Keep a reported range separate from a single result

Identify what the range describes

Preserve both bounds and their meaning so the next user can choose a valid analysis rather than inherit a hidden assumption.

Before paying for technical-document extraction, inspect how the source meaning survives in the output. ParseMill’s public material-register example gives you a synthetic PDF and downloadable evidence to compare. Bring your own document type and acceptance checks to a scope discussion; the self-service invoice plan is a separate offer.

If the destination expects one number, do not automatically choose the midpoint or an endpoint. Those choices add an interpretation that the source may not support. A range can remain a structured non-scalar result until the downstream task defines an appropriate treatment.

Keep the roles visible

This invented fixture uses similar characters for three distinct meanings. The final row also shows why a partial range should not become a complete interval. The unknown upper bound must remain explicit rather than being copied from a specification or neighboring record.

NIST: SI notation checklist discusses notation that makes ranges and their units clear. Use that principle when writing normalized fields, while retaining the source’s original typography.

Review before splitting

1. Read the field heading and any legend that defines the notation. Record whether the range is an observation, a requirement or another type of text.
2. Inspect both endpoints, signs and units. Check whether an endpoint has its own qualifier or a footnote marker.
3. Create lower and upper fields only for a confirmed range. Add a range-role field so requirements cannot be mistaken for observations.
4. Keep unsupported scalar conversion in an exception queue. Record who must decide the analysis method and which intended use requires it.

Microsoft: Power Query data types distinguishes text and numeric representations. Separate numeric endpoints can support later analysis, but their types alone do not capture whether the range is a result or a specification.

Avoid hidden meaning changes

An en dash can be mistaken for a minus sign, especially in copied PDF text. Two negative endpoints also need careful handling: the separator and the signs have different roles. Do not split on every hyphen without checking the source structure.

A midpoint is mathematically easy to calculate but may not represent any observed value. If an analyst creates one for a stated purpose, label it derived and document the rule. Never overwrite the reported interval with that value in the evidence register.

Define a successful handoff

For the fixture, the first row remains a result interval, the second a requirement, the third an identifier and the fourth unresolved. A recipient should see these roles without opening an undocumented calculation. Preserve the source references needed to verify them.

Use result intervals and specification intervals as explicit requirements in a supplier-report scope discussion. The worked example lets you inspect a simpler source-to-output relationship first. Range-heavy technical reports are not a promise of the invoice trial.