rewirebio.iobenchmarks
Source

When Edit Flows are Edit Jumps: replicating Edit Flows and EvoFlows (arXiv v1)

arXiv:2609.18745v1, 16 September 2026 · Preprint

Original source

What this source contributed

Records in this release that cite this source. A record can cite several sources.

Results

Values as the source printed them. Compare results only under compatible protocols, datasets and metrics.

Model or configurationMetricValueDatasetEvaluation
EditJumps editor, clock normalisation 40Covariance agreement (percent of the natural homolog ceiling)84.3 percentTwo-family reference sets, 200 homologs eachEditJumps (Ours), Covariance (% of ceiling), ceiling-normalised
EditJumps editor, clock normalisation 40Mutual information product agreement (percent of the natural homolog ceiling)78.8 percentTwo-family reference sets, 200 homologs eachEditJumps (Ours), MIP (% of ceiling), ceiling-normalised
EditJumps editor, clock normalisation 40Levenshtein distance (pooled reduction, expressed as a ratio to natural homologs)1.070 ± 0.011 unitlessTwo-family reference sets, 200 homologs eachEditJumps (Ours), Pairwise Distance‡, ceiling-normalised
EditJumps editor, clock normalisation 40Jensen shannon divergence (positional formulation chosen by the EditJumps authors, expressed as a multiple of the floor)2.68 ± 0.10 unitlessTwo-family reference sets, 200 homologs eachEditJumps (Ours), Positional JS† (× floor), ceiling-normalised
EditJumps editor, clock normalisation 40Covariance agreement (residue covariance agreement against a 300-sequence natural homolog ceiling)0.93 unitlessHER2-VH reference holdout, 200 homologsEditJumps on HER2-VH, cov. agr., 200 references
EditJumps editor, clock normalisation 40Edit count (mean realised edits per generated sequence, 400 generated sequences per family)4.08 countHER2-VH reference holdout, 200 homologsEditJumps on HER2-VH, edits, 200 references
EditJumps editor, clock normalisation 40Jensen shannon divergence (per-position formulation chosen by the EditJumps authors, against a 200-sequence natural holdout)1.03 unitlessHER2-VH reference holdout, 200 homologsEditJumps on HER2-VH, JS pos., 200 references
EditJumps editor, clock normalisation 40Kullback leibler divergence (pooled amino-acid composition against a 200-sequence natural holdout)0.63 unitlessHER2-VH reference holdout, 200 homologsEditJumps on HER2-VH, KL pool., 200 references
EditJumps editor, clock normalisation 40Kullback leibler divergence (per-position amino-acid composition against a 200-sequence natural holdout)0.05 unitlessHER2-VH reference holdout, 200 homologsEditJumps on HER2-VH, KL pos., 200 references
EditJumps editor, clock normalisation 40Mutual information product agreement (mutual information product agreement against a 300-sequence natural homolog ceiling)0.87 unitlessHER2-VH reference holdout, 200 homologsEditJumps on HER2-VH, MIP agr., 200 references
EditJumps editor, clock normalisation 40Maximum mean discrepancy (k-mer spectrum MMD between the generated set and a 200-sequence natural holdout, unnormalised)1.02 unitlessHER2-VH reference holdout, 200 homologsEditJumps on HER2-VH, MMD, 200 references
EditJumps editor, clock normalisation 40Levenshtein distance (mean pairwise Levenshtein distance within the generated set, pooled over templates)24.73 countHER2-VH reference holdout, 200 homologsEditJumps on HER2-VH, pairw. Lev., 200 references
EditJumps editor, clock normalisation 40Log likelihood (profile log-likelihood of the generated sequences under the family alignment profile)-59.03 unitlessHER2-VH reference holdout, 200 homologsEditJumps on HER2-VH, prof. LL, 200 references
EditJumps editor, clock normalisation 40Entropy delta (positional entropy difference from the natural holdout; the target value is zero)0.08 unitlessHER2-VH reference holdout, 200 homologsEditJumps on HER2-VH, ΔH, 200 references
EditJumps editor, clock normalisation 40Covariance agreement (residue covariance agreement against a 300-sequence natural homolog ceiling)0.91 unitlessTy1 reference holdout, 200 homologsEditJumps on Ty1, cov. agr., 200 references
EditJumps editor, clock normalisation 40Edit count (mean realised edits per generated sequence, 400 generated sequences per family)4.58 countTy1 reference holdout, 200 homologsEditJumps on Ty1, edits, 200 references
EditJumps editor, clock normalisation 40Jensen shannon divergence (per-position formulation chosen by the EditJumps authors, against a 200-sequence natural holdout)1.26 unitlessTy1 reference holdout, 200 homologsEditJumps on Ty1, JS pos., 200 references
EditJumps editor, clock normalisation 40Kullback leibler divergence (pooled amino-acid composition against a 200-sequence natural holdout)0.50 unitlessTy1 reference holdout, 200 homologsEditJumps on Ty1, KL pool., 200 references
EditJumps editor, clock normalisation 40Kullback leibler divergence (per-position amino-acid composition against a 200-sequence natural holdout)0.07 unitlessTy1 reference holdout, 200 homologsEditJumps on Ty1, KL pos., 200 references
EditJumps editor, clock normalisation 40Mutual information product agreement (mutual information product agreement against a 300-sequence natural homolog ceiling)0.83 unitlessTy1 reference holdout, 200 homologsEditJumps on Ty1, MIP agr., 200 references
EditJumps editor, clock normalisation 40Maximum mean discrepancy (k-mer spectrum MMD between the generated set and a 200-sequence natural holdout, unnormalised)0.98 unitlessTy1 reference holdout, 200 homologsEditJumps on Ty1, MMD, 200 references
EditJumps editor, clock normalisation 40Levenshtein distance (mean pairwise Levenshtein distance within the generated set, pooled over templates)24.47 countTy1 reference holdout, 200 homologsEditJumps on Ty1, pairw. Lev., 200 references
EditJumps editor, clock normalisation 40Log likelihood (profile log-likelihood of the generated sequences under the family alignment profile)-64.01 unitlessTy1 reference holdout, 200 homologsEditJumps on Ty1, prof. LL, 200 references
EditJumps editor, clock normalisation 40Entropy delta (positional entropy difference from the natural holdout; the target value is zero)0.07 unitlessTy1 reference holdout, 200 homologsEditJumps on Ty1, ΔH, 200 references
EvoDiff-MSA under a matched mutation budgetCovariance agreement (residue covariance agreement against a 300-sequence natural homolog ceiling)0.91 unitlessHER2-VH reference holdout, 200 homologsEvoDiff on HER2-VH, cov. agr., 200 references
