Zentari

Product 03 · Quantum Energy

Schedule flexible compute with the counterfactual visible.

An advisory control room places checkpointable workloads against price windows while retaining forecast, workload, decision and replay evidence.

Savings estimates across twelve deterministic public workload trace slices
FIG 05 Public-trace counterfactual replay12 × 1,000 jobs · mean 21.51% · 95% CI on mean 17.64–25.38% Ordered sample slices with a GB price proxy; not live-site, tariff or settlement savings.
Open evidence →
Engineering stateDeterministic advisory scheduler and Control Room
Scientific evidenceTwelve-slice public-trace counterfactual replay
Next gateLive shadow operation before any dispatch authority

The unavailable future should not be mistaken for the decision.

Workflow

Evidence-linked scheduling, not autonomous dispatch.

The current product advises; live operational control remains outside the evidence boundary.

  1. 01Ingest

    Checkpointable jobs, duration, deadlines and price windows.

  2. 02Forecast

    Construct the decision-visible price context.

  3. 03Schedule

    Optimise placement under workload constraints.

  4. 04Execute in replay

    Apply one fixed policy to each retained slice.

  5. 05Compare

    Baseline, day-ahead and hindsight are kept distinct.

  6. 06Explain

    Evidence drawer retains input and decision provenance.

Actual results

Twelve deterministic slices. One qualified replay claim.

The connected line orders sample slices; it is not a time trend. The shaded band is the confidence interval on the twelve-slice mean.

Savings estimates across twelve deterministic public workload trace slices
FIG 05 Public-trace counterfactual replay12 × 1,000 jobs · mean 21.51% · 95% CI on mean 17.64–25.38% Ordered sample slices with a GB price proxy; not live-site, tariff or settlement savings.
Open evidence →
Counterfactual cost bars comparing baseline, day-ahead policy and hindsight upper bound
FIG 06 Day-ahead policy value in replaybaseline £6,082.35 · day-ahead saving £1,478.36 · hindsight £1,997.98 Hindsight is a non-causal upper bound. The replay uses a price proxy and is not realised value.
Open evidence →
Jobs selected
12,000
twelve public-trace slices
Jobs compared
9,748
2,252 were excluded
Mean saving
21.51%
slice mean; 17.64–25.38% CI
Day-ahead value
£1,478.36
24.31% aggregate weighted replay
Hindsight bound
£1,997.98
32.85%; unavailable to policy
Live sites
0
no settlement or realised saving

Methods

A Control Room backed by retained projections.

The interface shows price, forecast, flexible load, savings and execution timing with an evidence drawer.

Workload model

Checkpointability and timing constraints determine which jobs can move.

Excluded jobs remain in the denominator record.

Optimisation

Day-ahead placement uses only the decision-visible window.

Hindsight is reported as an upper bound, never a policy input.

Reporting

JSON, Markdown, HTML, PDF and Control Room views share evidence-linked projections.

Interface screenshots are not performance evidence.

A replay saving is not a utility bill.

Bring the next claim into contact with reality.