--- id: VOLT-HOME-WP-008 title: "What should an optimizer do when one price interval is missing?" slug: what-should-an-optimizer-do-when-one-price-interval-is-missing description: "A synthetic interval-removal test of a next-cheapest-valid fallback for quarter-hour household scheduling." published: 2026-08-30 cluster: "Market intervals, clocks, and data integrity" status: measured evidence_url: /research-data/home-papers/what-should-an-optimizer-do-when-one-price-interval-is-missing.json figure_url: /research-media/home-papers/what-should-an-optimizer-do-when-one-price-interval-is-missing.webp figure_alt: "Chart for What should an optimizer do when one price interval is missing?: robust fallback cost when one selected quarter-hour is unavailable, shown as P10, Median, P90." source_ids: - google-scaled-content - entsoe-sdac - ec-sdac-15m - acer-sdac-products - volt-architecture peer_reviewed: false --- # What should an optimizer do when one price interval is missing? ## Abstract An optimizer should not silently invent a price when a selected interval becomes unavailable. This paper tests a bounded fallback: after optimization, one chosen quarter-hour is removed and replaced by the next-cheapest valid interval. Across 3,336 synthetic events, the mean added wholesale cost is €0.0014782808752997602 per event. The regenerated evidence reports no interval for this estimand. This result describes the observed curves under a specific stress. The missing interval is imposed after selection; it is not a measured upstream outage rate. Wholesale energy only is counted, with taxes, supplier margin, and network charges excluded. The experiment does not prove that every missing-data case is safe to repair, because timestamp ambiguity, a larger gap, infeasible power constraints, or missing adjacent intervals may require a halt rather than substitution. It is not a household bill study. Analysis and drafting were model-assisted; the working paper is not peer reviewed and is not trading advice. ## Plain-language answer If one interval selected by a price optimizer cannot be used, a reasonable first fallback is to preserve the required energy and move that portion to the next-cheapest interval that is still valid and feasible. In this synthetic test, that rule added an average wholesale cost of about fifteen-hundredths of a euro cent per event. The measured penalty is small, but it belongs only to the registered event design and observed curves. The word “valid” carries most of the safety requirement. The substitute interval must exist physically, fall inside the device's availability window, respect its power and energy limits, and not duplicate energy already scheduled. A controller should not select a cheap timestamp that belongs to another local day, has an unresolved revision, or appears only because a missing value was filled. The test removes one chosen interval after optimization. That is an adversarial scenario, not evidence that production data loses exactly one interval or that failures happen after a schedule is chosen. Real systems should first determine whether the interval is genuinely absent, delayed, revised, or merely part of a legitimate daylight-saving shape. Only then should the fallback policy run. ## Research question The research question is what wholesale cost is added when one selected quarter-hour becomes unavailable and the optimizer uses the next-cheapest valid interval. The primary metric is the robust fallback cost per synthetic event. This does not compare every possible imputation or rescheduling policy. It does not estimate missing-data frequency and does not test multiple simultaneous gaps. The narrow goal is to quantify a simple feasibility-preserving substitution in otherwise eligible observed curves. ## Data and provenance The evidence reports daily prices from 2021-01-01 through 2026-08-29, detailed intervals from 2025-10-01 through 2026-08-29, and long history from 2015-01-01 through 2026-08-29. Detailed interval analysis uses ten representative European bidding zones from 2025-10-01. The primary sample contains 3,336 synthetic interval-removal events. The source contract lists `zones`, `day_ahead_prices`, `grid_revisions`, `auction_publications`, and `ingestion_runs`. Zone and interval tables establish complete physical curves. Revision and publication lineage distinguish unavailable data from a later observation state. Ingestion state supports operational completeness rather than treating every empty cell as a market fact. The publication cutoff is 2026-08-30T00:00:00Z. The analysis