acceptodds
Under review as a conference paper at ICLR 2027

MOF-Select: Benchmarking Selection and Confidence in Conditional MOF Generation

Abstract

Conditional generative models predict metal–organic framework (MOF) structures from specified building blocks. Their value for materials design depends on converting candidate pools into reliable recommendations, a transition that candidate coverage alone cannot assess. We introduce MOF-Select, a dataset and benchmark that makes this transition a measurable learning problem. MOF-Select aligns outputs from different generators under shared chemical conditions and provides candidate-level reference-recovery supervision. Controlled protocols separate available coverage from successful selection and acceptance under changes in generator source, chemistry, and sampling budget. The resource comprises 245,760 generation attempts and supports comparisons between heuristic and learned decision rules. Across three generators, the best selector in each generator/test setting leaves 7.85–24.74 percentage points of reference recovery unused at ten attempts per condition. Learning improves selection and confidence ranking on MOF-LLM pools, while larger pools can displace correct choices. Better confidence ranking does not ensure certified acceptance, and source-calibrated rules can incur high acceptance error after a generator change. MOF-Select establishes a common framework for evaluating and improving how generative coverage translates into reliable recommendations, making downstream decision quality an explicit target for progress in MOF generation.

Primary area:
datasets and benchmarks
Keywords:
metal-organic frameworks, conditional generation, datasets and benchmarks, candidate selection, confidence calibration, selective prediction

Reviews

Among:
Shape:
1234this paper, Top 49%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 2.33 over 3 reviews, above 51% of the others.
Reviewer_tUoFOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
1
Soundness
1
Read review
Paper summary
  • Summary
    • Evaluated different strategies for selecting the right sample from a set of generated MOFs, but honestly very difficult to understand the way it is currently written
    • I think the title invites misreading; the paper should state in the first paragraph that it benchmarks post-generation selection rather than generators
    • Review is very jargon heavy/dense and difficult to follow, and this is coming from someone who has worked in both the MOF-space and ML space for >10 years
Critical strengths
  • Strengths
    • Hard to say, the question being answered is interesting (e.g., many people work on generation, but how do you prioritize/select the right sample for further testing). I did not think this was fully clear from the prose though, I had to work hard to figure out what the authors were doing here.
Critical weaknesses
  • Weaknesses
    • I found the paper very convoluted and difficult to read, what terms were was not defined and it felt like a lot of nonstandard terms were used and never defined. Inventing lots of new terminology where good terminology already exists (e.g., "success rate" instead of "oracle@K")
    • For instance, what does this sentence mean? "Their value for materials design depends on converting candidate pools into reliable recommendations, a transition that candidate coverage alone cannot assess." This is the second sentence in the abstract and it only gets more confusing from there.
    • There is another sentence in the abstract "MOFSelect aligns outputs from different generators under shared chemical conditions and provides candidate-level reference-recovery supervision", what even is "candidate-level reference-recovery supervision"? I have worked in both MOFs and ML for over 10 years and I don't know what this is talking about.
    • Figures do not help to understand what was done at all
    • Advice: terms like pool, candidate, condition, recovery, coverage, and selection should be defined in the intro with a concrete example, and a figure showing an actual pool (e.g. 10 structures with one marked as matching) would help far more than Figure 1's current schematic
    • The paper is on benchmarking MOF generation/selection and not a single MOF structure was shown....
    • The related work cites some irrelevant things. For instance, DiffDock is cited but the connection to MOFs or this work is not clearly made
Questions for authors
  • Questions
    • What are candidate "pools"? Neither what a "candidate" is (is it a building block? a topology? a full MOF structure? something else) nor a "pool" is ever defined. Does "pool" refer to a set of some kind? Why not just say "set" if so (or if it is not a set, can you explain it please)? Is it different? Just define the terms used throughout. The term pool appears 105 times in the paper yet it is never defined that I could tell.
    • Hard for me to evaluate the technical soundness of this work because it took me over an hour even to understand what the authors were doing. For instance, how was data leakage controlled for across the 3 generators? were they trained the same? Then, what was the variance in the predictions? Were different seeds tried? Are the conclusions statistically significant?
5 comments
AuthorsOpenReview ↗

Response to Reviewer tUoF (1/2): Post-generation MOF selection: definitions and a concrete example

Thank you for reading the paper so closely and for pointing to the specific passages that were difficult to understand. Your detailed comments were particularly helpful in revising how we introduce the task and explain the terminology.

