portfolixir

ADR-0043: an ADR that closes a decision gate names the asks it answers — and the ones it does not

• Status: Accepted (owner sign-off 2026-08-18; this is itself a change to how the project decides, so the signature is the owner’s, per ADR-0026)
• Date: 2026-08-17

Context

ADR-0026 step 1 requires a decision gate — an ADR or spec with acceptance criteria — signed off before a batch starts. In practice a gate is opened by an owner asking for something, often several things at once, recorded in a triage document or a product brief. It is closed by an ADR written later.

Nothing checks that the closing ADR answers what the gate was opened for.

This is not hypothetical. It was diagnosed on 2026-08-15 (feedback triage, Round 7) with a worked example, and the example is the reason this ADR exists rather than a note in a retrospective.

What it cost, once

Gate B3.2 was opened on two asks, both in the owner’s words in the 2026-08-12 triage: derived values should be durable, and recomputation should be triggered by the write that invalidated it, or by a schedule — “so that a read is never the thing that pays”.

ADR-0039 decided durability. It is silent on the push half. Nobody removed that ask and no one argued against it; it fell out between the gate and the ADR.

The consequence is what makes this worth a rule. The drop survived:

• the batch — Sprint 6 Lane C shipped C1–C5 against the ADR;
• the agentic review closing act — six roles, every confirmed finding fixed;
• the close-out — ADR-0026 step 5, with gates green and a retrospective.

Every one of those holds the work against the ADR, and the ADR was internally complete. There was no artifact anywhere in the chain that still knew what the gate had been opened for. The owner discovered it by using the product and asking why a “computing” cue still appears — three weeks and one release later, and it cost the more visible half of the feature.

Why the existing rules do not catch it

• The Story Workflow’s steps hold a story against its acceptance criteria, which come from the ADR.
• The agentic review holds the branch against the ADR and the design spec.
• The close-out reconciles issues, the FR map and the tracker.
• ADR-0026 step 1 requires that a gate be signed off; it says nothing about what the signature is for.

Each is correct and none of them looks upstream of the ADR. The gap is structural, not a lapse.

Decision

An ADR that closes a decision gate carries a short list of the asks the gate was opened on, each marked answered or deferred, with a one-line reason for every deferral.

That is the whole rule. Concretely:

1. The list names its source — the triage round, product brief section or issue where the asks were recorded, so a reader can check the list against what was actually asked.
2. Every ask gets a verdict: answered by this ADR (with the section), or deferred (with the reason and, where one exists, the gate or issue it moves to). “Not mentioned” stops being an available outcome.
3. A deferred ask is a written deferral, which means the close-out and the next planning round can see it. Today a dropped ask is invisible by construction.
4. It is checked at the gate signature, not later. The owner signing a gate is the last person who still remembers what they asked for, and the list is what makes the signature about that rather than about the ADR’s internal coherence.

What this is not

• Not acceptance criteria. Those belong to the stories and stay there. This is a list of asks, usually three to six lines.
• Not a template section for every ADR. Only ADRs that close a gate carry it. An ADR recording a design choice nobody gated has no asks to list.
• Not a new gate, workflow step or artifact. It is a required section in a document that already exists, which is why it is cheap enough to actually survive.

Retroactive application: exactly one

ADR-0039 gains the list, since its missing half is the reason this rule exists and the half is already filed (#710, #711). No other ADR is reopened — auditing the back catalogue would cost more than the rule saves, and the rule is preventive by design.

Consequences

• A gate-closing ADR is a few lines longer, and the lines are the ones a reviewer most wants.
• The owner’s signature gets a checkable object. The recurring complaint behind this — that sign-offs are the real load, four in one day on 2026-08-15 — is not made worse: reading a five-line list of one’s own asks is faster than reconstructing them from memory, which is the alternative.
• The close-out gains a place to look for deferred asks, which currently has no source.
• A rule can be satisfied hollowly. Someone can write “answered” against an ask the ADR does not really answer. This does not prevent that; it makes it a visible false statement rather than an absence. That is the whole of the improvement, and it is enough — the ADR-0039 drop was invisible, not disputed.

What this does not decide

• Whether gates should be smaller. B3.2 carried two loosely related asks; one could argue for one gate per ask. That is a different decision and this rule works either way.
• Anything about scope, the scope ladder or the permanent non-goals. This is a rule about how a decision is recorded, not about what may be decided.

This site is open source. Improve this page.