Gas data in depth: UK Link, Gemini and UIG

Gas centralised its data services roughly a decade before electricity seriously tried. Where the electricity market is only now routing settlement messages through a shared platform under MHHS, gas has run its retail data through a single central agent for two decades and through one rebuilt central system since 2017. If you want to understand what a centralised energy data spine looks like when it has been operating at scale for years, gas is the place to look.

The operator is Xoserve, which holds the Central Data Service Provider (CDSP) role under the Uniform Network Code (UNC). The CDSP is a contracted role, not a company type: it exists because the UNC parties needed one agent to hold the shared record rather than each transporter and shipper keeping its own. That single point of custody is why gas queries have a single obvious answer to the question “who has the data?”, which is more than electricity can say for most of its estate.

Three systems carry the load, and they are routinely confused with one another. Keeping them straight is the first competence this module asks for.

UK Link: the retail backbone

UK Link is the retail settlement and registration system. It holds the supply point record for every Meter Point Reference Number (MPRN) in GB, the shipper and supplier relationships attached to it, the meter reads submitted against it, the demand estimates derived from it, and the daily allocation and reconciliation that follow. It is the system rebuilt under Project Nexus and brought into service in 2017, and it is the system that produces the numbers shippers are invoiced against. When a gas registration, read or allocation issue is raised, UK Link is almost always where the evidence sits.

Gemini: the transmission system

Gemini is a different animal serving a different market. It handles capacity booking, nominations and balancing on the National Transmission System (NTS): the daily business of shippers telling the transmission operator how much gas they intend to put in and take out, buying the entry and exit capacity to do it, and being balanced against what actually flowed. Gemini is operated by Xoserve on behalf of National Gas. The common error is to treat Gemini as the gas equivalent of a settlement system for households. It is not. Gemini is wholesale and transmission; UK Link is retail and distribution.

The Gas Enquiry Service (GES): the front door

GES is the enquiry channel through which market participants look up supply point information: who is registered, what the meter details are, what the address record says. It replaced the older Data Enquiry Service on 18 July 2022, which matters because a great deal of second-hand industry documentation still names the Data Enquiry Service as though it were current. If you see DES quoted in a process document, treat that document as out of date until proven otherwise.

The division of labour is clean once you see it: UK Link is the record, Gemini is the transmission market, GES is the way you ask the record a question. The figure below places all three against the parties they serve.

The takeaway for a data practitioner is that gas questions resolve to systems faster than electricity questions do. Ask where a household's gas consumption record lives and the answer is UK Link, held by the CDSP, governed by the UNC. Ask the same about a household's electricity consumption and you have to work through the supplier, the retrieval agent, the DCC and the settlement chain before you can name a custodian. Centralisation buys clarity. What it does not buy, as the rest of this module shows, is accuracy.

Gas has no half-hourly settlement. The settlement unit is the gas day, and the question the system has to answer every day is how much of the gas that entered each distribution network belongs to each shipper. Electricity answers this with measured half-hours; gas answers it with a mixture of measurement and estimation, and the mixture is the reason UIG exists.

Supply points fall into two families. Daily-metered points are read every day, typically large industrial and commercial sites where a daily read is worth the equipment and the effort. Their contribution to the daily total is measured, not guessed. Everything else is non-daily-metered (NDM): the domestic and small commercial population, read monthly, quarterly or on customer submission, and allocated daily by estimation. NDM points dominate by count and matter enormously by aggregate volume, so the quality of the estimation model is a first-order driver of settlement quality.

NDM demand estimation works from an annual quantity for each supply point, shaped across the year by weather and by the behaviour of the end use category the point belongs to. Cold days pull more gas; a domestic heating load has a different daily shape from a bakery. The model produces a daily estimate for every NDM point, the estimates are summed, the daily-metered reads are added, and the total is compared with what the transporters actually measured entering the Local Distribution Zone (LDZ). Later, as actual meter reads arrive, the estimates are reconciled and the differences are settled.

Volume is not energy

A gas meter measures volume in cubic metres. Customers are billed, and settlement is run, in kilowatt hours. The bridge between them is the calorific value (CV), the energy content of the gas, which varies with the composition of what is flowing and is therefore measured and published daily for each LDZ. The conversion is the standard UNC formula:

Energy (kWh) = Volume (m³) × CV (MJ/m³) × 1.02264 ÷ 3.6

The 1.02264 is the volume correction factor that adjusts metered volume to standard temperature and pressure conditions. The 3.6 converts megajoules to kilowatt hours. Neither is negotiable and neither is a rounding convenience: they are code-mandated constants, and getting one of them wrong scales every number downstream of it.

This is where gas differs structurally from electricity as a data problem. Electricity settlement takes a measured quantity in the settlement unit and uses it. Gas settlement takes a measured quantity in the wrong unit, applies a daily published factor to convert it, and for most supply points estimates the daily split as well. Two extra transformations sit between the meter and the money, and each one is a place where error can enter. That is the setup for the next section.

Unidentified Gas is the daily difference, in each LDZ, between the gas the transporters measured entering the network and the gas that has been allocated to supply points. It is a residual produced by a mass balance, and that single fact explains almost everything about its behaviour. UIG is not a measurement of theft, or of leakage, or of any one cause. It is whatever the arithmetic cannot place.

