximately $1.3\%$ top-$1$ accuracy, while existing source-anchored CTTA methods continue applying the same anchor strength. We call this failure mode blind anchoring and propose RMemSafe, a reliability-gated extension of ROID that uses the frozen source's normalized predictive entropy to attenuate all explicit source-coupled uses in the objective. When the source posterior approaches uniformity, the gate closes: the source anchor and agreement filter vanish, and the objective reduces to a source-agnostic fallback comprising ROID's base losses plus marginal calibration. Combined with ASR, RMemSafe achieves the lowest error on $8$ of $9$ matched-split continual-corruption cells and is the best reset-based method on all $9$, improving ROID+ASR by $1.05$~pp on ResNet-50 and $0.48$~pp on ViT-B/16. A controlled source-degradation sweep shows a $1.13{\times}$ shallower harm slope than ROID+ASR, consistent with the graceful-decay prediction. The entropy gate detects high-entropy source collapse, not confidently wrong low-entropy sources; this scope is explicitly evaluated and discussed.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

RMemSafe adds an entropy gate to ROID that turns off source anchoring when the frozen model collapses to uniform predictions, delivering modest gains on continual corruption streams.

read the letter

The core contribution here is a reliability gate that uses the source model's normalized predictive entropy to dial down all explicit source terms in the objective. When entropy rises, the anchor and agreement filter drop out and the loss falls back to ROID's base terms plus marginal calibration. This directly targets the blind-anchoring failure mode on CCC-Hard, where a ResNet-50 source drops to roughly 1.3% accuracy yet prior methods keep pulling toward it at full strength. The paper reports that the gated version plus ASR beats ROID+ASR by 1.05 points on ResNet-50 and 0.48 on ViT-B/16, wins on eight of nine matched-split cells, and shows a 1.13 times shallower degradation slope in a controlled source-harm sweep. Those numbers are the main empirical payload. The technique itself is a clean, lightweight extension rather than a wholesale redesign of the CTTA pipeline. The fallback objective is defined independently of the target metrics, so there is no obvious circularity. The authors are explicit that the gate only catches high-entropy collapse and does not claim to handle confidently wrong low-entropy predictions; they state this scope was checked. That honesty is useful. The main soft spot is that the abstract and stress-test note leave the gate's activation frequency and isolated ablation impact somewhat opaque. Without seeing the full tables or activation histograms it is hard to judge how often the gate actually fires versus staying open, and whether the reported gains are driven mostly by the fallback or by the gating logic itself. The entropy assumption is reasonable for the corruption setting they study, but it remains an assumption that could be stress-tested further on other non-stationary streams. This is the kind of targeted, engineering-style paper that people working on deployed continual adaptation will want to read and try. It is not a foundational rethinking of test-time methods, but the fix is practical and the experiments are aimed at the right failure case. I would send it to peer review; the central claim is narrow enough to be checked and the method is simple enough to reproduce.

Referee Report

2 major / 1 minor

Summary. The paper proposes RMemSafe, a reliability-gated extension of ROID for continual test-time adaptation. It uses normalized predictive entropy from the frozen source model to attenuate source-coupled terms (anchor and agreement filter) when the source posterior approaches uniformity, reducing the objective to a source-agnostic fallback of ROID base losses plus marginal calibration. On the CCC-Hard continual-corruption benchmark, RMemSafe combined with ASR achieves the lowest error on 8 of 9 matched-split cells and is the best reset-based method on all 9, improving ROID+ASR by 1.05 pp on ResNet-50 and 0.48 pp on ViT-B/16. A controlled source-degradation sweep reports a 1.13× shallower harm slope, consistent with graceful decay. The entropy gate is explicitly scoped to high-entropy collapse and does not claim to handle confidently wrong low-entropy predictions.

