Multi-input intelligence rules

Open Alerts → Intelligence Rule Studio. Give the rule a name, choose inputs, combine conditions with ALL (AND), ANY (OR) and optional NOT groups, test the current evidence, then save and arm. Nested groups allow patterns such as A AND (B OR C).

Execution and privacy

Rules and reports are account-scoped in this browser. They evaluate about every ten seconds while the page is visible, live and account verification is current. They pause during replay, hidden-page intervals and uncertain account access. They do not run when the page is closed and do not send email, Telegram messages, webhooks or push notifications. Cross-device synchronization and unattended server monitoring are not implemented by this feature.

Account changes clear the visible rule/report state. Same-account re-verification can preserve a draft while access is suspended. Definitions are saved separately from existing simple alerts; imports are disabled until reviewed and armed. Export files contain your rules, notes and evidence, so choose their recipients deliberately.

Inputs and freshness

The catalogue reads scalar fields from the app’s existing loaded public-source registry. It does not fetch arbitrary URLs or access unconfigured providers. Private/credential-bearing data branches and private report stores are excluded. Source arrays are bounded to 500 candidate rows, traversal to six levels and 70,000 visited nodes, and the catalogue to 12,000 fields. The interface displays omitted-count information; this is not a claim to expose every record or licensed dataset in Watchtower.

Array fields use stable unique identifiers rather than row offsets. Missing, stale, failed, future-dated or over-age inputs evaluate as unknown. NOT unknown stays unknown. AND with a known false is false; OR with a known true is true, with unavailable branches still shown in the report.

Timestamp basis is explicit: a reported observation date, source snapshot time, or browser receipt time. Maximum age applies to that basis. A recent receipt is not proof that the underlying economic observation is recent. Units remain source-defined; there is no automatic currency or scale conversion.

WHEN, duration and cooldown

First match triggers on the first known match, then rearms only after a known false. Unknown data does not silently rearm it. While matching can repeat only after cooldown. “Observed true for” requires matching checks over the chosen duration; unknown inputs or a missed-check gap reset that duration. This is sampled monitoring, not proof of uninterrupted real-world conditions between checks.

Crosses above/below needs two comparable timestamped raw values. Missing history cannot establish a crossing. Duplicate timestamps do not create extra samples.

Reporting and forecasting

A generated intelligence report records the exact rule, thresholds, values, units, timestamp basis, evidence classification, match states and your optional scenario note. A scenario note is your hypothesis, not a model conclusion or verified event.

Experimental linear projection uses at least three distinct timestamped samples observed during this session, spanning at least one minute. The horizon cannot exceed the observed span. It is ordinary least-squares extrapolation of local observed snapshots, not a calibrated event forecast, historical backtest or probability. It may extend beyond physically plausible ranges. Revisions, sparse updates and source changes limit interpretation. No synthetic history is created on reload.

Future server monitoring

Unattended delivery would require an authenticated owner-isolated rule service, a typed and entitlement-aware field registry, persistent time-series observations, replayable evaluation, versioned rule changes, quotas, notification destinations and delivery auditing. Those are separate from the current browser evaluator. Existing server-side exposure-monitor rules continue to use their own contracts.