MOF-Select benchmarks post-generation selection and confidence: which generated structure should be recommended for a chemical request, and when should the method abstain? We address the possible ambiguity in the title by stating this scope directly in the revised abstract and first introduction paragraph. The evaluated methods choose among frozen generator outputs; the contribution is a dataset and protocol for those decisions.

Candidates and pools. A condition specifies building-block identities and counts. A candidate is a complete periodic MOF, with atom identities, coordinates, and a unit cell. We use “pool” to mean the ordered record of generation attempts for one condition and generator, rather than a set of unique structures. Each attempt remains a separate entry, including repeats and failures. At each budget K (1, 5, or 10), all selectors are evaluated on the first K attempts from that same record.

Recovery means matching a known structure for that condition; candidate coverage asks whether any attempt matches, whereas selection chooses one candidate without the test reference. The ordered-pool formulation appears in the original §4.1 (p. 5, Eq. 1); we now place its plain-language definition and the other five terms together in the introduction. Section 4 introduces Oracle@K as pool success rate and Selected@1 as selected success rate. Accepted coverage separately measures the fraction of conditions with an accepted recommendation. These qualifiers distinguish finding a match, choosing it, and accepting the choice.

A concrete MOF example. Figure 3(a) shows the BW hypothetical reference for condition bw_2151 alongside two MOF-LLM candidates. Candidate 1 does not match the reference, while candidate 6 meets both the primary and stricter matching criteria. Figure 3(b) gives the matching results for all ten candidates and the actual choices: Full selects 6, Consensus 7, and Energy 10. Four candidates meet the primary criterion; two also meet the stricter criterion. All three methods select a primary match here. The introduction explains how this pool distinguishes coverage from selection: a match is available, but choosing one of the six nonmatching candidates would miss it. Appendix H.3 (Figure 8) shows the reference and all ten candidate structures.

The case was chosen by a fixed, reproducible rule without considering which structures the selectors chose. Candidates 1 and 6 are the first nonmatching and first strictly matching candidates in generation order. This example illustrates the task; Section 5 compares method performance across test conditions. Figures 1 and 2 retain the decision overview and benchmark protocol.

The two abstract sentences you highlighted are now expressed directly: generating a reference match does not ensure a method can identify it or assess confidence in its choice; the dataset groups complete structures generated from the same building-block identities and counts, then labels whether each matches a known reference. These labels support training and evaluation, while test references remain hidden from scoring.

Related work and contribution. DiffDock is relevant because it trains reference-based confidence to rank generated poses and evaluates selective accuracy. Revised §2 makes this connection explicit. We cite it as a precedent for the decision problem, rather than a MOF generator or direct baseline. MOF-Select supplies reusable supervision and fixed candidate pools for developing decision models without rerunning generation. Its paired protocols separate candidate availability, selection, and acceptance reliability across MOF sources. As detailed below, decision errors persist after structural screening and exclusion of identifiable upstream training exposure. Reference recovery measures structural agreement, not physical stability or synthesizability.

We appreciate the time and care you put into this review. Your suggestions helped us explain the task more clearly and make the paper easier to follow. We would be grateful to hear whether the revised explanations and concrete MOF example address your concerns about clarity.

Our second comment answers the questions about leakage, training, variability, and significance.

AuthorsOpenReview ↗

Response to Reviewer tUoF (2/2): Data leakage, training, and statistical evidence

Data leakage controls. Selectors are compared on identical frozen candidates from each generator. Chemical-condition groups are disjoint across selector training, development, calibration, and testing; the family test holds out 23 linker skeleton groups. Scorer inputs exclude reference coordinates, target cells/topologies, labels, matching runtime, and generator identity. Leave-one-generator-out runs exclude that source from fitting, development selection, and calibration. Appendix F.2 also checks cross-role reference neighbors.

