Balancing market surveillance from published offers

Day-ahead surveillance reconstructs a merit order, because the order book is not published, and every finding carries the reconstruction's own error. Here the regulatory feed publishes the individual balancing energy bids, so this module reads the ladder rather than estimating it. There is no model error to disclaim.

What the publication does not carry is identity. It can show that one price step carried the volume that set the price; it can never show whose step it was. Every finding here attributes to a step in a ladder, never to a participant.

Each quarter-hour is screened in both directions against its own trailing band, so a period is measured against how that period has behaved rather than against a fixed constant. Both directions are screened on every quarter-hour, whichever one was actually called, because a control that chose one would hide the other.

The ladder

Every offer published for the quarter-hour, cheapest first, per balancing product. Products are called in sequence and each keeps its own price axis, so nothing crosses on the chart: only one direction is called in a period, and the reference lines are settled marginal prices rather than a point where supply met demand.

The trailing band

The distribution of the same period across preceding delivery days. A screen fires when a reading clears that band, so a volatile period needs a larger move than a quiet one to be flagged.

What is not on the ladder

Not everything that balances the zone comes off it. Activation by component splits what answered, and a product that supplied nothing has no settled price to be measured against, so its price-relative readings are absent rather than zero.

Merit order

The individual offers behind each quarter-hour, screened. This is the only one of the four that applies a threshold.

Settlement replay

What the day cost and what the price was built from: the system's own position minute by minute, and the marginal price and penalty component behind each settled quarter-hour. The only arithmetic performed is the subtraction in the residual column, and it is shown so it can be checked.

Day-ahead carry-over

What the auction left open and what balancing charged for it. The spread is what one MWh left open past the day-ahead gate was exposed to; it is not anyone's settlement figure, because what a participant actually paid depends on the position they held, which neither publication carries.

Capacity auction

What reserve was procured for the tender date and at what price, across the aFRR and mFRR tenders. The bid identifier does not thread across tenders - verified: 0 of 25,701 bid ids recurred across 14 tender dates - so nothing here can be attributed to a participant. Nothing on this view is screened.

The bids are as published, so the ladder is not in doubt. The inference from it is. A unit already at minimum stable load, a network constraint, an outage not yet disclosed, a position already hedged: each produces the shapes these screens look for, and none is visible in this data. Verify against outage disclosures before drawing any inference.

Settlement figures are restated as they validate, and only the latest version is served. Nothing here holds a copy, so a figure read today can differ from the same figure read tomorrow. The console shows how much of a day has validated.

Belgium only, and individual bids only from the market platform change of 2024-05-22. Delivery days before that were published under a schema this does not read, which is not a gap in collection. The publisher holds a rolling two years; a delivery day past that is no longer served, and nothing here stored a copy of it.