How We Tell Whether Your Geodata Will Still Answer Questions in Five Years
MovingLayersThe awkward questions rarely arrive at the end of a project. They arrive three years later, and from a direction nobody anticipated. An expert witness wants to know which pipeline route was documented at the time of the permit. A colleague is looking for the elevation value that stood in the records before the correction, because the correction itself has now been called into doubt. And at some point somebody asks about an attribute that evidently nobody measured, and nobody can say where it came from.
Whether such questions can be answered is not decided on the day they are asked. It is decided years earlier, when the target data model is cut, in the specification of a tender, and in the seemingly harmless choice of whether a correction replaces the old value or takes its place beside it. In our projects we regularly see holdings that are well maintained in professional terms and still cannot give any account of how they came about, because that choice was never made deliberately.
That one choice has three prerequisites, and they sit in three different professional worlds: in standardisation, in data storage, and since this summer in law. Taken separately they look like three projects, and we have described them separately over recent months. They are one.
How You Can Tell Whether Your Data Can Answer
The test question runs: who will still be able to explain, in five years, how a given data state came about?
That sounds like a documentation task, and it is usually treated as one in projects, which is why it so often fails. If you want to answer the question seriously, you need three things at once. The terms in your holdings must mean the same as in the holdings you want to compare them with. The earlier state must still exist. And for every value it must be possible to say whether it was measured or derived. If one of the three is missing, the other two no longer carry.
An unbroken history without defined semantics produces a sequence that nobody can reconcile with a second source. A clean statement of origin on a value whose predecessor was overwritten documents only the last step. And a structure modelled to standard but without a time axis answers every question in the present tense, including the one that refers to 2023.
Why We Take Part in Standardisation
Dr Jürgen Hahn has been a member of Austrian Standards Committee 084 “Geoinformation and Surveying” since June 2026. For a mid-sized company that is, at first sight, effort without immediate return, and we are asked from time to time why we put ourselves through it.
The answer has to do with the shift in content over the past three decades. What the committee negotiates today is data modelling, metadata, data quality, exchange formats and the interfaces of digital services; symbol and drawing keys for plan production have moved into the background. That means the very decisions on which integration projects later founder are taken there. Anyone wanting to bring a protected-area layer, a BIM structural model and a network operator’s as-built data together in one analysis negotiates semantics, structure and quality either beforehand in a committee or afterwards in the project, under time pressure, once only and without a record. The second variant is the more expensive one, and it leaves nothing behind for the next cycle to build on.
→ Why we take part in the work of Austrian Standards
Why We Overwrite Nothing
The second article, in August, concerned the time axis. A GIS usually shows the current state and nothing else, and the common workarounds only appear to help: backups, copied layers and archived interim states do not answer the question about the history of an individual object. An auditor does not want to see the dataset as it stood the day before yesterday. They want the history of this foundation, this pipeline, this parcel, with geometry, attributes and a record of who changed what and when.
Versioning records that an object has several successive states. Historisation makes those states queryable along the time axis. In practice the change is less a technical matter than an organisational one, because it touches a habit: a correction feels like putting something right, and its natural reflex is to overwrite. In MoversSurvey, therefore, even the correction made in the field takes its place in the object history, and as a QGIS solution the history sits in the database, queryable in the familiar environment and without specialist knowledge of how the data is stored. Both are in progress for Q4 2026 (as of September 2026).
→ Geospatial data needs a memory
What We Anchor Obligations To
The third article had a legal occasion. The transparency obligations under Article 50 of the AI Act have applied since 2 August 2026; for systems placed on the market before that date, the transition period under Article 50(2) ends on 2 December 2026; and Article 4 on AI literacy applies on an ongoing basis regardless of risk class in any case.
Working through the regulation along our own projects showed that the system list does not work as an assessment grid. The same segmentation remains a convenience function as long as it is configured and used internally, and it can trigger provider obligations as soon as the same procedure is made available externally under your own name. A network model that was documentation yesterday and feeds a switching logic tomorrow changes category without a single line of code changing. The classification hangs on the procedure and on its context of use, and both can only be described if it is recognisable in the holdings which value was measured and which was generated. Which brings us back to historisation, and to the question of whether the terms were defined when the value came into being.
→ AI in Geospatial Data Processing: What Actually Needs Doing Since 2 August 2026
How We Run the Geodata Cycle
This mutual dependency is what shapes the way we describe the geodata lifecycle. For us it is a closed loop and not a chain of project phases, because a chain ends with a handover and after that nobody is responsible. A cycle is one complete pass for a defined scope, from capture through the decision to the implemented measure, and its value comes from the comparison with the cycle before it: what has changed, where, and since when.
