url	short_title	label	code	confidence	rationale	evidence
https://arxiv.org/abs/2601.19944	Classifier Calibration at Scale	wheat	P4	high	Empirical study of post-hoc calibration incl Venn-Abers using proper scores; honest no-method-dominates	"distribution-free validity under exchangeability"; "no method dominates uniformly"
https://pubs.acs.org/doi/10.1021/ci5001168	CP as Applicability Domain Alternative	wheat	P1	med	CP used to define applicability domain with honest confidence-level interpretation	"confidence level 0.8 means at most 20% errors"
https://arxiv.org/abs/2007.10438	Conditional calibration FDR under dependence	wheat	P4	high	Finite-sample FDR control via conditional calibration under dependence; correct theory	"finite-sample false discovery rate control with dependent test statistics"
https://link.springer.com/chapter/10.1007/978-3-662-44722-2_29	Trajectory Anomaly Detection via CP	wheat	P3	med	Conformal anomaly detection with p-values and avg-p efficiency criterion; distribution-free test	"kernel density as non-conformity measure; average p-value as efficiency"
https://link.springer.com/chapter/10.1007/978-3-662-44722-2_25	Aggregated Conformal Prediction	wheat	P4	med	Ensemble CP for efficiency; community acknowledges validity is empirical not fully theoretical	"aggregated conformal predictors improve efficiency; validity empirical"
https://arxiv.org/abs/2001.09225	Strong validity consonance and CP	wheat	P4	high	Theory linking CP to consonant plausibility/imprecise probability; honest validity proofs	"Type-2 validity via consonant plausibility contour function"
https://arxiv.org/pdf/2010.07147.pdf	Distribution-Free Test of Covariate Shift via CP	wheat	P3	high	Conformal p-values for distribution-free covariate-shift test with valid Type-I control	"valid inference for conditional distributional testing without model specification"
http://proceedings.mlr.press/v105/morsomme19a/morsomme19a.pdf	CP for Students Grades Recommender	wheat	P1	med	ICP intervals for grade prediction; reports valid intervals with width as reliability	"prediction intervals constructed with ICM are valid; width gives reliability"
https://papers.nips.cc/paper/2021/hash/46c7cb50b373877fb2f8d5c4517bb969-Abstract.html	Locally Valid Discriminative Intervals	borderline	A4	med	Claims finite-sample local coverage guarantees; relaxed local notion but framed as conditional validity	"finite-sample local coverage guarantees (contrasted to marginal)"
https://arxiv.org/pdf/2201.12006.pdf	Improving Expert Predictions with CP	wheat	P5	high	Prediction sets as decision-support deliverable to improve expert performance; honest tradeoff	"provides sets via CP; trade-off miscoverage with set size to help experts"
https://arxiv.org/pdf/1912.06116.pdf	E-values calibration combination applications	wheat	P4	high	Theory of e-values, calibration and combination; correct by construction	"calibration and combination of e-values and applications"
https://proceedings.mlr.press/v139/xu21h.html	Conformal interval for dynamic time-series (EnbPI)	wheat	P4	high	Beyond-exchangeability method for time-series; approx marginal coverage, exchangeability dropped honestly	"does not require data exchangeability; finite-sample approximately valid marginal coverage"
https://arxiv.org/pdf/2207.01609.pdf	Recommendation systems with distribution-free reliability guarantees	wheat	P1	Set-of-items is the deliverable with finite-sample FDR control, distribution-free, honest about guarantee	"return a set of items rigorously guaranteed to contain mostly good items"; "finite-sample control of the false discovery rate regardless of the unknown data distribution"
https://arxiv.org/pdf/2206.06584.pdf	Probabilistic Conformal Prediction Using Conditional Random Samples	wheat	P4	Methodology paper; new CP variant with proven finite-sample marginal coverage, claims sharper sets not quality	"PCP guarantees correct marginal coverage with finite samples"; "provides sharper predictive sets"
https://arxiv.org/pdf/2208.08401.pdf	Conformal Inference for Online Prediction with Arbitrary Distribution Shifts	wheat	P4	Beyond-exchangeability theory; adaptive CP for online shift with provable regret bounds, honest framing	"adaptive to both size and type of distribution shift"; "provably small regret over all local time intervals"