Significance. If the empirical claims hold under full verification, the work provides a concrete mechanism to mitigate blind anchoring in source-anchored CTTA, a practical failure mode when pretrained sources degrade under continual distribution shift. The explicit scope limitation, fallback to established losses, and controlled sweep constitute strengths that could inform robust online adaptation designs. The approach is parameter-free in its gating logic and directly testable via the reported degradation slope.

major comments (2)

[Abstract] Abstract: the headline claims of lowest error on 8/9 cells, 1.05 pp improvement on ResNet-50, and 1.13× shallower harm slope are presented without reported standard deviations, number of runs, or statistical significance tests; this information is load-bearing for establishing that the gains are reliable and attributable to the gate rather than run-to-run variance.
[Abstract] Abstract and experimental description: no ablation isolating the entropy gate (e.g., RMemSafe vs. ROID+ASR with gate disabled) or statistics on gate activation frequency across the CCC-Hard stream is described; without these, the causal contribution of the gate to the graceful-decay result cannot be verified and remains a load-bearing gap for the central claim.

minor comments (1)

The abstract is information-dense; splitting the method description and results into separate sentences would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below and will revise the paper accordingly to strengthen the empirical support for our claims.

read point-by-point responses

Referee: [Abstract] Abstract: the headline claims of lowest error on 8/9 cells, 1.05 pp improvement on ResNet-50, and 1.13× shallower harm slope are presented without reported standard deviations, number of runs, or statistical significance tests; this information is load-bearing for establishing that the gains are reliable and attributable to the gate rather than run-to-run variance.

Authors: We agree that reporting standard deviations, the number of runs, and statistical significance tests is necessary to substantiate the headline claims. In the revised manuscript we will add these details to the abstract and the experimental section: all reported improvements will be means over three independent runs with different random seeds, accompanied by standard deviations, and we will include the results of paired t-tests against the relevant baselines to establish statistical significance. revision: yes
Referee: [Abstract] Abstract and experimental description: no ablation isolating the entropy gate (e.g., RMemSafe vs. ROID+ASR with gate disabled) or statistics on gate activation frequency across the CCC-Hard stream is described; without these, the causal contribution of the gate to the graceful-decay result cannot be verified and remains a load-bearing gap for the central claim.

Authors: We acknowledge that an explicit ablation isolating the entropy gate and statistics on its activation frequency are required to confirm its causal contribution. We will add a controlled ablation comparing RMemSafe to ROID+ASR with the gate disabled (i.e., source terms always active) and will report the fraction of timesteps in which the gate activates across the CCC-Hard stream. These results will appear in the experimental section with a concise reference in the abstract. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected; derivation is self-contained.

full rationale

The paper explicitly defines RMemSafe via normalized predictive entropy gating that attenuates source-coupled terms (anchor, agreement filter) when the frozen source posterior approaches uniformity, reducing the objective to ROID base losses plus marginal calibration. All performance claims (lowest error on 8/9 cells, 1.05 pp improvement, 1.13× shallower harm slope) are presented as empirical results on CCC-Hard and controlled degradation sweeps, not as predictions derived from fitted parameters or self-referential definitions. No self-citation chain, ansatz smuggling, or renaming of known results is used to justify the central gating mechanism; the entropy-based reliability signal is stated as an independent design choice whose scope (high-entropy collapse, not low-entropy errors) is explicitly scoped in the text.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only review provides no explicit free parameters, axioms, or invented entities beyond standard CTTA assumptions; the entropy gate is a computational mechanism rather than a new postulated entity.

axioms (1)

domain assumption Normalized predictive entropy from the frozen source model indicates when the source posterior is approaching uniformity and should be down-weighted.
Central to the gating logic described in the abstract.

pith-pipeline@v0.9.0 · 5609 in / 1253 out tokens · 34086 ms · 2026-05-15T05:25:43.615263+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages · 2 internal anchors

[1]

