What "calibrated" actually means.
Most safety platforms are calibrated against a feature checklist. Pear MS is calibrated against twelve years of structured incident, risk and progress data from the projects that built the offshore-wind industry. This page tells you exactly what the corpus contains, how it was assembled, and what it deliberately leaves out.
8 → 72% — risk-assessed work, before / after.
When the same workforce moved from a homegrown HSG48 system to a corporate Tripod-Beta one, the proportion of work covered by an active risk assessment at the moment the work was performed nearly doubled. Pear MS is the third generation of that curve — designed to lift it again, by removing the ergonomic frictions that suppressed it in the first two.
The shift was not a software win. It was a workflow win. The Tripod-Beta system happened to make the right thing easier and the wrong thing harder — and the curve moved. Pear MS encodes the lessons of that shift directly: every design rule traces back to something the historical data showed.
7,315 events. Every one structured.
The corpus is not a folder of accident reports. It is the structured residue of a decade-plus of operational records: incidents, near-misses, risk-assessment versions, CAPA actions, daily-progress entries, downtime cause codes. Captured in their original contractor systems, then normalised into a single relational schema for analysis.
Twenty-one calendar years from the early UK Round 1 wind farms through to in-construction US East Coast projects.
Named offshore-wind installations the IMS has been calibrated against or designed to operate within. UK, Continental Europe, and US East Coast.
Includes accidents, incidents, near-misses, RA changes, CAPA actions, downtime entries, and daily progress events.
Capture originated in five different contractor operational systems across the corpus period. All normalised into one consistent structure for analysis without losing per-system traceability.
Methodology, in plain language.
What we kept
Every record where the event date, the activity code, the location and the responsible party could be reconstructed unambiguously. Records were retained even when the original causal analysis was thin — those records still tell us what happened and where, even if they can't tell us cleanly why.
Where the same event appeared in two systems (e.g., recorded as both a near-miss and a downtime entry), it was de-duplicated against a composite key of date, location, activity and contractor — but the original system tags were preserved so the cross-system patterns could still be analysed.
How we structured it
Source records from each project were normalised into a single structured form. The structure unifies records across the dimensions that matter for analysis — when, where, what activity, by whom, against which controls, under which contract — while preserving traceability back to the project and operational context the record originally came from.
The same structural rigour that built the corpus shapes how Pear MS handles operational data today. Methodology and product are deliberately consistent — what the corpus learned, the live system encodes.
What we excluded
Records where the date or the activity could not be reliably reconstructed. Records flagged by the originating contractor as commercially sensitive. Records from non-offshore-wind projects (oil & gas crossover work has been excluded for calibration purposes — different hazard profile, different regulatory regime).
PtWIssued (Permit-to-Work issued) records were de-prioritised in calibration because the field is overwhelmingly default-false in the source data — its presence or absence carries no signal in the historical record.
What it does not claim
The corpus is not a complete census of UK / EU / US East Coast offshore-wind incidents. It is a deep slice of the projects the contributing contractors worked on, normalised to a single schema. Where the site copy says "calibrated against 2,000 turbines," that is the rough installed-capacity scope of the projects represented — not the count of turbines whose every event is in the corpus.
Inferences from the corpus are most reliable for offshore-wind installation and early-operations phases, less reliable for long-tail O&M, and weakest for decommissioning (limited data so far).
31 named projects, grouped by sea.
The projects below are the named offshore-wind installations against which the methodology was calibrated. They are the calibration domain, not the scope of the product — the controls, the evidence patterns, and the risk taxonomy transfer cleanly to oil & gas, deep-sea mining, and the offshore ventures coming next. Only the asset descriptions change.
North Sea — UK 10
- Burbo Bank
- Dudgeon
- Gunfleet Sands
- Gwynt y Môr
- Kentish Flats
- Lincs
- London Array
- Lynn–Inner Dowsing
- Sheringham Shoal
- Westermost Rough
Irish Sea — UK 4
- Rhyl Flats
- Robin Rigg
- Teesside
- West of Duddon Sands
German Bight 10
- Amrumbank West
- Baltic 2
- Borkum Riffgat
- Borkum Riffgrund 1
- Butendiek
- DanTysk
- Gode Wind 2
- Meerwind
- Sandbank
- Veja Mate
Other European 4
- Anholt (Kattegat, DK)
- Formosa (Taiwan Strait)
- Gemini (NL)
- Westermeerwind (IJsselmeer, NL)
US East Coast 3
- CVOW (Coastal Virginia)
- Empire Wind (NY Bight)
- Ocean Wind (NJ)
Calibration depth varies by project. The earlier UK and German Bight installations contribute the deepest historical event data; US East Coast projects contribute design context for an in-construction regulatory environment. We do not claim equal-depth coverage across all 31 — we claim that every project listed has shaped at least one concrete design rule in the IMS.
Want the methodology in writing?
We can share a one-page methodology brief and the canonical schema definition under NDA for prospective IMS auditors, certification bodies, and procurement evaluators. Tell us who you are and what you need to see; we'll send the right artefact.
Request the brief or read the broader story → About Pear ms.