https://www.sciencedirect.com/science/article/pii/S0925231219316170	Audio-visual domain adaptation using conditional semi-supervised GANs	unclear	NA	Paper does not use conformal prediction at all; GAN-based cross-modal emotion-recognition domain adaptation	"semi-supervised adversarial network for knowledge transfer from labeled videos to audio domain"; no CP content
https://arxiv.org/pdf/2210.04166.pdf	Test-time recalibration of conformal predictors under distribution shift	wheat	P6	Explicitly caveats that unlabeled recalibration cannot guarantee reliability in general; honest UQ layer	"impossible in general to guarantee reliability when calibrating based on unlabeled examples"; "provably works for a specific model of distribution shift"
https://arxiv.org/pdf/2210.10161.pdf	Nonparametric Quantile Regression: Non-Crossing Constraints and Conformal Prediction	wheat	P4	Theory/methodology; CP intervals adaptive to heterogeneity with derived validity/length bounds vs oracle	"conformal prediction intervals fully adaptive to heterogeneity"; "good properties in terms of validity and accuracy"
https://openreview.net/pdf?id=Koug1i2HpH	Engineering Uncertainty Representations to Monitor Distribution Shifts	borderline	P3	Uses CP non-conformity to flag uncertain zones for shift monitoring; no formal coverage claims, loose framing	"Non-Conformity Analysis exploits results of conformal prediction... to display uncertain zones"; empirically validated, no guarantee claims
https://arxiv.org/pdf/2301.11136.pdf	Conformal Prediction for Trustworthy Detection of Railway Signals	wheat	P2	Safety/containment application; conformalized bounding boxes meet predefined probability of success, honest	"adjust predicted bounding boxes to comply with a predefined probability of success"; safety-critical railway UQ
https://arxiv.org/pdf/2211.01539.pdf	Conformal prediction for STL runtime verification	wheat	P2	Safety verification; distribution-free prediction regions for CPS failure with formal confidence guarantees	"first formal guarantees for a predictive runtime verification algorithm"; "without distributional assumptions"
https://arxiv.org/abs/2302.04019	Fortuna: A Library for Uncertainty Quantification in Deep Learning	borderline	A2	Software library; markets CP as producing "reliable uncertainty estimates" on any net without conditional caveat	"conformal prediction applied to any trained neural network to generate reliable uncertainty estimates"
https://proceedings.mlr.press/v152/bostrom21a/bostrom21a.pdf	Mondrian Conformal Predictive Distributions	wheat	P4	Methodology; conformal predictive distributions (full distribution output), evaluated by CRPS and interval tightness	"forms distributions from multiple Mondrian categories"; "significantly outperforms... with respect to the continuous-ranked probability score"
https://arxiv.org/pdf/2304.00432v1.pdf	Multi-Agent Reachability Calibration with Conformal Prediction	wheat	P2	Safety/containment; CP + reachability for probabilistically safe prediction sets, certifies planner safety	"probabilistic assurances on prediction error with calibrated confidence intervals"; "probabilistically safe and dynamically feasible prediction sets"
https://arxiv.org/pdf/2303.01422.pdf	Design-based conformal prediction	wheat	P4	0.9	Methodology extending CP to non-exchangeable complex-survey data; distribution-free finite-sample coverage is the objective and it is honest about the exchangeability limitation.	Abstract: "assumption-lean... guaranteed finite-sample coverage"; handles non-exchangeable survey data.
https://www.sciencedirect.com/science/article/pii/S0887233318300237?via%3Dihub	Predicting skin sensitizers (GARD applicability domain)	wheat	P2	0.65	CP used to determine the applicability domain / valid prediction region for a toxicology classifier; containment of valid region is the genuine objective. Abstract paywalled (403); judged from title+venue+knowledge.	403 paywall; title/abstract describe CP for applicability domain of GARD assay with confidence.