Chen, W., Singh, V ., Rahmani, Z., Ganguly, D., Hariri, M., and Chaudhary, V . (2025). K4: Online log anomaly detection via unsupervised typicality learning.arXiv preprint arXiv:2507.20051

work page internal anchor Pith review Pith/arXiv arXiv 2025
[2]

A., and Yang, B

Döbler, M., Marsden, R. A., and Yang, B. (2023). Robust mean teacher for continual and gradual test-time adaptation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 7704–7714

work page 2023
[3]

Ganguly, D., Iyengar, S., Chaudhary, V ., and Kalyanaraman, S. (2024). PROOF OF THOUGHT : Neurosymbolic program synthesis allows robust and interpretable reasoning. InThe First Workshop on System-2 Reasoning at Scale, NeurIPS’24

work page 2024
[4]

R., Yu, A

Ganguly, D., Morningstar, W. R., Yu, A. S., and Chaudhary, V . (2025a). Forte : Finding outliers with representation typicality estimation. InThe Thirteenth International Conference on Learning Representations

work page
[5]

Ganguly, D., Sankar, S., Zhang, B., Singh, V ., Gupta, K., Kavuru, H., Luo, A., et al. (2026). Trust the typical: An out-of-distribution safety detection framework.arXiv preprint arXiv:2602.04581. ICLR 2026

work page arXiv 2026
[6]

Ganguly, D., Singh, V ., Sankar, S., Zhang, B., Zhang, X., Iyengar, S., Han, X., et al. (2025b). Grammars of formal uncertainty: When to trust LLMs in automated reasoning tasks.arXiv preprint arXiv:2505.20047. NeurIPS 2025

work page arXiv 2025
[7]

Gong, T., Jeong, J., Kim, T., Kim, Y ., Shin, J., and Lee, S.-J. (2022). NOTE: Robust continual test-time adaptation against temporal correlation. In Oh, A. H., Agarwal, A., Belgrave, D., and Cho, K., editors,Advances in Neural Information Processing Systems

work page 2022
[8]

and Dietterich, T

Hendrycks, D. and Dietterich, T. (2019). Benchmarking neural network robustness to common corruptions and perturbations. InInternational Conference on Learning Representations

work page 2019
[9]

M., and Do, M

Hoang, T.-H., V o, D. M., and Do, M. N. (2024). Persistent test-time adaptation in recurring testing scenarios. InAdvances in Neural Information Processing Systems (NeurIPS)

work page 2024
[10]

and Matsuo, Y

Iwasawa, Y . and Matsuo, Y . (2021). Test-time classifier adjustment module for model-agnostic domain generalization. InAdvances in Neural Information Processing Systems (NeurIPS)

work page 2021
[11]

Lee, J., Jung, D., Lee, S., Park, J., Shin, J., Hwang, U., and Yoon, S. (2024). Entropy is not enough for test-time adaptation: From the perspective of disentangled factors. InInternational Conference on Learning Representations (ICLR). Spotlight (top 5%)

work page 2024
[12]

and Chang, J.-H

Lee, J.-H. and Chang, J.-H. (2024). Continual momentum filtering on parameter space for online test-time adaptation. InThe Twelfth International Conference on Learning Representations

work page 2024
[13]

Liang, J., He, R., and Tan, T. (2023). A comprehensive survey on test-time adaptation under distribution shifts.International Journal of Computer Vision

work page 2023
[14]

Lim, T., Hwang, J.-W., and Lee, K. (2026). When and where to reset matters for long-term test-time adaptation.arXiv preprint arXiv:2603.03796

work page arXiv 2026
[15]

Liu, Y ., Kothari, P., van Delft, B., Bellot-Gurlet, B., Mordan, T., and Alahi, A. (2021). TTT++: When does self-supervised test-time training fail or thrive? InAdvances in Neural Information Processing Systems (NeurIPS). 10

work page 2021
[16]