Generator training and exposure. We use released checkpoints with their original training recipes, without unified retraining, and align chemical requests and sampling budgets. Cross-generator results characterize these released systems. We audited the public AtomMOF training/validation data, MOF-LLM task-pretraining, supervised fine-tuning and reinforcement-learning data, and the MOFFlow-2 inventory. A common union-exposure mask removes 73 core-test and 25 family-test conditions from every source. On the remaining 951 core and 487 family conditions, screened-pool selection gaps remain 6.38–13.35 pp under primary matching and 3.29–11.64 pp under strict matching, with all paired 95% intervals above zero. On unfiltered pools, MOF-LLM Full's family gain over Consensus remains 4.24 pp [1.63, 6.75]. Thus the selection findings persist after excluding identifiable training exposure. The audit covers released task datasets; undocumented checkpoint histories are outside its scope (§3.3, Appendix F.2).

Learned selectors use 2,048 training conditions and three seeds; neural models receive 5,120 updates and tree families search six configurations (§4.3).

Prediction variability across seeds. Table 20 reports the original three Full seeds. MOF-LLM recovery is 44.53 ± 0.78% on core and 40.17 ± 0.63% on families (sample standard deviation). Existing predictions show about 55% pairwise disagreement in selected attempt IDs, but 4.36%/4.95% in core/family recovery outcomes. Different IDs need not denote different geometries. Across generators, outcome disagreement is 11.20–22.11% on pools containing both matches and nonmatches (Appendix H.3). Thus stable aggregate performance does not imply identical individual predictions.

Statistical support for the conclusions. Section 4.4 uses 10,000 paired bootstrap replicates, averaging seeds within condition; family sensitivity resamples skeleton groups. These intervals describe condition or family variation given the three fixed seeds; Table 20 separately reports seed variability.

  • Selection opportunity: after the requested structural filtering, surviving-pool coverage exceeds the best of three evaluated rankers by 6.15–13.48 pp under primary matching and 3.13–12.11 pp under strict matching. All six generator/test settings have paired 95% intervals above zero; family intervals resample skeleton groups (Table 19).
  • Budget expansion: expanding Full's input from five to ten candidates loses previously correct recommendations on 1.46–5.86% of conditions across the six settings, averaging three training seeds (Table 4). Added family-group resampling gives positive 95% intervals for all three generators' losses (Appendix C.2). This is displacement, not evidence that larger pools always reduce net recovery: new successes can offset losses.
  • Confidence transfer: a source-certified rule targeting 10% risk makes 174 errors among 533 accepted held-out MOF-LLM core predictions: 32.65% error, with a descriptive 95% binomial interval of [28.68%, 36.81%]. The other two seeds cannot certify the source target and accept nothing; their risk is undefined, not zero (§5.4).

Learning gains are evaluated separately. MOF-LLM Full-minus-Consensus gains are:

EvaluationGain (pp)Interval
Original family, primary match3.84[1.43, 5.98], 95%
Original family, strict match1.30[−0.88, 3.29], 95%
Original frozen confirmation, 2,048 conditions3.34[1.95, 4.72], nominal 97.5%
Post-hoc corrected-input sensitivity, same conditions3.11[1.76, 4.48], nominal 97.5%

The strict-family advantage remains inconclusive. The corrected-input result is a post-hoc implementation sensitivity on the same confirmation conditions (Table 3). Two additional generation streams, distinct from training seeds, reuse 256 core and 256 family conditions. Averaging streams within condition gives gains of 4.30 pp [1.43, 7.23] and 5.01 pp [1.82, 8.27], using condition-level bootstrap (§5.5; Appendix D.4, Table 12). Together, these controls establish selection quality, budget use, and acceptance reliability as measurable targets for decision models.

We would appreciate your reassessment of soundness and the overall recommendation in light of these clarifications and revisions.

Thank you for the detailed response. However, I still feel substantial revisions are needed to the original submission in order to make it understandable and to make it clear why this work is significant. I maintain my score at 1 and encourage the authors to resubmit to a venue that can better assess the significance in the domain of MOF optimization, as well as to better place the work in context. Major revisions needed.

Reviewer_yXw1OpenReview ↗
Rating
2
Confidence
2
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

This paper introduces MOF-Select, a benchmark for candidate selection and confidence estimation in conditional MOF generation, which collects 245,760 generation attempts from AtomMOF-M, MOFFlow-2, and MOF-LLM under shared chemical conditions.

