TallyVision defect labels need image-to-lot lineage

The image must retain its production identity

Each observation should preserve the source camera and configuration, line and station, capture time and time zone, image identifier and integrity, inspected item or sequence, raw-material lot, work order, finished-product lot where assigned, supplier, species and product form, shift, operator context, and any merge, split, rework, or repack event. A filterable lot field is useful only when the join from image to physical production remains testable.

Continuous inspection also needs a denominator. Retain expected and observed item counts, line speed, camera uptime, unreadable or rejected images, skipped frames, duplicate captures, manual bypasses, and periods outside the validated operating range. A high number of labeled images does not establish that every relevant fillet or every portion of the lot was observed. Missing vision coverage should remain visible in the lot record.

Classification needs model and threshold provenance

For every result, preserve the model and software version, label taxonomy, size or severity method, color scale and version, decision thresholds, preprocessing, confidence or score where available, inference time, output, manual review, correction, and reason. A term such as bruise, softness, or color class can have operational meaning only under defined acceptance criteria and product context.

Training performance and production performance are different evidence. Lighting, camera angle, water, ice, species, cut, line speed, overlapping product, equipment drift, and seasonal variation can alter results. Keep validation populations and known limits with the deployment. When a model or threshold changes, do not silently relabel the historical images used for a prior hold, grade, release, or supplier discussion.

A defect label is not the disposition

The machine result should route into a governed quality decision that identifies product and quantity, specification and version, reviewer, corroborating evidence, applicable sampling or inspection plan, hold status, segregation, regrade, trim, rework, release, destruction, supplier action, and approval time. One labeled fillet may affect only that item, a sampled population, a work-order segment, or a wider lot depending on validated rules and evidence.

Traceability records should connect the affected material to receiving, transformation, inventory, packing, and shipping events without absorbing the quality decision. A complete genealogy does not prove conformance, and a quality rejection does not prove where all affected product moved. Preserve both evidence lanes and their reconciliation, including quantities and unresolved material, before narrowing a hold or recall scope.

Test coverage, drift, and a disputed label

A representative evaluation should run two supplier lots through one work order, create a line interruption, miss images during camera downtime, change lighting, include overlapping fillets, apply a new model version mid-shift, and route a low-confidence defect to a human who disagrees. Reviewers should reproduce the image-to-item-to-lot path, coverage denominator, original output, override, affected quantity, and final disposition.

Then trace released, held, reworked, and shipped quantities through a later customer or supplier inquiry. The current ThisFish page establishes the described image capture, size, color, defect-classification, dashboards, filters, and report positioning; the registered URL redirects to that current TallyVision page. It does not establish classification accuracy, complete inspection, lot linkage, specification conformity, disposition, recall scope, regulatory compliance, or food-safety outcome.

Enterprise buyer test

Translate this change into the exact population, record type, workflow stage, decision owner, effective date, and evidence that could be affected. Ask current or prospective providers to demonstrate the named workflow with representative data and an exception—not a polished feature tour. Record what official documentation establishes, what a provider states, what the team observes, and what remains unresolved.

A defensible review also identifies the dependency outside the product. Authority interpretation, policy configuration, data quality, integrations, human judgment, approval rights, release governance, training, and retained evidence may remain customer or service responsibilities. The evaluation should preserve those boundaries instead of treating a technology claim as the complete operating model.

What we will watch next

Food Traceability Ledger will watch the named source and affected market records for later evidence that changes status, scope, availability, implementation timing, workflow consequence, or the limits of the initial report. A later announcement does not silently overwrite this dated account; the change ledger preserves the sequence.