Decision Record

Context

The initial queue recorded priority, question, decision, evidence, and status in five rows. It could not represent stable identity, maturity, parent/child relationships, dependencies, evidence links, confidence, remaining uncertainty, enabled engineering, review triggers, or completion consequences. Priorities therefore could not be reconstructed or challenged systematically.

Decision

Each portfolio item receives a stable RQ-COMM-NNN identifier and records:

• parent and child questions;
• maturity from Idea through Monitoring;
• priority tier and scoring rationale;
• supporting and contradicting evidence;
• dependencies and related research;
• artifacts and engineering work enabled;
• confidence and largest remaining uncertainty;
• next action, exit evidence, and review trigger.

context/RESEARCH-QUEUE.md remains the compact current-state view. research/frontier/PORTFOLIO-ROADMAP.md becomes the richer, mutable graph view. Neither is accepted scientific evidence; conclusions remain in EV-, HY-, TH-, and RP- records.

Alternatives

1. Keep the checklist and add prose notes. Rejected because relationships remain implicit and item identity unstable.
2. Create a database and new validator immediately. Rejected as premature engineering before the representation proves useful.
3. Put all portfolio fields into mission records. Rejected because early ideas and monitoring questions are not yet executable missions.

Evidence

The observed five-row queue omitted every graph and maturity field required by the portfolio mandate. Existing ROS lifecycle rules already separate idea, mission, research artifact, and accepted knowledge states.

Consequences

Prioritization becomes inspectable and queue items can move backward when confidence weakens. Manual duplication between the compact queue and roadmap is a drift risk until a schema and generator are justified.

Reversibility

High. The identifiers and fields are Markdown conventions; no accepted research claim or software interface depends on them.

Validation

After two completed research cycles, audit whether item identity, dependencies, and exit evidence improved selection and handoff. If not, simplify or replace the model. Do not automate before that review.