EvoDiff-MSA under a matched mutation budgetEdit count (mean realised edits per generated sequence, 400 generated sequences per family)2.61 countHER2-VH reference holdout, 200 homologsEvoDiff on HER2-VH, edits, 200 references
EvoDiff-MSA under a matched mutation budgetJensen shannon divergence (per-position formulation chosen by the EditJumps authors, against a 200-sequence natural holdout)1.18 unitlessHER2-VH reference holdout, 200 homologsEvoDiff on HER2-VH, JS pos., 200 references
EvoDiff-MSA under a matched mutation budgetKullback leibler divergence (pooled amino-acid composition against a 200-sequence natural holdout)0.42 unitlessHER2-VH reference holdout, 200 homologsEvoDiff on HER2-VH, KL pool., 200 references
EvoDiff-MSA under a matched mutation budgetKullback leibler divergence (per-position amino-acid composition against a 200-sequence natural holdout)0.06 unitlessHER2-VH reference holdout, 200 homologsEvoDiff on HER2-VH, KL pos., 200 references
EvoDiff-MSA under a matched mutation budgetMutual information product agreement (mutual information product agreement against a 300-sequence natural homolog ceiling)0.82 unitlessHER2-VH reference holdout, 200 homologsEvoDiff on HER2-VH, MIP agr., 200 references
EvoDiff-MSA under a matched mutation budgetMaximum mean discrepancy (k-mer spectrum MMD between the generated set and a 200-sequence natural holdout, unnormalised)1.11 unitlessHER2-VH reference holdout, 200 homologsEvoDiff on HER2-VH, MMD, 200 references
EvoDiff-MSA under a matched mutation budgetLevenshtein distance (mean pairwise Levenshtein distance within the generated set, pooled over templates)23.23 countHER2-VH reference holdout, 200 homologsEvoDiff on HER2-VH, pairw. Lev., 200 references
EvoDiff-MSA under a matched mutation budgetLog likelihood (profile log-likelihood of the generated sequences under the family alignment profile)-55.03 unitlessHER2-VH reference holdout, 200 homologsEvoDiff on HER2-VH, prof. LL, 200 references
EvoDiff-MSA under a matched mutation budgetEntropy delta (positional entropy difference from the natural holdout; the target value is zero)0.04 unitlessHER2-VH reference holdout, 200 homologsEvoDiff on HER2-VH, ΔH, 200 references
EvoDiff-MSA under a matched mutation budgetCovariance agreement (residue covariance agreement against a 300-sequence natural homolog ceiling)0.89 unitlessTy1 reference holdout, 200 homologsEvoDiff on Ty1, cov. agr., 200 references
EvoDiff-MSA under a matched mutation budgetEdit count (mean realised edits per generated sequence, 400 generated sequences per family)2.96 countTy1 reference holdout, 200 homologsEvoDiff on Ty1, edits, 200 references
EvoDiff-MSA under a matched mutation budgetJensen shannon divergence (per-position formulation chosen by the EditJumps authors, against a 200-sequence natural holdout)1.38 unitlessTy1 reference holdout, 200 homologsEvoDiff on Ty1, JS pos., 200 references
EvoDiff-MSA under a matched mutation budgetKullback leibler divergence (pooled amino-acid composition against a 200-sequence natural holdout)0.61 unitlessTy1 reference holdout, 200 homologsEvoDiff on Ty1, KL pool., 200 references
EvoDiff-MSA under a matched mutation budgetKullback leibler divergence (per-position amino-acid composition against a 200-sequence natural holdout)0.07 unitlessTy1 reference holdout, 200 homologsEvoDiff on Ty1, KL pos., 200 references
EvoDiff-MSA under a matched mutation budgetMutual information product agreement (mutual information product agreement against a 300-sequence natural homolog ceiling)0.78 unitlessTy1 reference holdout, 200 homologsEvoDiff on Ty1, MIP agr., 200 references
EvoDiff-MSA under a matched mutation budgetMaximum mean discrepancy (k-mer spectrum MMD between the generated set and a 200-sequence natural holdout, unnormalised)0.97 unitlessTy1 reference holdout, 200 homologsEvoDiff on Ty1, MMD, 200 references
EvoDiff-MSA under a matched mutation budgetLevenshtein distance (mean pairwise Levenshtein distance within the generated set, pooled over templates)22.85 countTy1 reference holdout, 200 homologsEvoDiff on Ty1, pairw. Lev., 200 references
EvoDiff-MSA under a matched mutation budgetLog likelihood (profile log-likelihood of the generated sequences under the family alignment profile)-59.15 unitlessTy1 reference holdout, 200 homologsEvoDiff on Ty1, prof. LL, 200 references
EvoDiff-MSA under a matched mutation budgetEntropy delta (positional entropy difference from the natural holdout; the target value is zero)0.02 unitlessTy1 reference holdout, 200 homologsEvoDiff on Ty1, ΔH, 200 references
EvoDiff-MSA, published valuesLevenshtein distance (printed under the column heading 'diversity'; mean pairwise Levenshtein distance, under conditions the replication could not recover)93.06 countThe six evaluation datasets behind EvoFlows' published figuresEvoDiff-MSA on HER2-VH, diversity, EvoFlows published values
EvoDiff-MSA, published valuesEdit count (mean edits per sequence as the EvoFlows paper prints it, under conditions the replication could not recover)13.80 countThe six evaluation datasets behind EvoFlows' published figuresEvoDiff-MSA on HER2-VH, edits/seq, EvoFlows published values
EvoDiff-MSA, published valuesKullback leibler divergence (pooled amino-acid composition, under conditions the replication could not recover)10.16 unitlessThe six evaluation datasets behind EvoFlows' published figuresEvoDiff-MSA on HER2-VH, KL x10^3, EvoFlows published values
EvoDiff-MSA, published valuesMaximum mean discrepancy (k-mer spectrum MMD, under conditions and a reference set size the replication could not recover)0.66 unitlessThe six evaluation datasets behind EvoFlows' published figuresEvoDiff-MSA on HER2-VH, MMD, EvoFlows published values
EvoDiff-MSA, published valuesLevenshtein distance (printed under the column heading 'diversity'; mean pairwise Levenshtein distance, under conditions the replication could not recover)30.01 countThe six evaluation datasets behind EvoFlows' published figuresEvoDiff-MSA on Ty1, diversity, EvoFlows published values
EvoDiff-MSA, published valuesEdit count (mean edits per sequence as the EvoFlows paper prints it, under conditions the replication could not recover)4.43 countThe six evaluation datasets behind EvoFlows' published figuresEvoDiff-MSA on Ty1, edits/seq, EvoFlows published values

Claims

Statements this source supports, each with where in the source it was checked.