https://link.springer.com/content/pdf/10.1007/s10472-023-09847-0.pdf	Conformal Predictive Distribution Trees	wheat	P4	0.8	Methodology combining interpretable tree models with conformal predictive distributions to communicate per-instance confidence; correct-by-construction CPD, no overclaim.	Abstract: interpretable models plus a measure of confidence per prediction via conformal predictive distributions (Johansson, Lofstrom, Bostrom).
https://arxiv.org/pdf/2306.17815.pdf	Bayesian Optimization with Safety via Online CP	wheat	P2	0.92	Online CP provides formal safety guarantees with a controllable (non-zero) violation rate, dropping strict assumptions of SAFEOPT; containment/safety is the deliverable, guarantee stated honestly.	Abstract: SAFE-BOCP "satisfies safety requirements irrespective of properties of the constraint function" at cost of controllable violation rate.
https://arxiv.org/pdf/2308.09647.pdf	Robust UQ via Conformalised Monte Carlo (MC-CP)	borderline	A2	0.6	Hybrid MC-dropout+CP for UQ; CP backbone gives valid sets but framing emphasizes "robust uncertainty quantification" and "significant improvements over UQ methods" without conditional-coverage caveat.	Abstract: "yielding robust prediction sets/intervals", "significant improvements over advanced UQ methods... in classification and regression".
https://arxiv.org/pdf/2309.08313.pdf	Heteroskedastic conformal regression	wheat	P4	0.9	Methodology study of normalized/Mondrian CP for heteroskedastic noise; explicitly distinguishes marginal vs conditional validity and ties conditional validity to data-generating assumptions. Honest.	Abstract: "focus on marginal coverage... connecting conditional validity to implicit assumptions about how data is generated".
https://proceedings.mlr.press/v204/canete23b/canete23b.pdf	Online NoVaS Conformal Volatility Prediction	wheat	P4	0.78	Combines NoVaS studentization with ACI's adaptive alpha for realized-volatility forecasting; ACI is the appropriate distribution-free tool for non-exchangeable time series, so the guarantee object is correct.	Abstract (search): NoVaS + adaptive alpha of ACI; no stored history; realized volatility forecasting (COPA 2023).
https://proceedings.mlr.press/v204/giovannotti23a/giovannotti23a.pdf	MT Quality with Conformal Predictive Distributions	wheat	P1	0.82	Conformal predictive distributions yield intervals with guaranteed coverage for MT quality scores; set/interval is the deliverable and exchangeability requirement is stated explicitly.	Abstract (search): guaranteed coverage at significance eps; outperforms baseline on coverage and sharpness; "requires data exchangeability assumption to hold".
https://arxiv.org/abs/2310.15641	Guaranteed Coverage PIs with Gaussian Process Regression	wheat	P1	0.82	CP wraps GPR to guarantee marginal coverage even under model misspecification; coverage is genuinely the deliverable. Slightly strong "guaranteed coverage" branding but standard and correct (marginal).	Abstract: "guarantees the production of PIs with the required coverage even when the model is completely misspecified".
https://openreview.net/pdf?id=xDCmlkSavR	Beyond Confidence: Atypicality	borderline	A1	0.6	Primary aim is improving UQ/calibration/accuracy via an atypicality signal; CP appears only as one UQ instrument rather than the core object, and gains are framed as performance/accuracy improvement.	Abstract: "incorporating atypicality improves uncertainty quantification and model performance"; improves accuracy/calibration.
https://arxiv.org/abs/2312.08604	Verification of Neural Reachable Tubes via CP	wheat	P2	0.92	CP gives probabilistic safety guarantees for neural reachable tubes and is proven equivalent to scenario optimization; containment/safety is the objective and the trade-off is stated honestly.	Abstract: "probabilistic safety guarantees for neural reachable tubes"; split CP reduces to scenario-based approach.
https://arxiv.org/abs/2401.13744	Conformal Prediction Sets Improve Human Decision Making	wheat	P1	0.9	The prediction set itself is the deliverable, evaluated in a pre-registered RCT; honest claim that variable-size CP sets beat fixed-size sets with equal coverage. No overclaim.	Abstract: pre-registered RCT; accuracy improves with conformal sets vs fixed-size sets of same coverage guarantee.
