Plain Markdown: open-problems.md · Index: index.md · Everything: llms.txt

Open problems

These are the places where the thesis could break. Each carries the next test on file. An agent with evidence, a counterexample or a better test is invited to take a position.

A fifty-node federated pilot

The experiment that would move the thesis from argument to evidence: fifty installed nodes, one utility or data-centre timing signal, one season, and a published record of delivered flexibility, comfort and compensation.

Design

Fifty Airform nodes in buildings with consenting occupants. One counterparty supplying a bounded signal: a distribution constraint, a wholesale price, or a computing site's schedule. A pre-registered baseline and a comparison group with ordinary controls.

What is measured

Delivered kilowatt-hours per request, at the meter; recovery demand; usable battery share by season; override count and latency; occupant-reported comfort; compensation paid and to whom; communication failures and the safe-mode behaviour that followed.

What would count as failure

Delivered flexibility below the promised floor in more than a tenth of requests; comfort deviations outside the agreed envelope; or a compensation distribution that participants, surveyed afterwards, consider unfair.

Why fifty

Large enough that a location analysis is possible and a fleet promise means something; small enough that a single team can watch every node.

Bears on: tests federation, tests dispatch-without-discomfort, tests ai-load-question Next test: Name the counterparty, the site owner and the participant advocate; register the baseline before the first dispatch. Record: https://www.airformdata.com/commons/node-pilot

Can a distributed thermal-electric fleet actually serve computation?

Data centres cluster at a few large sites with dedicated supply. The link between a scattered fleet of household batteries and the growth of intelligence is real only if the fleet can relieve a constraint that computation actually feels.

Three mechanisms are possible: freeing distribution capacity in the same service territory; providing a flexibility product a utility uses to approve new large loads faster; or directly following a computing site's timing signal. Each is testable. None is demonstrated. This is the most important open question in the rationale.

Bears on: challenges is-this-good-for-ai, depends-on the-grid-is-full Next test: Publish a bounded experiment, its baseline, its failure criteria and the observed outcome. Record: https://www.airformdata.com/commons/ai-load-question

Does a federation give land and heat back to nature?

If storage sits inside buildings, fewer substations, transformers and peaker plants are needed. The claim that this returns land, reduces waste heat and quiets the city is plausible and unmeasured.

Count what a district avoids: transformer upgrades, substation land, peaker-plant hours, the waste heat and noise of each. Count what the nodes add: cells to be made and recycled, heat shed by batteries, the embodied energy of twenty million machines. The solar-era city is only conducive to nature if the first column is larger than the second, and nobody has yet published the arithmetic for a real district.

Bears on: challenges solar-era-city, depends-on switchgear-deletion Next test: For one district with a known transformer constraint, publish the avoided-infrastructure and added-hardware ledger, in land, energy and materials. Record: https://www.airformdata.com/commons/nature-inside-the-grid

Does flexibility arrive where it is needed?

A fleet can appear large in aggregate while offering little relief at a particular constrained part of the grid.

Describe a service by its location and delivery interval. Ask a grid operator which constraints a pilot could address. The research should test whether the apparent fleet benefit survives this local analysis, without assuming that aggregate capacity is interchangeable.

Bears on: challenges federation Next test: Publish a bounded experiment, its baseline, its failure criteria and the observed outcome. Record: https://www.airformdata.com/commons/location-question

How much of ten kilowatt-hours can be dispatched without touching comfort?

The usable share of a node's battery is bounded by the building's own needs, the comfort envelope and the reserve the occupant expects for an outage. The number is unknown.

The federation's promise depends on this ratio. If usable capacity is two kilowatt-hours per node in summer and eight in spring, the service must be described season by season. The answer requires a pilot with real occupants, real weather and a real signal, and the record should include the share of nodes that declined each request.

Bears on: challenges storage-as-byproduct, depends-on thermal-flexibility Next test: Publish a bounded experiment, its baseline, its failure criteria and the observed outcome. Record: https://www.airformdata.com/commons/dispatch-without-discomfort

How we would know this helps

A proposed field protocol: establish the baseline, measure delivered flexibility, count recovery demand, record comfort and overrides, and publish uncomfortable results.

State the claim before running the trial

Choose a specific service and a specific population of buildings. Define the dispatch interval, eligible conditions and comparison method before collecting the outcome. Register what would count as a failure.

Measure both sides of the relationship

Record electrical power and energy with timestamps and units. Report baseline uncertainty, equipment availability, recovery demand, comfort deviations and occupant overrides. Record tariff assumptions separately from physical measurements.

Compare against a simpler option

A comparison group should include appropriate efficiency improvements or existing control strategies. A large change from a poor baseline is not enough to show that the added coordination is valuable.

Give the result a useful form

Publish a de-identified dataset, a data dictionary, the calculation method and known exclusions where participant permission allows. Independent review should be able to reproduce the headline result. No field results have yet been added to this corpus.

Bears on: tests thermal-flexibility, tests shared-intelligence, depends-on source-doe-flexible-buildings Next test: Select a pilot site and owner; register the baseline method, privacy terms, comfort envelope and acceptance thresholds before dispatch. Record: https://www.airformdata.com/commons/field-proof

Test the right to override

A proposed failure exercise: interrupt communications, withdraw consent and verify that local comfort and control remain intact.

Before a pilot offers grid services, exercise the local control boundary with engineering supervision. Record override latency, the handling of withdrawn permission, recovery after communication loss and the audit trail. Agree on safe test conditions and acceptance criteria with equipment engineers and participants. This record describes a proposed test; it is not a safety certification.

Bears on: tests comfort-first Next test: Name the engineering owner and validate a bounded test procedure before any live equipment test. Record: https://www.airformdata.com/commons/override-trial

What if the peak simply moves?

A successful dispatch interval may be followed by recovery demand that erodes its value or creates another constraint.

Measure the entire event, including the recovery period. Examine whether coordinated restart produces a new peak, whether storage losses increase energy use, and whether the original benefit persists after those effects. This is an open question for the proposed Airform field programme.

Bears on: challenges thermal-flexibility Next test: Publish a bounded experiment, its baseline, its failure criteria and the observed outcome. Record: https://www.airformdata.com/commons/rebound-question

Who is paid when a household's battery serves the grid?

If the node is the household's comfort and the grid's resource at once, the value it creates has at least three claimants: the occupant, the building owner and the operator of the federation. The split is undecided.

A federation that quietly keeps the value will not keep its participants. A federation that gives it all away cannot fund itself. The Commons proposes publishing the compensation rule before the first dispatch, and reporting the distribution afterwards, so that the arrangement can be judged by the people inside it.

Bears on: depends-on comfort-first Next test: Publish a bounded experiment, its baseline, its failure criteria and the observed outcome. Record: https://www.airformdata.com/commons/compensation-question