Think² logoThink²
Sign in

Guides (Spanish)

A total differs by a few hundredths: did you round the lines or their sum?

Three amounts of 0.335 do not always produce the same two-decimal total. Under one rule, rounded lines add up to 1.02 and the rounded total is 1.01; under another, that total is 1.00. Before attributing a difference to a payment, fee or error, identify original amounts, the rounding stage and tie-breaking mode. This guide reproduces six synthetic Decimal cases: it inspected no invoices, banks or applications and prescribes no accounting or tax rule.

The waitlist does not guarantee immediate access or a free plan. The waitlist destination is in Spanish.

Situations to review

The same three amounts yield a difference of 0.01 or 0.02.

R1 and R2 use 0.335 three times and a 0.01 unit. Rounding each line gives 0.34 + 0.34 + 0.34 = 1.02. The original sum is 1.005. ROUND_HALF_UP rounds it to 1.01: lines minus total is +0.01. ROUND_HALF_EVEN rounds it to 1.00: the difference is +0.02.

‘Two decimal places’ alone does not identify a rule. Keep both the rounding stage and tie-breaking mode; none of these results alone proves a system is wrong.

Changing the mode can also change the sign of the difference.

R3 and R4 use two amounts of 0.345. Their original sum is 0.690 and the rounded total is 0.69 under either mode. Per line, HALF_UP yields 0.35 + 0.35 = 0.70, a +0.01 difference; HALF_EVEN yields 0.34 + 0.34 = 0.68, a −0.01 difference.

Do not turn the difference into an absolute value or assume it is always an excess. Its sign preserves the direction of comparison; it does not determine the accounting treatment.

Negative amounts and cases with no difference belong in the test too.

R5 repeats −0.335 three times with HALF_UP: lines −1.02, total −1.01 and difference −0.01. R6 uses 0.10 and 0.20: both routes yield 0.30, with a 0.00 difference. All six results matched our own expectations fixed before the local test.

This check covers these inputs, not every magnitude, currency or application. Zero difference validates neither data provenance nor tax treatment; a negative amount does not itself establish a refund.

The waitlist does not guarantee immediate access or a free plan. The waitlist destination is in Spanish.

How to evaluate the work

  1. Keep an authorised copy of original amounts, their representation, unit, currency where relevant and scope. Do not replace originals with already rounded display values. Establish whether the same lines are compared and whether discounts, taxes or fees are included; do not mix currencies or invent missing values.
  2. Document the system rule or ask its owner: precision, line-level or total-level rounding, tie-breaking mode and the order of other operations. The 0.01 unit is a choice for this exercise, not a rule for every currency or document. If the applicable rule is missing, preserve that question rather than choosing the mode that balances the figures.
  3. Reproduce both routes separately with the same input and context: round each amount then sum; sum original amounts then round once. Our test used Decimal from strings, precision 28 and an explicit mode in quantize. Decimal does not remove the need to configure precision, limits and error handling for other inputs.
  4. Record original amounts, rounded lines, line sum, rounded total and signed difference. Investigate whether the rule could explain a real discrepancy without declaring it the proven cause. The competent person decides any adjustment and its treatment; calculating, approving, recording and verifying the record are separate steps. Do not alter originals to hide the difference.

Reproduce six cases without opening your accounts

The table uses the decimal point found in the data file. ‘Lines’ sums already rounded amounts; ‘total’ rounds the original sum; difference = lines − total. HALF_UP resolves ties away from zero; HALF_EVEN towards the even neighbour. These are technical modes, not tax recommendations.

Download the exercise and run the local checker on synthetic data. It opens no connections and does not modify the cases. Python 3 and its standard library are sufficient for this exercise; the original test used Python 3.13.5. The 6/6 match is a check against our own expected values, not an independent evaluation or a test of Think² or accounting software. The data and code are shared by both language versions; the download includes English and Spanish instructions.

CaseOriginal amountsModeLinesTotalDifference
R10.335 + 0.335 + 0.335HALF_UP1.021.010.01
R20.335 + 0.335 + 0.335HALF_EVEN1.021.000.02
R30.345 + 0.345HALF_UP0.700.690.01
R40.345 + 0.345HALF_EVEN0.680.69-0.01
R5-0.335 + -0.335 + -0.335HALF_UP-1.02-1.01-0.01
R60.10 + 0.20HALF_UP0.300.300.00

Download rounding cases and checker

Ask for an explanation, not a forced balance

In Think² you can assign a collaborator a comparison with authorised sources, a stated rule and visible exceptions. This is a proposed use case requiring compatible connections, permissions and operations, not an implemented automated reconciliation. Ask it to preserve the difference before proposing what to investigate.

Bring together the data owner and the rule owner

People and collaborators can coordinate in a group to clarify which amount is original and which rule applies. A group replaces neither the accounting owner nor confidentiality or isolation controls. Share only permitted information and check who can see it.

Choose tools without treating calculation as authority

Think² supports authorised compatible models and permitted tools; each collaborator may have an optional computer. None grants extra access or replaces professional judgement. Review and approve each specific external message or change, then verify its outcome; this guide recorded no adjustments and measured no savings.

Frequently asked questions

Which rounding method is correct?

This guide does not determine the rule for your document or system. Check contractual terms, configuration, the operation and accounting or tax requirements with the competent person. Both calculations are reproducible under their stated rules; that makes them neither interchangeable nor permission to choose whichever balances.

Does Decimal prevent all precision problems?

No. Constructing values from strings preserves the exercise's written decimal values; converting a float preserves its exact binary value, which may differ from that decimal. Operations still follow precision and limits in a context. This small sample does not establish that every input can be processed without intermediate rounding or errors.

Does the difference prove a fee or authorise an adjustment?

No. It explains an arithmetic possibility using fictional data. A real discrepancy requires sources, matching scope, the system rule and competent review. No bank was queried and no journal entry or tax document was created. Any external action requires its own authorisation and outcome check.

Sources and scope

Keep exploring

    The waitlist does not guarantee immediate access or a free plan. The waitlist destination is in Spanish.