One governed record should serve the shift and the model

A record created beside the work preserves evidence that later reconstruction cannot reliably recover. The supervisor can inspect a partial bundle, confirm a stop cause or resolve a meter identity while the physical condition still exists. An end-of-week clerk sees only totals and memories. Capture at source therefore means capturing at the responsible point of work, not merely typing an old sheet into a computer sooner.

Garment research demonstrates that detailed source capture is technically possible. One study describes RFID recording front-line production data in varied, small-batch garment production.[S19] Another uses sewing-machine power monitoring to collect consumption data and extract working times for sewing tasks.[S20] These are capture methods, not automatic proof of operational meaning. An RFID read still needs a defined bundle and boundary. A power-derived interval still needs validation against the work event it is intended to represent.

Manual and sensed capture should consequently share the same event contract. A handwritten completion sheet and a machine message can both supply an operation-completion record if both preserve identity, quantity, event time and source. The manual route may carry a later entry time. The sensed route may carry device health and signal-quality flags. Neither route should invent missing context to make the row look complete.

Four timestamps or time attributes answer different questions. Event time records when work occurred. Entry time records when the account was created. Receipt time records when an authoritative service received it. Availability time records when it became eligible for a named review or calculation. A lightweight manual system may not need all four as separate fields, but it must not pretend that entry time is always event time. Late and backdated entries remain visible rather than being silently absorbed into an apparently contemporaneous history.

The shift calendar supplies another essential boundary. Work crossing midnight belongs to the shift defined by the operating calendar, not to whichever calendar date a device happens to print. The source record retains that shift identity alongside the raw timestamp. The review cut-off then determines whether a valid late record belongs to the next operational review, a later reconciliation or both, with each use labelled accordingly.

Source capture should also make correction possible without making falsification easy. The person closest to the work can request a correction and state the reason. The authorised reviewer accepts it where the operating procedure requires separation. The original value, revised value, author, approval and timestamp remain linked. Overwriting a count because the shift total looks wrong destroys the evidence needed to discover why it was wrong.

Good capture design reduces decisions at the point of work. Controlled lists replace free-text defect and downtime codes. Default values are limited to facts the system can know, not assumptions about production. Partial entries are marked unfinished rather than coerced into valid rows. The operator sees a clear route for an unusual event, while the steward sees an exception queue instead of discovering a problem only after the plant report changes.

Automation earns its place through a measured failure or burden. If manual transcription repeatedly loses event time, a station terminal may help. If bundle identity is unreliable, RFID may merit a controlled trial. If stop causes are missing, machine states may support but not replace supervisor classification. Each trial compares the captured record with physical work and tests exception handling. The source technology changes only after the event contract remains intact under both routes.

Production events acquire meaning through the reference records attached to them. A style code identifies a product only if its definition is stable. An operation code supports comparison only if the underlying method and routing remain distinguishable. NIST's smart-manufacturing review states that the exchange, understanding and use of product, production and business data depend critically on information standards.[S21] In a greenfield factory, the practical expression of that principle is a small, owned master-data register with effective dates.

Master data differs from transaction data. The order, style, operation, defect, downtime cause, machine and meter registers change through controlled releases. Production, inspection, interruption and resource events then reference the versions in force when work occurred. A revised routing does not rewrite the events recorded under its predecessor. A retired defect code remains readable in historical records even though new entries cannot select it.

Effective dating is more useful than overwriting. Each master version carries a start, an end where applicable, an approver and a reason for change. The routing register links style, operation sequence, work-content basis and method revision. The line register preserves changes in physical layout or organisational assignment. The meter register identifies boundary, unit, reading convention and successor where hardware changes. History resolves against the applicable version rather than today's labels.

Identifiers should not be reused after their meaning changes. If an operation is split, renamed or given a materially different method, the owner decides whether this is a new version or a new entity and records the relationship. A code that meant one operation last month cannot quietly mean another this month. Such reuse creates a continuous time series from discontinuous work and can contaminate both operator comparisons and model features.

Taxonomies need the same discipline. Defect codes describe observable conditions before suspected causes are assigned. Downtime codes separate waiting, changeover, failure and no-work conditions where those distinctions lead to different responses. Meter and resource-factor registers preserve units and accounting boundaries. Adding a code requires an operational definition, examples where useful and an owner. Merging codes requires a migration rule that does not pretend the original detail never existed.

Exhibit 47 assigns the minimum register to accountable roles. The roles are illustrative and may be combined in a small organisation, but the accountabilities should not disappear. The table also distinguishes the authoritative record from the many reports that may read it.

The register is a working control, not a documentation annex reviewed once a year. Capture forms draw their allowed values from it. Validation rejects inactive codes at event time while allowing historical reads. Reviewers can see when a method or organisational change interrupts comparability. A migration team can map old identifiers without guessing from free text. These uses justify the maintenance effort.

Stewardship should sit near operational knowledge. Industrial engineering understands routing and method revisions. Quality understands defect meaning. Production understands shift and line boundaries. The data steward coordinates the rules and tests, but should not be expected to invent operational definitions alone. Escalation resolves conflicts between convenience and meaning, such as a request to reuse a code to avoid changing a report.