A., Döbler, M., and Yang, B

Marsden, R. A., Döbler, M., and Yang, B. (2024). Universal test-time adaptation through weight ensembling, diversity weighting, and prior correction. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 2554–2564

work page 2024
[17]

Mishra, H. (2026). Rdumb++: Drift-aware continual test-time adaptation

work page 2026
[18]

K., Hutmacher, R., Rambach, K., Levinkov, E., Brox, T., and Metzen, J

Mummadi, C. K., Hutmacher, R., Rambach, K., Levinkov, E., Brox, T., and Metzen, J. H. (2021). Test-time adaptation to distribution shift by confidence maximization and input transformation. arXiv preprint arXiv:2106.14999

work page arXiv 2021
[19]

Niu, S., Wu, J., Zhang, Y ., Chen, Y ., Zheng, S., Zhao, P., and Tan, M. (2022). Efficient test-time model adaptation without forgetting. InInternational Conference on Machine Learning, pages 16888–16905. PMLR

work page 2022
[20]

Niu, S., Wu, J., Zhang, Y ., Wen, Z., Chen, Y ., Zhao, P., and Tan, M. (2023). Towards stable test- time adaptation in dynamic wild world. InInternational Conference on Learning Representations

work page 2023
[21]

H., and Dokania, P

Prabhu, A., Torr, P. H., and Dokania, P. K. (2020). GDumb: A simple approach that questions our progress in continual learning. InEuropean Conference on Computer Vision (ECCV), pages 524–540

work page 2020
[22]

Press, O., Schneider, S., Kümmerer, M., and Bethge, M. (2023). RDumb: A simple approach that questions our progress in continual test-time adaptation. InAdvances in Neural Information Processing Systems, volume 36, pages 39915–39935

work page 2023
[23]

V ., Bringmann, O., Brendel, W., and Bethge, M

Rusak, E., Schneider, S., Pachitariu, G., Eck, L., Gehler, P. V ., Bringmann, O., Brendel, W., and Bethge, M. (2022). If your data distribution shifts, use self-learning.Transactions on Machine Learning Research (TMLR)

work page 2022
[24]

Schneider, S., Rusak, E., Eck, L., Bringmann, O., Brendel, W., and Bethge, M. (2020). Improv- ing robustness against common corruptions by covariate shift adaptation. InAdvances in Neural Information Processing Systems (NeurIPS)

work page 2020
[25]

Singh, V ., Cassel, D., Weir, N., Feng, N., and Bayless, S. (2026a). VERGE: Formal refinement and guidance engine for verifiable LLM reasoning.arXiv preprint arXiv:2601.20055

work page internal anchor Pith review Pith/arXiv arXiv
[26]

Singh, V ., Ganguly, D., Yu, H., Zhou, C., Singh, P., Lee, B., Chaudhary, V ., and Datta, G. (2026b). Toward guarantees for clinical reasoning in vision language models via formal verification.arXiv preprint arXiv:2602.24111

work page arXiv
[27]

S., and Choi, S

Song, J., Lee, J., Kweon, I. S., and Choi, S. (2023). EcoTTA: Memory-efficient continual test-time adaptation via self-distilled regularization. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

work page 2023
[28]

A., and Hardt, M

Sun, Y ., Wang, X., Liu, Z., Miller, J., Efros, A. A., and Hardt, M. (2020). Test-time training with self-supervision for generalization under distribution shifts. InProceedings of the 37th International Conference on Machine Learning (ICML)

work page 2020
[29]

Tzeng, E., Hoffman, J., Saenko, K., and Darrell, T. (2017). Adversarial discriminative domain adaptation. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 7167–7176

work page 2017
[30]

Wang, D., Shelhamer, E., Liu, S., Olshausen, B., and Darrell, T. (2021). Tent: Fully test-time adaptation by entropy minimization. InInternational Conference on Learning Representations

work page 2021
[31]

