Skip to content

Assumption Audit

An audit procedure — instantiates Premise–Action Decoupling

Sweeps a whole plan or decision for the assumptions it silently rests on, keeps the load-bearing ones, tests their support, and names what would have to be true instead where support is thin.

Assumption Audit is the reasoning-level pass that takes an entire plan, argument, or decision — not a single sentence — and drags its unstated premises into daylight. Every plan runs on assumptions; most are never written down, and the dangerous ones are exactly those confident enough to go unspoken. The audit enumerates them, then does the two things that make it more than a brainstorm: it triages by materiality, keeping the load-bearing few and discarding the trivia, and it tests each survivor's support, grading how well the evidence actually backs it. For the load-bearing assumptions that turn out thin, it names the alternative — what would have to be true instead, and what that would do to the plan. Its whole point is to decouple the plan's commitments from its premises: to know which beliefs the action is leaning on, so a wrong belief can be caught before it takes the action down with it.

Example

A relief agency is finalizing a pre-positioning plan for cyclone season: warehouse stock forward, deliver cash grants so families buy locally. The plan reads as a set of actions, but the audit rewrites it as the set of beliefs those actions require. Enumerated: the coastal road stays passable; forward warehouses have spare capacity; caseload will resemble the last event; and — the quiet one — local markets keep functioning, since the entire cash-grant approach assumes families can actually buy food.

Triaged by materiality, two assumptions are load-bearing enough to decide the plan's fate: road passability and market function. Tested for support, they split. Road passability has a recent engineering survey behind it — well warranted. Market function has almost nothing but hope; in the last severe cyclone, markets collapsed for weeks. So the audit marks that assumption thin and load-bearing and states the alternative plainly: if markets don't function, cash aid fails and an in-kind pipeline is needed. That single line converts a hidden bet into an explicit contingency the plan can pre-build and watch for, rather than discover in the wreckage.

How it works

  • Enumerate the premises — walk the plan and list what must be true for each step to work, forcing out the assumptions confident enough to have gone unwritten.
  • Triage by materiality — keep the assumptions the plan's outcome actually hinges on; set aside the ones that could be wrong without moving the result. Effort goes only to the load-bearing few.
  • Test support — for each survivor, ask what evidence backs it and grade the warrant honestly: solid, contested, or hope dressed as fact.
  • Name the alternative — for every thin-but-load-bearing assumption, state what would be true instead and what that does to the plan, so the exposure is concrete rather than a vague worry.

The distinguishing discipline is that it operates on a decision's whole premise-set, prioritizing by consequence — not decomposing one utterance, and not testing a claim in isolation for its own sake.

Tuning parameters

  • Materiality threshold — where you draw the line between load-bearing and ignorable. Set it low and the audit drowns in trivial premises; set it high and a quietly critical assumption slips under the bar.
  • Evidence bar — how much warrant an assumption must have to pass. A strict bar surfaces more exposures but can stall a plan on unavoidable uncertainty; a lax bar clears the plan but launders hope as fact.
  • Search breadth — how hard you hunt for hidden assumptions versus cataloguing the stated ones. Broad search catches the dangerous unspoken premises but costs time and can feel adversarial.
  • Disposition routing — what a flagged assumption gets marked for: drop, test now, build a contingency, or hand to standing monitoring. Tunes the audit between a one-off scrub and the front-end of ongoing governance.

When it helps, and when it misleads

Its strength is catching the fatal premise while it is still cheap — before the plan is built on it — and doing so in priority order, so scarce scrutiny lands on the assumptions that can actually sink the effort. It converts "we just assumed that would hold" from a post-mortem line into a pre-mortem checklist.

It misleads when the triage or the evidence bar is off. A materiality threshold set wrong is the classic failure: the whole value is in catching the load-bearing assumption, and an audit that spends itself on trivia while missing the one that matters is worse than none, because it manufactures false confidence. It also tends to over-scrutinize the assumptions that are easy to articulate and under-scrutinize the ones so deep they never get listed. And it is readily run backwards — staged to bless a plan's premises after the plan is chosen, rather than to stress them. The discipline that keeps it honest is to fix the materiality bar and the evidence standard up front, and to treat the audit as a live instrument re-run as conditions change, not a one-time sign-off.[1]

How it implements the components

Assumption Audit realizes the premise-testing components — the reasoning-layer work of deciding which beliefs a decision may safely lean on:

  • materiality_threshold — its triage step: it separates the load-bearing assumptions from the ones that could be wrong without changing the outcome, so scrutiny is rationed by consequence.
  • evidence_test — it grades each material assumption's warrant against the evidence actually available, distinguishing solid from contested from hope.
  • alternative_assumption — for each thin-but-critical premise it states what would be true instead and what that does to the plan, turning an exposure into a concrete, plannable alternative.

It does not extract premises from a single utterance (embedded_premise, surface_action) — that is Literal Content Parsing; and it does not own or record premises over time (assumption_owner, consequence_record) — that is the Decision Premise Register, which consumes the audit's findings.

Notes

The audit is a point-in-time sweep; it grades warrant as of now. The premises it flags for watching do not monitor themselves — a "test later" or "hold under review" disposition is only as good as the standing ledger it is handed to. That ongoing tracking is the Decision Premise Register's job; the audit feeds it rather than replacing it.

References

[1] Assumption-Based Planning (RAND; James Dewar) is the established practice of identifying the load-bearing assumptions under a plan, judging their vulnerability, and setting signposts to watch — the same load-bearing-first logic this audit applies, and the reason materiality triage precedes evidence testing. ↩