Five stages, in sequence. Each states the work to be done, the record that should exist on leaving it, and the conditions to verify before proceeding. The software implements this sequence; the sequence itself does not depend on the software.
The question, its explanation, and the feasible alternatives
Begin with the reason the decision is being taken: an event, a request, or a change in conditions. Recording it before anything else matters because the initial formulation of a problem constrains every alternative subsequently considered.
Set out next the explanation of the situation. The same evidence supports different courses of action depending on the causal account adopted, which makes the explanation part of the decision rather than a preface to it. Observations should be distinguished from inferences, and propositions resting on belief rather than measurement identified as such.
A decision requires at least two feasible alternatives. A single alternative is a proposal, and a proposal presented with its supporting reasons already attached tends to accumulate further support rather than attract comparison. The set should therefore be generated before it is evaluated, and should include continuation of the current course of action wherever that remains feasible — the alternative most often omitted from the set and most often adopted by default.
A statement of the decision to be taken.
The evidence that the situation obtains, together with any evidence to the contrary.
The explanation of its causes, the constraints in force, and the parties affected.
The assumptions relied upon, recorded as assumptions.
At least two materially different alternatives, including the status quo where continuation is feasible.
For each alternative: the supporting reasons, the opposing reasons, and its requirements.
Is the decision to be taken stated, as distinct from the problem observed?
Is there evidence that the situation obtains, rather than a report that it does?
Is any proposition presented as established in fact an assumption?
Are there at least two materially different alternatives?
Were the alternatives generated before a preference was formed?
Are the requirements of each alternative — authorisation, funding, personnel, time — recorded?
The decision is the first item on its Stage, and every further proposition is recorded as a node with a stated relation to it: supporting, opposing, required, or refining. Alternatives are the nodes that answer it. AI drafts three candidates at a time under the selected node, and nothing is stored until it is accepted.
The criteria, and how the alternatives differ on them
Establish feasibility first. An alternative whose requirements cannot be met is not a lower-ranked option but an infeasible one. The exclusion and the unmet requirement should both be recorded rather than the alternative removed, since requirements change and an excluded alternative may become available again.
Specify the criteria next, and compare the remaining alternatives on them. Where one alternative is at least as good as another on every criterion and better on at least one, the dominance is worth stating explicitly. Elsewhere the comparison yields a trade-off, which is a matter of judgement rather than computation. 1D2A does not aggregate criteria into a single score unless weights are supplied, because an aggregate conceals the disagreement about weights that the comparison exists to surface.
The criteria, related to the outcomes they are intended to measure.
Values for each criterion across the alternatives, with unavailable values recorded as missing.
The alternatives excluded, and the requirement each failed to meet.
The trade-offs, stated explicitly.
Do the criteria measure the intended outcomes, or what is most readily measurable?
Where a value is unavailable for an alternative, is its absence visible?
Are estimates and measurements distinguished, or treated as equivalent?
Criteria are defined once per log and given a value for each alternative. The comparison matrix presents criteria against alternatives and transposes on request; an unavailable value is shown as missing rather than as zero. Each value offers the same three routes — enter it, request an AI proposal across the alternatives that have none, or ask a named person for it.
Whether the rationale withstands examination
A comparison becomes a rationale only once it has been examined. Before it is adopted as a recommendation it should be tested for claims without supporting evidence, evidence narrower in scope than the claim it is cited for, assumptions carrying more weight than their support allows, and disagreement reconciled informally rather than recorded.
Examination is most usefully directed at the preferred alternative. It is the one least likely to have been contested, and the one whose supporting arguments are most readily mistaken for findings.
This stage is also where further participants contribute. A request is more informative when directed at the specific proposition in doubt and to the party holding the relevant information than when circulated as a general invitation to comment, and the response is recorded as a reason attached to the proposition it concerns.
Supporting and opposing reasons for each remaining alternative, recorded together.
Claims without supporting evidence identified as unsupported.
The assumptions on which the preferred alternative depends, and the conditions under which each holds.
The evidence not currently available, and what it would establish.
Disagreement recorded as opposing reasons, including from parties who subsequently approve.
Does each alternative carry opposing reasons as well as supporting ones?
Has the preferred alternative been examined as rigorously as the others?
Does any argument rest on evidence narrower in scope than the claim it supports?
Is unresolved uncertainty recorded, or absorbed into the recommendation?
Supporting and opposing reasons for a node occupy the same row of its relation block, so a proposition with nothing recorded against it is identifiable as such. AI drafts candidates under the selected node, three at a time, which the author retains or discards. A request to a person remains open until the requested material is added; it cannot be closed by an assertion that it is complete.
Authority, selection, and the commitment
Commitment is an act distinct from analysis, taken by a party holding the authority to take it. That authority is best assigned before the selection: allocating decision rights while a decision is contested introduces a second dispute alongside the first.
The alternative selected and the reasoning for selecting it should both be recorded, and the rejected alternatives retained. A decision from which they have been removed cannot be reconstructed, and reconstruction is the principal reason for recording it at all.
Commitment with checks outstanding is permitted. Deadlines are binding, and a procedure that does not accommodate them is abandoned at the first conflict. What matters is that proceeding is itself recorded as a decision — which check was outstanding, who proceeded, and on what grounds — rather than passing without record.
The party holding the decision, named.
The alternative selected, and the reasoning for the commitment.
The approvals the selected alternative requires, and the response of each approver.
Any outstanding check proceeded past, with the grounds.
Is the deciding party identified, and does it hold the relevant authority?
Have all required approvers been asked, and have they responded?
Is disagreement recorded, or reconciled without record?
Is evidence against the selected alternative represented alongside evidence for it?
A decision’s Stage carries three further rows: the party holding the decision, the approvers and their responses, and the lock. The lock is not a control — it closes once every approver has approved, and only the deciding party can reopen it. The method’s checks appear alongside those rows, marked open or met, so committing with one outstanding is recorded rather than unremarked.
Implementation, and the ex post record
A commitment takes effect through assigned work. The selected alternative is decomposed into tasks, each with an owner, a date, and its dependencies recorded. Retaining that work alongside the reasoning is what makes a subsequent failure diagnosable: a disappointing outcome admits three distinct explanations — the situation was misdescribed, the alternative was misselected, or the alternative was sound and implementation failed — and they can be distinguished only where implementation is recorded next to the rationale.
The outcome is then recorded and compared with what was expected. The expectation is already on record, since it is the criterion values on which the alternative was selected. Where outcome and expectation diverge, the divergence should be attributed: to an assumption that did not hold, an incomplete explanation, an alternative that underperformed, or work not carried out.
This is what makes a body of records worth consulting. The tenth decision of a given type should be better informed than the first, which requires that the outcomes of the earlier ones were recorded.
Tasks with owners and dates, and the dependencies between them.
A reference from each task to the decision it implements.
The outcome observed, expressed in the same terms as the expectation.
The divergence, and its attribution.
Assumptions the outcome contradicted, marked as contradicted.
Tasks are requests with an owner and a date, and an open request closes when the requested material is added. Every change to a log is retained: the history block on a node shows its own record, and the Trail view presents the log’s record in full, so the state of the decision at the time it was taken remains recoverable. Structured comparison of expected against observed values is not yet implemented — an outcome is currently recorded as an ordinary node linked to the assumption it bears on — and this page does not represent it as available.
This page avoids the vocabulary the software is built from. For work on the data itself — an export, an import, or the API — the correspondence is as follows.