https://arxiv.org/pdf/2402.07407.pdf	Conformal Predictive Programming for Chance Constrained Optimization	wheat	P4	0.85	Uses CP quantile lemma to reformulate chance constraints; containment/feasibility is genuinely the objective and CP applied for what it does. Methodology, honest about quantile guarantee.	"utilizes ... the quantile lemma - central to conformal prediction - to transform the chance constrained optimization problem into a deterministic problem"
https://arxiv.org/abs/2403.01216	API Is Enough: CP for LLMs Without Logit-Access	wheat	P1	0.85	Prediction set with user-defined coverage guarantee is the deliverable for API-only LLMs; "outperform baselines" means smaller sets at fixed coverage (standard efficiency), not coverage=quality.	"ensures a statistical guarantee of the user-defined coverage" and "minimizes the size of prediction sets"
https://arxiv.org/abs/2403.15527	Conformal Online Model Aggregation	wheat	P4	0.85	Methodology for aggregating CP sets while retaining coverage; explicitly states the negative-correlation assumption needed and verifies it. Honest about guarantee scope.	"As long as the input sets have (distribution-free) coverage guarantees, COMA retains coverage guarantees, under a negative correlation assumption"
https://arxiv.org/abs/2404.02722	On-line conformalized NN ensembles for day-ahead electricity prices	borderline	P6	0.6	Coverage (passing coverage tests) is genuinely the objective and online recalibration is the standard TS mitigation, but abstract does not explicitly flag exchangeability/temporal dependence; mild A3 risk.	"deployed within an on-line recalibration procedure ... improved hourly coverage and stable probabilistic scores"
https://www.sciencedirect.com/science/article/pii/S2667113124000093	Enhancing reliability of probabilistic PV forecasts using CP	borderline	P6	0.6	Builds CP intervals/CDFs as deliverable (uncertainty quantification is the object) and uses weighted/Mondrian variants, but framing on "reliability" plus headline 14% interval-score gain leans toward sharpness/accuracy and no explicit exchangeability caveat for TS.	"transform the point predictions into rigorous uncertainty intervals ... improving the weighted interval score by 14%"
https://proceedings.mlr.press/v238/jager24a.html	Conformal Data Cleaning (imputation to cleaning)	chaff	A1	0.75	CP repurposed to detect/correct erroneous values to boost downstream predictive performance; the deliverable is accuracy, not coverage. Coverage guarantee is not the objective.	"combine imputation techniques with conformal prediction ... CDC improves predictive performance in downstream ML tasks in the majority of cases"
https://arxiv.org/abs/2404.16970	CarbonCP: Carbon-Aware DNN Partitioning with CP	borderline	P5	0.55	CP serves as an uncertainty-aware layer informing partitioning decisions (risk-control flavor), which is defensible, but objective is carbon/latency and reporting centers on error rate; no coverage/exchangeability detail given.	"uncertainty-aware AI inference framework built upon conformal prediction theory ... 9.9% error rate"
https://dl.acm.org/doi/pdf/10.1145/3589334.3645595	Robust Route Planning under Uncertain Pickup Requests (ROPU)	borderline	A4	0.55	Spatio-temporal "conformal interval with high confidence" feeds RL for robustness (risk-control intent), but applied to dependent spatio-temporal data with no exchangeability caveat and framed for operational accuracy gains; conditional/coverage claim loose.	"unified spatial-temporal conformal interval with high confidence ... improvements of at least 30.49% in the pickup overdue rate"
https://jmlr.org/papers/v25/22-1218.html	Conformal Inference for Online Prediction with Arbitrary Distribution Shifts	wheat	P4	0.9	Theory/methodology extending ACI with provable local-interval regret; explicitly distribution-free and honest that the guarantee is regret over local windows, adaptive without knowing shift rate.	"a novel procedure with provably small regret over all local time intervals of a given width ... without the need for distributional assumptions"
https://openreview.net/pdf?id=wu3JIjKmXQ	Robustness via Conformalized Randomized Smoothing (time series)	wheat	P4	0.7	Methodology extending randomized smoothing to CP for certified robustness/bounding performance under perturbations; robustness certification is the object. Minor "reliability/accuracy" framing but core is sound.	"generalize randomized smoothing to arbitrary transformations and extend it to conformal prediction ... bound the performance on new domains"
https://arxiv.org/pdf/2405.02140	An Information Theoretic Perspective on Conformal Prediction	wheat	P4	0.9	Theory connecting CP set size to conditional entropy via proven upper bounds, with methodology (training objectives, side info); careful framing, validated by reduced inefficiency.	"prove three different ways to upper bound the intrinsic uncertainty ... lower inefficiency (average prediction set size)"
