# Operating review (/docs/app/operating-review)

> Study: a study of the work. Defines nothing. No package exists.

> App study: where a Business steers itself by its Goals and numbers, on a cadence that ends in a Decision.



An Operating review is the App where a Business steers itself by its numbers. Its readers answer for outcomes: the owner, operator cofounders, the accountable Role behind each Metric, the Role agents that read their measures on every wake, and the parent's Watchers.

It holds the Goal tree with pace, exceptions, watches that act, Experiments, product analytics and forecasts. It shows agent and work performance only at the grain the measurement policy allows. Its review cadence ends in a Decision, and its scope is one Business, a Portfolio or a Studio.

Earlier studies folded executive and KPI dashboards into Admin as Views. This type undoes that. The study describes the work; how an App is defined is open, see [What every App needs](https://next.mdxui.dev/docs/app/every-app).

## Why it is a type

| Question | Operating review | [Admin](https://next.mdxui.dev/docs/app/admin) |
| --- | --- | --- |
| Sides | Owners, accountable Roles, Role agents as readers and as measured subjects, Watchers | Operators over records whose subjects never sign in |
| Unit of work | The Goal tree with its readings, and the review that ends in a Decision | A record and a named Action with a reason |
| Ball | The accountable Role of the Key Result furthest off pace; the reviewer at the close | The operator |
| Money | Moves none; reads money under a declared Policy | Commands money on a record |
| Authority | Who may set or move a target, override a status, see which Metric, conclude an Experiment, arm a watch | Per Action |

A Metric is always derived and never typed in. An Admin form over Metrics lets an operator type a value, and Event, Metric, Funnel and Goal as CRUD tables is what flattening analytics into Admin produces.

The type is expressible two ways: as the analytics part of the operator's App, beside Admin and the [Worklist](https://next.mdxui.dev/docs/app/worklist), or as an App of its own. A Portfolio review spans Startups no single shell holds. A law practice reading its morning view needs the type without any operator shell.

## Actors

| Actor | Human or agent | Work here |
| --- | --- | --- |
| Owner or founder | Human, or an agent founder | Reads the Situation, sets and moves targets, names objectives, closes reviews with Decisions, arms watches |
| Teammates with Plan access | Human | Edit Goal names, targets, periods and the Hypothesis link; never values |
| Accountable Role | Either | Answers for one Metric; its seat is reviewed weekly |
| Role agents | Agent | Read their Role's View as a context block on every wake; are measured by their Role's results |
| Office Hours agent | Agent | Holds the weekly review with the founder and writes the Decision |
| Analyst | Usually an agent | Defines and watches Metrics; answers questions with the query shown |
| Watcher | Either, in the parent namespace | Pauses, redirects or cancels pending Actions, each a Decision |
| Studio operator | Human with agents | Reviews the Portfolio by exception; trips and lifts brakes |
| Owner-private reader | Human | Values the Business, runs scenarios, solves for "what must be true" |

Every reader is also a subject: the founder's Key Results are read by the Office Hours agent, and a Role agent reads the Metric it is judged by. That is why the measurement policy is part of the type, not a filter added later.

## Interfaces

| Reader | Screens | Renderings |
| --- | --- | --- |
| Founder, every reader | The Situation: Needs you from the Worklist, Goals at pace, Exceptions, Asks, with its address, window and `asOf`; money in and what the Business keeps; the drill path from a number to its formula, Events, Actions and Verbs | Chat briefing and "why is this number X?"; a daily email; device glances that notify only on blocked, missed, stale or broken |
| Founder, Plan teammates | The Goal tree with derived progress, pace, machine status beside a human override with reason and owner, confidence as a named judgement | |
| Analyst, accountable Role | Metric page: unit, polarity, additivity, grain, window, freshness, empty-denominator rule, counting guards, Policy, owner, Version history, the series with typed gaps; watch rules | MCP and CLI over the same readings |
| Weekly review | Weekly Test and Office Hours; Role reviews per filled seat; the review run with its frozen reading and one disposition per signal; the closing Decision | Slack thread for the pulse; one Decision request per missed result |
| Product and Growth Roles | Experiments with the Hypothesis, assignment, exposures, stopping rule and readout; the workbench: funnels as conversion Metrics, activation per Flow, retention, Visits and Recordings | |
| Founder, Finance Role | Forecast and runway, the spend state it sets, projections with method and interval, scenarios labelled as scenarios | |
| Watcher, Studio | Pending feed of Actions in a delay window; Portfolio review with sample sizes, Cohort against control, brakes, Traction | |
| Role agents, native clients | none | A context block of about 150 tokens, a briefing of 600 and a review packet of 2,000, all the same reading; JSON for native clients |

A Dashboard is the View this App renders at a scope's address, not the App. Editing a Metric's definition or a Goal's target is a Change in Admin, and this App links to it and shows its effect on readings.

