Concept

Order-of-magnitude / gut-check

analytical

Overview

An order-of-magnitude (or gut) check is a fast, deliberately crude, independent estimate of a quantity — aiming to get its magnitude right to within a factor of a few — used to test whether a model’s result is plausible before it is trusted. Its job is not precision but to act as a tripwire that catches gross errors: unit slips, misplaced factors, and answers off by a power of ten.

Body

What it is. A gut check recomputes roughly the same quantity by a different and simpler route than the detailed model: rounded inputs, dropped second-order terms, dimensional reasoning. It targets only the magnitude — the power of ten, or a factor of ~2–3 — not the exact figure. A clean spreadsheet will return a confident number whether or not its inputs and wiring are right; the gut check is the independent recomputation that tests it from outside.

Independence is what gives it power. The check must take a different path than the model. A rough estimate that reuses the model’s own suspect input, or repeats the same erroneous factor, only reproduces the error and appears to confirm it. The value of the check is precisely that an independent route will not share the model’s specific mistake.

Common forms.

Build-up / Fermi estimate — assemble the answer from a few rounded first-principles terms (energy per unit × price) and see if the total lands in the same place.
Dimensional / units check — confirm the units resolve and the input magnitudes combine sensibly (a frequent catch for kWh-vs-MWh or kg-vs-tonne slips).
Anchor comparison — set the result beside a known value for a comparable thing (a real plant’s cost per tonne, a commodity price). When the anchor is a characterized incumbent, this shades into benchmarking.
Ratio / share check — ask whether each component share sits in a believable band (feedstock and energy being the majority of a commodity cost; the capital share being a tenth-ish of capex spread over output).

What “passing” means. Agreement to within the check’s own accuracy class — a factor of ~2 at this level — means no gross error was detected, not that the model is right to its last digit. A clear failure — off by 10× — is a strong signal of a real mistake to be hunted down. The gut check is a guardrail at the same ±30%, FEL-grade resolution as the estimate it guards, which is what makes it proportionate rather than pedantic.

Limits & typical error

It rules out blunders, not inaccuracy. A factor-of-2 check passes a model that is 30% wrong, so a passed gut check confirms only the absence of a gross error — reading it as validation of the detailed figure overstates what it can do.
A non-independent check is worthless. If the rough estimate borrows the model’s own questionable input or repeats the same wrong factor, it reproduces the error and falsely “agrees.” Independence from the thing being checked is the entire mechanism.
A wrong anchor passes a wrong answer. Comparing against a misremembered reference, or one on a different basis (an old cost without escalation, a nameplate-vs-operating mismatch), can validate a wrong result or reject a right one. The anchor needs its own provenance to be worth checking against.
Magnitude reasoning hides compounded mid-size errors. Two ~2× errors in opposite directions cancel in the order of magnitude but not in the answer; a check that looks only at the power of ten can wave through compounded factor-of-2–3 mistakes large enough to move a decision.
It sets the resolution the answer should be stated at. A result that survives only an order-of-magnitude check cannot honestly be reported to several significant figures — claiming that precision is the mirror-image failure of false precision, rooted in the same gap between a method’s resolution and the number quoted from it.
Confirmation bias turns it into a rubber stamp. Reaching for the anchor that matches the expected answer, and stopping once it agrees, defeats the check; its diagnostic value depends on running it as a genuine test rather than as after-the-fact justification.

Mini-example

Gut-checking green ammonia’s levelized cost of ~$800/t by an independent build-up — a different route than the layered capex/CRF model that produced it. The dominant cost is power: ~10 MWh/t × ~$40/MWh ≈ $400/t for electricity alone. For a new, capital-intensive plant, capital and fixed costs roughly double that, landing around $800/t — the same magnitude as the detailed model, so no gross error is flagged. A ratio check agrees: power ~$400 (~half), capital ~$355 (~44%), fixed ~$50 (~6%) are all believable shares for an electrified commodity process.

Separately, to show the check catching an error: suppose the detailed model returned ~$80/t. The power term alone is ~$400/t, so any total below that is impossible — the gut check flags a ~10× error at once (the classic kWh-for-MWh or kg-for-tonne unit slip), without needing to find where in the spreadsheet it happened.

And to show a wrong anchor: “checking” the ~$800/t green figure against a ~$x/t gray-ammonia market price treats a different route on a different basis as the reference — an anchor mismatch that could wrongly condemn a correct green number or excuse a wrong one. The anchor has to be on the same basis as the thing it checks.