ran in a read-only transaction with a 180-second timeout. The frozen snapshot hash is `7e97489fc8528c8cc8c38830e05b48d949ce1f67b98575dff26f5d7c321e4c67`; analysis-code hash, `57c57de79cdab2b5b6d6c54c485cb5162598c5ba0b0bfe995da40d75e6c52ba9`; protocol hash, `adb36bf6b447af9f96339249b8becaefc20422499cca1977242866347a97bd4b`; paper-registry hash, `7bcb91d7476d0a69fe9fa75a5c7782f8117e0153f82f9112b7e1d307d3943717`; source-registry hash, `07949550ac443ff673fda5c0209b99f137544f3ffecf6775f109bb9d09663bd6`. The evidence manifest records evidence SHA-256 `8e4c3e4e91ab3535b9efb9359d5ba2ae8fe8efb2dcb12b24c82e49afe5b6756e` and figure SHA-256 `77dc6b279bd064dc335f378ed9321ded0e9882dd0501c670a804a3e56a99ff9a`. The public JSON reports only paper-level aggregates and the source contract. It does not expose raw licensed market payloads, and its licensing note governs downstream use. ## Method The method is part of the registered paired interval reconstruction, clock-safe counterfactual, and day-block bootstrap family. On each eligible complete native-resolution curve, an optimizer selects a schedule. The test then removes one selected quarter-hour after optimization. The fallback chooses the next-cheapest interval that remains valid. The imposed removal makes this a controlled stress. It targets an interval that mattered to the original decision rather than deleting an irrelevant expensive interval. The comparison is paired on the same curve: original optimized wholesale cost versus fallback wholesale cost after replacement. Wholesale energy is the only cost component. Taxes, supplier margin, and network charges are excluded. The paper does not assume that a missing price can be interpolated, carried forward, or assigned zero. Those approaches would fabricate an objective value and could select an interval whose economics are unknown. The regenerated artifact stores `bootstrap_95_interval` as null and labels its interval method `not reported for this estimand`. Within-family Holm control applies to inferential claims, and the evidence labels this a descriptive result without an unadjusted significance claim. ## Results Across 3,336 events, the mean robust fallback cost is €0.0014782808752997602 per event. No uncertainty interval is reported. The evidence interpretation is that selecting the next-cheapest valid interval produces a bounded fallback in the observed curves. The figure labels `P10`, `Median`, and `P90` carry exact values €0.0, €0.0002959999999999937, and €0.003552 per event. These distributional summaries do not form a confidence interval and must not be substituted for one. “Bounded” here describes the measured registered sample and fallback rule. It is not a universal maximum. The public artifact does not report the largest event penalty, the distribution by zone, or the effect of multiple missing intervals. This paper therefore does not claim a hard service guarantee. The result supports preserving feasibility through explicit substitution rather than guessing a price. It does not show that the substitute will always be cheap in a future curve or that the household experiences the wholesale difference on its bill. ## Robustness and placebo checks The original schedule is the matched control, and removal of a chosen interval is the stress. Because both costs use the same observed curve and event, the comparison isolates the fallback change from day-level price differences. Requiring the substitute to be valid is a structural check. A timestamp outside the availability window, a duplicate interval, an interval from another local day, or a physically missing interval is not eligible merely because a stored number appears cheap. Clock-safe reconstruction must precede ranking. The next-cheapest rule is transparent and deterministic. It avoids optimization instability caused by arbitrary imputation. The current evidence reports no interval for the estimand, and within-family Holm control governs any inferential claims. The evidence does not report alternative placebos such as zero filling, forward filling, random replacement, or halting. This paper does not invent their outcomes. For ambiguous or larger failures, halt-and-fallback may still be the safer operational policy. ## Limitations The missing interval is synthetic and is removed after optimization. Production failures can arise before publication, during ingestion, after revision, or at device execution. They may affect several intervals or the entire curve. The detailed sample covers ten representative European bidding zones beginning on 2025-10-01. The