This page is a working reference for the language we use in the room. The glossary mixes charterparty terms with prompting and verification, because that mix is what a maritime desk needs. The three notes, all dated 2026, cover scanned statements of facts, verification habits, and emissions evidence assembled before a verifier asks.
Desk papers before a clinic
Glossary
Working glossary
Terms sit in one alphabetical list on purpose. A post-fixture operator and a prompt pattern belong in the same conversation, because the model is only useful if it can be held to a clause, a log entry or a reading. Definitions are teaching notes, not legal opinions, and they assume British English as used on Singapore desks that work international fixtures.
Charterparty recap
The short-form recap of fixture terms circulated after agreement and before, or alongside, the full form. Exercises treat the recap as a working text to be read against the governing form, not as a substitute for it, and they flag gaps that a model is apt to paper over.
CII
Carbon Intensity Indicator, the operational rating applied to ships under the IMO framework. Training uses it as a commercial as well as a technical figure: desks practise explaining a rating to colleagues who need the cause, not a slogan, and tying the explanation to noon-report and bunker evidence.
Demurrage
The sum payable when laytime is exceeded, calculated from the charterparty’s laytime clause and the statement of facts. Models draft narratives readily; the course insists that every hour claimed cites a clause and an SOF event, and that totals are re-derived by hand once before anything leaves the desk.
EU ETS (shipping scope)
The European Union Emissions Trading System as it applies to shipping, including allowance narratives that commercial and operations staff must be able to follow. Exercises assemble the trail from voyage data; they do not treat a chatbot summary as a substitute for the operator’s own figures and verifier queries.
Hallucination
A fluent output that is not supported by the source documents. In this practice the usual illustration is an invented shifting time on a poorly scanned SOF. The remedy taught is not a stronger adjective in the prompt, but a verification drill that refuses any fact without a visible origin.
Human-in-the-loop
The rule that a named person on the desk remains responsible for anything that leaves the office. Tools may draft; they do not sign. Courses require a human check against clause, log or reading before correspondence, claims, work orders or regulatory narratives are issued.
Laytime
The time allowed for loading or discharging under the charterparty, subject to exceptions and to how NOR and SOF events are treated. AI is used to extract candidate events; the desk still applies the clause, including weather working days, shifting and notice mechanics, and re-checks the arithmetic.
MRV/DCS
Monitoring, reporting and verification under EU rules, and the IMO Data Collection System. The teaching task is an evidence trail: noon reports, bunker delivery notes and port logs reconciled so that a verifier query can be answered from documents the company already holds, not from a generated paragraph.
NOR
Notice of Readiness, the tender that the vessel is ready to load or discharge. Validity often turns on place, free pratique and the precise wording of the NOR clause. Prompt patterns for NOR checklists are trained to list conditions from the governing form, not from a generic port-operations template.
Noon report
The vessel’s regular position, consumption and condition report. Anomaly review compares successive noons against expected speed and bunker figures. Models are useful for spotting gaps; they are not allowed to invent a missing ROB or to smooth a jump that the log does not explain.
Prompt pattern
A reusable instruction, tied to a document type and a verification step, rather than a one-off chat. Cohorts leave with patterns for their own recaps, SOFs, defect notes or circulars. A pattern that cannot name the source to check against is not considered finished work.
PSC inspection
Port State Control inspection of a vessel against international conventions, shaped by port, flag and recent deficiency history. Preparation exercises build a jurisdiction-specific pack from class status, previous findings and the SMS, and they treat a model’s “likely deficiencies” list as a prompt for checking, not as a prediction.
Redaction
The removal of counterparty names, rates, crew personal data and privileged incident detail before any document meets a public tool. Navigatewithai works from client-supplied redacted samples or from a synthetic corpus. Redaction is a desk habit, not a setting left to the vendor’s default.
