Skip to content

Benchmark Deconstruction Grid

A comparison matrix — instantiates Design-Principle Extraction and Reapplication

Arrays several successful sources against a shared feature grid so the design logic that recurs across all of them separates from the quirks local to any one.

The Benchmark Deconstruction Grid is a matrix with one column per source and one row per feature, filled in for several successful examples at once. Its whole reason for existing is triangulation: a pattern that shows up across three or four independent, differently-built sources is far more likely to be the load-bearing design logic than any feature you would spot by staring at a single admired example. Where a single-source analysis can't tell essential structure from local accident, the grid lets the co-occurrence pattern itself do the sorting — features that recur are candidates for the principle, features unique to one column are candidates for local implementation, and the reader can see which is which at a glance.

Example

A mid-size city studying bus rapid transit doesn't want to copy one famous system wholesale, so it decks out a grid with four: the columns are Bogotá, Curitiba, Guangzhou, and a mid-size peer, and the rows are features — dedicated median lanes, level-boarding platforms, off-board fare collection, signal priority, distinctive branded stations, articulated buses, a specific fare-gate vendor. Filling the cells, a shape emerges. Dedicated lanes, level boarding, and off-board fare collection appear in every column — that recurrence is the signal that they carry the "make a bus behave like a train" logic. The branded station architecture and the particular fare-gate hardware appear in only one column each; the grid quietly reclassifies them as local implementation, not the principle.

The output isn't a ranking of who's best. It's a separation: a short list of what recurs across all of them (extract this) from a longer list of what's specific to one (leave it behind), with the evidence for each classification sitting right there in the row.

How it works

The grid's distinguishing discipline is comparison across independent sources rather than depth on one. Sources are chosen to be genuinely different in materials, scale, or context, so that a recurring feature can't be explained away as shared heritage. Each feature is scored per source (present / absent / partial), and then the read is columnar-to-rows: high-recurrence rows become essential candidates, single-column rows become incidental candidates, and rows that recur but with visible variation flag a principle expressed in different local forms. Recurrence is treated as evidence, not proof — a strong recurring feature still has to earn "essential" against the objective it's supposed to serve.

Tuning parameters

  • Source diversity — how different the benchmarked examples are. Diverse sources make recurrence a much stronger signal; near-clones inflate false agreement (they share features because they copied each other, not because the feature is essential).
  • Number of sources — more columns sharpen the recurrence signal but cost analysis effort and dilute depth per source.
  • Feature granularity — coarse rows ("fast boarding") vs fine rows ("off-board fare collection with proof-of-payment"). Finer rows catch real differences but explode the grid.
  • Recurrence threshold — how many columns a feature must appear in to count as essential. A high bar is conservative (misses principles expressed differently); a low bar over-includes.

When it helps, and when it misleads

Its strength is that it defeats single-example dazzle: by demanding a feature prove itself across independent sources, it strips out the idiosyncrasies that a one-source teardown would earnestly copy. It is the cheapest reliable way to separate "the logic" from "how this one did it."

Its central trap is equifinality — different structures can reach the same outcome, so a feature that fails to recur is not thereby incidental; it may be an equally valid alternate solution that only one source happened to choose.[1] The grid also mistakes shared ancestry for shared principle when the benchmarked sources are secretly copies of one another, and it invites "benchmark to the average," blending several designs into a mush that optimizes nothing. The discipline that keeps it honest is to pick sources that are independent and dissimilar, and to check each recurring feature against why it recurs before promoting it to a principle.

How it implements the components

  • multi_source_principle_triangulation — the grid is the triangulation instrument: it structures several sources so shared design logic can be read off the recurrence pattern.
  • essential_incidental_feature_filter — it separates essential (recurring) from incidental (single-source) features across the benchmarked set, a cross-source complement to a within-source classification.

It classifies features by cross-source recurrence, not by within-source functional necessity (that's the Essential/Incidental Feature Sort); it does not open up any one source's internal structure (the Function-Behavior-Structure Matrix and Teardown Workshop do that) or elicit why a feature exists (Constraint Laddering Interview).

Notes

Recurrence across the grid is a hypothesis generator, not a verdict. A feature that recurs everywhere should still be checked against the objective it supposedly serves before it is promoted to "essential" — otherwise the grid can canonize an industry-wide bad habit simply because everyone shares it.

References

[1] Equifinality — from general systems theory (von Bertalanffy) — is the principle that a system can reach the same end state from different initial conditions and by different paths. It is why a non-recurring feature is not automatically incidental: it may be an alternate route to the same result rather than a mistake. ↩