  • Agreement scores move with reference-set size alone: Reference set sensitivity Agreement metrics between generated and reference distributions exhibit severe sample-size dependency. Scoring real homologs against real homologs yields covariance scores of 0.790 at N = 20, 0.921 at N = 100 and saturates at 0.985 at N = 800, so the paper states that raw scores from evaluations with differing reference set sizes are mathematically non-comparable, and that an apparent agreement score of 0.79 falls entirely within the random sampling variance of finite sets. (Appendix E 'Sample-Size Dependency and Grounded Baseline Controls')
  • All evidence is developer-reported: Evidence status Every value in this batch was produced and reported by the developers of EditJumps. None of it has been independently reproduced by Rewire, no model was run and no number was recomputed here. (Section 4; Tables 1, 2 and 4; Appendix E)
  • All reported evaluation is computational: Measurement scope The paper states 'we note that our evaluation is strictly in silico'. Every reported number is a comparison between generated sequences and natural sequences: edit counts, pairwise and template Levenshtein distances, k-mer spectrum MMD, composition KL and Jensen-Shannon divergences, covariance and mutual information product agreement, entropy delta and profile or ESM-2 log-likelihood. No binding affinity, specificity, expression, stability or retained function is measured anywhere in the sources. (Section 1, the paragraph ending 'reflect single-domain constructs'; Table 2 column headings; section 5)
  • An undocumented clock normalisation sets the realised edit count: Undocumented hyperparameter The clock normalisation hyperparameter is omitted from the EvoFlows text and governs the realised edit budget. A sweep across both families and three seeds finds that a clock rate of approximately 95 replicates the published edit ratio of 0.43, while the reported evaluation uses 40 for a budget of about 5 edits, and a clock normalisation that optimises MMD sits around 80 or about 8.5 edits. Reconciling published and replicated edit distributions required reverse-engineering this parameter. The paper is not self-consistent about the replicating rate: the body gives approximately 95 while Figure 2's own annotation reads about 94, and neither value is reconciled with the other. (Abstract; section 2.3 'Length-invariant clock normalization'; section 4 'Sensitivity to the clock normalization hyperparameter'; Figure 2 and its caption; Table 5)
  • Best validation loss differs between the paper and the model card: Validation loss The model card gives a best validation loss of 130.48 at step 18,500 and marks the step with its own 'TODO: confirm step' comment, while Appendix D of the paper gives 135.2 at step 19,999 and no value below it. The two figures are not reconciled in either source. (Model card 'Model details', Best val loss row; paper Appendix D 'Training Protocol and Compute Resources')
  • Both upstream methods are one pure-jump process: Framing The paper identifies Edit Flows and EvoFlows as instances of the pure-jump case of generator matching with a common jump kernel, and states the two regularity conditions a variable-length space requires: a non-explosion condition for well-defined waiting times, which Edit Flows achieves with a hard length cap while EvoFlows leaves the space nominally unbounded, and a length-dependent clock normalisation to stop the realised jump count growing with candidate length. (Abstract; sections 1 and 2.3; section 7)
  • Canonical disulfides are retained in every numberable generation: Structural constraint retention EditJumps preserves the canonical structural disulfides at IMGT Cys23 and Cys104 in 100% of numberable generations, 393 of 393 on Ty1 and 389 of 389 on HER2-VH, against 89.0% and 93.8% under the random mutation baselines. (Appendix E 'Biophysical Analysis: IMGT Regions and Disulfide Retention')
  • Ceiling values depend on the alignment frame: Ceiling sensitivity Ceilings are highly sensitive to multiple sequence alignment projections: computing ceilings in an unaligned or mismatched frame shifts mutual information product agreement by 2.3 percentage points, so ceilings must be evaluated in identical coordinate frames. (Appendix E 'Ceiling Normalization and Alignment Frame Sensitivity')
  • Code and weights licence: Licence All code, model weights and evaluation pipelines for EditJumps are released under an MIT licence. The repository and the model repository serve byte-identical MIT licence files, and the model card states MIT for both the code and the weights. (Appendix D 'Operational Integrity and Artifact Verification'; repository LICENSE; model card 'License')
  • Covariance recovery moves with the reference partition: Reference set sensitivity Covariance recovery shifts from 83.6% to 85.1% simply by varying reference partitions, which the paper gives as a reason that ungrounded metrics risk rewarding sequence conservatism rather than genuine biophysical fitness. (Section 5, the paragraph beginning 'In computational protein engineering')
  • Euler sampling at 50 steps adds negligible discretisation error: Sampler Against an exact Gillespie mean of 71.29 plus or minus 0.15 edits on a fixed synthetic rate field over 2,000 trajectories from a sequence of length 120, Euler tau-leaping over-fires by 4.44 plus or minus 0.20 edits at 2 steps, 0.76 plus or minus 0.21 at 10 steps and 0.08 plus or minus 0.21 at 50 steps, the setting used throughout all experimental runs, a relative error under 0.3%. (Appendix B 'Empirical Discretization Analysis: Euler vs. Exact Gillespie')
  • High aggregate fidelity is dominated by conserved framework residues: Region resolved fidelity Partitioning antibody sequences by IMGT numbering puts 76% of residues in the structural framework and 24% in complementarity-determining regions. Within the CDR loops, which dictate antigen binding, EditJumps closes 18.7% on Ty1 and 23.9% on HER2-VH of the composition KL distance between random mutation and natural homologs, which the paper's Limitations section gives as under 25%. In framework positions all methods approach the natural floor, 0.0001 to 0.0011. The paper states fidelity is substantially driven by conserved framework scaffolds. (Section 5 'Limitations'; Appendix E 'Biophysical Analysis: IMGT Regions and Disulfide Retention')
  • More than half of the Ty1 reference set had been seen in training: Training overlap Evaluation references were 200 sequences per family drawn from held-out family members, but because the families were mined from the Observed Antibody Space, 113 of Ty1's and 103 of HER2-VH's reference sequences had been present in the training set. Re-evaluating with strictly disjoint reference sets shifted spectrum MMD by -36.7% on Ty1 and -12.6% on HER2-VH and increased covariance agreement by 12 to 17 points, driven by greater sequence homogeneity in the disjoint subset, 23.00 against 24.90 Levenshtein distance. (Appendix E 'Reference Construction and Overlap Impact')
