Pay-as-clear: numerical residual can set the market-clearing price

[tomissg]

PayAsClearRole.clear() calculates the uniform clearing price as the maximum bid price across all accepted_supply_orders:

clear_price = float(max(map(itemgetter("price"), accepted_supply_orders)))

The calculation does not exclude supply orders with an economically zero accepted volume.

In an EOM run with storage, floating-point SOC/ramping arithmetic produced an accepted BESS volume of 1.1247e-14 MW. Although this is physically zero, the BESS offer remained in accepted_supply_orders and set the uniform price.

Observed examples:

 Timestamp           Published price    BESS accepted volume    Highest materially accepted supply bid
━━━━━━━━━━━━━━━━━━  ━━━━━━━━━━━━━━━━━  ━━━━━━━━━━━━━━━━━━━━━━  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 2027-01-03 19:00      1000.00 €/MWh           1.1247e-14 MW                              290.36 €/MWh
──────────────────  ─────────────────  ──────────────────────  ────────────────────────────────────────
 2027-10-26 19:00       778.18 €/MWh           1.1247e-14 MW                              281.74 €/MWh

This is economically material: in a pay-as-clear market, a numerical residual with no physical dispatch should not become the marginal unit and determine settlement for all accepted
supply.

Suggested fix: canonicalize near-zero accepted volumes using a configurable volume tolerance, and exclude orders with abs(accepted_volume) <= tolerance from price formation (and preferably from accepted-order/dispatch output). The tolerance should apply generically to all units, not only storage.

A regression test should cover a high-price supply bid with an accepted volume below tolerance and assert that it cannot set the clearing price.

[maurerle]

For calculations without floating point errors, you can set a proper price_tick and/or volume_tick in your market config.
This ensures, that only integer values/fractions with these minimum "tolerances" exist and can be selected by the market.

I guess that this is helpful here.
In general, floating point calculation does have issues, which means that money can be missing at the end of the simulation due to such effects.

I do not think that the LLM suggested fix is wanted in the long run, as it introduces another band aid fix with side effects and adds complexity.

It would also be interesting if the BESS did bid such a small volume instead of being turned off or if the market somehow had this small amount of demand left.