# Views (/docs/views) A **View** decides *what* an agent sees when the runtime wakes it; the agent decides *what to do* with it. A Kestrel pod is not one agent but a small organization of **seats** — roles distinguished by the deliberation budget they can afford — and each seat reads a different screen because each seat does a different job at a different speed. This section is the library of those screens. It is organized on two axes, the same two the engine's own template-cell key turns on: * **By seat / role** — one page per View, each explaining its layout in terms of the seat's task and timescale: the [Strategist View](/docs/views/strategist), the [Watcher View](/docs/views/watcher), the [Scanner View](/docs/views/scanner), and the [PM View](/docs/views/pm). * **By market variant** — instrument class, timescale, and strategy archetype. The Views you see today are the equity-index options variant; the same View grammar carries the other instrument classes, described under [The variant axis](#the-variant-axis) below. ## What a View is [#what-a-view-is] A seat's screen is not a monolith. It is assembled through four layers, each answering a different question, each authored on its own: 1. **Seat** — *who is this?* A role in the pod org, fixed by its affordable deliberation budget: a PM and a Strategist are slow, deliberate, frontier-class judgment (dollars, a few times a day); a Watcher is the fast reflex (pennies, seconds); a Trigger is a non-LLM reflex below even the Watcher. Seats' latency envelopes differ by **orders of magnitude**, which is why no single model fills them all well. 2. **Tasks** — *what must it decide?* The seat's job: allocate an envelope, author a leaf, manage held inventory, summon a bigger brain. The tasks fix what information is decision-relevant. 3. **View config** — *what does it see?* The role-keyed View — the panes, in order, at the band the wake zooms to. This is the layer this section is about. Perception is tuned here. 4. **Persona / Brief** — *how does it lean?* The soft, directional English the seat reasons *from* (a "disciplined risk manager," a "momentum chaser"). Personas tune **policy**; Views tune **perception** — different channels, both role-attachable, and the Brief never enters admission (it directs, it never authorizes). The seat and its persona are policy; the View is perception. This section fixes layer 3 and leaves layers 1, 2, and 4 to their own documents. ## The pod [#the-pod] Reading the stack: **Risk is the L0 layer**, not a seat — strictly subtractive, above everyone. The **PM** runs a Pod (allocates children's envelopes, arms and de-arms, assigns Coverage, watches aggregates) and splits into an **Allocator** half and a **Scanner** half. The **Trader** runs a Book (a leaf, the only place positions live) and splits into a **Strategist** and a **Watcher**. The **Trigger** is the pod's third live tier below the Watcher — a non-LLM reflex that *summons*, never decides, and so reads the frozen Frame's features directly rather than a screen. The **Historian** works off-tape after the close, reading Blotters and Grades rather than a live Frame; it curates the pod's memory and evolves the Brief. Trigger and Historian have no View by construction, for opposite reasons — one reads below language, the other reads after the session. The pod-seat roster is closed — `pm`, `strategist`, `watcher` — and it is the `role` axis of the engine's evidence cell key. A seat is admitted only where its sub-job genuinely diverges: by latency of at least an order of magnitude, by a typed boundary a document crosses, and by being independently measurable. That filter is why Risk stays a mechanism, the Grader stays an engine, and the Trigger stays a tier — none of them a seat. ## How a View is chosen [#how-a-view-is-chosen] Every wake resolves to exactly one View through a strict precedence: ``` explicit authored View > seat founder View > phase default ``` * An **explicit `view`** — one an agent authored, scheduled, or a Wake forced — always wins. A seat never overrides an authored lens. * Otherwise, when the driver knows the acting **seat**, that seat's founder View for the frame kind is used (the Strategist's at the open, the Watcher's and the Scanner's on a wake). * Otherwise the frame kind's **phase default** panes render — byte-identical to a run with no seat at all. Seat Views are **opt-in**. By default a Session runs the phase-default panes and consults no seat; set `seatViews: "founder"` on the `AgentConfig` to have each acting seat read its View. The two settings are distinct experimental subjects and fold into the run's `ConfigId` automatically, so a `founder` run and a `phase-default` run are always separable in the record. Where a seat has no View for a frame kind — a Strategist at a wake, a Watcher at the open, the PM's allocation half — resolution falls through to the phase default; absence is a first-class value, never a silently fabricated screen. Every rendered screen in this section is **byte-real** — generated from the live renderer, never hand-typed, so a renderer change re-pins the screens or CI goes red. Every View block is written in real Kestrel syntax and round-trips byte-stable against its golden corpus. ## The seat axis [#the-seat-axis] | View | Seat | Task | Timescale | Frame | | ----------------------------------------- | -------------------- | ------------------------------------------------------------------------- | ------------------------------------ | ----- | | [Strategist View](/docs/views/strategist) | Strategist | Frame the session, author the plans, price the exits | A few times a day, at dollars a call | OPEN | | [Watcher View](/docs/views/watcher) | Watcher | Manage held inventory inside its plans; escalate at the edge of certainty | Seconds, at pennies a call | WAKE | | [Scanner View](/docs/views/scanner) | PM (discovery half) | Read one name deeply when a universe-wide scan fires | Minutes, single-name | WAKE | | [PM View](/docs/views/pm) | PM (allocation half) | Allocate children's envelopes; watch aggregate exposure | Slow, aggregate | — | The Strategist and the Watcher are the two halves of the Trader that runs a Book; the Scanner and the PM's allocation cockpit are the two halves of the PM that runs a Pod. Each half is a distinct layout because each does a distinct job at a distinct speed — which is exactly the "different layout by task and timescale" this section is organized around. ## The variant axis [#the-variant-axis] A View is composed once and rendered against whatever market the instrument declares. The market a screen speaks is set by the instrument's **SessionScheme** — the registry attribute that declares an instrument's boundary stream (its opens, closes, and session splits). Four schemes ship today: | Variant | SessionScheme | Status | | -------------------------------- | ------------------- | --------------------------------------- | | US equity & index options (0DTE) | `equity-rth` | **Shipped** — the Views in this section | | CME futures | `cme-rth-overnight` | On the same View grammar | | Crypto perpetuals | `perp-utc-day` | On the same View grammar | | Spot FX | `fx-week` | On the same View grammar | The founder Views documented here are the **equity-index options** variant, and their panes are options-native: the chain pane with its intrinsic-floored fair, the pin, the expiry-relative session ordinals, the theta the Watcher pays to hold. They render against `equity-rth`, the default scheme every equity and index instrument resolves to. The other three schemes ride the **same View grammar and the same seats**. A perpetual's rolling UTC day has no daily close, so its `prior-context` pane names that absence instead of inventing a midnight bar; a CME instrument splits its RTH session from the overnight; spot FX runs a Wellington-open-to-NY-Friday-close week. As each variant's panes land, its layouts join this section as those Views — futures, perpetuals, and FX pages slot onto the variant axis without disturbing the seat axis, because a variant changes the market a View speaks, not the seat that reads it.