  • Most unforced evotuning predictions leave the residue unchanged: Control Because the unforced masked language model frequently re-predicts the template residue, 71.7% unchanged on Ty1, the paper also evaluates a forced variant that may not return a mask to its previous value, isolating how much of evotuning's fidelity reflects generative editing rather than non-intervention. (Section 3 'The evotuning baseline')
  • Only two of the upstream targets could be evaluated: Evaluation targets Antibody-wide pretraining restricts downstream evaluation to the two accessible antibody targets of the EvoFlows paper: Ty1, an anti-SARS-CoV-2 VHH, and HER2-VH, an anti-HER2 heavy chain, each evaluated across 20 templates and 200 held-out homologs. Non-antibody targets have no homologs in the Observed Antibody Space and anti-EphA2 is unrecoverable in public repositories. (Section 3 'Data curation and evaluation targets')
  • Reference-set size can reverse the ordering of methods: Reference set sensitivity Cross-study benchmarking frequently normalises scores. The paper reports that this holds for relative edit distance but fails for distributional estimators such as spectrum MMD: expanding the reference alignment from 200 to 800 homologs alters the normalised ratio by 50% on Ty1 and 26% on HER2-VH, 'a difference sufficiently large to invert method rankings'. Because both the numerator and the baseline floor contract with sample size, internal normalisation cannot replace standardised reference sets. (Section 5, the paragraph beginning 'Furthermore, cross-study benchmarking frequently normalises scores')
  • Released checkpoint training corpus: Training data 1,660,105 unordered homolog pairs from the Observed Antibody Space paired corpus of 2,586,927 VH.VL sequences. Clusters were built with MMseqs2 easy-cluster at minimum sequence identity 0.5 and coverage 0.8 and capped at 20 pairs per family. (Model card 'Model details', Training data row; paper Appendix D 'Corpus Curation and Homolog Clustering')
  • Sequence recovery of 0.21 is 300-fold over chance: Grounded baseline An exact sequence recovery rate of 0.21 appears modest in isolation yet represents a 300-fold increase over chance at matched edit budgets for antibody leads, which the paper gives as its reason that in silico surrogates are easily misinterpreted without grounded empirical baselines. (Section 5, first paragraph; Appendix E 'Sample-Size Dependency and Grounded Baseline Controls')
  • Six of the ten upstream metrics are not reproducible from the text: Specification completeness Of the ten evaluation panels in Figure 3 of the EvoFlows paper, six cannot be deterministically recomputed from the text alone. Four are well specified, one of them only modulo an unstated MMD kernel parameter. Of the remaining six, three are referenced solely by name but have standard mathematical formalisation, two define pairwise matrices without specifying scalar reductions, and one admits multiple incompatible reductions. Table 3 lists each panel and what is omitted. (Section 4 'Identifying EvoFlows' 10 evaluation metrics'; Table 3)
  • Symmetric pair enumeration replicated without discrepancy: Verification The EvoFlows paper states that training pairs are formed by unordered enumeration of homolog pairs. In a random sample of 2,000 training pairs, 44.9% showed a higher property score for the second sequence, and 44.3% in an independent 4,000-pair audit, consistent with unbiased symmetric enumeration at about 50% and contrasting with oriented pairing at 100%. The paper states this protocol replicated cleanly without discrepancy. (Appendix D 'Algorithmic Verification: Symmetric Pair Enumeration'; section 5)
  • The authors' statement on AI assistance: Ai use The paper states that LLM-based assistance was used across code development, data analysis and manuscript preparation under direct author supervision, that all code implementations, algorithmic workflows and metric definitions were audited, verified and executed under the authors' direct oversight, and that all empirical claims and recovered data points were independently validated against raw artifacts by the authors. ('AI Use Statement', all three paragraphs)
  • The edit budget holds in expectation, not per sequence: Edit budget The edit budget is calibrated on average rather than guaranteed per sequence. The paper states that clock normalisation has no closed-form mapping to realised edits, because jump rates vary with time and sequence context, so achieving a specific edit budget requires empirical calibration rather than deterministic control. The repository states the --edits parameter 'provides approximate guidance via clock calibration' and that realised counts vary per sequence, and the README states '--edits is a target, not a guarantee'. (Section 5 'Limitations'; docs/limitations.md, third bullet; README 'Quick Start' step 5)
  • The published comparison values were recovered from a figure, not a table: Value provenance Because the EvoFlows paper reported quantitative results exclusively in a figure without tabulated values, the numerical markers were recovered by extracting vector paths from the published PDF and calibrating against axis tick marks, with the six series resolved by tick pitch, by domain endpoints and against a legend printed in a later figure. (Appendix E 'Vector Graphic Extraction and Method Resolution')
  • The scored arm and the released weights state different head parameterisations: Head Section 3 of the paper lists 'linear rather than MLP rate heads' as one of four operational adaptations of the scored arm, while the model card's Heads row gives three MLP rate heads and its quickstart loads the released weights with rate_head mlp and q_head esm_lm_head. Appendix E reports the two parameterisations give different edit ratios, 0.19 for the published Appendix A heads against 0.22 to 0.28 for lightweight linear heads, so they are not interchangeable. (Section 3 'Supervision discrepancies and deviations'; Figure 3 caption; Appendix E 'Non-Explanations for the Edit Budget Discrepancy'; model card 'Model details', Heads row and the quickstart command)
  • The two sets of numbers are not comparable: Comparability The paper states, before interpreting Table 1: 'our computed values and published EvoFlows values are not comparable and we could not recover the original experimental conditions from the paper'. Table 1's caption states the two blocks are not comparable and that each row compares against its own block's floor and ceiling. The paper also states its released artefacts 'are a specified baseline rather than a match to the published values, which the reported conditions do not allow us to recover'. (Section 4, first sentence; Table 1 caption; section 1, contribution (ii))