The decisive line in this diagram is the arrow running from phase 4 back to phase 1. A decision does not yet change reality; the measure does. And because it does, it writes the capture brief for the next pass: which objects have to be recorded anew after the works, which reported deviation triggers a re-survey, which area rests for the time being. That is why the centre reads “one cycle” and not a target state.
The four circles describe what happens to a record, not which department is responsible; the field capture, the sensor report and the legacy holdings from a replaced system all pass through the same phase. The labels outside the circles deliberately name categories rather than product names, because the process is yours and not ours. MovingLayers equips it, sets it up together with you and keeps it running. Where you have your own tools, we integrate them; where tools are missing, we supply them. The dark bar along the bottom sits beneath all four phases for the same reason, rather than inside the loop: consulting and operation is not a fifth phase, it is the level on which the other four run at all.
1 · Capture
Capture has two forms, and most holdings need both. The site visit is bounded in time and planned, your people travel to the objects and record what is there; MoversSurvey gives them a tool that works offline, can be used without prior GIS knowledge and delivers structured data. The data stream runs without a crew and without a planned end, with sensors reporting location and condition even where mobile coverage cannot reach, using VirtualShepherd based on LoRaWAN. The two belong together, because as a rule it is the data stream that shows a site visit is due.
2 · Validate and prepare
Validation happens before admission. What fails validation is not lost, it is kept, flagged and reported, and it simply does not become part of what your holdings assert about reality. Quality checking, standardisation, historisation and versioning sit here, and it is where the two time stamps on every record are created: when the world was in that state, and when you came to know it. A site visit on 12 June that syncs on 15 June is both, and which of the two a report needs depends on the question being asked. Anyone who does not separate them from the outset notices at the first cut-off date report, and by then it is reconstruction work.
3 · Understand and decide
Captured data is not yet a decision. Kermit GeoAI adds an analysis layer to your existing GIS, with spatial questions in plain language, operated locally, and with a traceable chain from the data source through the rule applied to the result. At the end there is a decision you can evidence.
4 · Implement and improve
This is the phase most often missing from cycle models, and it is the one that decides whether the model really is a loop. Built, refurbished, relocated, re-surveyed, switched off: the measure changes reality, and the dataset falls behind it again. The difference from mere drift is that this time you know where.
This phase gives two things back to the cycle. It writes the capture brief for the next pass. And it reveals which validation rules were drawn too narrowly or too widely, because whether an attribute the validation demands can actually be determined in the field becomes clear during implementation and not at the desk. Which is exactly why the phase is not called implement alone.
What a Cycle Has to Hold
Four rules decide whether the cycle stays able to answer questions over time. We build them in from the start, because adding them afterwards is expensive, and because the cost of that rework almost always lands in a project that was not costed for it.
Nothing is overwritten, every state stays reconstructible, and how far back is agreed by contract, because erasure obligations and retention obligations pull in opposite directions and cannot be resolved in general terms. Data is validated before it counts, and no path into the data bypasses that validation; an import path that skips it is a defect and not a shortcut. When the data structure changes, we keep the mapping as its own versioned artefact, so that older records remain explainable after a model change. And there is one state and no second path: field device, web map and desktop GIS read from the same source, offline work is the planned exception, and the device records which state it set out from.
What We Anchor Our Offering To
Our products sit where data comes into being and is read within the cycle, which means MoversSurvey in the site visit, VirtualShepherd in the data stream and Kermit GeoAI in the analysis. The services apply where a tool on its own holds nothing, and they follow the lifetime of a system:
- Tenders and procurement applies before the first cycle, to the question of what is procured in the first place. Historisation, validation rules and evidence obligations become verifiable requirements in the procedure here, while they are still negotiable.
- Requirements and migration accompanies the transition when a system is replaced and the history has to come along. This is where the target data model is created, and with it the data classification that later carries both architecture and AI decisions.
- Operations and maintenance secures phases 2 and 4, that is the ongoing validation and the updating that follows a measure, with logs and a documented rule set. Which is exactly the evidence asked for months later.
What We Advise for the Coming Months
Your next deadline is 2 December 2026. In our view the faster route to it runs through taking stock along the process rather than through a system inventory, because the guiding question is not where AI is in use but at which point in your cycle values arise that were generated rather than measured. In our experience the answer is hidden in convenience functions that nobody has recorded as an AI system. Part of this is recording the classification per procedure, with intended purpose, safety function, deployment environment and the role as provider or deployer; that is four fields on one sheet, administrative work in normal operation and the decisive piece of paper in a dispute.
Independently of any deadline, there is one choice we would bring forward while your holdings are still growing: whether a correction creates a new version. That single decision determines which answers will be possible in the years ahead, and it belongs in the data model rather than in a working instruction.
We do not know this sequence from a collection of methods. We run it ourselves, with network operators, utilities and in public administration, and we sit at the table where the foundations for it are defined.
Let us talk about where your geodata cycle is stalling today.
Book an initial consultation →