Critical strengths
  • Having a correct candidate in the generated pool does not mean that the model can select it, the paper evaluates these two problems separately.
  • The presentation is clear.
Critical weaknesses
  • See the questions.
Questions for authors
  • Existing tools such as MOFChecker can detect atomic overlaps, abnormal coordination, and disconnected fragments. Could the authors compare the learned selectors against a baseline that first filters candidates using these checks and then ranks the remaining structures by energy or Consensus?
3 comments
AuthorsOpenReview ↗

Response to Reviewer yXw1: Structural filtering baselines

Thank you for your review and for suggesting the MOFChecker-based baselines. Your suggestions motivated the additional comparisons reported below and helped us clarify the role of structural screening in candidate selection.

We have added the requested MOFChecker → Energy and MOFChecker → Consensus baselines, checking all 46,080 candidates from the three generators and both test sets. We also apply the same filter to Full, the main learned selector, to compare rankers on identical eligible candidates.

The basic filter checks atomic overlaps, overcoordinated C/N/H, and disconnected molecular fragments. Consensus is recomputed among survivors; Full keeps its frozen scores and original pool context. No reference geometry or recovery labels enter filtering. Conditions with no survivors count as unsuccessful selections. The table reports primary reference recovery.

MOF-LLM selectorCore recovery (%)Family recovery (%)
Energy43.5539.06
Filter + Energy40.8238.09
Consensus43.1636.33
Filter + Consensus40.6336.33
Full44.5340.17
Filter + Full41.4138.02

Reference recovery and the strength of Energy. On MOF-LLM families, filtering lowers primary recovery for Full and Energy and leaves Consensus unchanged. Among the same eligible candidates, filtered Full exceeds filtered Consensus by 1.69 pp [0.19, 3.00], but differs from filtered Energy by −0.07 pp [−1.71, 1.54]. With strict matching, filtered Energy exceeds filtered Full by 1.76 pp [0.14, 3.66].

As a supplemental, post-hoc diagnostic, we require the selected candidate to both match a reference and pass the checks. On MOF-LLM families, filtering improves Energy's own joint success rate by 2.54 pp [0.84, 4.65], reaching 38.09%; filtered Full reaches 38.02%. This measures the outcome of enforcing the specified checks alongside reference agreement. Primary and strict reference recovery remain the benchmark targets.

Screened pools retain substantial selection headroom. Coverage of surviving candidates exceeds the highest mean recovery among the three filtered rankers by 6.15–13.48 pp under primary matching and 3.13–12.11 pp under strict matching. All six generator/test settings have paired 95% intervals above zero. Restricting evaluation to conditions whose declared references all pass the same checks leaves primary gaps of 6.19–13.25 pp.

An additional gate-sensitivity diagnostic rejects undercoordinated C/N. On MOF-LLM families, it leaves 30.08% of pools empty, versus 16.41% for the basic gate. Both configurations are fixed; neither was selected by test recovery. One parsing failure is rejected; retaining that candidate changes no condition-level recovery result.

The revised §5.6 states these findings. Appendix H.2 and Tables 18–19 report all generators and settings. Intervals are exploratory, paired within condition, with skeleton-group resampling for family tests. The checks address specified defects, including C/N/H coordination, rather than a universal metal-coordination or synthesizability criterion.

The requested baseline shows that screening and strong energy ranking still leave recoverable candidates unselected. We would appreciate your reassessment of the remaining concern in light of these comparisons.

Thank you again for your time and constructive feedback. We would welcome your assessment of whether these additional comparisons address your concerns, and we look forward to hearing your thoughts and continuing the discussion.

Thank you again for your constructive review. We would like to briefly follow up on our response regarding the MOFChecker-based baselines. We have added the requested filtering followed by Energy or Consensus comparisons, with results reported in the response and revised §5.6 and Appendix H.2.

When you have a chance, we would appreciate your thoughts on whether these comparisons address your concern. We would be happy to clarify any remaining questions. Thank you for your time and consideration.

Dear Reviewer yXw1,

We would like to follow up once more on our response to your review. If you have a chance before the discussion closes, could you please review the additional MOFChecker-based comparisons and share your assessment? We would appreciate hearing your thoughts and would be happy to clarify any questions.

Thank you for your time and consideration.

