Metering

01Measuring usage without influencing the service #sec_317780ad

Metering is REVIEWIT's system for measuring consumption. It counts technical and functional events — API calls, queries addressed to RITA, exported volumes or processed objects — to produce a reliable reading of usage by subscription.

It does not calculate prices, invoices or access rights. It does not decide that a subscription should be suspended and never blocks a business action. This separation is structural: metering observes, it does not govern. A counting error must be detected and treated, but must not prevent a reviewer from making a decision or an administrator from operating his workspace.

The consumption event is measured as close as possible to its source, consolidated locally in the region, then transmitted to the control plane. The region pushes its counters; the control plane never sucks up regional business data.

02Counters, never people #sec_213beff9

The central unit is a daily counter: a metric, a scope, a UTC day and a quantity. The scope stops at the level necessary for usage governance, notably the subscription or the workspace. Counters carry neither reviewer identity, nor campaign content, nor conversation text.

The metrics catalog is versioned like data. It indicates what is measured, the unit and the aggregation mode. A sum of requests does not obey the same rules as a daily maximum or a state snapshot; the mode is therefore explicit and does not depend on a subsequent interpretation.

When the product already has a durable trace, the measurement is derived from it. For example, an export manifest allows counting the broadcast events without adding a write on the path of each event. Online instrumentation is reserved for facts that leave no reliable trace, such as an API call or a RITA query.

This rule limits the load and avoids double truths: only one durable source is authoritative for the fact and for its counting.

03An idempotent and replayable measurement #sec_64dede5c

Each counter has a deterministic identity derived from its dimensions. Replaying the same processing therefore produces the same logical object instead of adding a second consumption. Processing can resume after an incident without artificially inflating volumes.

Counters remain open during their collection day, then are sealed and transmitted. The transport accepts replays and technical duplicates; the recipient neutralizes them thanks to this stable identity.

A catch-up window allows recalculating derivable measurements from dated traces. On the other hand, a disappeared state snapshot cannot always be honestly reconstructed. The system then prefers to declare a lack rather than invent a value. The principle is simple: an absent measurement is not guessed.

Transmission defects feed a replay queue and operational alerts. They never reach the user who performs the measured business act.

04Closing a period without rewriting the past #sec_817a7b16

Measurement periods follow the UTC calendar: days and months thus have the same definition in all regions. A monthly period goes from open to closing, then to closed after a grace window for normal late arrivals.

A closed period is immutable. A late fact does not silently modify its totals: it produces a revision or a regularization visible in a still open period. This rule protects the stability of reports and makes corrections explainable.

Metering also distinguishes client volume from supplier cost. For RITA, for example, the number of queries or the functional volume consumed is not confused with the technical tokens billed by an AI model. Both can be observed, but they answer different questions and are never substituted for each other.

05Residence and availability #sec_591c95b7

The measurement is aggregated in the regional cell before leaving the region. Only the counters necessary for steering are transmitted; no documentary detail or review decision accompanies the flow.

Local storage allows absorbing a temporary unavailability of the control plane. Counters wait, then restart upon resumption. Indicators monitor the age of the oldest batch, the depth of the queue, the periods that fail to close and the abnormally silent regional activity.

The region's activity report uses the same channel, but its absence never constitutes an automatic authorization to cut the service. Here again, a missing observation triggers an investigation, not a business decision.

06In summary #sec_434f4b3f

Metering provides a durable, minimal and explainable view of usage. It measures as close as possible to the source, consolidates counters without personal data, supports replay without double counting and keeps an absolute rule: a measurement failure must never become a REVIEWIT failure.