Computer-Vision Ore-Grade Model Diligence Director

The mandate

The acquisition case assumes that a target’s camera and spectral models will transfer from two reference mines to a broader ore and lighting portfolio while preserving recovery and reject quality. The defined problem is to test that assumption against raw production evidence, sensor configuration, calibration practice and geology rather than value the company or certify the equipment.

The deliverable is an Ore-Grade Vision Diligence Opinion containing claim-to-evidence lineage, dataset and site coverage, model and calibration inventory, production confusion economics, sensor and illumination dependencies, transfer tests, maintainability assessment, downside scenarios, red flags and a 180-day mitigation budget. It must separate addressable engineering debt from failure of the acquisition thesis.

Work starts 11 January 2027. Milestone one on 22 January is the evidence sufficiency map and urgent red-flag notice; milestone two on 5 February provides site observations, replay tests and transfer findings; milestone three on 5 March delivers the signed opinion, committee presentation, sensitivity model and conditions for any transaction close.

Acceptance requires geology and finance sponsors to reproduce sampled classifications from retained imagery, spectral inputs, labels, calibration and model versions and then recalculate value under false-reject and false-accept costs. Each critical dependency needs an owner, lead time and cost; vendor demonstrations, hand-picked belts or aggregate recovery without denominator evidence will not pass.

The client will provide secure data-room access, raw sensor streams, laboratory and production assays, belt conditions, model artefacts, maintenance records, customer contracts and target interviews. Legal, tax and corporate valuation, new drilling, hardware certification and post-close remediation are excluded; the consultant may recommend but not negotiate transaction terms.

Why this is external work

The target controls most evidence, while the buyer’s operations team is invested in the promised recovery improvement. Internal data scientists have not industrialised ore-grade vision across changing geology and sensor conditions. Independent technical authorship gives the investment committee a falsifiable transfer opinion before it commits acquisition capital.

What you will own

• Reconcile claimed tonnes, recovery and reject quality to mine, belt, sensor, calibration, model, sampling and assay evidence.
• Map training and validation coverage across ore type, particle size, moisture, dust, illumination, belt speed, sensor state and operating intervention.
• Test transfer through held-out site and time samples, calibration perturbation, sensor degradation and geologically distinct material.
• Translate false accept, false reject and abstention into metal loss, dilution, rehandling, throughput and downstream processing economics.
• Challenge maintainability through label turnaround, assay latency, drift detection, recalibration, model release and mine-operator capability.
• Model acquisition downside under slower site transfer, sensor replacement, limited data rights, lower availability and greater expert support.
• Deliver a signed investment opinion stating deal conditions, invalidating evidence, mitigations, residual uncertainty and first 180-day priorities.

Candidate qualifications

• Led technical diligence or industrialisation of computer vision, spectral classification or sensor-based sorting in mining or mineral processing.
• Linked model classifications to laboratory assays and production economics across ore variability, not solely curated image benchmarks.
• Exposed a transfer or calibration failure between mine sites and can quantify its effect on recovery or reject quality.
• Evaluated camera, illumination, spectral or mechanical dependencies deeply enough to separate model weakness from sensing and process conditions.
• Built an investment sensitivity model that changed transaction conditions, valuation assumptions or the decision to stop.
• Conducted confidential target and site work while preserving evidence lineage suitable for investment-committee challenge.

Non-negotiables

• The named director must attend the Johannesburg reviews and all four target or mine-site evidence days within eight weeks.
• Must be independent of the target, its investors, key sensor providers and advisers compensated on transaction completion.
• Will sign the final opinion and state where unavailable data prevents a transfer or production claim from being supported.
• No target model, geological record, imagery or assay evidence may leave the approved diligence environment.

1. 49 words maximum. Describe an ore or industrial vision model that failed to transfer and the production evidence that exposed it.
2. 49 words maximum. How would you convert classification errors into mine economics when assays arrive later than sensor decisions?
3. 49 words maximum. Disclose every target, investor, sensor-vendor or transaction relationship relevant to your independence.