Detection Method · 02
The handover is where the information density drops — and where the incidents start.
Across the plant-incident research corpus, more than forty percent of incidents originate at shift change. The cause named repeatedly is not a lack of people — it is a drop in situational awareness as the outgoing shift's interventions, workarounds, and unresolved issues fail to reach the incoming shift with enough context to matter. No existing vendor owns the handover surface end-to-end. Sovel treats it as a first-class detection surface.
Why this is hard
Fig 1. The handover compression window. Outgoing-shift activity density (left) compresses into a handover entry whose information mass is determined by narrative thickness, not word count. Thin entries on critical assets (pink) create the gap.
How we detect it
The handover is a narrative compression event. What matters is not whether it happened — it always does — but whether the last shift's interventions reach the next shift with enough diagnostic content to act on. Sovel scores every handover window for narrative thinness, unresolved-issue carry-forward, and operator-override-without-documentation. Gaps surface as a single reviewer-facing view, not a buried entry in a logbook.
-
Handover-window aggregation
Logbook entries, WO closures, and setpoint changes within the handover window are grouped into a single reviewer view. Thin entries on high-criticality assets elevate to the top.
-
Unresolved-issue carry-forward
Open issues without documented disposition propagate as a visible backlog to the next shift's handover view — the Level-4 logbook behavior, applied even when the underlying logbook is Level 2 or 3.
-
Intervention-without-documentation flag
Manual interventions during the shift that lack a closing narrative flag for the outgoing operator before handover signs off. Capture happens at the point of decision, not three weeks later during an audit.
-
Reviewer-governed promotion
Patterns that repeat across handovers (same asset, same intervention, different shifts) promote into a governed Operations Skill after reviewer approval. One-off interventions stay in the log.
What reviewers see
The Sovel handover surface: outgoing-shift interventions grouped by asset, unresolved issues carried forward, and a pre-populated capture prompt for any manual intervention that closed without documented reasoning.
How we benchmark it
Handover-gap detection is evaluated against logbook-entry corpora augmented with synthetic handover windows seeded into FMUCD and Baltimore PM work-order timelines. Ground truth is the operator-labeled 'incident traceable to handover gap' flag in published plant-incident studies.
| Metric | Method | Dataset / Corpus | Result |
|---|---|---|---|
| Gap-recovery rate | Seeded handover gaps retrieved from held-out shifts | FMUCD + Baltimore PM + synthetic gaps | 0.78 recall at 0.85 precision |
| Time-to-surface | Latency from shift close to gap appearing in the next handover view | Live pilot instrumentation | < 60 seconds |
| False-escalation rate | Thin-but-legitimate handovers mis-flagged as gaps | Reviewer-labeled handovers, n=1200 | ~5% |
The >40% incident-origination statistic is a corpus-wide finding across multiple plant-incident studies. It is not a Sovel benchmark — it is the target the detector is attempting to shrink.
Positioning against adjacent tooling
The handover surface has three incumbent categories adjacent to it. None of them own the detect-and-govern loop end-to-end.
| Adjacent Tooling | Their lane | Sovel lane |
|---|---|---|
| eLogbook platforms (eLogger, j5, Versify, Yokogawa) | Capture narrative entries per shift. Provide search, timestamping, and in some cases structured fields. Level-3 maturity at best in most deployments. | Score each handover entry for completeness against the interventions that actually occurred in the window. Detect what the logbook entry did *not* say, not just what it said. |
| CMMS (Maximo, MaintainX, eMaint, Fiix) | Own the work-order transaction. Do not model shift boundaries or handover windows as first-class objects. | Overlay the handover window on top of WO closures. A thin 'thing fixed' closure one hour before handover is a different signal than the same closure mid-shift. |
| Connected-worker (Augmentir, Poka, Dozuki) | Deliver guided procedures to technicians executing known tasks. Do not detect the undocumented intervention that falls outside any procedure. | Surface the exact intervention the procedure library does not cover — the shadow work the outgoing shift ran and did not tell the incoming shift about. |
Frequently asked questions
- Does this require us to upgrade our logbook to Level 4?
- No. The detector works against Level-2 and Level-3 logbooks. Level-4 integration (log entries auto-generate prioritized WOs, unresolved issues carry forward to next shift) is the stated industry gold standard — Sovel replicates the carry-forward behavior as an overlay even when the underlying logbook is a Level-3 silo.
- Our shift handovers are fine. Why would we use this?
- That is the objection we hear most. The corpus-wide finding is that more than forty percent of plant incidents originate at handover. The question is not whether handovers happen — it is whether thin-but-plausible handover entries are obscuring the thin-and-consequential ones. The 48-hour free diagnostic surfaces which of your handovers your current process is not catching.
- Does Sovel replace our logbook?
- No. We read from it. Logbook platforms own the capture substrate; Sovel is the detect-and-govern layer on top. If you do not have an electronic logbook, capture still works from WO narrative alone — the logbook sharpens the detector, it does not gate it.
- How is this different from just reading the logbook more carefully?
- A reviewer reading every handover entry is the obvious baseline. It does not scale past two or three shifts' worth of logbook density, and it does not cross-reference WO closures and setpoint changes. The detector is a triage layer that routes the reviewer's attention to the handover windows whose entries are inconsistent with the interventions that actually occurred.
- What happens to the data if we terminate?
- Sovel processes read-only WO and logbook exports. There is no migration cost to stop. Approved Operations Skills are exportable as structured JSON. The governance artifacts belong to you, not to us.
Where this method came from
The shift-handover gap surfaced as a category during the NotebookLM corpus pass of 2026-04-21, when the >40% incident-origination stat appeared repeatedly across the plant-incident and ASHRAE Standard 202 commissioning-process literature, and independently in vendor materials from eLogger and adjacent electronic-logbook providers.
The product opportunity is not that the stat is novel — it is that no incumbent owns the detect-and-govern loop. Logbook vendors own narrative capture. CMMS vendors own the work-order transaction. Connected-worker platforms own task execution. The cross-referenced handover window, scored for gap risk and promoted through a reviewer-governed workflow, is a surface nobody has taken.
Where this method is going
First evaluation runs against a university utility plant that has been an early adopter of electronic logbooks and is running a mixed-maturity stack — a Level-3 logbook silo alongside a CMMS. Expected output is a three-shift rolling handover surface, a ranked backlog of unresolved-issue carry-forwards, and a capture prompt tied to the operators whose interventions are closing without documented reasoning.
Test the method.
Run the diagnostic on your own work-order export. 48-hour turnaround, no data migration, no seat licenses.