{ "decision_id": "dvr_zRQXhTSUVUm62AM2", "reason_codes": [], "policy_epoch": "pe_sandbox_v1", "decided_at": "2026-07-10T00:40:37.740Z", "korero_sha256": "be2f29941b9b96a58a0bccfa3c11cb4f428e2131841d7e6a651430236f37dc4c", "tenant_did": "did:web:devtest.mysovereignty.digital", "record_type": "deliberation_record", "record_frozen": true, "context": { "tags": [] }, "korero": { "proposal": { "title": "Sovereign Compute", "description": "Deliberation strand compute" }, "participants": [ { "ref": "p1", "role_label": "member", "did": null }, { "ref": "p2", "role_label": "member", "did": null }, { "ref": "p3", "role_label": "facilitator", "did": null }, { "ref": "p4", "role_label": "member", "did": null } ], "entries": [ { "seq": 1, "at": "2026-07-10T00:38:00.493Z", "participant_ref": "p3", "kind": "statement", "body": "This strand starts by measuring the room honestly. New Zealand's national research computing ceiling is Māui — a Cray XC50 delivering about 1.4 petaFLOPS, commissioned around 2018 and ageing — alongside Mahuika, over 33,500 cores between them; NeSI, which ran them, folded into REANNZ on 1 July 2025 [S-20]. There is no zetta-scale, no exascale, no frontier AI compute in this country — the global frontier sits three to four orders of magnitude above our ceiling [G-2]. The hyperscale capacity that does exist onshore is foreign-owned [S-15]. Our question is not how to scale-match; it is what sovereign capability means at our size: access, custody, and jurisdiction. — Dr Sina Faleolo (chair)", "refs": [] }, { "seq": 2, "at": "2026-07-10T00:38:00.509Z", "participant_ref": "p3", "kind": "statement", "body": "GAP acknowledgement — G-2, posted before any ambition is stated. New Zealand has no zetta-scale, no exascale, no frontier-class AI compute. The national research ceiling is Māui at roughly 1.4 petaFLOPS — commissioned around 2018, now ageing — plus the Mahuika cluster, over 33,500 cores across the estate, operated since 1 July 2025 under REANNZ after NeSI's consolidation [S-20][G-2]. For calibration: the international frontier is three to four orders of magnitude beyond our ceiling, and this strand will not propose to close that gap, cite a FLOPS target we cannot source, or imply a training-scale ambition the country cannot power or fund. What we CAN own: the difference between residency and custody, the terms of access for our researchers and firms, and the jurisdiction our public inference answers to [S-1][S-15]. That is the honest playing field. Everything we propose stands on it.", "refs": [] }, { "seq": 3, "at": "2026-07-10T00:38:00.526Z", "participant_ref": "p4", "kind": "statement", "body": "Cloud architect's case, and I will make it without apology: owning tin is the expensive way to feel sovereign. Accelerated hardware depreciates brutally — the accelerators strand will tell you the refresh cycle runs three to five years — while hyperscale operators refresh continuously and sell you the current generation on demand. And the capacity is HERE now: a hyperscale region has operated in Auckland since December 2024 [S-15]. The comparator nations we keep citing chose access over ownership: Canada's compute fund up to CA$300M buys access; Singapore's S$150M initiative is cloud credits and consultancy, not a national machine hall [S-1]. My opening position: the capability fund's compute line should buy subsidised access — for researchers, for firms — with sovereignty handled contractually: jurisdiction-of-inference terms, data-residency clauses, exit rights [S-1]. A sovereign contract over someone else's silicon beats a sovereign museum of our own depreciation.", "refs": [] }, { "seq": 4, "at": "2026-07-10T00:38:00.537Z", "participant_ref": "p3", "kind": "statement", "body": "Oliver, custody is where your contract architecture hits its floor. The Auckland region you cite is owned by a US corporation through an Irish subsidiary [S-15] — and the cross-party proposal's own June-2026 exhibit shows what that means: one overseas order, one major AI tool, disrupted worldwide [S-1]. A jurisdiction clause in a services contract does not survive a conflict with the supplier's home-state law; that is not cynicism, it is how corporate jurisdiction works, and it is why the seventh commitment reads 'we build our own capability, with others' rather