Evidencing availability guarantee compliance, and where the argument actually goes

An availability guarantee commits a party, usually the operations and maintenance contractor or the equipment supplier, to keeping the asset capable of operating for some proportion of a defined period, with a liquidated damages mechanism if the proportion is missed and often a bonus if it is exceeded.

Everything interesting sits in the definitions rather than in the percentage. What counts as the period. What counts as capable. Which hours are excluded and on whose evidence. How partial capability is treated. What notice a planned outage requires to stay planned. And what happens when more than one cause contributed.

The percentage is negotiated in an afternoon. The definitions decide what it means, and they are frequently drafted by reference to a template rather than to the asset. The executed contract governs in every case. The measurement problem sits underneath whatever it says.

Curtailment and grid unavailability. A plant that is fully capable but not producing because the offtaker or the network told it to stop is usually available under the definition, and demonstrating that requires the dispatch instruction, the grid signal or the market record, timestamped and aligned with plant state. Sites routinely retain the plant side and not the grid side.

Partial derating. A unit running at seventy per cent is neither available nor unavailable in most templates, and the treatment has to be stated. Proportional treatment, banded treatment and all-or-nothing treatment produce materially different annual figures from identical operation, and derating is far more common than full outage on a renewable-coupled plant.

Planned against unplanned. Most definitions exclude planned maintenance provided notice was given within a stated window and the work stayed inside the notified scope. An outage that overruns, or that expands once the equipment is open, migrates across the boundary, and the evidence for which side it lands on is the notification record and the work order.

Force majeure and external cause. Weather, feedwater supply, grid faults and third-party damage sit outside the guarantee if they can be evidenced. Ambient and resource data at plant resolution is what turns a claim of extreme conditions into a demonstrated one.

Multi-cause events. A trip with a contributing instrument fault, a contributing operating decision and a contributing supply issue is the case every template handles badly. Whoever holds a continuous, aligned record of plant state, alarms and external conditions sets the version of the sequence that gets argued from.

Plant state at a resolution that survives the question. If the definition operates on hours, a record averaged to hours cannot resolve an event that began at twenty past. Availability arithmetic is usually done on a coarse period, but the disputes are always about a specific transition.

The state of the exclusions, not only the state of the plant. Grid signals, dispatch instructions, ambient conditions, resource availability and feedwater or utility supply all need to be recorded alongside plant state and on the same clock. This is the half most sites miss, because it is not their equipment.

Alarm and event history with sequence. What annunciated, in what order, and what the operator did. First-out logic and time synchronisation across the control system, the protection system and the historian decide whether a sequence is reconstructable.

Work orders and notifications joined to the same timeline. The maintenance management system usually lives apart from the process historian, and the joining of the two after the fact is where a defensible position becomes an afternoon of spreadsheet work.

A sequence assembled from systems with unsynchronised clocks is not a sequence. Where the control system, the protection relays, the analyser and the historian each carry their own time, an event ordering can be wrong by seconds, and seconds are frequently what the causation argument is about.

Synchronising them to a common source, and recording that the synchronisation was in place, is inexpensive at commissioning and effectively impossible to assert retrospectively. Where sub-second ordering matters, as it does for electrical protection events, this is the difference between a record and an anecdote.

It is also one of the first things a technical advisor or an expert will check, because a demonstrated time base is what makes everything downstream of it credible.

The record was averaged. A historian applying compression on the way in discards points it judges uninformative, and a signal stored as a five-minute average cannot be un-averaged. This is the most common and least recoverable failure.

The retention window expired. Availability is usually tested annually and disputed later, and a system configured with ninety days of retention at commissioning has already lost the evidence by the time the argument starts.

The exclusions were never instrumented. A site with excellent plant data and no grid or ambient record can prove what it did and not why, which is exactly the wrong half for an exclusions argument.

The provenance is contested. Where the party being measured produced the report, the other side is entitled to ask how the figure was derived. Nobody is accusing anyone of dishonesty; the incentive simply is not neutral, and a traceable figure is easier to accept than a stated one.

The definitions were never mapped to tags. The single most useful piece of preparation, and the one most often skipped, is writing down which measured signals implement each term in the definition, before anyone needs them.

Map every defined term to a signal. Take the availability definition clause by clause and write, next to each, which tag or record demonstrates it. Terms that map to nothing are the ones that will be argued about, and finding them before signature is considerably cheaper than finding them afterwards.

Set retention to the contract, not to the default. The useful horizon is the guarantee period plus any dispute window, which for a financed asset usually means the tenor of the debt rather than the historian's configured retention.

Record the exclusions. Grid signals, dispatch instructions, ambient and resource data at plant resolution, and utility supply status, on the same clock as plant state.

Agree the calculation, not just the target. A worked example against a month of real data, produced jointly before the first reporting period, removes most of the room for later disagreement. Doing it once at the start is an afternoon; doing it during a dispute is a claim.

Where the same figures feed the finance package, they also feed reporting to lenders after commercial operation, and the retention decision covers both.

It helps with attribution. Where an event has several contributing causes, a model that can say which readings are mutually consistent and which are not narrows the argument from an opinion to a shortlist. That is the same reasoning applied to separating stack degradation from the operating regime, and it works here for the same reason.

It helps with completeness. A system holding plant state, exclusions and alarm history on one clock removes the assembly work that otherwise happens under time pressure.

It does not decide the question. Availability is a contractual construct and the contract decides it. Data supports a position; it does not settle a definition. Yunify is built to hold the record and to attribute movement to a mechanism, so that a figure in a report can be followed back to something physical rather than asserted, and the parties argue about the clause rather than about the number.