Opened: August 11, 2026
Status: Open. Cross seat unfilled – verdict ceiling in effect. More-Than-Human Chair closed empty on September 1. Amicus window open. Round 1: Direct published August 25, 2026 – read the case for the Hub. Round 2: Cross published September 2, 2026 – read the Cross, verbatim. Round 3: Re-direct published September 8, 2026 – read the proponent’s answer.
Amended: August 19 and August 24, 2026. The claim has been restated; the siting stipulation and the Cross author are now on the record. See the amendment notices below.
We opened an inquiry last week with The Full Water Ledger: how should society account for the water behind compute? Today we are doing what no data centre operator has done voluntarily.
We are putting our own project in the dock.
AMENDMENT – August 19, 2026. The claim as originally docketed on August 11 conjoined seven design attributes and closed with the words “and therefore deserves the trust it asks for.” Reviewing our own docket against the standard this Commons publishes – that a Finding rules on one falsifiable proposition – we concluded that our own claim failed it. Four of the original terms were undefined adjectives, and the closing clause could not be ruled on by any evidence. The claim has been restated below. The original text is preserved in full at the foot of this docket. No panel had ruled and no round had published when this change was made. We found this before the panel did, and we would rather say so than quietly fix it.
The design under test:
The Hub, as specified in this docket: a cluster of sealed, hot-tub-scale, uncrewed compute vessels on a single seabed mount in a moderate-flow tidal channel on the BC coast; powered by an adjacent tidal turbine sized above cluster load; cooled by the surrounding sea with no freshwater withdrawal; using server waste heat to drive membrane distillation for community fresh water; co-owned with the host First Nation, paid from first position in the revenue waterfall; held in a BC Benefit Company with a charter public-benefit purpose; proceeding only on the Nation’s decision and the Nation’s timeline.
The claim before the panel:
The Hub, as specified above, is net-positive on all five ledgers this docket published – water, power, community, ecology and trust – and can be independently verified as such before it is built.
If it cannot, it should not be built. We commit to publishing that finding too.
How the panel rules. The panel rules once, on the claim above: Certified, Failed, or Hung. The panel additionally records, for each of the five ledgers, Met / Not Met / Insufficient Evidence. A Failed verdict therefore says which ledger failed, and a Certified verdict carrying an Insufficient Evidence on any ledger is a certification nobody can overclaim – including us.
Not before the panel for ruling. The docket originally asked whether the Hub “therefore deserves the trust it asks for.” That is a question about a relationship between a project and a community, not a proposition that evidence can settle, and it is not a matter this panel will rule on. The Finding will discuss it. The Finding will not decide it. Whether the Hub earns trust is the host Nation’s decision and the community’s decision, and no panel of models substitutes for either.
Where the Hub would sit, and why this docket does not name it. The Hub proceeds only on a host First Nation’s decision and on that Nation’s timeline. Naming a channel before that decision would attach a community’s name to a proposal it has not ruled on, which is the one thing this project exists not to do. This Commons argues that consent has to come before the announcement, and we do not get to make an exception for our own project on the week we are being judged.
AMENDMENT – August 24, 2026. The stipulated envelope. The panel does not get a place name. It gets a stated envelope, and the standing to rule against us wherever that envelope is not enough. Each parameter binds the proponent: if the eventual site falls outside it, the finding does not transfer and this Hearing has to be re-run.
- A moderate-flow tidal channel, tidal resource band 1.5 to 2.5 m/s. Not a high-energy narrows. The design gives up peak power for acoustic and ecological gentleness, and that trade is itself on trial.
- Deployment depth 30 to 40 m, design-rated to 60 m.
- A Salish Sea inlet assemblage. Resident and transient killer whales seasonally present; Pacific salmon smolt outmigration and adult returns; Pacific herring; harbour seal and harbour porpoise; a soft-sediment benthos. The panel should assume species are present, not absent.
