RETRIEVAL LAYERS

A record ID and recognized words do different jobs.

Structured index
record_id
ILLUSTRATIVE: ENG-024
series
Plans
date
1982-04-20

Agreed fields support exact retrieval.

OCR text layer

Recognized words can support keyword search.

Illustrative record ID and text shown for explanation only. OCR usefulness varies with source quality and does not verify critical fields.
TWO WAYS TO FIND A RECORD

Indexing and OCR work together.

Structured indexrecord_id: DRAW-0124
date: 1982-04-20
source: CARD-017
Retrieve by agreed, reviewed fields.
OCR text layer

Recognized words from the image can support keyword search.

Recognition varies with source quality.
Fictional identifiers shown for explanation. OCR is not a replacement for an agreed index or verification of critical fields.

Begin with how users retrieve records

Ask what someone knows at search time: a document date, case number, account number, drawing number, title, report period, location, source box, or a frame reference. Then ask whether that value is present on the item, in a register, in a header, in a punch-code map, or only in an external system. A field without a defined source becomes difficult to trust and maintain.

Also decide what level the field describes. A series title might apply to an entire collection. A reel identifier applies to a roll. A fiche header identifies a sheet. A frame, page, or drawing record needs its own source reference. Mixing these levels into one undifferentiated row makes later retrieval and imports harder.

Use collection, carrier, and item layers

Collection-level fields can include record group, department, system, title, date span, and retention-relevant categories supplied by the customer. Carrier-level fields can describe box, drawer, reel, fiche, jacket, channel, strip, card, polarity, or reduction when known. Item-level fields can include document date, title, case/account/drawing number, source page, report period, or frame position.

Not every project needs every field. For a report-style COM collection, report title, period, fiche sequence, and frame may support retrieval. For aperture cards, drawing number, sheet, revision, and title may be more useful. For a roll of correspondence, date, sender/recipient fields may be sensitive or not suitable for a public index; the approved use case governs the map.

Record layerExample fieldsSource examples
CollectionSeries, department, date spanFinding aid, existing database
CarrierReel ID, jacket/channel, card IDLabel, header, container, observed item
Item/frameDate, title, case/drawing numberDocument image, title block, approved register
DeliveryRecord ID, path, OCR statusProduction manifest and data dictionary

Store where an important value came from

Source provenance makes an index easier to correct. For a field such as revision, possible sources could be “drawing register,” “card label,” “title block OCR,” or “manual review.” This does not weaken the index; it tells later users how the value was derived and where to look if evidence conflicts.

Fictional example — no PII: a card record may show `drawing_number=ME-2047` from `card label`, `revision=B` from `title block`, and `location=Plant 3` from `drawing register`. A stable card ID links each value to the captured source image rather than hiding all evidence in one text string.

Document the fields before delivery

A data dictionary should list source field, normalized field, target field, data type, allowed values, date format, repeatability, required status, and transformation rule. This is the bridge between inventory language and a repository’s content types, templates, document classes, or object properties.

National Microfilm can review customer-defined fields, file relationships, OCR status, and destination needs as part of a project scope. The final map follows the approved retrieval and import workflow; it is not assumed from a generic microform template.

CLEAR ANSWERS

Common questions

Should every visible label become an index field?

No. Choose fields that answer real retrieval needs and have a stable, documented source. Other information can remain in images or notes.

What is a source-linked field?

It is a value associated with where it was observed or derived, such as a reel label, card text, title block, register, or approved review step.

Can OCR populate index fields?

It can assist extraction, but critical fields should use an agreed source and exception process rather than assuming OCR is infallible.

Sources & further reading

These references explain technical formats, standards or destination requirements. They do not imply endorsement of National Microfilm.