Because it is a residual, UIG absorbs every kind of error at once. Theft and unregistered consumption push it up. Meter reading errors, keying errors and misread registers move it in whichever direction the error runs. Shrinkage and network losses contribute. Identifier problems contribute: a supply point registered to the wrong MPRN, or a site whose meter is physically connected but whose record is missing, distorts the allocated side of the balance. And the NDM estimation model contributes systematically, because an estimation model that is even slightly biased against the weather on a given day will misplace a large volume across a very large population of meters.

Two consequences follow that surprise people new to gas. First, UIG can be negative. If more gas has been allocated to supply points than was measured entering the LDZ, the residual is below zero, and negative UIG days are a normal feature of the record rather than a fault report. Second, UIG is volatile: it is a difference between two large numbers, so a small proportional error on either side produces a large swing in the residual. The rebuilt UK Link arrangements from 2017 made this visible in a way the previous arrangements had not, which is why UIG became an industry issue at that point rather than earlier.

Why an expert, and what the expert produces

Somebody has to pay for the unidentified gas, because the transporter measured it and it was consumed. The question is how to split it across shippers when, by definition, nobody knows whose customers used it. Splitting it by market share would be crude and would ignore the fact that different portfolios contribute different amounts of the underlying problem: a portfolio weighted towards small non-domestic sites with a history of theft is not the same risk as a portfolio of domestic smart-metered homes.

The UNC parties therefore appoint an independent expert, the AUGE, through the Joint Office of Gas Transporters. The AUGE produces an annual Unidentified Gas Statement containing weighting factors that allocate UIG across shipper portfolios by supply point category, so a shipper's share reflects the composition of its book rather than only its size. Engage Consulting was reappointed to the AUGE role in August 2025. The methodology is consulted on, the statement is published, and the factors then drive real invoiced amounts.

This is the argument worth carrying out of the module. UIG is what happens when identifier quality problems and meter read quality problems meet a mass balance equation. The mass balance refuses to let the errors disappear; it converts them into a number, and the industry then has to govern that number. Most sectors bury their data quality problems in reconciliation accounts nobody reads. Gas prices its data quality problem, publishes it daily, and appoints an independent expert to divide it up. Whatever else you think of the outcome, that is a more honest treatment of data quality than most of the energy system manages.

The claim that one bad read moves everyone's costs sounds rhetorical until you run the arithmetic. Take a Class 4 supply point: a non-daily-metered site, read periodically rather than every day, which is the class most small commercial and domestic gas meters sit in. A meter reader visits, the index is transposed when it is keyed, and a read of 12,345 cubic metres is submitted as 21,345 cubic metres. The figures below are an illustration built to show the mechanism, not a published case.

The submitted read is 9,000 cubic metres higher than the truth. Convert that error to energy with an illustrative daily LDZ calorific value of 39.5 megajoules per cubic metre, using the UNC formula from section 21.2:

9,000 × 39.5 × 1.02264 ÷ 3.6 = approximately 100,985 kWh

So a single keying slip has injected roughly 101,000 kWh of phantom consumption into the record. Now follow it through the four consequences it triggers.

One: the customer's own bill

The obvious effect is the visible one. The site is billed for gas it never used, and the shipper's expected revenue from that supply point jumps. This is the part that generates a complaint and eventually a correction, and it is also the least interesting part, because it is self-limiting: somebody notices.

Two: the annual quantity, and every future estimate

The over-read does not only affect the period it lands in. The supply point's annual quantity is derived from its read history, and the estimation model uses that annual quantity to produce the site's daily allocation for every day until the next actual read corrects it. An inflated annual quantity means an inflated daily estimate, day after day. One bad read therefore contaminates a stream of subsequent estimates, not a single number.

Three: the LDZ mass balance, and everyone else

This is the step that matters for this module. The gas measured entering the LDZ is fixed by the transporters' measurement; it does not care about a keying error. The allocated side of the balance has just gone up by around 101,000 kWh. UIG for that LDZ, being measured minus allocated, therefore goes down by the same amount, and may go negative. Every shipper operating in that LDZ sees its UIG share move, weighted by the AUGE factors, on a day when nothing about its own customers changed. The error has left the shipper who made it and become everybody's number.

Four: the reversal

When the next actual read arrives and the index is corrected, the phantom volume is removed and reconciliation reverses the effect. But the reversal lands on a different day from the error, so the LDZ balance is now wrong twice: too low on the day of the error, too high on the day of the correction. Neither day is accurate, and both days were invoiced. This is why gas data quality work concentrates so hard on the read itself. Downstream correction restores the total but never restores the daily profile, and the daily profile is what settles.

Scale that up. There is no single dramatic failure in the gas record; there is a continuous background of transposed digits, estimated reads standing in for actual ones, registrations attached to the wrong MPRN, and model bias on unusual weather days. Each is small. The mass balance adds them all together every day, in every LDZ, and hands the sum to the AUGE to divide. That is the whole system in one sentence, and it is why gas is the best data quality case study the GB energy sector has.