Back to Cases

Testwork

Testwork Atlas: testwork records with provenance

Turn a pile of testwork reports into reviewed, comparable and traceable records. Every value keeps its source table and cell, every sample keeps its lineage, and nothing becomes canonical until a person publishes it.

Request a Free Session

Demo with made-up data · In development

Who it is for

Built around one practical workflow first

  • Study, due-diligence and consulting teams who hold many clients’ reports, did not commission most of the testwork, and are accountable for statements about it.
  • Process engineers and testwork coordinators who need every locked-cycle, Bond or bottle-roll result across campaigns in one comparable table.

Workflow pain

Where the work usually slows down

  • Results are re-typed from scanned PDFs, and a corrupted micron symbol turns 40 µm into 40 m without anyone noticing.
  • A Qualified Person must say whether the tested samples represent the deposit, and the sample origin is buried in prose three reports back.
  • A modified Bond test and a standard Bond test share a name, so unlike results get compared as if they were the same measurement.

Implementation design

What Eleticle builds and controls

What is built

  • Import of PDF, DOCX, XLSX, CSV, image and ZIP report packs, with deterministic adapters for Bond series and drop-weight summaries, batch and locked-cycle flotation, and bottle-roll, CIL and CIP bench leach.
  • A three-pane review screen: source viewer, extracted candidates and the canonical form, with every field linked to its source span.
  • Sample lineage, comparability warnings, a test matrix, the supported plots, and a full export so a published recovery equation can be re-derived from its inputs.

What data is used

  • Your testwork reports, laboratory workbooks and legacy archives, uploaded by your team.
  • Your programs, campaigns, sample codes, and the tolerances your administrator sets.

What AI may do

  • Propose candidate values from prose and irregular tables, each citing its source span.
  • Summarise approved evidence and state reported differences between runs.
  • Search approved records and return a sample’s lineage.

What it does not do

  • Not a flowsheet simulator: it does not size equipment, scale up or recommend testwork.
  • Heap and column leach, gravity, dense-medium, thickening, filtration and mineralogy families are reported as unsupported, never parsed by a neighbouring adapter.
  • It compares. It never states that unlike tests are equivalent.

Human review

Extraction proposes. People verify and publish.

The assistant is scoped around clear boundaries: what it can read, what it can prepare, what must be reviewed, and what should stay manual.

A granted member reviews, edits or rejects every candidate; an organisation administrator publishes, singly or as one audited batch.
A correction creates a new version and marks dependent reports stale; the prior version stays visible.

Before and after

Example workflow change

Before

Numbers are copied from reports into a master spreadsheet, and the question of where a recovery came from takes a day to answer.

After

Every value opens its source cell, every sample shows its origin, non-comparable runs are flagged, and the whole result set exports with its provenance.

More workflows

Industrial operations

Private AI for industrial operations and technical workflows

QA/QC

Assay Guard: laboratory QA/QC review

Metallurgy

Met Daily: the daily metallurgical review, signed

Comminution

Grinding & Classification Monitor: what the circuit did

Start with one free session

Share the workflow you want to improve. Eleticle will help identify the safest useful first step.

Request a Free Session