Asset Zero · Open Questions Register (D1)
An open question, honestly held, is a feature. Each entry carries its origin in the evidence archive, where it gets worked in the framework cycles, and what would settle it.
Register v0.1 — last updated 16 August 2026 · 19 open · 0 settled
ID reconciliation: the W4 results file tags "the number" as OQ2 and the reversibility premium as OQ5 using the W4 pack's internal numbering — those map here to OQ-07 and OQ-11. This register's IDs are now canonical. Cycle references follow Cycle-Plan v2 (compressed, 2 Aug): cycle 1 = A2+B2+B1 (10–14 Aug) · cycle 2 = B3+B4+B5 (17–21 Aug) · cycle 3 = final pass, C+D+A (24–28 Aug) · v1.0 freeze 31 Aug; W5 date TBC (program office).
How much explainability/confidence is "enough" before acting on an AI output — and how should it be expressed per asset class?
[W5B] §9.1 · W1 q10 accuracy/trust themes
cycle 2 (boundary test Q4)
An evidence-threshold table per decision class, validated by WG; W5 vote on the contested rows
When is human oversight meaningful rather than symbolic — and how do we test for it?
[W5B] §9.2 · audit-guess median 0% [W3] · rubber-stamp 14% [W1 q8]
cycle 3 (final pass, C3)
An adoptable oversight-audit method (spot-check protocol + challenge test); pilot in Asset Zero's own ledger audits
Who is accountable for vendor/black-box AI, and how do members secure audit rights in practice?
[W5B] §9.3
cycle 3 (C4; relief valve: ships Draft (AI) for v1.x review)
A tested procurement clause set + at least one member-reported negotiation outcome
How should small organisations and local government validate AI on asset data with limited resources?
[W5B] §9.4 · W1 room skew: government-heavy
cycle 3 (C5 small-org path)
A minimum-viable control set (the "smallest version" answers from reviewer packets feed this)
How do we avoid deskilling the professionals the boundary depends on?
[W5B] §9.5 · W1 q2 skill-loss worries · P7 (least-evidenced principle)
cycle 1 decides P7's fate; cycle 3 (C3) drafts practice
Cycle-1 WG call on P7 + a capability-impact screening question in the boundary test
Exactly where is the recommend→trigger line for each sector and asset class?
[W5B] §9.6 · trigger re-measure: act-w/-sign-off 5%→24% cross-room [W2→W4 §7]
cycle 2 (matrix rows + envelopes)
Matrix rows populated per work class with vote data; sector deltas flagged where the room mixes differed
How should volume/alarm budgets for machine-created work be set — and who holds the stop, fast enough? (the W4 "OQ2")
Referred to as OQ2 in the W4 results file — this register's IDs are canonical.
[W5B] §9.7 · the number: median 7.5 WOs/planner/wk, 30% never [W4 §5] · pause-reality: 36% within-hour [W4 §3]
cycle 2 (B5 worked example)
WG validation of 7.5/wk as the published default + the per-review-period reframing; EEMUA-191 analogy check
How should AI-supported decisions be recorded and audited to satisfy SOCI / FAR / Privacy Act expectations?
[W5B] §9.8
cycle 3 (C1 records)
Record-content spec mapped to each regime by a reviewer with compliance background; legal review window is now days, not weeks — if it slips, the spec ships flagged "pending legal review"
How do we preserve tacit field knowledge as experienced workers retire?
[W5B] §9.9 · reviewer case theme (the retired engineer)
cycle 3 (C3)
At minimum: a named practice pattern; possibly deferred to v1.x with dissent recorded
What asset-management decisions should never be automated?
[W5B] §9.10 · abstract bans 0% everywhere vs 19% never on live 66kV isolation [W4 §6 · SES F3]
cycle 2 (matrix "never" rows, written as concrete work classes)
A concrete never-list; if W5 postdates v1.0, the list ships 31 Aug as community-drafted and W5's vote confirms it for v1.x (the evidence says concrete cases, not categories, produce honest "nevers")
What does the 38pp reversibility premium mean for envelope design — how far may reversibility widen auto-trigger? (the W4 "OQ5")
Referred to as OQ5 in the W4 results file — this register's IDs are canonical.
