Skip to content

Shown-vs-Unshown Audit

Audit — instantiates Salience-Significance Decoupling

Sets a display's visible items beside the relevant ones it leaves out, so the gap between what is shown and the full field becomes something you have to look at.

Shown-vs-Unshown Audit takes one concrete artifact — a report, deck, feed, dashboard, or search page — and reconstructs two lists side by side: the items it actually shows, and the relevant items it leaves out. Its defining move is to treat the omission as the product: most reviews study what is on the page, while this audit exists to surface what is not. To do that it must draw the "relevant universe" from somewhere other than the display itself, because a display audited only against its own contents can never reveal its own blind spot. The output is the second column — the material a reader would have weighed had it been visible.

Example

A quarterly board deck opens with the metrics that chart well and flatter: signups, uptime, an NPS score. A Shown-vs-Unshown Audit lists those shown items, then builds the relevant universe independently — from the board's charter and last quarter's risk register — and lays the omissions beside them: the churn-reason breakdown, the enterprise renewal that slipped, the swelling support backlog, a security near-miss nobody slotted onto a slide. It then inspects the editorial selection log: the slides were inherited from a template, and the section on churn was built by the team whose numbers churn made look worst. The outcome is not a new metric but a visible line item where silence used to be — "this deck says nothing about churn, and here is who chose the slides."

How it works

What sets it apart is that it runs against a specific artifact and produces the omission column, not commentary on the visible one:

  • Inventory the shown set exactly as presented, in the artifact's own units.
  • Reconstruct the relevant universe from an outside source — a charter, a checklist, a stakeholder who did not build the display — never from the display itself.
  • Subtract to get the omissions, and record the editorial selection log: who chose the shown set and by what rule (habit, template, self-interest).

Tuning parameters

  • Universe boundary — how wide you draw "relevant." Too narrow inherits the display's frame; too wide buries the real omissions under trivia.
  • Omission granularity — flag whole missing topics, or missing line-items within a shown topic. Finer catches quiet edits; coarser stays legible.
  • Selection-log depth — merely note what is shown, or trace who chose it and why. The deeper cut catches motivated omission but costs goodwill.
  • Cadence — a one-off audit versus a standing checklist bolted onto every recurring report so the omission column is refreshed each cycle.

When it helps, and when it misleads

Its strength is that it makes absence auditable — the failure mode where a decision-maker weighs only the visible subset because the rest never showed up. It is the standard guard against survivorship bias,[1] the error of reading a story off the survivors while the casualties sit silently off-page.

Its own soft spot is the relevant universe: that boundary is itself a judgment, and a lazily scoped universe simply relocates the blind spot one step back. The classic misuse is running the audit to manufacture a "we considered everything" reassurance while quietly scoping the universe to exclude the inconvenient. The discipline that guards against this is to derive the universe from a source outside the display's authors — and to keep the selection log, so a suspiciously tidy shown set has to explain itself.

How it implements the components

Shown-vs-Unshown Audit fills the inventory-and-omission side of the archetype — the components an audit of a concrete display can populate:

  • visible_item_inventory — builds the shown column verbatim and pairs it against the reconstructed relevant universe.
  • missingness_and_selection_check — the omission column is this check: what important material is absent from this specific display.
  • editorial_selection_log — records who selected the shown set and under what rule, so the pattern of omissions is attributable.

It does not model the population a sample was drawn from (comparison_set_restoration — that's Sample Frame Reconstruction), define what makes an item significant (significance_criteria_register — that's Salience-Significance Matrix), or gate any resulting action (decision_weight_gate — that's Notification Priority Review).

Notes

The audit is deliberately upstream: it produces the item list that Salience-Significance Matrix later scores, and the omission gaps that a Salience Red Team probes. It works on a curated display; when the visible set is a statistical sample rather than an editorial choice, reach for Sample Frame Reconstruction instead — the two share the missingness question but attack a different generating process.

References

[1] Survivorship bias — reasoning from the visible survivors while the non-survivors are silently absent from the data. The wartime aircraft-armor analysis (reinforcing where returning bombers were not hit, because the planes hit there never returned) is the textbook illustration of restoring the unshown column. ↩