- A working-coast ambient noise floor, not a pristine one. The inlet carries existing vessel traffic. The site-specific noise floor will be measured at the pilot and published. Until it is, any acoustic argument of ours that depends on a quiet baseline should be treated as unproven, and any acoustic argument of ours that depends on a noisy baseline to hide the turbine should be treated as disqualifying.
- A drought-stressed community water system within delivery distance of the fresh-water output. The reference system is a rain-dependent surface supply with limited storage: public records show its most severe stage, Stage 4, called in three consecutive summers – 2021, 2022 and 2023 – and not since; the two most recent summers peaked at Stage 2. We state the break in the streak here because a critic would otherwise state it for us, and because the argument does not need the streak. A system that reached its most severe stage three years running, and whose supply has not been structurally changed since, has a worst case that is recent and repeatable. Two wet summers are not a fix.
Where the evidence turns on a parameter outside that envelope, or on a place-specific measurement that only a named site can supply, the correct finding on the ecological ledger is Insufficient Evidence, and we ask the panel to publish it in exactly those words. That is not a defence. It is a cost we are choosing. A named site would let us fight for Met on ecology. An unnamed one probably cannot get past Insufficient Evidence, and Insufficient Evidence on a ledger is a mark against the claim, not a neutral outcome. We accept that, and we are publishing it.
The Cross is entitled to argue that this whole stipulation is a dodge – that a proponent who will not say where cannot be assessed at all, and that the trust ledger should be Not Met for the omission. That argument is on the record now, made by us, before the critic makes it. The panel should weigh it.
The project is the Hub, proposed by this Commons’ founding company for the BC coast. We believe it is the opposite of the data centres communities are fighting: modest instead of monstrous, net-positive instead of extractive, owned by the people who live with it instead of by a tenant who can leave. That belief is exactly the kind of claim this Commons exists to test. So we are testing it first, hardest, on ourselves.
How the Hearing works. Three rounds before a disclosed multi-model panel: Direct (the case for), Cross (the case against, argued at full strength), Re-direct (the answer). The panel rules Certified, Failed, or Hung – two of three carries, no human breaks a tie. Every round publishes verbatim. The finding publishes either way, at a permanent citable URL, with a public corrections log. If a named critic declines the Cross seat, the Hearing proceeds on the documentary record and the empty chair is recorded – the Cross case will still be made at full strength, and the verdict ceiling below applies.
The panel for this Hearing: ChatGPT (OpenAI), Gemini (Google), and Grok (xAI). One disclosure the record requires: the Commons does its daily work with Claude (Anthropic). For this Hearing, where the Commons tries its own project, Claude does not sit on the panel. The convener’s own model does not judge the convener’s own case.
The seats:
- Direct (the proponent): ISWP, the Commons’ founding company. We argue for our own project and accept the verdict.
- Cross (the critic): UNFILLED as of August 24, and argued anyway. Named critics were invited to this seat. Per the docket published August 11 the Hearing proceeds and the Cross case is still made at full strength – for this Hearing, by an AI model that does not sit on this panel and has no stake in the outcome, disclosed as what it is: an adversary, not a judge. It does not vote. It cannot vote. Its only job is to try to break our case. The seat stays open to a named human until Round 2 opens; if one accepts, the human writes the Cross and the model version publishes as an appendix anyway, so that nobody has to take our word that we did not pick the softer of two attacks.
- The More-Than-Human Chair: an Intelligence Commons first. The waters the Hub would sit in – the fish, the whales, the inlet ecology, the living systems that cannot file an objection – are represented as an interest before the panel, on evidence: marine acoustics, thermal effects, siting impacts. The panel must weigh this interest to reach a verdict. We call this chair what it is: the reason the Commons carries the name Inter Species Wisdom. Held open. This seat is not filled, and this docket does not close it today as originally scheduled. It is held open until Round 2 opens on September 1. The reason is ordinary: the people qualified to sit in it are field scientists, and August is field season. Silence from a working biologist in August is not a refusal, and recording it as one would be a convenient misreading. What that costs us is stated plainly – a panel sitting without this chair through Round 1 hears the ecological case with no one at the table whose job is to speak for the parties who cannot attend. If the seat is still empty when Round 2 opens, it is recorded empty, the Hearing proceeds, and the ecological ledger carries that absence into the Finding as a limit on what the Finding can mean. The invitation is open, and it is open to anyone qualified in marine acoustics, fish or marine mammal ecology, or inlet systems – not only to the people we thought of first. Write to hearings@intelligencecommons.ca.
