prioritization

without one ranked list, prioritization happens by default in chat threads, request volume, and whoever speaks loudest — none of which are evidence. the invariant: every candidate earns its rank from an evidence link, and the ordering is visible, owned, and re-ranked as evidence changes. the alternative compounds quietly: an unranked "later" column grows forever, work ships on request count instead of impact, and real pain with weaker signal goes unaddressed.

all candidate work lives on one ordered list. the rank answers one question: does this move a product question we have evidence for? rank by evidence-weighted impact over cost. the list has one owner, and it is re-ranked weekly from feedback loops, probe answers (continuous discovery), and analytics.

an item with no evidence link is a hypothesis — label it as one and give it a cheap probe (continuous discovery) instead of implementation rank.

• one list; no shadow lists in chat threads or heads.
• every item names its evidence link or its hypothesis label.
• shipped items record outcomes next to their question — the same evidence trail iteration judges by.
• stale items decay: anything unranked for a month is deleted or re-earned.

• rank by volume of requests alone — frequency and cost of the pain are different facts.
• keep a "later" column that never gets re-ranked; later means deleted until re-earned.
• let the loudest channel outrank the strongest evidence.
• split one list per stakeholder; one list, one owner.

read the list today: every item carries evidence or a hypothesis label; the top five can each state their question and expected movement. the last re-rank date is within cadence. one shipped item has an unbroken chain: evidence → rank → change → outcome.

bad

later:
  - redesign user dashboard
  - add email notifications
  - integrations: Slack, Teams, Zapier
  - export workflow results
  - dark mode

an unordered "later" column has no evidence, owner, or ranked reason to become work. "notifications" gets built because three people asked; it ships to 2% of users. "export" — requested in 30 support emails — stays in "later." shipped work moved no metrics.

good

1. reduce failed first workflow runs (move: 15% → 25% completion)
   evidence: 4/5 onboarding interviews, all stopped at connection setup
   blocker: none
   proof after ship: completion rate

2. fix stalled workflows (move: reduce timeout incidents)
   evidence: support ticket theme, 18 emails in last month
   blocker: depends on #1 rollout
   proof after ship: incident count

3. add email notifications (move: unknown)
   evidence: hypothesis, 3 feature requests
   next: one-week probe with power users
   proof after: engagement on notified workflows

each item names evidence, declares expected movement, shows dependencies, schedules proof. a hypothesis gets a cheap probe instead of a six-week build. the owner re-ranks weekly as new evidence arrives.

feedback loops · continuous discovery · iteration · analytics