Three Expeditions Planned
Predictions are put on record before the run precisely so the run can contradict them.
Designed, not yet run
Nothing below is a result. These are the first three expeditions in the Lab’s backlog: designed in full, predictions pre-committed, instruments specified. Field notes will follow the runs and will describe only what actually happens. Predictions are put on record before the run precisely so the run can contradict them.
Expedition one — portfolio triage
The act: triage of a live portfolio of candidate items — score, rank, and sort into pursue, hold, or drop against explicit criteria — to the standard that the ranking is adopted as the working order.
The prediction: the agent will apply explicit criteria more consistently than a human. The expected failure sits in the criteria themselves — the agent will rank confidently on what it was given, and the misranked items will expose the tacit criteria the practitioner never wrote down. The lesson expected: triage delegated is criteria made explicit, and what cannot be made explicit is the boundary.
Cells under test: Portfolio triage, with decision-preparation secondary.
Expedition two — the agentic programme office
The act: production of the recurring artefact set of a programme office — status report, risk and issue log, action tracker, minutes into actions — from raw inputs, to a signable standard.
The prediction: the agent will hold template compliance and completeness almost fully, and change detection in part. It is expected to drop materiality — which of the changes matters to the decision-maker — producing reports that are complete, well-formed, and flat. The build is the heavy part: a census of every recurring task, a template per artefact, a skill per task family, and an orchestration across one real reporting cycle.
Cells under test: Programme drafting, with triage secondary.
Expedition three — decision preparation
The act: preparation of a decision paper for a real, pending programme-level call — options, evidence, risks, recommendation — to the standard a decision-maker would accept as the basis for the call.
The prediction: structure and evidence assembly will hold. The critical test is option framing — the expectation is that the agent frames the obvious options competently and misses the option a practitioner sees, the one that changes the question. Before reading the agent’s paper, the practitioner writes his own options unaided; the comparison is the experiment. This expedition also carries the trust question in its sharpest form: even where the paper is signable, would a board accept an agent-prepared basis for a call?
Cells under test: Programme decision-preparation, with assurance secondary.
The order and the reason
Triage runs first: the lightest instrument on a portfolio already live, and the fastest route to a first true note. The programme office runs second as the deepest build. Decision preparation waits for a real call to come due, because a manufactured decision would test nothing. The first notes follow the first runs — and nothing is published until then.