To encode an uncertain date in TEI, keep the human-readable text as the element's content and add machine-readable attributes that express the uncertainty: @notBefore/@notAfter for a date known only within bounds, @from/@to for a span an event occupied, and @cert or a certainty element to record your confidence. The principle is to make "circa" and "probably" queryable without pretending you know more than you do.

Which attribute fits which kind of uncertainty?

Choosing the right attribute is the whole skill. They are not interchangeable:

You know	Use	Example
The exact date	@when	when="1841-06-07"
A span the event occupied	@from / @to	a reign, a war, a tenure
A single date, only its bounds	@notBefore / @notAfter	"sometime in the 1840s"
One bound only	@notBefore or @notAfter alone	"after 1850"

The most common mistake is using @from/@to for an unknown single date. A birth happened on one day; "between 1845 and 1850" is @notBefore/@notAfter, not a five-year span the birth "lasted".

How do you encode "circa 1850"?

Keep the source wording, add a defensible window, and flag the confidence:

<date notBefore="1845" notAfter="1855" cert="medium">circa 1850</date>

Decide your window size as a project rule — many historians treat "circa" as plus or minus five years — and document it in encodingDesc so every encoder uses the same span. Consistency here is what makes cross-collection date queries trustworthy.

How do you record genuine confidence levels?

For a quick flag, @cert="high|medium|low" is enough. When you need precision, use the standalone certainty element pointing back at the date:

<date xml:id="d1" notBefore="1820" notAfter="1830">in his youth</date>
<certainty target="#d1" locus="value" degree="0.7"
  assertedValue="1825">
  <desc>Estimated from the subject's likely age in the portrait.</desc>
</certainty>

This separates the date markup from the reasoning, and @degree="0.7" gives downstream analysis a numeric confidence it can filter on. Use precision instead when the issue is granularity (year vs day), not doubt.

What date format must the attributes use?

All standard date attributes expect ISO 8601 / W3C values: YYYY, YYYY-MM or YYYY-MM-DD. A few rules that save grief:

• notBefore="1840" is valid; notBefore="1840s" is not.
• For "19th century", use notBefore="1800" notAfter="1899".
• For BCE or non-Gregorian dates, use @when-custom with a declared @datingMethod rather than forcing them into ISO.

<date when-custom="1.15.4.3.2" datingMethod="#maya-long-count">
  Maya Long Count date
</date>

A reusable end-to-end workflow

Here is a workflow you can apply to a whole archive:

1. Transcribe the date as written — never normalise away "Lady Day 1750".
2. Classify the uncertainty — exact, span, or bounded-unknown — and pick the attribute from the table above.
3. Set a documented window for vague terms like "early", "circa", "before".
4. Flag confidence with @cert, escalating to certainty where the reasoning matters.
5. Validate that every @notBefore is not after its @notAfter (a Schematron check catches reversed bounds).
6. Record the conventions in encodingDesc so the rules outlive any single encoder.

How do you keep dates consistent across a team?

The threat to a date corpus is drift: one encoder reads "circa" as plus or minus two years, another as ten. Lock the policy down in two places — a written rule in the header and an enforced rule in your ODD/Schematron. A Schematron assertion that flags @notBefore > @notAfter and another that warns when cert="low" lacks a certainty explanation turns good intentions into automatic gatekeeping.

Key Takeaways

• Keep the original date text as content; express uncertainty in machine-readable attributes.
• @notBefore/@notAfter is for a single unknown date within bounds — not @from/@to.
• Encode "circa" as a documented window plus @cert, with the window size fixed as a project rule.
• Use @cert for quick flags and the certainty element with @degree for reasoned confidence.
• All standard date attributes need ISO 8601 values; use @when-custom for other calendars.
• Enforce sane bounds and documented conventions with Schematron and encodingDesc.

Frequently Asked Questions

How do I encode an approximate date in TEI?

Put a machine-readable value in @when-iso and flag the approximation. Use @notBefore/@notAfter for a range, or @from/@to for a span, and add cert="low" or a precision element when you want to record how confident you are.

What is the difference between when, from/to and notBefore/notAfter?

@when is a single known point. @from/@to mark a span the event actually occupied (a reign, a war). @notBefore/@notAfter mark the boundaries within which a single unknown date must fall — the right choice for genuine uncertainty.

How do I record a date like 'circa 1850'?

Encode the text 'circa 1850' as content and set @notBefore="1845" @notAfter="1855" (or your chosen window) plus cert="medium". The window makes 'circa' queryable; document the window size you use as a project rule.

Can I express how confident I am about a date?

Yes. Use @cert with high, medium or low for a quick flag, or the standalone certainty element with @degree for a precise numeric confidence and a @target pointing at the date. precision records granularity rather than confidence.

Which date format should TEI date attributes use?

Use ISO 8601 / W3C dates: YYYY, YYYY-MM, or YYYY-MM-DD in @when, @notBefore, etc. For dates beyond the Gregorian range or in other calendars, use @when-custom with a declared @datingMethod.

How do I encode a date only known to a decade or century?

Use the coarsest ISO value you can defend — a bare @when="18" is not valid, so use @notBefore="1800" @notAfter="1899" for "the 19th century", and add a precision element if you want to label the granularity explicitly.

Related reading

• Getting started with TEI P5
• Encode named entities in TEI
• Extract historical dates
• Encode corrections and authorial edits
• Fix common TEI validation errors