Models, benchmarks and other records

Methods (5)

Configurations (13)

Tasks (1)

Protocols (26)

Datasets (6)

Dataset subsets (11)

Evaluations (175)

Sources (8)

Evidence

Source checking verifies the cited claim or transcription. It does not establish independent reproduction.

Evidence table

Inspect claims, sources and review details

Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.

One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.

31 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-11-21-c8bea25e5abb
Property and statementOriginal source and locationReview and provenance
Artifact SHA-256
4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1
Source metadata
When Edit Flows are Edit Jumps: replicating Edit Flows and EvoFlows (arXiv v1)

Original source ↗

No field-specific location recorded

Version: arXiv:2609.18745v1, 16 September 2026
Retrieved: 2026-11-16T00:03:00Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.artifact_sha256

Source artifact SHA-256: 4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Artifact url
https://arxiv.org/pdf/2609.18745v1
Source metadata
When Edit Flows are Edit Jumps: replicating Edit Flows and EvoFlows (arXiv v1)

Original source ↗

No field-specific location recorded

Version: arXiv:2609.18745v1, 16 September 2026
Retrieved: 2026-11-16T00:03:00Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.artifact_url

Source artifact SHA-256: 4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Evidence concerns
1 values
  • applies to
    • editjumps-20261115-config-published-random-pairing
    • editjumps-20261115-eval-published-edits-random-pairing-ty1
    • editjumps-20261115-result-published-edits-random-pairing-ty1
    • editjumps-20261115-eval-published-edits-random-pairing-her2vh
    • editjumps-20261115-result-published-edits-random-pairing-her2vh
    • editjumps-20261115-eval-published-pairwise-levenshtein-random-pairing-ty1
    • editjumps-20261115-result-published-pairwise-levenshtein-random-pairing-ty1
    • editjumps-20261115-eval-published-pairwise-levenshtein-random-pairing-her2vh
    • editjumps-20261115-result-published-pairwise-levenshtein-random-pairing-her2vh
    • editjumps-20261115-eval-published-spectrum-mmd-random-pairing-ty1
    • editjumps-20261115-result-published-spectrum-mmd-random-pairing-ty1
    • editjumps-20261115-eval-published-spectrum-mmd-random-pairing-her2vh
    • editjumps-20261115-result-published-spectrum-mmd-random-pairing-her2vh
    • editjumps-20261115-eval-published-kl-pooled-random-pairing-ty1
    • editjumps-20261115-result-published-kl-pooled-random-pairing-ty1
    • editjumps-20261115-eval-published-kl-pooled-random-pairing-her2vh
    • editjumps-20261115-result-published-kl-pooled-random-pairing-her2vh
    • editjumps-20261115-config-published-evoflows
    • editjumps-20261115-eval-published-edits-evoflows-ty1
    • editjumps-20261115-result-published-edits-evoflows-ty1
    • editjumps-20261115-eval-published-edits-evoflows-her2vh
    • editjumps-20261115-result-published-edits-evoflows-her2vh
    • editjumps-20261115-eval-published-pairwise-levenshtein-evoflows-ty1
    • editjumps-20261115-result-published-pairwise-levenshtein-evoflows-ty1
    • editjumps-20261115-eval-published-pairwise-levenshtein-evoflows-her2vh
    • editjumps-20261115-result-published-pairwise-levenshtein-evoflows-her2vh
    • editjumps-20261115-eval-published-spectrum-mmd-evoflows-ty1
    • editjumps-20261115-result-published-spectrum-mmd-evoflows-ty1
    • editjumps-20261115-eval-published-spectrum-mmd-evoflows-her2vh
    • editjumps-20261115-result-published-spectrum-mmd-evoflows-her2vh
    • editjumps-20261115-eval-published-kl-pooled-evoflows-ty1
    • editjumps-20261115-result-published-kl-pooled-evoflows-ty1
    • editjumps-20261115-eval-published-kl-pooled-evoflows-her2vh
    • editjumps-20261115-result-published-kl-pooled-evoflows-her2vh
    • editjumps-20261115-config-published-evotuning
    • editjumps-20261115-eval-published-edits-evotuning-ty1
    • editjumps-20261115-result-published-edits-evotuning-ty1
    • editjumps-20261115-eval-published-edits-evotuning-her2vh
    • editjumps-20261115-result-published-edits-evotuning-her2vh
    • editjumps-20261115-eval-published-pairwise-levenshtein-evotuning-ty1
    • editjumps-20261115-result-published-pairwise-levenshtein-evotuning-ty1
    • editjumps-20261115-eval-published-pairwise-levenshtein-evotuning-her2vh
    • editjumps-20261115-result-published-pairwise-levenshtein-evotuning-her2vh
    • editjumps-20261115-eval-published-spectrum-mmd-evotuning-ty1
    • editjumps-20261115-result-published-spectrum-mmd-evotuning-ty1
    • editjumps-20261115-eval-published-spectrum-mmd-evotuning-her2vh
    • editjumps-20261115-result-published-spectrum-mmd-evotuning-her2vh
    • editjumps-20261115-eval-published-kl-pooled-evotuning-ty1
    • editjumps-20261115-result-published-kl-pooled-evotuning-ty1
    • editjumps-20261115-eval-published-kl-pooled-evotuning-her2vh
    • editjumps-20261115-result-published-kl-pooled-evotuning-her2vh
    • editjumps-20261115-config-published-evotuning-forced
    • editjumps-20261115-eval-published-edits-evotuning-forced-ty1
    • editjumps-20261115-result-published-edits-evotuning-forced-ty1
    • editjumps-20261115-eval-published-edits-evotuning-forced-her2vh
    • editjumps-20261115-result-published-edits-evotuning-forced-her2vh
    • editjumps-20261115-eval-published-pairwise-levenshtein-evotuning-forced-ty1
    • editjumps-20261115-result-published-pairwise-levenshtein-evotuning-forced-ty1
    • editjumps-20261115-eval-published-pairwise-levenshtein-evotuning-forced-her2vh
    • editjumps-20261115-result-published-pairwise-levenshtein-evotuning-forced-her2vh
    • editjumps-20261115-eval-published-spectrum-mmd-evotuning-forced-ty1
    • editjumps-20261115-result-published-spectrum-mmd-evotuning-forced-ty1
    • editjumps-20261115-eval-published-spectrum-mmd-evotuning-forced-her2vh
    • editjumps-20261115-result-published-spectrum-mmd-evotuning-forced-her2vh
    • editjumps-20261115-eval-published-kl-pooled-evotuning-forced-ty1
    • editjumps-20261115-result-published-kl-pooled-evotuning-forced-ty1
    • editjumps-20261115-eval-published-kl-pooled-evotuning-forced-her2vh
    • editjumps-20261115-result-published-kl-pooled-evotuning-forced-her2vh
    • editjumps-20261115-config-published-evodiff-msa
    • editjumps-20261115-eval-published-edits-evodiff-msa-ty1
    • editjumps-20261115-result-published-edits-evodiff-msa-ty1
    • editjumps-20261115-eval-published-edits-evodiff-msa-her2vh
    • editjumps-20261115-result-published-edits-evodiff-msa-her2vh
    • editjumps-20261115-eval-published-pairwise-levenshtein-evodiff-msa-ty1
    • editjumps-20261115-result-published-pairwise-levenshtein-evodiff-msa-ty1
    • editjumps-20261115-eval-published-pairwise-levenshtein-evodiff-msa-her2vh
    • editjumps-20261115-result-published-pairwise-levenshtein-evodiff-msa-her2vh
    • editjumps-20261115-eval-published-spectrum-mmd-evodiff-msa-ty1
    • editjumps-20261115-result-published-spectrum-mmd-evodiff-msa-ty1
    • editjumps-20261115-eval-published-spectrum-mmd-evodiff-msa-her2vh
    • editjumps-20261115-result-published-spectrum-mmd-evodiff-msa-her2vh
    • editjumps-20261115-eval-published-kl-pooled-evodiff-msa-ty1
    • editjumps-20261115-result-published-kl-pooled-evodiff-msa-ty1
    • editjumps-20261115-eval-published-kl-pooled-evodiff-msa-her2vh
    • editjumps-20261115-result-published-kl-pooled-evodiff-msa-her2vh
    • editjumps-20261115-config-published-random-mutations
    • editjumps-20261115-eval-published-edits-random-mutations-ty1
    • editjumps-20261115-result-published-edits-random-mutations-ty1
    • editjumps-20261115-eval-published-edits-random-mutations-her2vh
    • editjumps-20261115-result-published-edits-random-mutations-her2vh
    • editjumps-20261115-eval-published-pairwise-levenshtein-random-mutations-ty1
    • editjumps-20261115-result-published-pairwise-levenshtein-random-mutations-ty1
    • editjumps-20261115-eval-published-pairwise-levenshtein-random-mutations-her2vh
    • editjumps-20261115-result-published-pairwise-levenshtein-random-mutations-her2vh
    • editjumps-20261115-eval-published-spectrum-mmd-random-mutations-ty1
    • editjumps-20261115-result-published-spectrum-mmd-random-mutations-ty1
    • editjumps-20261115-eval-published-spectrum-mmd-random-mutations-her2vh
    • editjumps-20261115-result-published-spectrum-mmd-random-mutations-her2vh
    • editjumps-20261115-eval-published-kl-pooled-random-mutations-ty1
    • editjumps-20261115-result-published-kl-pooled-random-mutations-ty1
    • editjumps-20261115-eval-published-kl-pooled-random-mutations-her2vh
    • editjumps-20261115-result-published-kl-pooled-random-mutations-her2vh
    • editjumps-20261115-protocol-published-edits
    • editjumps-20261115-protocol-published-pairwise-levenshtein
    • editjumps-20261115-protocol-published-spectrum-mmd
    • editjumps-20261115-protocol-published-kl-pooled
    • editjumps-20261115-eval-ceiling-published-covariance-evoflows
    • editjumps-20261115-result-ceiling-published-covariance-evoflows
    • editjumps-20261115-protocol-ceiling-published-covariance
    • editjumps-20261115-eval-ceiling-published-mip-evoflows
    • editjumps-20261115-result-ceiling-published-mip-evoflows
    • editjumps-20261115-protocol-ceiling-published-mip
    • editjumps-20261115-eval-ceiling-published-js-positional-evoflows
    • editjumps-20261115-result-ceiling-published-js-positional-evoflows
    • editjumps-20261115-protocol-ceiling-published-js-positional
    • editjumps-20261115-eval-ceiling-published-pairwise-distance-evoflows
    • editjumps-20261115-result-ceiling-published-pairwise-distance-evoflows
    • editjumps-20261115-protocol-ceiling-published-pairwise-distance
    artifact sha256
    4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1
    message
    The paper states in section 4 that its computed values and the published EvoFlows values are not comparable and that it could not recover the original experimental conditions from the EvoFlows paper, and Table 1's caption states that the two blocks are not comparable. The records carrying EvoFlows' own published values are therefore withheld from automatic comparison with the reimplemented block, and neither set may be pooled into one leaderboard.
    review date
    2026-11-16
    source id
    editjumps-20261115-source-arxiv-pdf-v1
    source locator
    Section 4, first sentence; Table 1 caption; Table 4 caption; Table 5
Source metadata
When Edit Flows are Edit Jumps: replicating Edit Flows and EvoFlows (arXiv v1)