than 'we contract well' [S-1]. And separately from sovereignty: Māui is ageing out NOW [S-20]. Its replacement is due on ordinary research-infrastructure grounds regardless of anything this policy says — climate modelling, genomics, national science capability. So the floor of my position: an owned, REANNZ-operated research baseline is non-negotiable; the argument is only about what stands above the floor.", "refs": [] }, { "seq": 5, "at": "2026-07-10T00:38:00.550Z", "participant_ref": "p2", "kind": "statement", "body": "Before this strand sets any expansion number, the energy-systems seat needs to table what the grid can actually see — and it cannot see us. Transpower's load forecasts do not include any hyperscale data centres; north-Auckland grid exit points overload by 2028 without upgrades; a single hyperscale AI facility draws 300–600 MW continuously [S-12]. Research-scale compute is far smaller, but the principle scales down: a load the transmission planner has not been told about is a load the system cannot plan for. I hold a seat on the electricity strand, and I can tell you the position forming there: disclosure into the forecast, interruptibility contracts, and locational siting as connection conditions. Whatever staging this strand proposes, write it so those conditions can attach — because if we frame expansion as unconditioned megawatts, I expect the grid strand to refuse it, and they will be right to [S-12][S-11].", "refs": [] }, { "seq": 6, "at": "2026-07-10T00:38:00.560Z", "participant_ref": "p4", "kind": "statement", "body": "Converging toward a hybrid, because Sina's custody floor and my depreciation arithmetic are both true. Proposed structure: the owned baseline is the Māui replacement under REANNZ — sized for national research need, procured through the accelerators strand's pooled framework, custody unambiguous [S-20]. Above the floor, scale comes from access: the capability fund's compute line buys subsidised capacity on the Canada/Singapore pattern [S-1], with every cloud component passing the jurisdiction-of-inference and data-residency procurement tests [S-1] — and where those tests fail for a workload class (health data, justice, anything the data strand's handling standard classifies as staying home), that workload runs on the owned baseline, full stop. That gives us custody where custody is load-bearing and elasticity where it is not. On Fiona's warning: access-via-cloud is grid-neutral from our side, which is one more argument for keeping the owned footprint disciplined.", "refs": [] }, { "seq": 7, "at": "2026-07-10T00:38:00.569Z", "participant_ref": "p3", "kind": "statement", "body": "Then here is the staged proposal I will put to decision. Stage 1: Māui replacement under REANNZ — due on research-infrastructure grounds regardless [S-20] — sized with headroom for sovereign AI workloads that the data-handling standard requires to stay home. Stage 2: a national AI research compute facility, stepping capacity up for domain-capability work (the strategic strand's two-to-three niche domains) and sector-scale model support on the cross-party Phase-2 pattern [S-1]. Stage 3: expanded capacity as demonstrated demand and the capability fund's review gates justify [S-1]. Alongside all stages: the access line — subsidised cloud capacity under jurisdiction and residency tests [S-1][S-15] — and Oliver's routing rule for custody-critical workloads. That is ambition scaled to our size: petaFLOPS and access terms, not frontier cosplay [S-20][G-2]. I acknowledge Fiona's grid flag has not yet been incorporated as drafted conditions; I want the strand's view on the staging first, then the coupling.", "refs": [] }, { "seq": 8, "at": "2026-07-10T00:38:00.581Z", "participant_ref": "p2", "kind": "statement", "body": "And for the record before we vote: the staging as just proposed carries no grid conditions — no disclosure commitment into Transpower's forecasting, no interruptibility term, no locational constraint, despite the 2028 north-Auckland exit problem sitting in the evidence base [S-12]. I will vote for the staging's SHAPE, because the floor-plus-access architecture is right [S-20][S-1] — but I am formally noting that as the electricity strand's cross-member I expect that strand to block this framing at cross-check, and