An investigation with a spine you can see — and leave.
Day-1 intelligence, but never a free-for-all. Every analysis moves through named stages, each run by its own specialist — surfaced to you as 10 steps where the next does not open until the one before it has produced something real.
Triage & Containment
Incident data and initial containment actions
Evidence Capture
Photos, measurements, and on-site observations
Is / Is-Not Analysis
Structured comparison to isolate the failure
Change Point Timeline
Changes in 72h before the incident
Hypotheses
AI-generated and engineer-added root cause candidates
Evidence per Hypothesis
Targeted evidence collection for each hypothesis
Root Cause Declaration
Confirmed root cause and evidence chain
Actions & Control Plan
Corrective actions and verification plan
Closure Record
Documented information, read-across and the effectiveness criterion
Lessons Learned Review
Approve AI-generated lessons before report generation
Save and exit mid-investigation, and it waits. Nothing is lost and nothing is quietly finished for you — the analysis pauses exactly where it stands and resumes there when the engineer comes back.
The quality gate can say no.
One review, four possible outcomes. The count is the point: a gate that can only ever wave work through is not a gate at all.
It clears, and moves on
When the chain of evidence holds up, the gate advances the analysis to the report on its own — no ceremony, no manual push.
It sends the work back
Where the analysis is thin, the gate returns it to be strengthened and names what is weak — so a gap is fixed here, not discovered in the audit.
You can proceed — it is recorded
You may take a mediocre analysis to a report, but the override is written down and carried into the report and its corrective actions. Proceeding is a decision with your name on it, not a silent default.
Sometimes there is no proceed
When the evidence floor itself is not met, the decision has no user-facing override — the only way forward is back to analysis to put real evidence behind the conclusion.
And a decision nobody ever makes is marked unmade — a review left hanging is flagged as unresolved, never quietly waved through to a report.
The honesty is in what never reaches you.
Five things the investigation refuses to do quietly — each one visible in a live demo, none of them a setting you can turn against yourself.
A named team, and a second pair of eyes
An investigation carries the people on it, each with a role. When it is done a named reviewer is assigned — told by email and in-app — and their approve-or-return decision is recorded. Every inline edit is tracked to whoever made it.
When the AI was degraded, the report says so
A hypothesis drawn while a model was unavailable is labelled provisional and its confidence colouring is withheld — an unlabeled percentage on stand-in content is false precision. Degraded content can never satisfy the evidence floor, so it can never carry an investigation to sign-off.
A cause with no evidence never reaches you
Candidate causes are drawn strictly from this investigation’s own evidence, and every one has to cite a specific piece of it. A cause that can cite nothing is discarded before it ever reaches your screen — not shown with a caveat pinned to it.
Lessons wait for an engineer to approve them
Every lesson the analysis proposes stops at a person. Approve it, edit it, or reject it — only the approved and edited ones publish, each stamped with who approved it and when. Nothing writes itself into what your plants will search tomorrow.
Recurrence stops the work and asks you to acknowledge it
When a failure is recognised as one you have investigated before, the pipeline holds. Before it goes on, it asks the engineer to acknowledge — in as many words — that the last set of corrective actions may not have held.
One click to an audit-ready report.
Every analysis resolves into the same five-section report — traceable, author-tracked, and exportable for your audit without reformatting.
Executive summary
A recurring process deviation on a primary production line was investigated, the systemic cause confirmed against prior history, and a sequenced corrective plan issued. No personal fault identified.
Root cause
A process-control gap allowed an upstream parameter to drift uncorrected between scheduled maintenance, producing the observed failure mode under specific operating conditions.
Evidence
Conclusions are traced to source records — operating logs, maintenance history, and matched prior incidents.
Corrective actions
Sequenced as SOP update → crew training → field verification, with closure gated on confirmed production evidence.