Usage and delivered value

The usage ledger, the daily pre-aggregate and index health: measure, don't estimate.

ValorBrain measures what it delivers per tenant from a ledger — every operation by channel (MCP, REST, hook, CLI), with latency and outcome. Nothing here is an estimate: counts come straight from usage_events.

`usage_report`

{ "tool": "usage_report", "arguments": { "period": "month" } }

period: today / week / month (default, 30 days) / quarter.
Returns: operations by channel, by tool, by person/agent, latency p50/p95, results returned and the daily series.

It is the source for "how many queries did we run this month" and for sizing value. The SaaS dashboard reads the same data — no interpretation layer in between.

The daily aggregate: `value_daily`

The nightly job pre-aggregates the ledger into value_daily (per tenant, per day). The public read is REST:

curl -sS ".../api/v1/tenant/value-summary" -H "Authorization: Bearer $VALORBRAIN_TOKEN"

Complete days come from the aggregate; today and missing days are computed live — and the stale_days field reports the holes. Value families (search / context / write / collab) group what each operation delivered.

Answers delivered

curl -sS ".../api/v1/usage/answers" -H "Authorization: Bearer $VALORBRAIN_TOKEN"

The cut by answers (not by calls): how many answers cited memory, with sources — the metric that matters for "is the product delivering knowledge or just latency?".

Index health

index_stats — distribution by content type, staleness (how long without a re-embed), overall health. Use it before blaming ranking.
memory_health — open proposals/conflicts, high-priority items and canonical sources with drift. It is the end of substantive session call: it measures knowledge debt, not infra.

The loop that closes the value

Measured usage without memory_used is half the story:

memory_retrieve / memory_prepare deliver context.
The answer cites its sources (docids).
memory_used declares what carried the answer — and the verdict (confirmed/corrected) feeds ranking.

The ledger records the operation; memory_used records the outcome. Without step 3, the report shows activity, not value.

In the SaaS

What the tenant sees on the dashboard is built on these same surfaces. If a dashboard number and an API usage_report diverge, the bug is ours — report it with feedback.

usage_report