Wang, N., Liang, T., Singh, V ., Song, C., Yang, V ., Yin, Y ., Ma, J., Singh, J., et al. (2026a). HugRAG: Hierarchical causal knowledge graph design for RAG.arXiv preprint arXiv:2602.05143

work page arXiv
[32]

Wang, Q., Fink, O., Van Gool, L., and Dai, D. (2022). Continual test-time domain adaptation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7201–7211

work page 2022
[33]

Wang, S., Yang, W., Ma, C., Ganguly, D., Singh, V ., Song, C., Li, X., Long, X., Chaudhary, V ., and Han, X. (2026b). Path-lock expert: Separating reasoning mode in hybrid thinking via architecture-level separation. 11

work page
[34]

Yang, W., Ganguly, D., Li, X., Song, C., Wang, S., Singh, V ., Chaudhary, V ., and Han, X. (2026). Mid-Think: Training-free intermediate-budget reasoning via token-level triggers.arXiv preprint arXiv:2601.07036

work page arXiv 2026
[35]

Yuan, L., Xie, B., and Li, S. (2023). Robust test-time adaptation in dynamic scenarios. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 15922–15932

work page 2023
[36]

Zhang, M., Levine, S., and Finn, C. (2022). MEMO: Test time robustness via adaptation and augmentation. InAdvances in Neural Information Processing Systems (NeurIPS)

work page 2022
[37]

does the gate help?

Zhang, Q., Bian, Y ., Kong, X., Zhao, P., and Zhang, C. (2025). COME: Test-time adaption by conservatively minimizing entropy. InInternational Conference on Learning Representations (ICLR). A Algorithm Algorithm 1 gives the full per-batch update rule of RMEMSAFE+ASR. The method combines the ROID backbone (soft-likelihood-ratio loss, diversity weighting, a...

work page arXiv 2025
[38]

The offset is approximately constant (∼7pp) across reset-based methods

(where available). The offset is approximately constant (∼7pp) across reset-based methods. Method Local Streamed Offset ROID86.07∼79 +7 ROID+RDumb86.77∼80 +7 ETA+ASR89.74∼83 +7 EATA+ASR88.89∼84 +5 ROID+ASR84.56 77.79 +6.8 RMEMSAFE+ASR (ours)83.81− − R Broader Impact and Limitations RMEMSAFEis designed forsafetyin continual test-time adaptation: it aims to...

work page
[39]

ViT-B/16 under a class-permuted source is the empirical witness (∆ = +1.14 pp, App

Scope of the reliability signal.Entropy-only; does not detect confidently miscalibrated sources. ViT-B/16 under a class-permuted source is the empirical witness (∆ = +1.14 pp, App. P)

work page
[40]

The reliability-gated reset trigger ( τgate = 0.40, App

Reset-paradigm failure on CCC-Hard ViT-B/16.Every reset-based method we evaluate underperforms non-reset ROID on this cell, across base adapters and reset mechanisms (§4.3). The reliability-gated reset trigger ( τgate = 0.40, App. O) recovers the non-reset mean but not per-split variance

work page
[41]

[14]; the offset is approximately constant across methods on ResNet-50

Local-data offset on CCC.Our shards yield CCC-Hard numbers ∼7 pp harder than the streamed numbers of Lim et al. [14]; the offset is approximately constant across methods on ResNet-50. Cross-study absolute comparisons on CCC-Hard should be interpreted with caution; the matched-split head-to-head is the unbiased estimator of relative method quality. 24

work page
[42]

Per-cell tuning would likely yield further small gains but is discouraged in the unlabeled test-time setting

Fixed hyperparameters.The five core hyperparameters are held constant across all nine benchmark cells. Per-cell tuning would likely yield further small gains but is discouraged in the unlabeled test-time setting

work page
[43]

Marginal-calibration EMA under abrupt label shift.The EMA prior ( ρ= 0.01 ) lags abrupt label-distribution shifts; our streams exhibit gradual rather than abrupt shift, so this regime is not exercised. 25

work page