Original source ↗

No field-specific location recorded

Version: arXiv:2609.18745v1, 16 September 2026
Retrieved: 2026-11-16T00:03:00Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.evidence_concerns

Source artifact SHA-256: 4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Limitations
4 values
  • A preprint; the repository badges name the NeurIPS 2026 AIDaR workshop, and no peer-reviewed version was found.
  • Figures 1, 2, 3 and 4 carry no printed numeric markers for their series, so the only values taken from them are the ones the captions and the body print in words.
  • Table 2 prints two arms, BF-DiT and BF-ESM, that the paper, the repository and the model card never define, so the system behind those 40 cells cannot be identified.
  • Table 1 heads the pairwise Levenshtein column 'diversity', while Figure 1's caption defines diversity as a ratio to real homolog pairs whose values run from about 1.0 to 1.4.
Source metadata
When Edit Flows are Edit Jumps: replicating Edit Flows and EvoFlows (arXiv v1)

Original source ↗

No field-specific location recorded

Version: arXiv:2609.18745v1, 16 September 2026
Retrieved: 2026-11-16T00:03:00Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.limitations

Source artifact SHA-256: 4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Media type
application/pdf
Source metadata
When Edit Flows are Edit Jumps: replicating Edit Flows and EvoFlows (arXiv v1)

Original source ↗

No field-specific location recorded

Version: arXiv:2609.18745v1, 16 September 2026
Retrieved: 2026-11-16T00:03:00Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.media_type

Source artifact SHA-256: 4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Missing metadata: doi: note
No DOI is printed in the preprint.
Source metadata
When Edit Flows are Edit Jumps: replicating Edit Flows and EvoFlows (arXiv v1)

Original source ↗

No field-specific location recorded

Version: arXiv:2609.18745v1, 16 September 2026
Retrieved: 2026-11-16T00:03:00Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.missing_metadata.doi.note

Source artifact SHA-256: 4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Missing metadata: doi: reason
unreported
Source metadata
When Edit Flows are Edit Jumps: replicating Edit Flows and EvoFlows (arXiv v1)

Original source ↗

No field-specific location recorded

Version: arXiv:2609.18745v1, 16 September 2026
Retrieved: 2026-11-16T00:03:00Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.missing_metadata.doi.reason

Source artifact SHA-256: 4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Missing metadata: licence: note
No licence statement appears in the PDF; the arXiv abstract page was not pinned as a source.
Source metadata
When Edit Flows are Edit Jumps: replicating Edit Flows and EvoFlows (arXiv v1)

Original source ↗

No field-specific location recorded

Version: arXiv:2609.18745v1, 16 September 2026
Retrieved: 2026-11-16T00:03:00Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.missing_metadata.licence.note

Source artifact SHA-256: 4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Missing metadata: licence: reason
unreported
Source metadata
When Edit Flows are Edit Jumps: replicating Edit Flows and EvoFlows (arXiv v1)

Original source ↗

No field-specific location recorded

Version: arXiv:2609.18745v1, 16 September 2026
Retrieved: 2026-11-16T00:03:00Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.missing_metadata.licence.reason

Source artifact SHA-256: 4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Publication status
preprint
Source metadata
When Edit Flows are Edit Jumps: replicating Edit Flows and EvoFlows (arXiv v1)

Original source ↗

No field-specific location recorded

Version: arXiv:2609.18745v1, 16 September 2026
Retrieved: 2026-11-16T00:03:00Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.publication_status

Source artifact SHA-256: 4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

How to read the evidence

Sources and history

Release 2026-11-21-c8bea25e5abb · Record review: source checked

The EditJumps paper: the preprint that reformulates Edit Flows and EvoFlows as a pure-jump process, describes the released editor and reports its evaluation against reimplemented baselines.

