Maintenance · Workload · Human judgement

Maintenance decisions are made by people, under operational conditions.

Full Ahead Maritime treats maintenance as a human decision system, not only a schedule of machinery tasks. Alongside what was due and what was done, it keeps the decision behind a change — who acted, in what role, when, and on what stated reason — and protects that record against later alteration. From the same maintenance records it then examines work patterns for a view of the maintenance workload those decisions were taken under.

What is in place today: the maintenance decision record

This part is built, exercised and running. It is the demonstrated half of the proposition.

What was due, and what was actually done

The job, its due position, the action taken and the resulting values are held together in one record, so the schedule and the work done against it can be read side by side rather than reconstructed.

Who acted, in what role, and when

The record carries the acting account, its role within your workspace and the timestamp, so a later reader can see who made the judgement and when they made it.

The reason, where the product requires it

When a maintenance job is re-timed or a tolerance is applied, a written reason above a minimum length is required before the change is accepted. Supporting evidence attached to the job stays with it.

Append-only, protected against alteration

The audit trail and the hash-chained ledger behind it are append-only in this implementation. Update, delete and truncate are refused at the database level, including for ordinary application roles, and that refusal is exercised as a test rather than assumed. Other systems keep history too; what is stated here is how this one behaves, verified in production.

What we are exploring: work patterns as a view of maintenance workload

Maintenance work has human workload consequences. These are the patterns the deterministic model reads from records the workflow already produces — no extra form, nothing to wear. They describe the work, not the person.

Night-window work

The share of technical work recorded during circadian night hours. This is derived from timestamps the workflow already produces, not from a rest-hours declaration filled in afterwards.

Task switching

How often recorded work moves between unrelated jobs within an active period. It is a work-pattern measure, not a measure of a person's concentration.

Sustained-duty streaks

Consecutive days with recorded work and no clear stand-down, aggregated per vessel and per role — never as a ranking of named seafarers.

Maintenance backlog and priority mix

How much critical work is open and how the priority mix is moving. This describes the load the vessel is being asked to absorb; it does not describe any individual's state.

Stated plainly: this is an emerging model, not a prediction engine

  • Three things are kept apart: the published research basis linking night work, task switching and sustained duty to degraded performance; the product implementation, which is fixed arithmetic over recorded timestamps and job data; and live-vessel validation, which has not been done.
  • It is not presented as validated prediction of human error, degraded judgement, sleep, alertness, incidents, near-misses, deficiencies, or any specific lead time.
  • The product has not established a reduction in errors or failures, a change in incident outcomes, or a monetary saving. No such outcome is claimed.
  • Nothing is published below the sufficiency floor of 12 signals across 3 distinct days for a vessel; below it the platform returns insufficient data, not a number.
  • The intent is to surface a question for a competent human — a superintendent or chief engineer who knows the ship — never to issue a verdict on a seafarer.

The current model holds no biometric, physiological, wearable, camera or medical fields. Capture is consent-gated, scoped to your workspace by row-level security, and always aggregated to vessel and role rather than scored per person. That describes what the product handles — it is not a legal conclusion about your obligations, and it does not make the model medically or scientifically validated. The final maintenance decision stays with the responsible human.

Frequently asked

Evidence base

Each object separates what independent published evidence establishes from what Full Ahead Maritime observes, and states what remains unvalidated.

The maintenance evidence library

Primary sources used by those objects: IMO Fatigue guidance (MSC.1/Circ.1598) and the ISM Code.

The consent-safe alternative

Fatigue monitoring without putting a sensor on your crew.

The wearable and camera-based fatigue category measures the seafarer. Full Ahead Maritime measures the work. That difference means there is no device to roll out and no conversation with the crew about being filmed on watch. What your own data-protection assessment concludes remains a matter for you and your advisers.

What is measured

Work patterns already recorded: night-window jobs, task switching, sustained-duty streaks, maintenance backlog.

The seafarer's body — heart rate, eye movement, sleep, head position.

Personal data captured

None beyond event type, role, session wall-clock duration, offline flag and vessel. No biometrics, no cameras, no keystroke capture — and none planned.

Health and biometric data, which the GDPR treats as special-category personal data.

Who it points at

The vessel and the role. Never a league table of named seafarers.

The individual, by design — that is what the sensor is attached to.

Consent and crew acceptance

Consent-gated per workspace; capture is off until the operator turns it on. Nothing to wear, so nothing to refuse to wear.

Requires informed individual consent that can be withdrawn, plus a device the crew must accept wearing on watch.

Hardware on board

None. It reads the maintenance and compliance work already flowing through the system.

Wearables or cameras to buy, charge, replace, and maintain across crew changes.

If the link drops

Records the underlying work offline on the vessel and syncs later.

Typically streams to a cloud analytics service to produce a score.

General description of the wearable and camera-based fatigue-monitoring category, not a claim about any named product. Verify any specific vendor’s data handling with that vendor.

Honest AI · what this feature actually is

Three kinds of “AI”. We name which one you are getting.

The human-factor indices are fixed arithmetic over timestamps you already record. No model is trained, and no language model sees crew data.

Deterministic rules

Published arithmetic over your own records. Same inputs, same score, every time — traceable back to the row that produced it.

Used here

Statistical calibration

Bounded, auditable corrections to lead times, computed inside your workspace from your own outcomes. No model weights change and nothing is trained on your data.

Not used here

General-purpose language model

Third-party models draft text only. Nothing they produce is applied automatically — a qualified person accepts, edits or rejects it.

Not used here

Read the full methodology note