The day

flowchart LR
    goals([goals: the direction]) --> pass
    pass --> view{{"plan view: you approve"}}
    pass["plan pass<br/>(any number per day)"]
    view --> work["builders, gates, shepherd"]
    work --> evidence{"evidence<br/>complete?"}
    evidence -- gaps --> work
    evidence -- yes --> merge["merge"]
    merge --> checkpoint["checkpoint<br/>report + retro"]
    checkpoint -- "standards, vaccines, proposals" --> pass

Goals and the plan view

Your goals live in .wuwei/goals.md: outcome, measure, target, date and priority. They set the direction the lead ranks against.

Each plan pass ends with the plan as a picture: wuwei plan view writes .wuwei/plan.html (light and dark, readable on a phone) with the ranking formula, the estimated build hours (one WSJF size point is one hour), and a card per goal with its items by score. Then one question: approve or change. New items wait on the board as awaiting plan approval until you approve; work you already approved keeps running.

Why: you own the direction and the trade-offs, and you should see them at a glance, not in a two-kilobyte paragraph. One picture, one answer.

Plan passes

A plan pass (/wuwei-run) adds, drops or re-ranks items in .wuwei/plan.json. Run it as often as the day needs: in the morning, when the queue runs dry, when you add something, when priorities shift. The plan is a running queue that carries across days, not a daily document.

The lead does the thinking in a pass. It reads every source it can reach (issues, red checks, review threads, follow-ups, regressions), returns only ready candidates (a promise, acceptance scenarios, the repository and files), and red-teams the plan: it names the riskiest item and the cheapest way to learn early whether it works, and ranks that learning first.

Why: days change. A plan you can only make once goes stale by lunch, and a team that cannot re-plan either stalls or improvises.

Ranking

The configured framework ranks the queue, never a hand order.

• WSJF (default): value to the goal, time criticality, risk reduction, each on 1, 2, 3, 5, 8, 13, 20, divided by job size.
• RICE: reach times impact times confidence, divided by effort.

Every score carries one line of evidence. Why: a ranking you can read the reasons for is a ranking you can correct.

Checkpoints

A close (/wuwei-close) writes the report for the day so far and launches the steward for the retro. It locks nothing: the next plan pass reopens the day. A second checkpoint starts with what changed since the first. There is no end time.

Why: you stop when you stop. A checkpoint gives you the picture whenever you want it without pretending the day has a fixed shape.

Tiers

Each item has a tier that sets how much evidence it needs:

Tier	For	Gates
light	a trivial, local change	quality
standard	the default	arch, quality, security
full	contracts, schemas, boundaries, auth, secrets, agent tooling, infrastructure	arch, quality, security, goal

When unsure, the lead goes up. A repository over its error budget raises every tier one step, and a light item whose code touches a trust path is raised to standard.