average is not a universal bound for all zones or regimes. The artifact reports no event maximum or tail-risk statistic. Wholesale energy excludes taxes, supplier margin, network charges, device losses, comfort, and hardware costs. This is not a household bill study. A small wholesale fallback penalty does not mean missing data is operationally harmless. The next-cheapest interval may violate constraints not represented in the paper. A real controller must consider device availability, energy deadlines, power, state, comfort, and command reliability. If no valid substitute exists, the correct outcome is not forced optimization. ## Practical implication Controllers should separate data validation from optimization. First establish a complete, clock-safe curve and version. If a selected interval later becomes unavailable, invalidate only the affected plan component, preserve the energy requirement, and rerun against the remaining valid feasible set. The fallback should be deterministic and auditable. Its record should include the missing interval identity, reason, curve version, replacement identity, and cost delta. Idempotency prevents a repeated missing-data event from issuing duplicate commands. When validity is uncertain, safety dominates price. A home system can retain a nonoptimized baseline schedule, device-native charging rule, or comfort-preserving heating plan. The measured next-cheapest penalty supports one bounded substitution case; it does not justify fabricating prices to keep optimization alive. The controller should also distinguish a missing market interval from a temporarily unavailable command slot. In the former case, the objective value is unknown; in the latter, the price may be known while the device cannot act. Both can lead to the same next-cheapest substitution, but they require different incident records and recovery checks. Keeping those causes separate prevents a device outage from being misreported as faulty market data. If a replacement changes the planned completion time, the system should revalidate the full energy deadline rather than checking only the substituted slot. A locally cheap fallback is not robust if it leaves the event incomplete. Feasibility, explicit uncertainty, and a safe baseline remain prior to optimization. ## Reproducibility The public evidence JSON is `/research-data/home-papers/what-should-an-optimizer-do-when-one-price-interval-is-missing.json`. It includes exact assumptions, windows, source tables, metric, sample size, null interval, figure values, limitation, disclosure, and provenance hashes. A reproduction should verify the snapshot, code, protocol, and registry identities; preserve the publication cutoff and read-only boundary; and reconstruct complete native quarter-hour curves with unambiguous timestamps. Run the registered optimizer, remove one selected interval, and choose the next-cheapest remaining valid interval without imputation. Compute the paired wholesale cost difference and aggregate over the original eligibility set. Reproduce the exact figure vector `[0.0, 0.0002959999999999937, 0.003552]` and preserve the null interval. Any test involving multiple gaps, a different fallback, retail tariffs, or device constraints should be issued as a separate analysis rather than merged into this result. ## Disclosure Analysis and drafting were model-assisted; sources, code, assumptions, and evidence hashes are disclosed. This public working paper is not peer reviewed, is not a household bill study, and is not trading advice. Volt has zero live traders and zero live trading capital; C0R is the only paper strategy. Production weather forecasting is a non-trading service. This fallback study authorizes no order, capital, or live trading. ## References - [Spam Policies for Google Web Search — Scaled content abuse](https://developers.google.com/search/docs/essentials/spam-policies) — Google Search Central. - [Single Day-ahead Coupling (SDAC)](https://www.entsoe.eu/network_codes/cacm/implementation/sdac/) — ENTSO-E. - [EU electricity trading in the day-ahead markets becomes more dynamic](https://energy.ec.europa.eu/news/eu-electricity-trading-day-ahead-markets-becomes-more-dynamic-2025-10-01_en) — European Commission. - [ACER Decision 13-2024 on SDAC Products](https://www.nemo-committee.eu/assets/files/ACER%20Decision%2013-2024%20on%20SDAC%20Products-702e63479704f5a1b75b75aa71c45ec8.pdf) — Agency for the Cooperation of Energy Regulators. - [Voltcast Architecture](https://github.com/ossedk/voltcast/blob/main/docs/voltcast/ARCHITECTURE.md) — Voltcast.