Retrieval over your own documents
The practice of grounding a model in a defined set of files — a charterparty, an SOF, a class status extract — rather than in the open web. Training stresses that retrieval still hallucinates when the scan is poor or the wrong form is in the set, so citation and a human pass remain mandatory.
SIRE 2.0
The OCIMF Ship Inspection Report Programme in its current inspection model, with observation responses that must address the finding as written. Exercises draft responses from the observation text and the SMS; they forbid invented corrective actions and require every statement to be checkable against a procedure or a record.
SOF
Statement of Facts, the timed record of port events used for laytime. Scanned and photographed SOFs are the hard case: column drift, stamps and handwritten amendments defeat casual OCR. The course treats the image as the source of truth and will not let a model invent a time that cannot be pointed to on the page.
Notes
Working notes from 2026
These notes are teaching material, written for desks that already live with NOR, SOF and noon-report paper. They describe failure modes we build exercises around, not results claimed for any company. Dates are the dates the notes were cut for cohorts; the underlying paperwork problems are older than the tools.
14 January 2026
Why scanned SOFs break most AI workflows
A statement of facts that arrives as a phone photograph or a fax scan is a different object from a clean digital table. Optical character recognition will misread times, especially when columns have drifted after a photocopy, and it will treat a handwritten amendment in the margin as decoration rather than as the operative time. Models then compute laytime from the wrong events and present the arithmetic with unearned confidence. The check that catches this is dull and reliable: a human reads the SOF against the image, lists every event with a time and a source line, and only then allows a model to draft. If a shifting time cannot be pointed to on the page, it does not enter the statement. We train desks to refuse any laytime output that cannot be walked back to a visible entry.
Paper beside the working screen
3 March 2026
Three verification habits worth keeping
The first habit is clause citation. A demurrage narrative, a NOR validity note or a SIRE observation response is not finished until each material statement points to a clause, a procedure or a log line the reader can open. The second is re-derived arithmetic: laytime totals, bunker differences and CII inputs are worked once by the model and once by hand, on paper or in the desk’s own sheet, before anyone sends the figure. The third is a dated source list — which recap, which SOF page, which noon, which BDN — so that a later dispute does not rest on a chat history. None of these habits requires a particular vendor. They are slower than an unchecked paste, and they are the difference between a draft and something a claims handler or a superintendent can stand behind.
22 April 2026
Preparing an emissions evidence trail before the verifier asks
Verifier queries arrive when noon reports, bunker delivery notes and port logs do not tell the same story. The useful work is therefore done before the query: reconciling ROB figures with BDNs, checking that steaming hours and distance in the noons can support the consumption claimed, and recording where a port log disagrees rather than smoothing the disagreement in prose. Models can line the documents up and draft a narrative of the gaps; they cannot be allowed to invent a missing delivery or to average away a jump. Cohorts practise assembling an MRV/DCS pack in which every number has a home in a named file, and in which EU ETS or FuelEU Maritime text is written as an explanation of those files. The trail is the product. The paragraph is only a covering note.
Reading
Reading we point cohorts to
We do not publish a private bibliography of blogs. Cohorts are pointed at instruments they already have a duty to read, because an AI draft that cannot be checked against those instruments is a liability. The list below is by type, not by invented hyperlink; desks use the editions their flag, class and charterparties actually require.
IMO guidance and circulars that govern the desk’s work — including MARPOL Annex VI material, PSC procedures and the current CII and DCS texts as published by the Organisation.
Class society circulars and survey status extracts for the vessels in the fleet, used as source documents in technical and PSC exercises rather than as decoration.
BIMCO clause libraries and the governing charterparty form, against which recaps, NOR wording and demurrage narratives are read in the room.
The company’s own safety management system, including the parts that constrain how documents may be processed, who may sign, and what must never leave the office.
If a tool’s behaviour changes, the reading does not. Clauses, class rules and the SMS remain the check. Prompt packs are re-cut when the model shifts; the underlying paper does not.