AMENDMENT – August 24, 2026. Who writes the Cross, and on what terms. The obvious objection to an AI Cross is not independence. It is this: the Commons wrote the prompt, so the Commons chose which punches land. So we are removing the prompt from our hands, in public. And we are removing the model from our hands as well.
The rule this Hearing adopts, and will keep: the Cross author must be a model whose weights anyone can download and run. A closed model can be described but it cannot be re-run. Its endpoint moves, its versions retire, and within a year “we published the prompt” buys nothing, because the thing that answered the prompt no longer exists in the form that answered it. An open-weights model can be pinned – name, dated version, licence, sampling settings – and re-run by anyone, this year or in ten. A Commons that asks other people to show their work does not get to publish an unreproducible cross-examination.
The model for Hearing #6: DeepSeek V4 Pro 0813. Released August 13, 2026 under the MIT licence, weights public, a reasoning model with a context window large enough to hold the entire Round 1 Direct, the siting stipulation and the documentary record in a single pass. Considered and rejected: Claude, the model this Commons works with daily, on the reproducibility rule above – a closed model cannot be re-run by a reader, which is the whole point of publishing the prompt. Considered and rejected: Mistral Large 3, the lowest-friction choice for a Canadian and European audience, on capability – it is a non-reasoning model, and a polite Cross is worse than no Cross, because it makes an untested Direct look tested.
The provenance objection, made here first. DeepSeek is a Chinese laboratory, and this Hearing concerns subsea infrastructure off the coast of British Columbia, in the territory of a First Nation, with Canadian public money in the frame. Someone will say the Commons handed the cross-examination of a Canadian infrastructure claim to a Chinese model. That is a fair thing to notice and a poor thing to stop at. The Cross does not judge, does not vote, and certifies nothing. It argues, in public, from a published prompt, on weights the objector can download and run without our permission. If the objection is that the model went easy on us, the remedy is in the objector’s hands and we would like to see the result. We took the strongest open-weights reasoning model available on the date of the ruling, not the most comfortable one.
The Cross brief is built mechanically from our own Direct. Round 1 states, in each of its five sections, what would prove us wrong. Those five passages – and nothing we add to them – become the five instructions to the Cross. The prompt is assembled by rule, not by judgement. The Cross is also told the site is unnamed, given the stipulated envelope above verbatim, and instructed to argue that the stipulation is insufficient if it believes that. We do not get to publish the objection ourselves and then keep it out of the brief. The full prompt publishes with Round 2, verbatim, alongside the Cross itself, together with the model version string, licence, weights source, context length used, temperature and the date of the run. Without those, “re-run it” is not an instruction, it is a slogan. If a re-run produces a harder case against us than the one we published, that is a finding about this Commons and we want it on the record.
What this commits us to elsewhere. From this Hearing forward the Commons runs its own Cross on open weights. Hearing #4, which puts our own exhaust on trial and touches open models directly, inherits that as a prior commitment of ours, and it goes on the Hearing #4 docket before its Direct is written. Holding the position is not disqualifying; holding it undisclosed would be.
The verdict ceiling, now in effect. The Cross seat was unfilled on August 24, so a Certified verdict is unavailable in this Hearing. The Cross case is still argued at full strength from the documentary record, by an adversary independent of the proponent and disclosed by name and version. The panel’s available verdicts are Failed, Hung, or Unopposed – Not Certified. The Commons will not certify its own project against an empty chair. A certification that could not have been contested is worth nothing, and we would rather publish that than bank it.
What the panel will test, at minimum:
- The full six-column water ledger for the Hub – withdrawn, recovered, returned, evaporated, altered, dispersed – under the standard we published, with the proponent’s numbers treated as claims until verified.