Reviewer_ficbOpenReview ↗
Rating
4
Confidence
2
Informativeness
2
Presentation
3
Soundness
3
Read review
Paper summary

The paper describes a benchmark, MOF-select, designed for assessing candidate selection and confidence estimation in conditional metal-organic framework (MOF) generation. The benchmark considers whether a correct candidate can be selected from a generated pool and whether the resulting recommendation can be accepted with sufficient confidence. The authors evaluate selection methods using metrics for oracle coverage, selected recovery, confidence ranking, and risk-constrained acceptance. The experiments examine generalization across generators and chemical families, the effects of sampling budget, and different training strategies. The results show that substantial gaps remain between oracle coverage and selected recovery, that increasing pool size can displace correct selections, and that confidence calibration can deteriorate under generator shift.

Critical strengths

The paper addresses an important gap in the evaluation of conditional MOF generation by distinguishing candidate coverage from successful selection and reliable acceptance. This is an important aspect of generation as the presence of a correct structure among generated candidates does not guarantee that it is flagged as recommended, The benchmark provides a useful framework for evaluating downstream decisions on existing and future models. I appreciate the use of shared chemical conditions and fixed candidate pools, which allows for controlled comparisons of selection methods independent of the generators. The evaluation covers several heuristic and learned baselines, sampling budgets, chemical family generalization, and generator transfer.

Critical weaknesses

None, I did not identify any major weaknesses with the paper’s methodology. However, I have one minor comment. One limitation is the sensitivity of the results to the structural matching criterion. In Table 10, the lowest-energy heuristic gets higher point estimates than Full on MOF-LLM when using stricter matching. Full’s improvement over Consensus on MOF-LLM family conditions decreases from 3.84 percentage points under the primary criterion to 1.30 percentage points under stricter matching (Table 12). The confidence interval for the latter includes zero, so the evidence for this advantage becomes inconclusive when matching is tightened.

Questions for authors

Table 10 evaluates the selectors under stricter matching using models trained on the primary labels. Have the authors investigated whether training directly on the stricter labels changes the relative performance of the learned selectors (particularly on MOF-LLM)? This would help clarify whether the reduced learning gains are due to a mismatch between the training and evaluation criteria or some general limitation.

3 comments
AuthorsOpenReview ↗

Response to Reviewer ficb: Strict-label training

Thank you for your strong support (4/4) and insightful questions that will help us improve the paper.

Thank you for highlighting sensitivity to the matching criterion. We have completed the requested control: training Full on strict labels and selecting checkpoints with strict development labels. Your observations are correct: the original family gain falls from 3.84 pp under primary matching to 1.30 pp under strict matching, whose interval includes zero; Energy also has higher strict-recovery point estimates.

We trained three seeds per generator, keeping the same 2,048 training conditions, 512 development conditions, architecture, inputs, and 5,120-update budget. Candidates and test labels are unchanged. On MOF-LLM, strict-evaluation recovery is:

SelectorCore (%)Family (%)
Energy20.3112.50
Consensus17.199.38
Full, primary-label training20.1510.68
Full, strict-label training19.7611.52

Effect of changing the training target. Strict-trained minus primary-trained Full is −0.39 pp [−1.24, 0.46] on core and +0.85 pp [−0.27, 2.02] on families. Both intervals include zero. We do not detect a clear effect of changing the training and development targets; the role of label mismatch in the smaller gains therefore remains unresolved.

Relative performance. Strict-trained Full exceeds Consensus by 2.57 pp [1.27, 3.97] on core and 2.15 pp [0.25, 4.23] on families. Energy retains higher point estimates: Full-minus-Energy is −0.55 pp [−1.60, 0.49] and −0.98 pp [−2.80, 0.78]. These comparisons distinguish the effect of changing supervision from the ranking of different selectors.

The revised §5.5 reports the direct target-switch result; Table 17 (Appendix H.1) gives all three generators under both evaluation criteria. Brackets are exploratory paired 95% bootstrap intervals: three seeds are averaged within condition, with skeleton-group resampling for family tests.

The result identifies an important benchmark requirement: selection methods should be evaluated across matching criteria, because their relative performance depends on the structural target. MOF-Select provides paired labels and fixed candidate pools to make this dependence measurable and guide improvements across candidate sources.

