Works-council mode
For: tenant admins and governance reviewers.
Works-council mode limits access to figures aggregated over people. Each job is still charged to its starter and team, as described in quotas per user and team. The mode starts on for every tenant and stays on until a tenant admin turns it off.
What is hidden, and from whom
A figure about people is a number summed or counted over a person's jobs, spend, approvals, questions or chats. While the mode is on:
| Figure | Who sees it |
|---|---|
A person's own figures (user:, an API key, an operator certificate) | Only that person |
| A team's figures | Everyone who may read the tenant, when the team has at least the threshold's people now (five by default) |
| A smaller team's figures | Nobody; they are in the tenant's totals |
A service identity's figures, such as a flow's webhook (service:webhook/<flow>) | Everyone who may read the tenant: it is not a person |
| The tenant's totals | Everyone who may read the tenant |
A team's size counts its user: members at the time you look, not its API keys or service identities. A team that drops below the threshold is hidden at the next query.
Where a figure needs that many people to have done something, such as the deciders behind the approver statistics, the quality scorecard's decision figures, an evidence pack's decision figures or a rollout's denial signal, only accounts (user:) count: an API key or an operator's certificate is never one of them. Their decisions stay in the figures, so a figure made of decisions with fewer people than the threshold behind it, none included, is hidden.
This applies to:
oz0 quota usage, which lists your own row, the teams that are large enough and service identities, then a(tenant total)line and a note that the mode is on;- the console's Cost page (By user and team, with a Tenant total row) and its Teams page, where a hidden team's or person's spending shows as hidden;
- the admin API's
GetOwnerUsage, which answers the rows the caller may see, the tenant'stotaland the mode as it applied (works_council); - every figure about people added later: budget warnings, scorecards, exports. They all ask the same rule.
What is counted about approvals
Approvals are figures about people too, and stricter than spending: while the mode is on, a group's approval figures need at least the threshold's people among those who decided, not only among its members.
| Figure | Who sees it |
|---|---|
| A team's approver statistics (decisions, approval and edit rates, decision times, the share decided within seconds of being shown, streaks, load, test approvals caught) | Readers of the tenant's usage, when the team has at least the threshold's people and at least that many of them decided in the period; test approvals caught also need that many people among those who decided the team's test cards |
| The tenant's approver statistics | The same readers, once at least the threshold's people decided in the period |
| Your own approver statistics | Only you |
| Another person's | Nobody, ever: no list, export or finding names a person |
| An agent's test approvals caught, and its agreement per action class (its deployment profile and scorecard) | Readers of the inventory or the scorecard, when at least the threshold's people decided them |
| The approval pattern finding | Readers of findings: it names the team's queue (while the team passes the rule), never a person |
Test approvals themselves are off by default, and turning them on is a governance change with a reason; whether your employee representatives must agree first is for you to find out.
What it does not cover
- Records. The audit trail, a budget request with who asked and who decided, who started a job on its page, and lists of jobs are records, not figures, and are shown as before. The audit trail stays complete in this mode: a refused start's entry keeps the quota that stopped it and its figures as the limit wrote them, a small team's too.
Soaudit.readis the control: give it, through the admin role or a custom role, only to the people your agreement with the employee representatives names, such as the tenant's administrators or an auditor, and never as a way for team leads to read their teams' figures. - Enforcement. Quotas per person and per team are still enforced. A spend refusal names the quota it hit and its figures only to a caller who may read them: a team under the threshold is withheld from its own members too.
- Figures that are not about people, such as cost by agent, model or node and the insights.
The mode does not stop someone from working a person's figure out of two groups built for it, such as a team of five and the same team with one more person, or the tenant's totals minus the teams shown. Changes to teams are in the audit trail as team.member.
Show or change it
oz0 tenant works-council default # show it
oz0 tenant works-council default --threshold 10 # teams of ten or more
oz0 tenant works-council default --off --reason "agreed with the employee representatives on 2026-10-01"
oz0 tenant works-council default --on # on again
In the console: Govern › Privacy › Works council. The tenant's readers see the setting; only its admins change it.
- The threshold is at least five. Raising it applies at once; lowering it needs a reason.
- Turning it off needs a reason (at most 500 characters). With the mode off, the tenant's readers see what each person's and each team's jobs spent and started, as before this mode existed.
What is audited
| Action | When | Detail |
|---|---|---|
tenant.works_council | Every change | on and the threshold before and after, whether it loosened, the reason given |
tenant.works_council.off | Turning it off | the reason |
tenant.works_council.on | Turning it on again | the threshold |
oz0 audit --action tenant.works_council.
After an upgrade
Tenants that existed before the upgrade get the mode on, so admins no longer see per-person rows on the Cost page, Budgets' quotas or in quota usage until one of them turns it off with a reason. See Upgrades.
Read outside text in a quarantine child
Read outside text in a quarantine child for tenant admins and governance reviewers.
Incidents
Keep one record of an incident, with the jobs, audit entries, kill switches and egress denials that tell what happened, deadline clocks for the regimes you say apply, reminders before each deadline, and a timeline that verifies offline.