HistoriaMP does not democratize truth

Truth does not emerge through voting. Not through AI either. And not because as many people as possible confirm a plausible reading.

For historical manuscripts, that would be the wrong claim.

The decisive point lies elsewhere: not every person has to be able to read a source completely in order to contribute meaningfully to its interpretation. But every observation, correction, doubt and proposed reading must have a traceable status.

Truth is not democratized. The path toward it is.

The Problem of Smooth Answers

Many digital methods begin with an apparently simple goal: a source should become text.

HTR systems can be very powerful. Especially for serial sources, church books, baptism registers, marriage lists or death registers, there are good examples of machine transcription and human correction working together. Non-academic contributors can also play an important role here, for example by correcting HTR outputs and thereby improving models.

But for more complex manuscripts, this approach is not always enough.

An HTR output can smooth the source. It can overlook special signs, resolve or lose abbreviations, turn uncertain places into apparently clear words and make graphic features invisible. The result then looks like a transcription, but it is already an interpretation.

The problem is not only that such outputs contain errors.

The problem is that they often look convincing.

The Source Before the Text

For this reason, machine-generated text must not automatically become the basis for further analysis.

The finding comes first:

• What is visible?
• Where is it visible?
• What structure is present?
• Which sign form can be recognized?
• What is secure?
• What is uncertain?
• What is only a candidate?

These questions are slower than a finished text. But methodologically, they are cleaner.

A historical reading is not reliable merely because it appears linguistically plausible. It becomes reliable only when it can be traced back to the source.

From Visible Evidence to a Justified Reading

At HistoriaMP, this idea does not remain abstract. It is already built into Pipeline v3.

The pipeline does not start from a finished transcription, but from visible evidence, documented uncertainty and traceable intermediate steps.

The principle is:

Visible Evidence → Uncertainty → Justified Reading

So not:

Image → finished text

But:

visible finding → documented uncertainty → justified reading

The reading does not stand at the beginning. It is the result of an auditable path.

Participation Without Arbitrary Claims

This opens an important space.

Many people are interested in historical sources: genealogists, local historians, archive users, technically interested people, open-source developers, or people who can still read certain old scripts. Not all of them are scholars. Not all of them can fully transcribe a difficult source.

But many can contribute meaningfully.

• They can mark conspicuous places.
• They can check segmentations.
• They can identify damaged areas.
• They can compare recurring sign forms.
• They can make HTR omissions visible.
• They can check place names and personal names as candidates.
• They can document uncertainty.

Such contributions are not finished truth. They are observations with status, image reference and review need.

That is the decisive difference between open participation and arbitrary assertion.

Status Instead of Authority

The central question is not only:

Who says it?

But:

On what basis is it being said?

An AI output can be useful, but it remains a candidate. An HTR transcription can be helpful, but it does not replace the finding. A community observation can be valuable if it is documented cleanly. A specialist reading still remains subject to review if it cannot be traced back to the source.

What matters is the status of a statement:

• Observation.
• Candidate.
• Uncertain.
• Checked.
• Rejected.
• Confirmed.
• Review required.

This creates participation without giving up scholarly control.

The Real Aim

The aim is not to turn historical sources into smooth text as quickly as possible.

The aim is to make visible how a reading comes into being.

That is less comfortable than automatic output. But it is more honest.

With historical sources, what is read at the end is not the only thing that matters. It also matters why this reading is possible, where it remains uncertain and which alternatives were rejected.

Conclusion

HistoriaMP does not democratize truth.

Truth does not arise from numbers, machines or assertion.

What is democratized is the auditable path toward a reading: through image reference, candidate status, uncertainty marking, review and traceable decisions.

The focus is not the fast answer.

It is the evidential path toward it.

The HistoriaMP pipeline demonstration shows what this looks like in practice: from visible evidence through uncertainty to a justified reading.