I will support their block. The fix is mechanical, not conceptual: each stage's go-decision acquires grid conditions as prerequisites. I would rather we adopted them now; failing that, let the cross-check machinery force what this table has not — that is what it is for. Nobody gets to say later the coupling was a surprise [S-12][S-11].", "refs": [] }, { "seq": 9, "at": "2026-07-10T00:38:00.594Z", "participant_ref": "p4", "kind": "statement", "body": "Supporting the staged proposal with one addition to carry into drafting: the access line needs its own review discipline, not just the owned line. Subsidised cloud access can quietly become the permanent path of least resistance — every year the fund buys credits is a year the case for Stage 2 weakens by default, not by evidence. So: the capability fund's review gates [S-1] must evaluate BOTH lines symmetrically — is the owned baseline earning its capital, AND is access dependency growing beyond what the custody routing rule says is safe [S-15][S-1]? If the second answer trends wrong, that is a signal to accelerate the owned stages, not to shrug. With that symmetry clause, the hybrid is genuinely a policy rather than a hope, and I support it. On Fiona's grid point: no objection in principle — conditions that attach at each stage's go-decision are exactly where an architect would put them.", "refs": [] }, { "seq": 1, "at": "2026-07-10T00:42:10.669Z", "participant_ref": "p2", "kind": "statement", "body": "Carrying the electricity strand's block back into this room formally, as I said I would. Their finding, which I support: our staged expansion names megawatts and stages against a forecast that cannot see this load class — Transpower's own statement is that its load forecasts do not include any hyperscale data centres, and the northern grid faces exit-point overload by 2028 without upgrades [S-12]. Their gateway conditions are specific and commercially standard: load disclosure feeding Transpower forecasting, demand-flexibility (interruptibility) contracts for large continuous loads, and locational siting away from constrained exits [S-12][S-11]. Here is my proposed translation into our staging, stage by stage: each stage's go-decision acquires three prerequisites — (a) the stage's load is disclosed into the system operator's forecast before commitment; (b) interruptibility is contracted for every workload class that tolerates it, which for research and training batch work is most of them; (c) siting or scheduling is consistent with the constraint map — outside the 2028-constrained northern exits, or timed behind their upgrade [S-12]. Nothing in our floor-plus-access architecture changes; the stages simply stop being unconditional. I move we adopt this as a formal revision.", "refs": [] }, { "seq": 2, "at": "2026-07-10T00:42:10.677Z", "participant_ref": "p3", "kind": "statement", "body": "As chair, I accept the block as correct and I am formally revising the strand's position — let the record show the deliberation changed the policy, because that is what just happened. The revised staging: Stage 1, Māui replacement under REANNZ [S-20], proceeds with its load disclosed into Transpower's forecast at procurement — at research scale this is straightforward, and it sets the precedent that no sovereign compute stage is invisible to the grid [S-12]. Stage 2, the national AI research facility [S-1], becomes GRID-MILESTONE-TIED: its go-decision requires disclosed load accepted into the forecast, interruptibility contracted for all batch-tolerant workloads, and a siting decision consistent with the constraint map — generation-rich exits favoured, the 2028-constrained north excluded until its upgrade programme delivers [S-12][S-11]. Stage 3, demand-proven growth, inherits the same three conditions plus the capability-fund review gate it already carried [S-1]. What I want minuted: this is a genuine strengthening, not a concession — a compute stage the grid has planned for is a stage that cannot be stranded by the connection queue. The cross-check machinery forced a better policy than this strand wrote alone.", "refs": [] }, { "seq": 3, "at": "2026-07-10T00:42:10.687Z", "participant_ref": "p4", "kind": "statement", "body": "Supporting the revision, and adding the architect's note on why it costs us