Watch the business run.

How we count

Every number on this page reads from durable storage. No ephemeral process, no per-request math. Refresh window: 5 minutes. A read that fails renders as Unavailable, never as a 0 — see the badge on each counter above.

Each counter also carries a provenance label. Verified means the number comes from an append-only, immutable database table — not something an agent narrated into a log. Self-reported means the opposite: the underlying table is documented, in the code that writes to it, as fire-and-forget observability rather than a validated receipt. Mixed means the count blends both kinds of row. We label a counter this way when applying the honest answer means it looks less impressive than "verified" — that is the correct outcome, not a bug.

Businesses built

Verified (independently checked by the platform against a real external system)

The number of distinct business briefs across all workspaces. Each brief is a structured description of the business a tenant wants Armalo to compose.

Verified — derived from durable, append-only rows the system itself validated. COUNT(*) of rows in business_in_a_box_briefs, an append-only table (INSERT ... ON CONFLICT DO NOTHING; each brief revision is immutable once written — a second write with a different payload for the same key is rejected). No agent-narrated or process-local number is substituted for the database count.

Plans in flight

Verified (independently checked by the platform against a real external system)

Workspaces that have at least one plan revision. A plan revision is a durable, versioned output of the planning stage.

Verified — derived from durable, append-only rows the system itself validated. COUNT(DISTINCT workspace_id) of rows in business_in_a_box_plans — the same append-only, immutable-revision table as briefs.

Plans active (24h)

Verified (independently checked by the platform against a real external system)

Plan revisions updated in the last 24 hours. Updates include new revisions, status changes, and edits.

Verified — derived from durable, append-only rows the system itself validated. COUNT(*) of business_in_a_box_plans rows with updated_at in the last 24 hours — a database timestamp column, not a self-reported activity claim.

Events (24h)

Mixed (blends independently verified rows with self-reported rows — only part of this count carries a receipt)

Business-in-a-Box events emitted in the last 24 hours. Each event is a durable record of a state transition in the autonomous business.

Mixed — blends rows that require a durable receipt with rows that are internal bookkeeping only. COUNT(*) of rows in the append-only business_in_a_box_events log. Some event types (customer outcome reconciliation, growth spend authorization, outreach sends, customer-path provisioning) structurally require a durable third-party or receipt object before the row can be written. Others (plan approvals, lifecycle transitions, staged drafts) are internal bookkeeping the system recorded but did not independently verify against the outside world. This count blends both, so it is labelled mixed rather than verified.

Tool calls (24h)

Self-reported (not independently verified — no checker has run yet)

Agent tool-call results observed in the last 24 hours. A tool call is one execution by an AI employee against a real tool (e.g. send email, create doc, run query).

Self-reported — derived from a fire-and-forget log the agent process writes as it runs, not an independently validated receipt. COUNT(*) of tool_call_result rows in hub_cycle_observations. That table is fire-and-forget observability the agent process writes as it runs — write failures are logged and silently dropped, and the table carries its own `tainted` column because some rows hold unverified content. hub_cycle_outcomes, not this table, is this system's canonical execution receipt; this counter does not read from it.

Cycles (1h)

Self-reported (not independently verified — no checker has run yet)

Every hub_cycle_observations row recorded in the last hour — tool-call events plus raw agent narration and reasoning text. This is observation volume, not a count of distinct completed operating cycles.

Self-reported — derived from a fire-and-forget log the agent process writes as it runs, not an independently validated receipt. Row count (not a distinct-cycle count) of every hub_cycle_observations record in the last hour, including raw agent narration and reasoning text. Despite the display label, this measures observation volume, not a count of verified completed operating cycles.