Is a tie between two figures in the same currency safe?

Same unit is the floor, not the proof. Two currency totals can answer different questions, a gross against a net or a booked against a shipped, and tie to the cent while meaning different things. The definition beside each number is what settles it.

What tolerance should a tie use?

One denominated in the value's own display unit, written down before the run. A count gets none at all: it matches or it fails, because a default tolerance of 1.0 on a count silently absorbs exactly the dropped or duplicated row the anchor exists to catch.

What if the two sides genuinely cannot be compared?

Then there is no tie, and the page says so. A figure the file states no total for is published labelled unverified with the reason, which is a smaller claim than a green mark and a truer one.

Failure catalogue · The verification is theater · 35 of 51

OBSERVED FAILURE MODE

The tie that compared revenue to seats.

A reconciliation tie is only a proof when both sides measure the same thing in the same unit. A tie between a revenue figure and a seat count is not agreement, it is arithmetic pointed at two different questions. The mark renders green, and it has proved nothing about either number.

What we saw

In our own audits a check row read ties, in green, over a pair of values with no shared unit: a currency total on one side, a count of seats on the other. Both numbers were computed correctly. Both came from the file. The row named neither side, so nothing on the page revealed the mismatch, and the tie sat on the certified surface beside real anchors where a reader scanning the check table had no way to tell that this one asserted nothing at all.

Why it passes a glance

Nothing in the arithmetic complains. Both values exist, both were read from the source, and the comparison runs without error. A tie renders as a green mark and a short label, so the reader sees agreement rather than the two operands. Unit checking is the step everyone assumes someone else did, and a tie between incomparables is the one form of agreement that can never fail.

What addresses it

The verification-practice skill requires every tie to name its two sides and their units, denominates the tolerance in the value's own display unit, and forbids an anchor that can never fail. Doctrine principle 2, Two independent reads beat one careful one, is the law underneath: two reads of different quantities are not two reads of one number, and a tie between incomparables is a failed check, not a green one.

Check your own file in two minutes

Read every tie row out loud as a sentence: this number, in this unit, against that number, in that unit.
Reject any tie whose row does not name both operands and both units.
Change one side by a known amount, rebuild, and confirm the tie turns red.
Confirm counts carry no tolerance, because a count matches exactly or it fails.

What this does not catch

Naming units catches the mismatch, not the meaning. Two figures can share a currency and still answer different questions, a gross against a net, a booked against a shipped. Principle 10 covers that gap, and the definition printed beside each number is what closes it.

Quick answers

Is a tie between two figures in the same currency safe?: Same unit is the floor, not the proof. Two currency totals can answer different questions, a gross against a net or a booked against a shipped, and tie to the cent while meaning different things. The definition beside each number is what settles it.
What tolerance should a tie use?: One denominated in the value's own display unit, written down before the run. A count gets none at all: it matches or it fails, because a default tolerance of 1.0 on a count silently absorbs exactly the dropped or duplicated row the anchor exists to catch.
What if the two sides genuinely cannot be compared?: Then there is no tie, and the page says so. A figure the file states no total for is published labelled unverified with the reason, which is a smaller claim than a green mark and a truer one.

Nearby failures

Both rails from one blueprintThe verification is theater The tolerance that ate the errorThe verification is theater The average of incomparablesThe metric means something else

Last updated 2026-09-02 · Dylan, founder · one of 51 observed failure modes, every one seen in a real build or in our own audits, none invented.

Get the free skill → The doctrine All 51 failure modes