Radar

Radar keeps one or more agents on duty for a single coin and fires real trades from each agent's bound strategy when the market qualifies.

Create an agent first. Deploy is separate: Radar is continuous coverage for that coin; Arena is timed prediction games; the trade terminal is you taking the wheel.

← Agents · Strategies

What a deployment is

A Radar deployment policy is scoped to (you, coin):

Field	Role
Coin	The anchor. Regime, qualification, and fires are always about this coin.
Deployment timeframe	Where the platform reads the coin's confirmed regime for slot selection
Enabled	Pause without deleting slots (false skips the policy at sweep)
Slots	Ordered agent rules (plus optional default catch-all) — at most 10 total
Revision	Optimistic concurrency token — echo it on the next replace, pause, or delete

Deploying does not spend another agent slot. The same agent can sit on several coins (subject to the platform coin cap) and still appear in Arena or on the terminal.

Standing Radar authority is what fires trades on confirmed-regime / qualification edges. Preview resolution is a dry-run against the same resolver; it writes nothing.

Slots

Each slot binds an agent and an optional min conviction trade bar.

Rule vs default

Kind	Priority	Conditions	Role
Rule	Unique positive integer; lowest wins	At least one; at most one hours and one regime	Match only when every attached condition is true
Default	null	Empty	Catch-all when no rule matches — optional, at most one

Resolution picks exactly one on-duty agent: first matching rule by priority, else the default, else idle (no agent).

Agents that are not eligible (for example not active) are filtered before selection. If every candidate is filtered and nothing remains, the policy is idle with reason all slots filtered.

Min conviction

Per-slot min conviction is Radar's participation dial for that agent on this coin (0–1 fraction, shown as a percent on the dial).

• A number sets a floor: capital only deploys when the live trade reading clears that bar.
• null means no Radar-side conviction floor on the slot — the agent's own rails still apply.
• The resolved bar (and which layer decided it) is server-computed. You read it; you do not re-derive it.

Conditions

A rule slot may carry hours, regime, or both. With both, both must match (AND). Radar owns this vocabulary; Arena session conditions are not accepted here.

Hours (UTC membership)

Wall-clock gate as two sets, not a from/to range:

• Hours — UTC hours 0–23 (non-empty, unique)
• Days — UTC weekdays 0 (Sunday) – 6 (non-empty, unique)

Membership is the whole predicate. A gapped schedule is natural (hours: [7, 13]). Overnight is a list across midnight ([23, 0, 1]), not a special wrap form.

Regime

Match against the coin's confirmed regime at the policy's deployment timeframe:

Regime	Typical shape
bull_expansion / bear_expansion	Directional expansion
bull_ranging / bear_ranging	Directional range
contraction	Compressed / quiet
volatile	High-volatility regime
fallback	Classifier fallback

Optionally set a regime min conviction floor on the condition (uncertain → low → medium → high → maximum). Reachable conviction tiers differ by regime (for example contraction only reaches medium; fallback reaches uncertain/low) — the platform gates options accordingly.

If no confirmed regime is cached for that timeframe, every regime condition is false and the live read reports cold regime.

Deployment timeframe

This is not the strategy's signal timeframe. It is the TF used to read confirmed regime for who is on duty.

Allowed pins: 1m, 3m, 5m, 15m, 30m, 1h, 2h, 4h, 8h, 12h, 1d, 3d, 1w.

Relative strategy timeframes (_ltf / _htf) still live on the playbook. Radar selection freezes the regime anchor to the coin at this deployment TF.

Live resolution (resolves now)

Every read of a policy carries a server-computed resolves-now projection: what the sweep would do right now. The client renders these fields; it does not recompute selection.

Card sections

Dashboard grouping priority — an open position outranks idle sub-states (live money wins, even under a paused policy):

Section	Meaning
In position	A live trading position is open on the coin
Blocked	An agent is on duty but its account admission is blocked — nothing is being scanned
Scanning	An agent is on duty, the coin is flat, watching for a qualified edge
Idle	No agent trading; the sub-state is in reason

Rotating means a position is open and its owner is not the current on-duty agent — the next selection has already moved while the incumbent still holds the coin. Manual / pre-fill positions have no owning agent and are never rotating.

Idle and blocked reasons

Reason	Meaning
Cold regime	No confirmed regime at the deployment TF — every regime condition is false
No rule match	Regime/time known; no rule matched and there is no default
All slots filtered	Every candidate agent was ineligible before resolution
Policy disabled	enabled is false — slots retained, sweep skips the policy
On duty but position blocked	On-duty and qualified, but a (user, coin) open-position mutex blocks a new fire
On duty but edge spent	On-duty and flat, live reading would qualify, but the edge baseline is already advanced — waiting to re-arm
Account blocked	On-duty agent's account admission episode suppresses triggers on every coin it covers
Platform paused	Platform Radar kill-switch is off — maintenance; not something you fix on the policy

Qualification and fire markers

When an agent is on duty and the coin is flat:

• Qualified — last known flip-baseline verdict for the on-duty agent (null if never evaluated or idle)
• Qualification block — which gate is failing when not qualified (aggregate, required-count, ATR floor, required condition false, verdict opposes direction, candidate levels unavailable, …). Copy is server-projected.
• Cooldown until — fire-cooldown expiry when cooling down
• Last flip / last fire — timestamps for the not-qualified→qualified edge and the last enqueued Radar fire

Held condition verdict and verdict dwell (how many consecutive resolutions the side has been held) surface debounce when the strategy carries a verdict — display-only.

Close decision (when the strategy decides on close)

If the on-duty agent's trigger decides on strategy bar close, the live per-minute qualification rail does not decide entry. Resolves-now then carries a close decision object for that pair:

Outcome	Meaning
Fired	Close qualified over an unconsumed baseline; fire enqueued
Not qualified	Close did not fire — either a failing gate, or a consumed edge (qualified but baseline already advanced)
Missed	Decision window passed before a sweep reached the bar — no measurement

State relative to the next decision is waiting (last close decided) or deferred (a closed bar still owes a sweep).

Activity (why it has not fired)

The activity summary answers "why has this agent not been firing?" in one read:

1. Cause groups — recurring (event, cause) rollups inside a reported window (gate crossed/blocking, idle reason, fire disposition, evaluation outcome)
2. Curve digest — recent evaluation ring for the on-duty agent (how far scores sit from the minimum) — the ring window is separate from the cause window
3. Recent events — last key transitions (not bounded by the cause window)

The journal is a transition log: a gate that blocks continuously without crossing again may have no group in the window. What is true right now is always resolves now on the deployment.

Pause vs withdraw

Action	Effect
Pause (enabled: false)	Sweep skips the policy; slots and revision stay
Resume (enabled: true)	Policy returns to discovery
Delete / withdraw	Removes the whole coin policy — standing autonomous authority for that coin is gone

Replacing a policy is a full slot-tree rewrite. Always pass the revision you last read (null only on first deploy); a stale revision is a conflict and writes nothing.

Cap and fleet

Fleet roll-up reports coins deployed (paused included — that is the cap-meter denominator), section counts, how many agents are on duty, and your live per-user coin cap (platform-configured). Stay under the cap when adding coins.

How this fits the stack

Layer	Owns
Strategy	Metrics, conditions, orders, signal rules, entry, levels, exit
Agent	Brain, bound strategy, risk rails, one account slot
Radar	Who is on duty for this coin, when, at which regime TF, and at what conviction floor

Commander and MCP use the same concepts: list / get / preview / upsert / delete Radar deployments and read activity.

← Agents · Next: Arena