Unlike empirical social-science papers built around correlation and regression tables, Operations Research exhibits are about structure, guarantees, and computation:
amsthm so environments render uniformly.Use TikZ/PGFPlots (or matplotlib exported to vector) for crisp figures; performance profiles and log-log scaling plots reproduce well as vectors. Keep raw data/scripts in the ORJournal repository so every exhibit regenerates.
| Referee remark on an exhibit | What it flags | Venue-specific fix |
|---|---|---|
| "Computational study lacks benchmarks" | no recognized baseline column | add a column for the closest prior method and a strong solver on standard instances |
| "Cannot tell what 'best' means" | undefined win marking | caption defines the metric and tie-breaking; bold/mark honestly |
| "Results not reproducible from the table" | missing units/time-limit/hardware | state units, time limit, termination, hardware, and the producing configuration |
| "Means hide variability" | bare means for stochastic output | error bars / confidence intervals on every random-output figure |
| "Theorem numbers don't match the text" | label drift | single numbering scheme via amsthm; reference by number everywhere |
| "Scaling claim unsupported" | linear-axis time plot | log-log size-vs-time plot or a performance profile across the instance set |
These are the exhibit-level versions of the two flags that sink OR computational sections: no benchmarks and structure-not-connected-to-application. Because Operations Research is the INFORMS flagship demanding both theorem-grade rigor and a credible computational study, the comparison table and the performance profile carry as much referee weight as the theorem environments.
A paper proves a 1.5-approximation and validates it. Two exhibits do the persuading:
Comparison table — Method × {assumption, factor, complexity}:
| Method | Assumption | Factor | Complexity |
|---|---|---|---|
| Prior (Author year) | bounded ratios | 2.0 | O(n²) |
| This paper | general weights | 1.5 | O(n² log n) |
Computational table — Instance set × {mean gap %, time (s), solver baseline}, e.g.
TSPLIB-derived instances showing 1.8% mean gap vs. 31% for the heuristic baseline at a
600 s limit on stated hardware, with the proved 50% bound never violated. The figure
pairs them: a log-log scaling plot with confidence-interval bands over 30 seeds. The
table corroborates the theorem's bound; the plot shows it holds as size grows.
Generate exhibits from the fitted result, not by retyping numbers (the usual source of
body-vs-appendix drift). Full map: execution-with-mcp. Operations Research is predominantly analytical / optimization / stochastic modeling; use the chain below only for its empirical/causal papers — modeling, optimization, and simulation are outside this causal-inference toolchain.
etable (multi-model columns) or did_summary_to_latex straight from the
result_id.plot_from_result / enhanced_event_study_plot / event_study_table —
axis units and the SE/clustering note baked in.See a full fitted-result → exhibit chain in the JF execution walkthrough.
【Formal statements】numbered consistently; INFORMS/amsthm style
【Comparison table】prior work × {assumptions, result, complexity}: drafted?
【Computational table】instances × {gap, time, baseline}; units/limits stated
【Plots】convergence / scaling / CIs present?
【House style】1.5-spaced, 11-pt; self-contained captions; e-companion split
【Next step】ors-writing-style