Thank you again for your time and thoughtful engagement with this work. Your feedback has encouraged us to pursue this direction further and deepen our study of selection and confidence in MOF generation. If you have any further questions, we look forward to discussing them and await your response.

Thank you for performing the additional requests experiments. I agree that the new results address whether the reduced gains under strict matching could be explained by a mismatch between the training and evaluation labels. I appreciate the additional analysis and the authors' acknowledgment of the remaining sensitivity to the matching criterion. My concern has been adequately addressed, and I will maintain my original rating.

Thank you again for your helpful feedback and for taking the time to review the new results. Your suggestions helped us improve the paper, and we sincerely appreciate your care and support.

Wishing you a pleasant week and all the best with your research!

Comments on the paper

1 comment
AuthorsOpenReview ↗

Rebuttal Summary: Contributions and Responses to Reviewers

Dear Area Chair and Reviewers,

Thank you for your time and suggestions. We summarize the contribution of MOF-Select, the additional experiments, and the remaining discussion below.

From generation to reliable recommendations. Materials screening requires choosing which generated structures to examine further. A generator may produce a MOF that matches a known reference, yet the selected recommendation may fail to match. Generation success alone therefore does not measure the quality of recommendations. MOF-Select addresses this evaluation gap with data and protocols for learning which structure to select and when to abstain. It makes progress at this decision stage measurable across generators, chemical families, and sampling budgets.

Main contributions.

  • Aligned supervision across generators. We provide 245,760 generation attempts from three generators under shared building-block conditions, with labels indicating which candidates match a reference. This alignment enables researchers to train selection models, compare them on the same candidates, and test transfer between generators without repeating generation.
  • Evaluation of selection and confidence under distribution shift. Our protocols measure candidate availability, selection success, and the error rate among accepted recommendations separately. Shared conditions and fixed candidates distinguish improvements in selection from differences in what was generated. Tests across generators, chemical families, and sampling budgets reveal when selection and confidence remain reliable.

Findings that guide model development. Learning improves selection and confidence ranking on MOF-LLM, showing that these decisions can benefit from supervision. However, more candidates can displace correct choices, and confidence calibrated on one generator can become unreliable on another. These findings show why selection accuracy and confidence must be evaluated as sampling budgets and generator sources change.

Responses to reviewers.

1. Strict-label training and confirmed resolution (ficb). The reviewer recognized an "important gap" and a "useful framework for evaluating downstream decisions on existing and future models." We completed the requested training on strict matching labels with three seeds per generator (§5.5; Appendix H.1). The analysis separates the effect of changing training labels from the relative performance of selectors. After reviewing the results, ficb stated: "My concern has been adequately addressed, and I will maintain my original rating." Their recommendation remains 4: Clear acceptance.

2. Structural-filtering baselines and selection gaps (yXw1). We implemented two baselines that filter candidates with MOFChecker and then rank them by Energy or Consensus, and applied the same filter to the learned selector. The evaluation covers all 46,080 test candidates from the three generators. We also varied the filter and repeated the analysis on conditions whose reference structures pass the same checks (§5.6; Tables 18–19). After filtering, we compare the fraction of conditions containing a reference match with the fraction for which a ranker selects one. Under primary matching, this gap is 6.15–13.48 percentage points even for the best of the three evaluated rankers in each setting. All six paired 95% bootstrap intervals exclude zero. Gaps also persist under strict matching. These results directly test the requested baselines and show that structural screening and energy ranking leave matching candidates unselected. We have not yet received the reviewer's assessment of these comparisons.

3. Task definitions and technical evidence (tUoF). We revised the abstract and introduction to state the task and define its terms, added actual MOF structures and a complete ten-candidate example (Figures 3 and 8), and explained the relevant prior work. We also answered the questions about generator training, data leakage, prediction variability, and statistical significance. The response includes audits of released training data, prediction variability across training seeds, and paired statistical comparisons (Appendices F.2 and H.3). The selection gaps remain after removing test conditions that overlap with publicly available generator-training data. We have requested further feedback on the revised explanations and evidence.

MOF-Select enables researchers to develop and compare the decision models that determine which generated structures are recommended and accepted. The data, evaluation protocols, and additional experiments provide a common basis for improving this stage of generative modeling. We appreciate your consideration of the contribution and the full discussion.

Thank you for your time and consideration.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.