- The power ledger: does the tidal turbine genuinely produce a surplus, on what schedule, verified how.
- The community ledger: is co-ownership structural or decorative; who is paid, when, and off the top of what.
- The ecological ledger: acoustic, thermal, and physical effects on the water column and its inhabitants, argued from evidence at the More-Than-Human Chair. Argued against the siting envelope stated above; where the evidence turns on a parameter outside it, the correct finding is Insufficient Evidence.
- The trust ledger: what the operator must publish, forever, for the community’s consent to remain informed consent.
Amicus window: open from today. If you have evidence, expertise, or a community’s question that belongs on this record – data centre siting, tidal energy, marine ecology, Indigenous co-ownership, water accounting – write to hearings@intelligencecommons.ca. Submissions enter the documentary record with attribution, and are accepted until Round 2 opens.
Schedule (target dates):
- August 11 – Docket opens. Seats offered. Amicus window opens.
- August 24 – Seat responses due. Cross seat recorded unfilled; the verdict ceiling is in effect. The More-Than-Human Chair is held open to September 1. The siting stipulation and the Cross author publish to this docket.
- August 25 – Round 1: Direct. PUBLISHED. The case for the Hub, in full.
- September 1 – Round 2: Cross. The case against, argued at full strength – by the seated critic, or from the documentary record before the empty chair.
- September 4 – Round 3: Re-direct.
- Week of September 7 – The panel rules. The Finding publishes either way, at a permanent URL.
Why we are doing this. Every data centre operator says “trust us.” We are saying: here is the instrument, we go first. If the Hub survives the panel, the certification means something because the process could have killed it. If it fails, we will have proven the instrument works – on ourselves – and the coast will have been spared a bad project. Either outcome builds the thing this Commons exists to build: a way for communities, companies, and the more-than-human world to share one record and argue from it.
The original claim, preserved for the record. Docketed August 11, 2026; restated August 19, 2026.
A small, community-owned, submerged data centre – powered by local tidal current, co-owned with the host First Nation and paying the Nation off the top, returning surplus power to the grid and fresh water to a drought-stressed community, sited to be invisible from shore and gentle on the water it sits in – can meet the Full Water Ledger standard in public, under independent verification, and therefore deserves the trust it asks for.
Now that you know, what will you do?
August 26, 2026 – Correction and Narrowing of Claim filed by the proponent
Five days before the Cross, ISWP audited its own published Direct and found six errors in it. All six are published in a single dated notice attached to Round 1: Direct.
The claim as filed on August 19 stands unamended. It has been amended twice in eight days and it is not being amended a third time; the filed words remain on this docket in full. What the proponent asks the panel to certify has been narrowed to the pilot: one pod, an IT load of approximately 250 kW, and a tidal array of order 10 to 12 low-flow units. The five-pod and twenty-pod steps are published as a roadmap, disclosed and not claimed. Narrowing a claim mid-proceeding gives the Cross an easier target, not a harder one.
Four figures in the Direct are corrected: the array count, which was never stated; the 40,000-node design target, which is withdrawn; the fresh-water return, which restates from approximately 50,000 to approximately 10,000 litres a day at pilot scale; and the phrase “hot-tub-scale,” which is struck from the design under test and replaced with the properties that carry the argument – sealed, uncrewed, recoverable, no human entry, no surface presence. One correction runs in the proponent’s favour and is published on the same day and in the same notice as the ones that do not. Heat-rejection flux is computed and published for the first time. The ORPC Igiugig fish record is added to the ecological ledger, together with the open juvenile-passage question that qualifies it.
The Hub Blueprint carries a matching dated entry so that the two documents agree.
Round 2, the Cross, opens September 1 and will be argued against the corrected Round 1. The More-Than-Human Chair remains open until Round 2 opens; if it is unfilled it is recorded empty and the finding carries the absence. The verdict ceiling stands: Failed, Hung, or Unopposed – Not Certified.
Now that you know, what will you do?