1 source record and release history

Read original source

Download this release (gzip)
Technical metadata and extraction receipts

Stable ID: editjumps-20261115-source-arxiv-pdf-v1

areas
protein-function; proteins-complexes
contexts
research
artifact sha256
4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1
artifact url
https://arxiv.org/pdf/2609.18745v1
evidence concerns
applies to: editjumps-20261115-config-published-random-pairing; editjumps-20261115-eval-published-edits-random-pairing-ty1; editjumps-20261115-result-published-edits-random-pairing-ty1; editjumps-20261115-eval-published-edits-random-pairing-her2vh; editjumps-20261115-result-published-edits-random-pairing-her2vh; editjumps-20261115-eval-published-pairwise-levenshtein-random-pairing-ty1; editjumps-20261115-result-published-pairwise-levenshtein-random-pairing-ty1; editjumps-20261115-eval-published-pairwise-levenshtein-random-pairing-her2vh; editjumps-20261115-result-published-pairwise-levenshtein-random-pairing-her2vh; editjumps-20261115-eval-published-spectrum-mmd-random-pairing-ty1; editjumps-20261115-result-published-spectrum-mmd-random-pairing-ty1; editjumps-20261115-eval-published-spectrum-mmd-random-pairing-her2vh; editjumps-20261115-result-published-spectrum-mmd-random-pairing-her2vh; editjumps-20261115-eval-published-kl-pooled-random-pairing-ty1; editjumps-20261115-result-published-kl-pooled-random-pairing-ty1; editjumps-20261115-eval-published-kl-pooled-random-pairing-her2vh; editjumps-20261115-result-published-kl-pooled-random-pairing-her2vh; editjumps-20261115-config-published-evoflows; editjumps-20261115-eval-published-edits-evoflows-ty1; editjumps-20261115-result-published-edits-evoflows-ty1; editjumps-20261115-eval-published-edits-evoflows-her2vh; editjumps-20261115-result-published-edits-evoflows-her2vh; editjumps-20261115-eval-published-pairwise-levenshtein-evoflows-ty1; editjumps-20261115-result-published-pairwise-levenshtein-evoflows-ty1; editjumps-20261115-eval-published-pairwise-levenshtein-evoflows-her2vh; editjumps-20261115-result-published-pairwise-levenshtein-evoflows-her2vh; editjumps-20261115-eval-published-spectrum-mmd-evoflows-ty1; editjumps-20261115-result-published-spectrum-mmd-evoflows-ty1; editjumps-20261115-eval-published-spectrum-mmd-evoflows-her2vh; editjumps-20261115-result-published-spectrum-mmd-evoflows-her2vh; editjumps-20261115-eval-published-kl-pooled-evoflows-ty1; editjumps-20261115-result-published-kl-pooled-evoflows-ty1; editjumps-20261115-eval-published-kl-pooled-evoflows-her2vh; editjumps-20261115-result-published-kl-pooled-evoflows-her2vh; editjumps-20261115-config-published-evotuning; editjumps-20261115-eval-published-edits-evotuning-ty1; editjumps-20261115-result-published-edits-evotuning-ty1; editjumps-20261115-eval-published-edits-evotuning-her2vh; editjumps-20261115-result-published-edits-evotuning-her2vh; editjumps-20261115-eval-published-pairwise-levenshtein-evotuning-ty1; editjumps-20261115-result-published-pairwise-levenshtein-evotuning-ty1; editjumps-20261115-eval-published-pairwise-levenshtein-evotuning-her2vh; editjumps-20261115-result-published-pairwise-levenshtein-evotuning-her2vh; editjumps-20261115-eval-published-spectrum-mmd-evotuning-ty1; editjumps-20261115-result-published-spectrum-mmd-evotuning-ty1; editjumps-20261115-eval-published-spectrum-mmd-evotuning-her2vh; editjumps-20261115-result-published-spectrum-mmd-evotuning-her2vh; editjumps-20261115-eval-published-kl-pooled-evotuning-ty1; editjumps-20261115-result-published-kl-pooled-evotuning-ty1; editjumps-20261115-eval-published-kl-pooled-evotuning-her2vh; editjumps-20261115-result-published-kl-pooled-evotuning-her2vh; editjumps-20261115-config-published-evotuning-forced; editjumps-20261115-eval-published-edits-evotuning-forced-ty1; editjumps-20261115-result-published-edits-evotuning-forced-ty1; editjumps-20261115-eval-published-edits-evotuning-forced-her2vh; editjumps-20261115-result-published-edits-evotuning-forced-her2vh; editjumps-20261115-eval-published-pairwise-levenshtein-evotuning-forced-ty1; editjumps-20261115-result-published-pairwise-levenshtein-evotuning-forced-ty1; editjumps-20261115-eval-published-pairwise-levenshtein-evotuning-forced-her2vh; editjumps-20261115-result-published-pairwise-levenshtein-evotuning-forced-her2vh; editjumps-20261115-eval-published-spectrum-mmd-evotuning-forced-ty1; editjumps-20261115-result-published-spectrum-mmd-evotuning-forced-ty1; editjumps-20261115-eval-published-spectrum-mmd-evotuning-forced-her2vh; editjumps-20261115-result-published-spectrum-mmd-evotuning-forced-her2vh; editjumps-20261115-eval-published-kl-pooled-evotuning-forced-ty1; editjumps-20261115-result-published-kl-pooled-evotuning-forced-ty1; editjumps-20261115-eval-published-kl-pooled-evotuning-forced-her2vh; editjumps-20261115-result-published-kl-pooled-evotuning-forced-her2vh; editjumps-20261115-config-published-evodiff-msa; editjumps-20261115-eval-published-edits-evodiff-msa-ty1; editjumps-20261115-result-published-edits-evodiff-msa-ty1; editjumps-20261115-eval-published-edits-evodiff-msa-her2vh; editjumps-20261115-result-published-edits-evodiff-msa-her2vh; editjumps-20261115-eval-published-pairwise-levenshtein-evodiff-msa-ty1; editjumps-20261115-result-published-pairwise-levenshtein-evodiff-msa-ty1; editjumps-20261115-eval-published-pairwise-levenshtein-evodiff-msa-her2vh; editjumps-20261115-result-published-pairwise-levenshtein-evodiff-msa-her2vh; editjumps-20261115-eval-published-spectrum-mmd-evodiff-msa-ty1; editjumps-20261115-result-published-spectrum-mmd-evodiff-msa-ty1; editjumps-20261115-eval-published-spectrum-mmd-evodiff-msa-her2vh; editjumps-20261115-result-published-spectrum-mmd-evodiff-msa-her2vh; editjumps-20261115-eval-published-kl-pooled-evodiff-msa-ty1; editjumps-20261115-result-published-kl-pooled-evodiff-msa-ty1; editjumps-20261115-eval-published-kl-pooled-evodiff-msa-her2vh; editjumps-20261115-result-published-kl-pooled-evodiff-msa-her2vh; editjumps-20261115-config-published-random-mutations; editjumps-20261115-eval-published-edits-random-mutations-ty1; editjumps-20261115-result-published-edits-random-mutations-ty1; editjumps-20261115-eval-published-edits-random-mutations-her2vh; editjumps-20261115-result-published-edits-random-mutations-her2vh; editjumps-20261115-eval-published-pairwise-levenshtein-random-mutations-ty1; editjumps-20261115-result-published-pairwise-levenshtein-random-mutations-ty1; editjumps-20261115-eval-published-pairwise-levenshtein-random-mutations-her2vh; editjumps-20261115-result-published-pairwise-levenshtein-random-mutations-her2vh; editjumps-20261115-eval-published-spectrum-mmd-random-mutations-ty1; editjumps-20261115-result-published-spectrum-mmd-random-mutations-ty1; editjumps-20261115-eval-published-spectrum-mmd-random-mutations-her2vh; editjumps-20261115-result-published-spectrum-mmd-random-mutations-her2vh; editjumps-20261115-eval-published-kl-pooled-random-mutations-ty1; editjumps-20261115-result-published-kl-pooled-random-mutations-ty1; editjumps-20261115-eval-published-kl-pooled-random-mutations-her2vh; editjumps-20261115-result-published-kl-pooled-random-mutations-her2vh; editjumps-20261115-protocol-published-edits; editjumps-20261115-protocol-published-pairwise-levenshtein; editjumps-20261115-protocol-published-spectrum-mmd; editjumps-20261115-protocol-published-kl-pooled; editjumps-20261115-eval-ceiling-published-covariance-evoflows; editjumps-20261115-result-ceiling-published-covariance-evoflows; editjumps-20261115-protocol-ceiling-published-covariance; editjumps-20261115-eval-ceiling-published-mip-evoflows; editjumps-20261115-result-ceiling-published-mip-evoflows; editjumps-20261115-protocol-ceiling-published-mip; editjumps-20261115-eval-ceiling-published-js-positional-evoflows; editjumps-20261115-result-ceiling-published-js-positional-evoflows; editjumps-20261115-protocol-ceiling-published-js-positional; editjumps-20261115-eval-ceiling-published-pairwise-distance-evoflows; editjumps-20261115-result-ceiling-published-pairwise-distance-evoflows; editjumps-20261115-protocol-ceiling-published-pairwise-distance; artifact sha256: 4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1; message: The paper states in section 4 that its computed values and the published EvoFlows values are not comparable and that it could not recover the original experimental conditions from the EvoFlows paper, and Table 1's caption states that the two blocks are not comparable. The records carrying EvoFlows' own published values are therefore withheld from automatic comparison with the reimplemented block, and neither set may be pooled into one leaderboard.; review date: 2026-11-16; source id: editjumps-20261115-source-arxiv-pdf-v1; source locator: Section 4, first sentence; Table 1 caption; Table 4 caption; Table 5
limitations
A preprint; the repository badges name the NeurIPS 2026 AIDaR workshop, and no peer-reviewed version was found.; Figures 1, 2, 3 and 4 carry no printed numeric markers for their series, so the only values taken from them are the ones the captions and the body print in words.; Table 2 prints two arms, BF-DiT and BF-ESM, that the paper, the repository and the model card never define, so the system behind those 40 cells cannot be identified.; Table 1 heads the pairwise Levenshtein column 'diversity', while Figure 1's caption defines diversity as a ratio to real homolog pairs whose values run from about 1.0 to 1.4.
media type
application/pdf
missing metadata
doi: note: No DOI is printed in the preprint.; reason: unreported; licence: note: No licence statement appears in the PDF; the arXiv abstract page was not pinned as a source.; reason: unreported
publication status
preprint
retrieval note
Downloaded twice from the arXiv version 1 PDF endpoint with identical bytes. The text layer used for the value assertions is pdftotext -layout output, poppler 26.08.0, SHA-256 08fa7d4b91c13ddcd5531b66cdec3629427ad1ef9ceef047d785bf1c408617f5.
retrieved at
2026-11-16T00:03:00Z
review
artifact sha256: 4211d8ef46c55efe8d89ba4ee6c0b4e57cca1e78d0856e9cdfbde37046a7f4e1; method: ai-assisted-source-review; source-hash-verification; method note: Independent review pass. Every pinned artifact was re-downloaded from the URL in its source record and its SHA-256 recomputed: all nine match. The arXiv PDF's text layer was regenerated with pdftotext -layout from poppler 26.08.0 and matches the hash notes.md records. npm run batch -- check-values over the whole batch reproduced the extractor's figures exactly (8 in_cell, 164 in_row, 3 locator_not_parsed, no wrong_column and no overclaimed_in_row). Because 164 results sit at in_row only, column assignment was then checked separately by review/column_check.py, written for this pass and using neither extract/tables.py nor extract/check_tables.py. It builds two independent column models: the PDF text layer, where Tables 2 and 4 are assigned by clustering printed field right edges and Table 1 by ordinal position within the row, and the arXiv HTML rendering parsed as a cell grid from the table markup. The two models agree on every cell, and all 172 table-located results match their locator's column and block in both. Four negative controls, two columns exchanged within a row, a row shifted one column left, a Ty1 and HER2 sub-column exchanged, and the Table 1 upper and lower blocks exchanged for one row label, are rejected by both models.; note: Re-retrieved from the URL in this record and the SHA-256 recomputed; it matches. The repository files were also checked against a tarball of commit d827d47c3bcd29e0f4c15a15f9ac92a6cd859e99, in which they are byte-identical.; retrieval url: https://arxiv.org/pdf/2609.18745v1; reviewed at: 2026-11-16T01:30:00Z; reviewer: claude; reviewer note: Claude (Opus 5) review agent for issue #244, independent of the extraction agent; no human review claimed. All nine pinned artifacts were re-retrieved and their SHA-256 re-computed; no value was reproduced and no model was run.
source locator
17 pages; Tables 1 to 5 and Figures 1 to 4
url
https://arxiv.org/abs/2609.18745
version
arXiv:2609.18745v1, 16 September 2026
Related records

Suggest a correction