[W4 §6]: 59% vs 21% allow on the same work order · "interpretation belongs to W5"
cycle 2 (B5) + W5
WG interpretation adopted into envelope anatomy (reversibility window as a first-class control); W5 validation vote
When is send back versus override the right human response to a machine-assembled plan?
[W3] Sim 3: split three ways on every plan, no majority to sign anything
cycles 1–2 (B1 human roles + matrix practice notes)
A decision rule drafted from the Sim-3 pattern; tested against reviewer cases
Who owns checking what an optimiser quietly defunded — and what does that check look like?
[W3 w3-s2-r7-owner]: named person 32% · nobody 12% · don't use 37%
cycle 2 (matrix planning rows) + cycle 3 (deferral ledger, C1)
The check named as a role in the matrix's accountability column + a ledger-based method
What ought the value weights be? The room set cost 26 · risk 25 · service 24 · equity 15 · carbon 10 — with no is/ought baseline ever captured.
[W3 §4] (pre-survey never ran)
candidate W5 live instrument (re-run weights with an is/ought split)
A measured is/ought comparison; until then the framework stays silent on what to value and firm on who signs it
How do close-out feedback loops stay alive — the control the community most expects to be quietly skipped?
[W4 §9] honesty poll: close-out loop 30% = the most-named clause to be skipped (37% said they'd keep all)
cycle 3 (C1: make close-out a ledger field, not a promise)
A close-out record requirement in C1 + an Asset Zero telemetry proof (our own loop, published — now an August-sized sample)
Should agents and AI systems be governed on a lifecycle basis, as data is — commissioning, change, retirement — with a lifecycle owner and a live register?
Cycle-1 MA3 (WG proposal) · 14 Aug call, received positively (A10)
cycle 3 (candidate section: C1 register fields + C4 vendor lifecycle)
A drafted lifecycle clause + register schema the WG can live with; else deferred to v1.x with the proposal recorded
Is competence assessed for the AI tool as well as the human — which tool for which task, and how is that demonstrated before use?
Cycle-1 MA3 (WG proposal, repeated in MA2 against P4) · 14 Aug call (A12)
routed to P1's screening test + C3 practice notes; the routing to be confirmed back to the proposer
Wording in P1/C3 that names tool-competence assessment as a screening step; proposer confirms it answers the input
How does AI make uncertainty visible rather than masking it behind confident presentation — and what must a human see before relying on an output?
Cycle-1 MA3 (WG proposal) · 14 Aug call (A12)
routed to P2 (as its spine) and P3 sharpening; overlaps OQ-01
A visible-uncertainty requirement in P2/P3 text; an evidence-threshold table (OQ-01) that makes it operable
How is the feedback loop closed — is any AI use ever judged by realised outcomes, and does that outcome record feed back into both the governance side and the model?
14 Aug call, chair's flag (A11) · candidate principle C1 "close the loop" (chair, 16 Aug) · [W3] audit-guess median 0%
cycle 2 disposition log (candidate eighth principle) + cycle 3 (C1 close-out record); overlaps OQ-15
A decision on whether it is a principle or a C1 record requirement with an owner and a schedule; either way, an outcome-record field
Closed entries land here — settled (pointing to the framework text and the evidence that settled it) or deferred to v1.x with the dissent recorded. Never deleted.
Nothing settled yet — the register opened 2 Aug 2026 and the first review cycle starts Tue 11 Aug.
Provenance: AI-drafted (Stream 1, 2 Aug 2026) from [W5B] §9, the four results files and the Series Evidence Synthesis; figure-checked against sources. Interim boundary registration: Recommend/draft-only — Kai reviews and commits.
