> ## Documentation Index
> Fetch the complete documentation index at: https://docs.raisegate.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Operations

> Run timing, alert email rules, classification stability, lead quality, and deletion.

## Tracker runs

Scheduled runs happen every `cadenceDays`. Each run scrapes the company website, company and founder LinkedIn (including company LinkedIn posts), company and founder X, Reddit and GitHub; filters for relevance; enriches newly discovered founders, scoring them against the VC's `thingsToTrack`; analyses the result into a signal; classifies the company; and evaluates alert rules.

A run holds a 30-minute lease; an expired lease is reclaimed automatically.

Pausing is durable: if a run is in progress when you pause, the run finishes and the entity stays paused.

## Timing observed in testing

| Step                          | Typical                  |
| ----------------------------- | ------------------------ |
| Create                        | 2-3 s                    |
| Background profile enrichment | \~10 s                   |
| First run                     | 1-6 minutes              |
| Manual refresh                | 60-90 s                  |
| Draft enrichment              | \~15 s (up to 4 minutes) |
| Warm read requests            | 160-500 ms               |

## Alert emails

Sent to `alertEmails` when a rule matches with source evidence and the content differs materially from the previous alert.

Repost-only activity, evidence-free events, duplicates and cooldowns are logged but not emailed; `decision.code` says which. Common values: `rule_matched`, `rule_not_matched`, `no_evidence`, `repost_only_event`, `duplicate_alert`, `cooldown_active`, `no_send_email_rules`, `alerts_disabled_or_no_recipients`, `email_delivery_failed`.

## Classification

Company and lead sectors come from RaiseGate's taxonomy classifier (retrieval over the taxonomy plus a Gemini re-rank). A label is recomputed only when the company's stored summary changes, so labels are stable between runs.

The stored summary is set when the company is first tracked and is not yet refreshed automatically if the company later changes direction; correct it by updating the entity or through RaiseGate support.

Filtering follows the taxonomy's grouping. For example, the taxonomy files *Data Center & Networking Infrastructure* under *Enterprise & Robotics Hardware*, so a data-centre company can appear under `robotics`.

## Lead quality

Leads come from the tracker's discovery pipeline, not the API. Most are relevant founders; some are adjacent (agencies, investors, media). Use `classification`, `stage` and `sourceQuery` to focus review.

The same person can appear as more than one lead row (for example the same founder listed with two websites); a decision applies to all of them.

## Deletion

Deletion is permanent and removes the entity's history (runs, signals, alert logs). If the entity was created from a lead, the lead stays accepted with `decision.trackedEntityId: null`, and it can be tracked again.