## Records and actions

| Record | States |
| --- | --- |
| Metric definition | declared, measured or unmeasured, versioned on change, retired; a rename maps every reader |
| Reading | present, absent with a reason, withheld by class, unconfirmed; provisional, partial, early, unset, failed, not run; none renders as zero |
| Key Result | value read from its Metric; machine status beside a human override with a reason |
| Review run | scheduled, gathering, dispositioned, objective named, closed with a Decision; an off-cycle run starts when a strategic watch trips |
| Watch rule | off or on; armed, fired (writes an Action), re-armed when false; per environment |
| Hypothesis, Experiment | stated, testing, supported or refuted; designed, running, stopped, concluded ship, kill or iterate |
| Role Review | one per filled Role per week, immutable, with the seat's rung after it |
| Forecast | computed weekly; sets the spend state free, tight or hold; scored on its read date |
| Brake | tripped with a reason and effects, lifted or expired, resolved by a Decision |

Verbs: define, query and watch a Metric; ask a question and get the query beside the answer; set or move a target; override a status; test a Hypothesis; conclude an Experiment; review a Role. A review closes with a Decision, and a Decision is superseded, never edited. A Watcher pauses, redirects or cancels, and the Studio brakes and lifts.

### Terms

| Term | In startupbuilder.dev | Definition, quoted from [startupbuilder.dev](https://startupbuilder.dev) or the gap |
| --- | --- | --- |
| Metric | [Metric](https://startupbuilder.dev/business/nouns/metric) | "A Metric is a named measure computed from Events or Entries, always derived and never typed in." |
| Goal | [Goal](https://startupbuilder.dev/business/nouns/goal) | "A Goal is what a Thing is trying to achieve: an objective with one or more Key Results." |
| Key Result | [Key Result](https://startupbuilder.dev/business/nouns/key-result) | "A Key Result is one measurable target inside a Goal: a Metric, a baseline, a target, and a deadline." No Verb sets or moves its target. |
| Dashboard | [Dashboard](https://startupbuilder.dev/business/nouns/dashboard) | "A Dashboard is one unified view of a scope, rendered as Markdown for Agents, screens for Humans, and JSON for native clients." What this App renders, not its name. |
| Decision | [Decision](https://startupbuilder.dev/business/nouns/decision) | "A Decision is the immutable record of a choice: its options, the choice, the reason, who decided, and the stakes." |
| Role Review | [Role Review](https://startupbuilder.dev/business/nouns/role-review) | "A Role Review is a dated check of one Role against its contract: its results, its standards and its autonomy rung." Over a seat one human fills, it is per-person attainment. |
| Hypothesis | [Hypothesis](https://startupbuilder.dev/business/nouns/hypothesis) | "A Hypothesis is a belief about a Startup stated so it can be proved wrong, with the Metric and threshold that settle it." Read for any Business. |
| Forecast | [Forecast](https://startupbuilder.dev/business/nouns/forecast) | "A Forecast is a dated week-by-week projection of a Startup's cash, revenue and Cost, computed from its records, that sets its spend state." |
| Policy | [Policy](https://startupbuilder.dev/business/nouns/policy) | "A Policy is a ruling a Startup declares on how its numbers are computed or its choices made, such as revenue recognition." |
| KPI | lacks a Noun | A relation: a Metric a Thing has chosen to watch. business-as-code makes it a record with typed-in values. |
| Watch rule | collides | startupbuilder.dev's [Watch](https://startupbuilder.dev/business/nouns/watch) is an outbound Event subscription. |
| Reading, pace, polarity, grain, review run, disposition, scenario | lack | |
| Review, Scorecard | collide, retired | Review has five senses; Scorecard is retired in favour of Role Review. |