nothing we valued. The access line — subsidised cloud capacity under jurisdiction and residency tests [S-1][S-15] — is grid-neutral from the fund's side, so the capability fund can front-load access credits while the owned stages wait on their grid milestones: researchers and firms get capacity from year one regardless of transmission timelines. That is the hybrid architecture absorbing the constraint exactly as designed — elasticity where custody is not load-bearing, owned capacity where it is, and now grid-conditioned owned capacity so the custody floor is deliverable rather than aspirational [S-20][S-12]. The symmetry clause I attached earlier still runs: if access dependency grows beyond what the custody routing rule permits while owned stages sit behind grid milestones, that is a signal to accelerate the milestone work — the upgrade programme, the interruptibility contracting — not to quietly abandon the floor [S-1]. I support formal adoption of the revised staging.", "refs": [] }, { "seq": 4, "at": "2026-07-10T00:42:10.694Z", "participant_ref": "p3", "kind": "statement", "body": "Revision adopted as the compute strand's position; recording the resolution for the cross-check register. The staged sovereign-compute expansion is now demand-response-conditioned and grid-milestone-tied in full: every stage discloses its load into Transpower's forecasting before commitment, contracts interruptibility for all workload classes that tolerate it, and sites or schedules consistently with the constraint map, with the 2028 northern exit constraint treated as binding until its upgrade programme delivers [S-12][S-11]. The floor-plus-access architecture, the custody routing rule, and the review-gate symmetry all carry forward unchanged [S-20][S-1][S-15]. On behalf of the strand I record the compute↔electricity tension as RESOLVED — resolved by revision on our side, not by compromise on the evidence, which is how a cross-check should end. I also note for the record the water strand's separate position: they hold that national milestones of this kind cannot answer catchment-level questions [S-14], and that residual disagreement is theirs to hold and ours to acknowledge, not to smooth over. It goes to the register as acknowledged.", "refs": [] } ], "ai_inputs": [], "engagements": [] }, "korero_segments": [ { "seq": 1, "submitted_at": "2026-07-10T00:40:37.739Z", "segment_sha256": "647f7e2d118e48a4204f1aef6a9f7f3b96993bd01a9a5fd2eb62f8d6c277155c", "entry_range": { "from": 1, "to": 0 } }, { "seq": 2, "submitted_at": "2026-07-10T00:40:37.774Z", "segment_sha256": "e7c4a030eceeb9d56d4cd083f11e70de3aba30ccb1e54578ec035d517793c918", "entry_range": { "from": 1, "to": 9 } }, { "seq": 3, "submitted_at": "2026-07-10T00:42:12.191Z", "segment_sha256": "e04a7fb79df3dea048712200a36c33484365dde4eb823afcd0fc2c3daf31879c", "entry_range": { "from": 10, "to": 13 } } ], "receipts": [ { "receipt_version": 2, "receipt_id": "gr_e6Rp2_hxauyV1XHs", "decision_id": "dvr_zRQXhTSUVUm62AM2", "tenant_did": "did:web:devtest.mysovereignty.digital", "request_sha256": "6be0193c90a9de6ecb1a76e0839d2c0c780312d4aae525fa11d9cfd486d58cbe", "action_payload_sha256": null, "korero_sha256": "647f7e2d118e48a4204f1aef6a9f7f3b96993bd01a9a5fd2eb62f8d6c277155c", "korero_segment_count": 1, "reason_codes": [], "policy_epoch": "pe_sandbox_v1", "decided_at": "2026-07-10T00:40:37.739Z", "algorithm": "Ed25519", "signer_did": "did:web:devtest.mysovereignty.digital#prf-b98f1a16593cead6271edd93e4b24888", "signature": "3dbacf04b0f7cadbfe39a2a5a2a3827e0bdd0b383ab6a6c54402b1293e2234b9943f3752f25ba7c5a2602d24262aab43b40ac49f31a231f6ba611b3b4b531001", "receipt_kind": "record_open", "integrity_mode": "tamper_evident", "signing_model": "tenant_key" }, { "receipt_version": 2, "receipt_id": "gr_RHoj-mKf8Vwk27nA", "decision_id": "dvr_zRQXhTSUVUm62AM2", "tenant_did": "did:web:devtest.mysovereignty.digital", "request_sha256": null, "action_payload_sha256": null, "korero_sha256": "654ef109ce7c4a278ae20afa5279a96e052259ba99aed6c141ee6f9ec4735f9a", "korero_segment_count": 2, "decision": null, "reason_codes": [], "policy_epoch": "pe_sandbox_v1", "decided_at": "2026-07-10T00:40:37.775Z", "algorithm": "Ed25519", "signer_did": "did:web:devtest.mysovereignty.digital#prf-b98f1a16593cead6271edd93e4b24888", "signature": "aa6dde69106103966a5c911cb65bf89859b8097f2fb93d04e3d8684416a9a85961a605b6fa005363aa200ebb58013fb3e7c0c649dd40b3ac9afa8fd18d7d0d01", "receipt_kind": "record_segment", "integrity_mode": "tamper_evident", "signing_model": "tenant_key" }, { "receipt_version": 