Managing Sources, Paradata and Metadata

The point of an EM graph is not the units, it is the reasoning chain behind each unit: which sources support it, what was extracted from each source, how partial extractions were combined, and what properties the combination justifies. This tutorial shows how to build that chain cleanly, how to expose it to the user in Blender, and how to keep your DosCo folder tidy enough that the chain still works two years from now.

Learning objectives

By the end you will be able to:

  • structure a DosCo folder so every source has a stable, citeable path;

  • author a complete paradata chain (Document → Extractor → Combiner → Property) that travels with the unit;

  • decide when an extractor is justified and when a property can be attached directly;

  • review and audit paradata in Blender via the Paradata Manager.

Prerequisites

DosCo discipline

DosCo (Documentation Source Collection) is the folder where every source the graph references lives. The contract is simple:

  • one folder per kind of source (photographs, reports, drawings, analyses);

  • file names that match the id of the corresponding Document node (DOC.012.pdf, DOC.012_p4_detail.png, …);

  • never reference files outside DosCo from a Document node — copy them in;

  • never rename files after they are referenced, only add.

If the contract holds, the EM Data Tree DosCo path field is the only setting that ever needs updating when the project moves.

The walk-through

We continue with the Casa di Esempio / Trench A example. Pick one US whose interpretation rests on more than one source — for instance, US.A.005 whose dating rests on a coin, a stratigraphic position, and an analogous context elsewhere.

Step 1 — Document nodes for each source

Data

Panel

Action

Output

the three sources for US.A.005

yEd

Add three Document nodes (DOC.005a, DOC.005b, DOC.005c); fill metadata; place in DosCo.

Three Document nodes resolvable to files on disk.

Step 2 — Extractors per source

An Extractor records what you actually pulled from a source. If the report says “the layer contained a coin of Hadrian and a fragment of late-Antonine sigillata”, the extraction is terminus post quem 138 CE — not the whole report.

Data

Panel

Action

Output

each Document

yEd

Add an Extractor node per source; fill method (e.g. temporal_tpq), value, confidence; connect Document → Extractor.

Three extractions, each citing exactly one source.

When not to use an extractor: if the property is directly observable from the source (e.g. a measured length on a drawing), attach the property to the Document directly. Reserve extractors for interpretive steps.

Step 3 — Combine the extractions

When several extractions converge on the same property, a Combiner records the act of synthesising them.

Data

Panel

Action

Output

the three Extractors

yEd

Add one Combiner node; connect all three Extractors into it; fill rationale.

One node that says “these three together imply X”.

Step 4 — Attach the property

Data

Panel

Action

Output

the Combiner

yEd

Add a Property node (dating, value 2nd c. CE, confidence high); connect Combiner → Property → US.

The unit now has a justified property whose chain you can audit.

Step 5 — Audit in Blender

Data

Panel

Action

Output

the loaded graph

Paradata Manager

Select US.A.005; enable Filter Paradata.

The three Documents, three Extractors, the Combiner and the Property surface in a connected list.

Self-check

  1. Every property on US.A.005 has a chain that ends in at least one Document.

  2. No Document referenced in the graph is missing from DosCo.

  3. The Paradata Manager’s filter view shows the chain in the expected order.

  4. Removing one of the three sources (in a copy of the graph) visibly weakens the chain — the Combiner rationale should make this auditable.

Common pitfalls

  • Stuffing everything into one Document node (“the report”) instead of one per logical citation. The chain becomes opaque.

  • Skipping the Extractor and connecting Document directly to Property when the inference is non-trivial — you lose the what was actually extracted layer.

  • Combiners with one input. If only one extraction supports the property, attach the property to the Extractor; reserve Combiners for genuine syntheses.

Where to go next