CIKM 2026 · Rome, Italy · Nov 7–11

Temporal fidelity is not implied by static fidelity.

Seq2Synth is a taxonomy-guided benchmark for synthetic sequential tabular data — evaluating the time axis itself, not just the pooled rows.

Kiwan Kwon*, Kangmin Kim*, Hojin Lee, Yeseong Jung, Hyeongwoo Kong, Vamsi K. Potluru, Saerom Park†, Yongjae Lee† UNIST · HUFS · Independent Researcher · LinqAlpha  |  * equal contribution · † corresponding authors

13datasets · 6 domains
8generators
4fidelity dimensions
27metrics

What static metrics miss

REAL t₀t₁t₂ t₃t₄t₅ t₆t₇ SYNTHETIC duplicate missing out of range
Real Synthetic Identical marginals · broken time axis
About the benchmark

Generators that match every marginal can still break the clock.

Sequential tabular data — transaction logs, patient visits, loan performance records, sensor streams — is the dominant format in high-stakes domains. Yet single-table benchmarks treat records as independent samples, and relational benchmarks measure inter-table dependencies while staying agnostic to temporal ordering and population dynamics. Both collapse time into a static distribution.

The consequence is measurable. A state-of-the-art relational generator can reach near-perfect marginal fidelity while emitting duplicate timestamps, misordered records, and incomplete observation grids — properties that are fixed by construction in the real data, so there is no ambiguity about the correct value. Existing protocols report none of it.

Seq2Synth replaces the fixed metric suite with a taxonomy-guided protocol. Each dataset is classified along four axes — time representation, sampling regularity, trajectory dependence, and schema structure — and that classification determines which temporal metrics are meaningful to compute at all.

On top of the four fidelity dimensions, we extend standard utility and privacy protocols to the trajectory level: downstream prediction that consumes history, and leakage that shows up as reproduced sub-sequences rather than matching rows.

TS

Timestamp fidelity

Is the generated time axis valid at all — before any repair or post-processing?

  • Order · Uniqueness · Range
  • Regularity · Grid completeness
  • Interval & duration similarity
CS

Cross-sectional fidelity

Does the population distribution hold at each aligned time slice, not only after pooling?

  • Per-slice numeric & categorical marginals
  • Statistics · range · category coverage
  • Correlation & contingency per slice
LG

Longitudinal fidelity

Do within-entity dynamics survive — step by step and across the whole trajectory?

  • First differences · state transitions
  • Autocorrelation similarity
  • MLD-TS · TT-Wasserstein
ST

Structural fidelity

Are relational cardinalities preserved as they evolve over time?

  • Sequence-length similarity
  • Temporal cardinality shape
  • Dynamic k-hop correlation
Evaluation protocol

Four ordered steps, run identically for every dataset.

Given a real relational dataset 𝒯 and a synthetic counterpart 𝒯̂, the benchmark decides what is measurable before it measures anything. The order matters: timestamp metrics are computed on raw generator output so that validity violations cannot be hidden by repair.

STEP 01

Classify the dataset

Assign it along the four taxonomy axes: absolute or relative time, regular or irregular sampling, dependent or isolated trajectories, and schema type.

STEP 02

Select applicable dimensions

TS and LG apply broadly. CS requires absolute time and dependent trajectories. ST depends on the schema, with extra metrics for multi-child and multi-parent.

STEP 03

Prepare metric inputs

Timestamp metrics use raw output. Everything else runs after a post-processing pass: records sorted, timestamps aligned to the grid, validity corrections applied.

STEP 04

Report dimension scores

Each dimension score is the unweighted mean of its applicable sub-metrics, so no a priori importance ordering is imposed. Sub-metric profiles carry the diagnosis.

Score conventions: every metric lands in [0, 1] and higher is better, except TT-WassersteinDistance, DCR, and CS-DCR, which are reported as raw distances. Static-distribution (SD) and TS use raw outputs; CS and ST use the sparse variant; LG uses interpolation-fill where complete trajectories are required.

Leaderboards

Per-dataset fidelity across five evaluation dimensions.

Seven core datasets, each supported by at least six of the seven broadly applicable baselines. Pick a dataset, then sort by any dimension — the rank column follows your sort. Higher is better throughout; “—” marks a dimension that does not apply to that dataset or a model that does not support it.

# Model SD TS CS LG ST
Table 3 of the paper. Blue fill marks the best score in a column, underline the second best. Ranking is by the sorted column only — the dimensions are diagnostic, not additive, and the paper does not define a single aggregate score.