https://arxiv.org/abs/2407.06867	Distributionally robust risk evaluation with isotonic constraint	unclear	NA	0.8	Shape-constrained distributionally robust learning paper; no mention or use of conformal prediction in the abstract. Cannot assess CP use because there is none stated.	"No mention of conformal prediction"; "shape-constrained approach to DRL ... density ratio ... is isotonic"
https://arxiv.org/abs/2407.15277	Conformal Predictions under Markovian Data	wheat	P4	0.95	Theory quantifying CP coverage gap under Markovian (non-exchangeable) data; honest about correlation effect, proposes K-split fix.	"quantify the gap in terms of coverage induced by the correlations... scales as sqrt(t_mix ln(n)/n)"
https://arxiv.org/abs/2408.10939	Conformalized Interval Arithmetic Symmetric Calibration	wheat	P4	0.9	Methodology extending CP to sums/averages of labels with proven validity under permutation invariance; coverage is the object.	"Under permutation invariant assumptions, we prove the validity of our proposed method"
https://openreview.net/pdf?id=eQ0Z2vrERs	Clinically Useful CP for Inherited Retinal Disease	wheat	P1	0.85	Prediction set is the clinical deliverable; reports coverage at stated confidence, flags clinicians for second opinion (containment).	"recommend statistically rigorous reliable prediction sets... coverage above 90% at a confidence level of 80%... flagging clinicians for a second opinion"
https://proceedings.mlr.press/v238/han24b/han24b.pdf	Conformalized Semi-supervised Random Forest	wheat	P1	0.85	Set-valued prediction with proven coverage under label shift; empty set as distribution-free outlier flag; honest theory.	"establish CSForest to cover true labels for previously observed inlier classes under arbitrarily label-shift... flag unseen outliers by generating an empty set"
https://arxiv.org/abs/2409.07902	Conformal Distributed Remote Inference Sensor Networks	wheat	P5	0.9	Online conformal risk control to guarantee target FNR under comms constraints; containment/risk-control is the objective.	"online conformal risk control... deterministic worst-case performance guarantees in terms of FNR"
https://arxiv.org/abs/2310.07850	Conformal prediction with local weights	wheat	P4	0.95	Methodology explicitly honest about marginal-vs-conditional coverage and pointwise impossibility; provides relaxed local + covariate-shift validity.	"these guarantees only ensure marginal coverage... impossibility of achieving pointwise local coverage is well established"
https://arxiv.org/abs/2410.16333	Conformal Predictive Portfolio Selection	borderline	A1	0.7	CP intervals feed portfolio selection; primary claim is superior returns, so object shifts from coverage to performance, but UQ use is defensible.	"selects the portfolio... showing that it delivers superior returns compared to simpler strategies"
https://arxiv.org/abs/2411.01289	Uncertainty for complex event prediction safety-critical	borderline	A5	0.6	CP used as generic interval-building UQ layer in safety-critical CEP; modest but vague claims ("very promising"), no coverage caveat discussion.	"we use conformal prediction to build prediction intervals... results, which are very promising"
https://arxiv.org/abs/2412.19511	UQ for radiomic radiation pneumonitis prediction	chaff	A1	0.7	CP treated as one of several UQ methods but evaluated/sold via AUROC/AUPRC and "predictive accuracy"; coverage is not the object.	"Highest AUROC is achieved by... conformal prediction... may improve both predictive accuracy and calibration"
https://arxiv.org/abs/2501.04823	Learning Robot Safety from Sparse Human Feedback	wheat	P2	0.92	CP identifies region containing user-specified fraction of policy errors; warning system with guaranteed miss rate (safety containment).	"region of states... guaranteed to contain a user-specified fraction of future policy errors... warning system... with guaranteed miss rate"
https://arxiv.org/abs/2501.18363	Robust Online CP under Uniform Label Noise	wheat	P4	0.92	Methodology/theory correcting CP coverage gap under label noise; honest about gap, proves convergence of coverage error.	"label noise causes a persistent gap... NR-OCP eliminates the coverage gap... convergence rate of O(T^-1/2)"
https://arxiv.org/abs/2501.13430	Wasserstein-regularized CP under Distribution Shift	wheat	P4	0.92	Methodology bounding coverage gap under joint distribution shift; explicitly honest that iid may fail and degrade coverage.	"i.i.d. assumption, which may not hold and lead to a gap... Wasserstein distance-based upper bound of the coverage gap"
https://arxiv.org/abs/2503.19068	Minimum Volume Conformal Sets	wheat	P1	0.9	Prediction set is the deliverable; new nonconformity score directly minimizes set volume while ensuring valid coverage; coverage is genuinely the objective.	Abstract: "construct predictive sets with finite-sample validity... directly learns minimum-volume covering sets while ensuring valid coverage."