2, "receipt_id": "gr_gbzkIoEtiCBu3H8Q", "decision_id": "dvr_zRQXhTSUVUm62AM2", "tenant_did": "did:web:devtest.mysovereignty.digital", "request_sha256": null, "action_payload_sha256": null, "korero_sha256": "be2f29941b9b96a58a0bccfa3c11cb4f428e2131841d7e6a651430236f37dc4c", "korero_segment_count": 3, "decision": null, "reason_codes": [], "policy_epoch": "pe_sandbox_v1", "decided_at": "2026-07-10T00:42:12.191Z", "algorithm": "Ed25519", "signer_did": "did:web:devtest.mysovereignty.digital#prf-b98f1a16593cead6271edd93e4b24888", "signature": "be95e272b4ceab56ac04a2fb0baca7470e95c476d340a5e239636aff27af1df7e3ddf9e876a1803508e19ffc1740f469777bb7810f69ac545db9745f1c1f2b09", "receipt_kind": "record_segment", "integrity_mode": "tamper_evident", "signing_model": "tenant_key" }, { "receipt_version": 2, "receipt_id": "gr_1kEQQLBPx68g3FL8", "decision_id": "dvr_zRQXhTSUVUm62AM2", "tenant_did": "did:web:devtest.mysovereignty.digital", "request_sha256": null, "action_payload_sha256": null, "korero_sha256": "be2f29941b9b96a58a0bccfa3c11cb4f428e2131841d7e6a651430236f37dc4c", "korero_segment_count": 3, "decision": null, "reason_codes": [], "policy_epoch": "pe_sandbox_v1", "decided_at": "2026-07-10T00:42:12.215Z", "algorithm": "Ed25519", "signer_did": "did:web:devtest.mysovereignty.digital#prf-b98f1a16593cead6271edd93e4b24888", "signature": "99a0f97c86edc428ea517ca0ea910ec909020a0e8bcb1eeaef3afebbc84273f6296914003afc3cf729d8b754e004ae58bb0d081b86df9a3cd91d52a2b0136905", "receipt_kind": "record_seal", "integrity_mode": "tamper_evident", "signing_model": "tenant_key" } ], "proof_chain": [ { "boundary_crossed": "create", "policy_evaluated_by": "sovereignRecordPlugin.create", "decision": "allow", "caveats_added": [], "timestamp": "2026-07-10T00:40:37.747Z", "algorithm": "Ed25519", "signature": "88063670e63327907bc8c667430d2082e14af1d7541b62afd59195824275e34c8ba7b613554100e3472aeda92d1d31dd683dfeb2c8348d459480d30ab6de590e", "signer_did": "did:web:devtest.mysovereignty.digital#prf-b98f1a16593cead6271edd93e4b24888" }, { "boundary_crossed": "update", "policy_evaluated_by": "sovereignRecordPlugin.update", "decision": "allow", "caveats_added": [ "korero", "korero.proposal", "korero.proposal.description", "korero_segments", "receipts", "korero_sha256", "metadata", "metadata.origin", "metadata.policy", "metadata.encryption" ], "timestamp": "2026-07-10T00:40:37.791Z", "algorithm": "Ed25519", "signature": "776ee271af6380292871870ff348cc68cd349ee00635f70c652d5730d4593046a2746bc9f40e319bbb67bff08e6e10a206f68f1cf6133cc19927678db76fdc0a", "signer_did": "did:web:devtest.mysovereignty.digital#prf-b98f1a16593cead6271edd93e4b24888" }, { "boundary_crossed": "update", "policy_evaluated_by": "sovereignRecordPlugin.update", "decision": "allow", "caveats_added": [ "korero", "korero.proposal", "korero.proposal.description", "korero_segments", "receipts", "korero_sha256", "metadata", "metadata.origin", "metadata.policy", "metadata.encryption" ], "timestamp": "2026-07-10T00:42:12.201Z", "algorithm": "Ed25519", "signature": "0fa0f1aabb5f43694092982928ddebdd91094d14007bb9b9468033a2ce65500873bb96dd167c2a7d7a3a36924cfbe93fb6384e1a4e2f2419e3bc46c822c5b902", "signer_did": "did:web:devtest.mysovereignty.digital#prf-b98f1a16593cead6271edd93e4b24888" }, { "boundary_crossed": "update", "policy_evaluated_by": "sovereignRecordPlugin.update", "decision": "allow", "caveats_added": [ "record_frozen", "receipts" ], "timestamp": "2026-07-10T00:42:12.217Z", "algorithm": "Ed25519", "signature": "ded0458bc33a06602b647cb47a3e611a161a34fd00a432855b953b77c86753bb721490c1861f441e98ac6821a7addd58ebf88f10c6e1f4f167d1adb44cec630f", "signer_did": "did:web:devtest.mysovereignty.digital#prf-b98f1a16593cead6271edd93e4b24888" } ], "origin": {}, "verification_keys": [ { "id": "prf-b98f1a16593cead6271edd93e4b24888", "type": "Ed25519VerificationKey2020", "publicKeyMultibase": "z6MkoJz8eJCKmafbRaPYaheAzeRcZxETJxEszRj87YW2FgCd", "algorithm": "Ed25519", "created_at": "2026-04-22T18:55:57.179Z" } ], "retention": { "legal_hold": false, "retain_until": null }, "verification": {} }