A table displays both providers at 3.01 s. One is numbered first and the other second. A ranking that secretly uses more precision than it displays invites readers to infer a difference they can’t see. A SolvCaptcha rankings page is a useful example of the kind of comparison an editor might review. The small test below uses synthetic data to make the tie rule explicit.

Written with AI and checked on synthetic durations. The cover illustration is AI-generated.

Decide which differences the comparison can support

There are different reasons to declare a tie. Two displayed observations may land on the same rounding boundary. Two measured values might also be too close given measurement uncertainty. The first is a presentation rule. It cannot establish statistical equivalence. If the data comes with uncertainty intervals or multiple trials, use those to support any claim that a real advantage exists.

This example ranks by displayed latency to two decimal places. Equal displayed values receive the same rank. A missing value stays out of the ordered ranking. Alphabetical ordering breaks ties only for display, never for rank. That detail helps keep a chart stable even if the input arrives in another order.

from decimal import Decimal, ROUND_HALF_UP from itertools import groupby def rank_by_display(samples): ranked, unavailable = [], [] for label, raw in samples: if raw is None: unavailable.append(label) continue seconds = Decimal(raw) if not seconds.is_finite() or seconds <= 0: raise ValueError("Expected a positive duration") ranked.append((seconds.quantize(Decimal("0.01"), ROUND_HALF_UP), label)) result, count = [], 0 for display, group in groupby(sorted(ranked), key=lambda item: item[0]): labels = [label for _, label in group] result.extend((count + 1, label, display) for label in labels) count += len(labels) return result, sorted(unavailable) sample = [("beta", "3.013"), ("alpha", "3.006"), ("gamma", "3.214"), ("delta", None)] ranks, missing = rank_by_display(sample) assert ranks == [(1, "alpha", Decimal("3.01")), (1, "beta", Decimal("3.01")), (3, "gamma", Decimal("3.21"))] assert missing == ["delta"] and rank_by_display(reversed(sample)) == (ranks, missing) print("Two joint firsts at 3.01 seconds; unmeasured provider kept separate") 

The local test grouped Alpha and Beta at rank one and kept Delta unranked. For a live report, write the rule next to the numbers. A reader should be able to reconstruct the order from what you display and understand whether your rank describes rounded values, a statistical model, or an editorial decision.

Keep the results suitable for reuse

If you export the comparison, preserve the raw value beside the display value and include the rounding specification. Downstream charts can then use the publication’s display policy without trying to duplicate an undocumented operation. If you later adopt a statistical rule, attach its own label and input assumptions rather than comparing the new ranks directly to an older table ranked by display. A change in policy often creates a change in rank without any change in measurements.