## Workflows

1. **The first reading.** The Founding Hypothesis becomes the root, and its Minimum Success Criteria its one Goal, with one Metric that reads honestly: unmeasured, source not live, or a valid zero.
2. **The weekly loop.** The Weekly Test ranks the Hypothesis's open questions and compiles the week's one Experiment. Role reviews run per seat, and Office Hours on Friday writes the Decision. The quarter ends with a review that freezes a reading.
3. **The monthly review.** It waits for the books to close, or runs on a flagged estimate. A late refund shows as a restatement, never a silent change.
4. **A watch fires.** On the edge, once, per environment: it opens a Worklist item, switches an agent to assist or pauses a campaign. A watch demotes freely and promotes only through the eval gate.
5. **Runway sets the spend state.** A Forecast after a large money Decision moves spend to tight or hold, and other Systems read it.
6. **The Studio reviews its Portfolio.** Per-Startup readings with sample sizes, Cohort against control, and brakes that skip the delay window.

## Boundaries

- Plan and Strategy are private to the owner and Plan teammates, never on the Site, a portal or the API.
- Customers appear in aggregate. A named account appears only in an owner-private rendering.
- Agents are measurable in full, except a rate on a key a named human deputized.
- Employees under the employer regime are read by object, step, Site and window, never per person. "Average hours per person" fails that rule.
- A Role Review of a human-filled seat published to every Team is per-person attainment by another route; the conflict is recorded, not resolved.
- A roll-up recomputes ratios, never averages them, and an aggregate over rows the reader may not see is withheld.
- Benchmarks across Startups need a floor the owner has not ruled.
- Every reading names its reporting entity and its Policy.

## Agents

Agents assemble every reading, rank exceptions, draft the review packet, run Experiments, Forecasts and Role reviews, and answer questions. A model never writes a figure: it routes the question to a declared Metric and cites it. An agent never varies the Metrics or Evaluators it is judged by.

A person names objectives and moves targets where a human founder exists, and approves spend beyond the envelope. A change to a figure an outside holder has seen goes to the [disclosure room](https://next.mdxui.dev/docs/app/disclosure-room) as a correction.

## Hard parts

1. **Honest readings.** Absent, withheld, unset, failed and zero are different; a source error never renders as 0, and a stale value never shows healthy.
2. **Target authority.** The founder sets targets, but no record says who else may, how an agent founder does, or that an agent never moves its own target.
3. **Two registers.** A goal register allows targets and pace; an evidence register refuses them. The refusal belongs to the View's audience, not to a field.
4. **One Task store, four measured subjects.** Agent track records, performers' Registers, Role reviews and the review read the same Events under different rules.
5. **Counting rules.** Outside revenue against run-rate revenue over everything; two NRR methods. Every reading names its Policy, and two north stars coexist without either replacing the other.
6. **A review that ends in no Decision is ceremony.** The close refuses an undispositioned red signal.
7. **Budgets for agents.** A 150-token block carries Needs you, pace and exceptions without truncating dishonestly.
8. **Physical work arrives as Events.** Gaps from unattested devices render as gaps, and grain stops at Site and window for employer-directed human work.

## Day one

For an idea-stage founder, day one holds:

- the Founding Hypothesis and its one Goal, read from one Metric, with no typed-in values;
- the Situation: Needs you, the Goal with pace, exceptions, money in;
- the context block for every Role agent, before any screen;
- the weekly loop: the Weekly Test, one change a week, Friday's Decision;
- drill to records for every number, and questions in chat;
- Test and Live apart, and a daily briefing that notifies only on blocked, missed or stale.

| Trigger | Turns on |
| --- | --- |
| The first filled Role | Role reviews and agent performance |
| The first Visitor | The product analytics workbench |
| The first reversible change of consequence | Experiments with assignment |
| The first Metric with a target and a source | Watch rules |
| The first money commitment | Forecast and spend state |
| The first month-end close | The monthly review on closed books |
| The first human in a Role, or a frontline Worker | The measurement policy for people |
| A raise, a valuation question or a sale | The owner's private instrument |
| A second Startup under one owner | The Portfolio review, benchmarks once the floor is ruled |
| The first outside holder | Sealed packs handed to the disclosure room |

An owner-operated practice uses the same type with no Studio: matters and deadlines by object, Goals that hold or break, and no review ritual beyond the owner's.

## Related studies

## Prototypes this study needs