backlog — Research backlog review

Close the loop that /persona-talk opens: questions land in Research backlog.md, this skill keeps that list useful instead of letting it silt up.

Trigger

/backlog              ← review: dedupe, group, flag stale, propose next steps
/backlog close <ref>  ← move a question to "Closed / turned into research"

The table

| Date | Persona | Question | Source / level | Status | Kind | Priority |

Kind is optional and holds qualitative, quantitative or mixed — the shape of the answer that would settle the question, not its topic. It decides the method, and the method decides where the answer lands: a session we run becomes a Signal, an aggregate (analytics, survey, published report) becomes Evidence. The app's Research backlog page writes this column and recommends 2–3 methods per row; an empty cell means nobody has decided yet — suggest a value, never backfill silently.

Priority — proposed by you, owned by the researcher

Holds critical, major or minor — how much rests on the answer, not how interesting the question is:

• critical — blocks a decision someone is making now (pricing, build/don't-build, a standing contradiction in the data, a blind spot nobody raised).
• major — shapes the work without blocking it: why and how people act.
• minor — good to know. Warm-up context, a plain fact, anything you would only ask because someone is already in the room.

Set it whenever you add a row. A question filed with no priority is a question nobody will schedule. Write the bare word (critical) — that marks it as machine-set, and a later run of this skill may revise it with a reason.

critical (locked) is a human decision. Never overwrite it, never clear it, never re-word it — not even when your own read disagrees. The suffix is written by the app the moment the researcher picks a priority by hand, and it means the same thing as affinity_lock: on a Signal or ==highlights== in a transcript: this field belongs to a person now. You may say you would rate it differently and why — once, in the review output, never in the file.

An empty Priority cell is not "minor": it means nobody has decided. The app shows a dashed guess read from the wording and writes nothing.

Status — three states, two tables

• open, in the Open questions table: to run.
• in study, also in Open questions: a round is booked and the answer is not in yet. The app's Board writes it when someone drags a card into "In a study". The row stays open, and the question still needs an answer.
• A row in Closed / turned into research is answered. Its status is closed — <what answered it>.

Treat in study rows as scheduled. Don't propose them again as the next round, and don't close them until the research has actually happened.

Review workflow

1. Read Research backlog.md (Open questions table). 1b. Check the split. A backlog that is nearly all quantitative usually means we are trying to measure something no one has understood yet — say so. Nearly all qualitative after several rounds means findings are never being sized.
2. Dedupe: near-duplicate questions (same underlying unknown, different wording) → propose merging into the sharper phrasing, keep the earliest date. Show merges before applying — the table is the researcher's document.
3. Group by persona and by theme — a cluster of 3+ questions around one theme is a research-round candidate; say so explicitly.
4. Flag stale questions — open for months with no linked activity: ask whether they're still worth answering or should be closed as "no longer relevant". 4b. Re-read the priorities. Rows you (or an earlier run) set are yours to revise — say what changed and why. Rows carrying (locked) are the researcher's: report a disagreement in one line, leave the cell alone. Rows with an empty cell get a proposed value with the reason drawn from the question's own wording.
5. Propose next steps, most valuable first — critical before major before minor, always:
   • a cluster ready for interviews → offer /interview-guide (it builds the discussion guide from these rows)
   • a question answerable by desk research → offer /researcher (result = Evidence, and say plainly it doesn't close the gap the way a real interview would)
   • a question answerable from product analytics → offer /analytics-sync
   • route by Kind where it is set: qualitative → /interview-guide; quantitative → /analytics-sync or /researcher, or a sized survey; mixed → qualitative round first, then size it. Never propose only one method — name the alternative and what each one cannot tell you.
6. Apply approved edits to the table only — never touch Signals/Evidence from here.

Closing workflow

When research actually happened (new transcript processed via /extract-findings, or Evidence created):

• move the row to Closed / turned into research with a pointer to what answered it (Signal/Evidence title).
• if the answer contradicts what a persona currently claims, flag it — that's /contradictions territory, not a silent persona edit.

Rules

• The backlog is append-honest: never delete a question silently; closing always states why.
• Source/level column keeps its origin (warm-up, known unknown (L1), blind spot — AI-inferred, not raised).
• A Priority you wrote is a proposal; a Priority ending in (locked) is a decision that has already been made. Proposals may be revised with a stated reason, decisions may only be questioned out loud.
• A Kind value is the researcher's call. You may propose one (with the wording that suggests it) and write it once approved — a guess left in the file is indistinguishable from a decision.
• This skill never creates Signals or Evidence — it routes questions toward the tools that do it legitimately.