Finding 1 · Temporal-aware dimensions are not interchangeable with static evaluation

Across the seven core datasets, rankings induced by timestamp, longitudinal, and structural fidelity diverge sharply from the SD-based ranking — mean Spearman ρ of 0.36, 0.46, and 0.53, with 36%, 30%, and 27% of model pairs reversing order. Cross-sectional fidelity is the exception (pooled ρ = 0.82, 13% inversions), because it aggregates slice-wise scores over time and therefore preserves much of the global static ranking while still exposing temporally localized failures.

Dataset SD vs TS · ρInv. SD vs CS · ρInv. SD vs LG · ρInv. SD vs ST · ρInv.
Table 4. Spearman ρ and inversion rates computed across models within each dataset; pooled ρ concatenates within-dataset ranks. Lower ρ and higher inversion rates mean static fidelity transfers worse to that temporal dimension.

Finding 2 · Temporal sub-metrics expose architecture-specific failures

Rossmann follows a daily grid with unique timestamps at one-day intervals, so TimestampUniqueness and RegularityConsistency equal 1.0 in the real data by construction. The joint-generation models score around 0.6 and 0.4 on those same properties — a violation read against a known reference value, not against another model's ranking. Autoregressive generators condition each event on the preceding sequence and preserve them; their strong SD scores show that plausible marginals can mask the failure entirely.

Model Order Unique Duration Regularity Grid
Table 5 · Rossmann, before post-processing. Columns are TemporalOrderConsistency, TimestampUniqueness, TrajectoryDurationSimilarity, RegularityConsistency, and GridCompleteness. Note RTF's near-zero Order score alongside near-perfect Unique/Regularity/Grid: it generates a clean, regularly spaced axis whose global chronological direction is not anchored.
Model KS-Comp TV-Comp StatSim CatCov ContSim
Table 6 · Cross-sectional sub-metrics on Rossmann, per time slice. RCTGAN and RelDiff recover per-slice categories (high CatCov) without their joint combinations (low ContSim).
Model FirstDiff KSComp TransMat TVComp AutoCorr Sim
Figure 3 · Longitudinal sub-metrics on Rossmann, interpolation-fill variant. 0.60 1.00

Structural sub-metrics are only distinguishing on multi-child schemas; among the core datasets that is Walmart alone. There, RTF's low ST is driven by SequenceLengthSimilarity and TemporalCardinalityShapeSimilarity — it does not reliably control how many child records each parent gets, or how that count evolves. Relational diffusion models (RDBDiff, ClavaDDPM) improve substantially under ST despite weaker SD rankings. On linear schemas, ST reduces to sequence-length similarity and saturates near 1.0, which is why those rows carry little signal.

Finding 3 · Temporal evaluation changes how privacy and utility read

Row-level nearest-neighbour metrics treat records as exchangeable across time, and can misread risk in both directions. Larger CS-DCR values than DCR mean the row-level nearest neighbours were coming from unrelated timestamps — incidental overlap in a bounded row vocabulary, not re-identification. In the other direction, single-row exposure cannot see copied consecutive sub-sequences: RTF and TabDiT look safe under DCR/NNDR and become the riskiest models under NGP(3). The autoregressive mechanism that improves within-trajectory fidelity also increases trajectory-level leakage.

Model DCRNNDRCS-DCRCS-NNDRNGP(1)NGP(3)
Table 7 · Privacy on Rossmann. DCR and CS-DCR are raw median nearest-neighbour distances; NNDR, CS-NNDR and NGP(n) = 1 − NgramExposure(n) are normalized to [0, 1]. Higher is safer in every column.
Dataset MLE-TS MLE-Temporal
ρSDρlongρall ρSDρlongρall
Table 8 · Correlation between fidelity dimensions and downstream utility, on the four datasets where MLE-Temporal applies (absolute time, regular grid, chronological split). As the task consumes more history, utility tracks longitudinal fidelity rather than static marginals.
Datasets

Thirteen datasets, six domains, classified before they are scored.

The taxonomy is not descriptive metadata — it is what decides which metrics run. Cross-sectional fidelity, for instance, is undefined without absolute time and dependent trajectories, which is why AirBnB and PTB-XL show “—” in the CS column of every leaderboard. marks the seven core datasets used for cross-model comparison.