https://arxiv.org/abs/2505.21658	STACI Spatio-Temporal Conformal Inference	borderline	A2	0.6	Provides a novel spatio-temporal conformal algorithm but abstract claims "statistically valid prediction intervals" in a spatio-temporal setting without explicit exchangeability/conditional caveat; defensible since it introduces a tailored ST algorithm.	Abstract: "novel spatio-temporal conformal inference algorithm... provides statistically valid prediction intervals."
https://arxiv.org/abs/2505.23592	Cross-Validation through Stability	wheat	P4	0.85	Theory paper on CV via stability; CP is a downstream application, treated correctly by construction; no overclaim.	Abstract/claim: stability lens yields theoretical results applicable to conformal prediction and selective inference.
https://arxiv.org/abs/2505.13118	Feature Contribution to CP Intervals	wheat	P4	0.8	Methodology using CP interval width/bounds as Shapley value functions for uncertainty attribution; honest, model-agnostic, CP used for what it does.	Abstract: "CP interval properties-such as width and bounds-serve as value functions... attribute predictive uncertainty to input features."
https://arxiv.org/abs/2402.05806	Temperature Scaling and CP of Deep Classifiers	wheat	P4	0.92	Careful empirical+theoretical study; explicitly distinguishes marginal vs class-conditional coverage; offers honest guidelines, no overclaim.	Abstract: "guaranteeing marginal coverage but not, e.g., per class coverage... establish a mathematical theory that explains the entire non-monotonic trend."
https://arxiv.org/abs/2505.24693	Conformal Prediction for Zero-Shot Models	wheat	P4	0.85	Split CP on CLIP; explicitly acknowledges domain drift degrades efficiency and proposes Conf-OT to maintain coverage guarantees; honest about the challenge.	Abstract: "domain drift negatively affects the efficiency... maintaining coverage guarantees" via optimal transport.
https://arxiv.org/abs/2510.02471	Why split conformal effective despite temporal dependence	wheat	P4	0.95	Theory directly confronting exchangeability violation in time series; introduces switch coefficient bounding coverage loss; sharp results; exemplary honesty about the guarantee.	Abstract: "even short-range temporal dependence is a strong violation of the exchangeability assumption... bound the loss of coverage... in terms of a new switch coefficient."
https://arxiv.org/abs/2507.10425	Non-exchangeable CP with Optimal Transport	wheat	P4	0.92	Explicitly tackles non-exchangeability/distribution shift; estimates coverage loss and mitigates shift; methodology honest about exchangeability requirement.	Abstract: "exchangeability... often violated in practice due to distribution shifts... estimate the loss in coverage and mitigate arbitrary distribution shifts."
https://arxiv.org/abs/2512.03298	Regime-Switching Forecasts with ACI	borderline	A3	0.6	Uses ACI/AgACI, appropriate tools for nonstationary TS, but phrases ACI's adaptive coverage as "finite-sample marginal guarantees under nonstationarity," which somewhat overstates ACI's long-run/adaptive nature; otherwise sound.	Abstract: "online predictive bands with finite-sample marginal guarantees under nonstationarity and model misspecification."
https://arxiv.org/abs/2602.01667	Quantifying Epistemic Predictive Uncertainty in CP	wheat	P4	0.9	Theory connecting CP to credal sets; proves characterization in split CP; proposes principled EPU measure; honest and methodologically careful.	Abstract: "any full CP procedure induces a set of closed and convex predictive distributions... we prove that this characterisation also holds in split CP."
https://arxiv.org/abs/2509.22240	COMPASS Feature CP for Medical Segmentation	wheat	P4	0.9	Metric-level CP for segmentation; explicitly states exchangeability assumption and proves marginal coverage; addresses covariate shift via importance weights; honest about guarantee.	Abstract: "We prove that COMPASS achieves valid marginal coverage under the assumption of exchangeability... recover target coverage under covariate shifts."
https://papertalk.org/papertalks/36577	Conformal Time-series Forecasting (NeurIPS 2021)	wheat	P4	0.8	Foundational extension of inductive CP to time-series forecasting with theoretical guarantees; established sound methodology; minor TS-exchangeability caveat not emphasized in summary.	Talk: "extend the inductive conformal prediction framework to the time-series forecasting setup... uncertainty estimates with theoretical guarantees."
https://openreview.net/pdf?id=jCdoLxMZxf	Copula CP for Multi-step TS	wheat	P4	0.8	Models temporal dependence via copulas for multi-step intervals; coverage/sharpness is the objective and honest about single-step CP limitation.	"existing conformal prediction algorithms for time series are limited to single-step prediction without considering the temporal dependency"; "more calibrated and sharp confidence intervals"
https://arxiv.org/abs/2402.05203	Bellman Conformal Inference	wheat	P4	0.85	Honestly claims approximately calibrated, long-term coverage under arbitrary shift/dependence; targets interval length via stochastic control. Coverage is the deliverable.	"approximately calibrated prediction intervals"; "long-term coverage under arbitrary distribution shifts and temporal dependence"
https://www.jmlr.org/papers/v25/23-1553.html	Split CP and Non-Exchangeable Data	wheat	P4	0.9	Theory paper that explicitly confronts exchangeability and quantifies the cost as a coverage penalty; honest about the guarantee.	"adding a small coverage penalty"; "the crucial assumption of data exchangeability, which hinders many real-world applications"
https://arxiv.org/abs/2503.21251	Dual-Splitting CP Multi-Step TS	borderline	P6	0.6	UQ for multi-step TS via CP is legitimate and Winkler is an interval-quality metric, but framing leans on improvement percentages and carbon-reduction with no conditional-coverage caveat.	"model-agnostic nature and statistical guarantees"; "Outperforms existing CP variants by up to 23.59% on the Winkler Score"; "11.25% carbon emission reduction"
https://openreview.net/pdf?id=RD9q5vEe1Q	Error-quantified Conformal Inference	wheat	P4	0.85	Online CP claiming long-term coverage under arbitrary dependence/shift; honest (long-term, online) and methodological contribution on adaptive feedback.	"long-term coverage guarantee for ECI under arbitrary dependence and distribution shift"; "valid miscoverage control and output tighter prediction sets"
https://proceedings.neurips.cc/paper_files/paper/2024/file/dbfb7b1443583fc7ab87e8b1b4f48c9c-Paper-Conference.pdf	Conformalized TS w/ Semantic Features	wheat	P4	0.75	Weighted CP with finite-sample coverage that explicitly addresses exchangeability failure in time series; honest, distribution-free framing.	"finite-sample coverage guarantees"; addresses "standard exchangeability assumption" failing in temporal contexts via weighting
https://arxiv.org/abs/2602.16537	Optimal training-conditional regret online CP	wheat	P4	0.9	Pure theory: minimax-optimal training-conditional regret, stability-based (not permutation) guarantees, non-asymptotic bounds. Correct by construction.	"provably achieves minimax-optimal regret"; "non-asymptotic regret guarantees" that "match the minimax lower bound"