DatasetDomainTime repr. PeriodicityTrajectoriesSchema
Table 2. Schema types: Single — one table, time-invariant attributes implicit in the entity id. Linear — one parent table linked to one child trajectory table. M-Child / M-Parent / M-C&P — multi-child, multi-parent, and both.
Metric catalog

Seven new metrics, sixteen adapted, four adopted as defined.

Every metric declares its provenance and its applicability conditions. Adapted metrics reuse an established measure — usually an SDMetrics estimator — but change what it is computed over: inter-arrival times instead of feature values, a time slice instead of the pooled table, first differences instead of raw levels.

Scope: metrics are univariate (per feature, averaged over the applicable set), bivariate (per feature pair, averaged), or multivariate (joint over all features or a trajectory embedding). DCR and NNDR are time-agnostic baselines and are not counted in the provenance totals.

FAQ

What the benchmark does, and what it deliberately does not do.

Why not report a single Seq2Synth score?

Because a model can be strong along one temporal axis and fail on another, and collapsing that into a leaderboard number would hide exactly what the benchmark exists to surface. Timestamp and longitudinal metrics test whether individual trajectories are valid; cross-sectional metrics test whether the population evolves correctly over absolute time; structural metrics test relational cardinalities. Use it as a diagnostic protocol and read the dimensions jointly.

Rankings disagree — does that make the temporal ranking the correct one?

Not on its own. Different metric families capture different properties, so disagreement only shows the dimensions are not interchangeable. The stronger evidence comes from properties with a fixed reference value in the real data: on a daily grid with unique timestamps, uniqueness and regularity must equal 1.0. A score of 0.4 there is a violation, not a preference.

Why compute timestamp metrics before post-processing?

Sorting and deduplication would repair precisely the failures we are trying to detect. Timestamp fidelity runs on raw generator output; every other dimension needs a well-defined time axis, so those run after records are sorted and aligned to the defined grid. Imputation ablations for that pipeline are in the online appendix.

What is NGP(n), and why is it not just DCR?

NGP(n) = 1 − NgramExposure(n) measures whether discretized synthetic trajectory fragments reproduce real length-n token subsequences. DCR and NNDR compare single rows, so they cannot see a copied run of consecutive events. NGP(1) isolates the row-level case; NGP(3) catches sequential reproduction. Models can look safe at n = 1 and be the riskiest at n = 3.

Which generators are evaluated?

Eight: ClavaDDPM, RCTGAN, RDBDiff, REaLTabFormer (RTF), RGCLD, RelDiff, SDV, and TabDiT. They split into autoregressive sequence models (RTF, TabDiT) and joint or relational generators (the rest) — a split that predicts their failure modes well. TabDiT covers only datasets whose structure it supports, hence the “—” rows.

What should generator designers take from this?

Three things, none of which need a new model family. Model timestamps as ordered objects rather than ordinary continuous attributes. Give autoregressive generators explicit absolute-time anchoring — calendar embeddings, anchor tokens, or auxiliary timestamp losses — so token position is not confused with calendar time. Condition on sequence length and relational cardinality directly instead of letting them emerge, especially where parent–child counts evolve over time.

Can I add a generator or a dataset?

Yes. The evaluation code, taxonomy classification, and post-processing pipeline are in the GitHub repository. A new dataset needs its taxonomy assignment first, since that determines which metrics are applicable; a new generator only needs to emit the expected table format. Open an issue or a pull request.

Citation

If Seq2Synth is useful in your work, please cite it.

@inproceedings{kwon2026seq2synth, title = {Seq2Synth: Benchmarking Temporal Fidelity in Synthetic Sequential Tabular Data}, author = {Kwon, Kiwan and Kim, Kangmin and Lee, Hojin and Jung, Yeseong and Kong, Hyeongwoo and Potluru, Vamsi K. and Park, Saerom and Lee, Yongjae}, booktitle = {Proceedings of the ACM International Conference on Information and Knowledge Management (CIKM '26)}, year = {2026}, address = {Rome, Italy}, publisher = {Association for Computing Machinery}, doi = {10.1145/3799682.3841108} }

Supported by the National Research Foundation of Korea (NRF) grants funded by the Korea government (MSIT) — No. RS-2025-24803208, No. RS-2024-00354727 — and by IITP grants (No. RS-2020-II201336, AI Graduate School Program at UNIST; No. RS-2026-25616928; RS-2024-00436936; RS-2026-25528781), and by the Hankuk University of Foreign Studies Research Fund.