Skip to content

Weekly report settings

The weekly report is enabled by default and runs deterministic checks for every account, including accounts with no review settings row. Account report opt-outs and recipient preferences still apply. Quiet exception weeks send no email.

Model analysis is off by default in 1.0.0 and is planned to return in 1.1.0. The global GOVERNANCE_REVIEW_MODEL_ANALYSIS_ENABLED setting defaults to false. Each account's governance_review_settings.model_analysis_override is nullable: true enables analysis, false disables it, and null inherits the global default. An account value wins even when the global default differs. Account admins choose Model analysis for this account on the review settings page; platform admins use the same selector on the account details page. This permits testing on one account while the global default remains off.

When the effective value is off, no analysis model is resolved, no model budget is reserved, and no session work is queued. Deterministic reports are complete and say “Model analysis is not enabled.” The same gate covers session digests, semantic policy findings and follow-up checks, the AI usage policy summary, and optional exception-email synthesis. The latter retains its separate account opt-in and caps. When analysis is enabled, existing model selection, consent, allowance admission and daily/weekly limits apply; unknown allowances refuse summaries, and exhausted allowances do not prevent deterministic findings.

Account settings expose independent switches through GET and PUT /api/v1/weekly-report/preferences:

  • account_enabled: weekly reports for owners and admins, on by default.
  • members_enabled: personal reports for regular members. The stored nullable field uses the edition's plan default until an admin explicitly changes it.
  • member_policy_findings_enabled: allow personal reports to include explicitly attributed stored governance findings, off by default. The automatic factual digest carries no policy findings, so the console no longer shows this switch; the field stays in the API for explicit previews.
  • synthesis_enabled and synthesis_model_id: an optional short summary of the selected facts, off by default. Only account administrators can change them, and the model must belong to the account. When the model, pricing or the account-week cap ($0.02, two calls, 12,000 tokens, 20 seconds) is not available, the digest sends the deterministic facts.

The automatic Enterprise sender is a factual exception digest of gateway failures and approval requests. It sends nothing when no actionable fact changed since the last delivered material state. Each emailed fact links to an authenticated confirmation that calls POST /api/v1/weekly-report/deliveries/{delivery_id}/acknowledge/{exception_key}; only the member it was sent to can acknowledge it, and opening the link alone does not.

user_enabled remains the recipient's personal unsubscribe preference. Account switches require the primary owner, a superuser, or manage_account; a member may update only their own unsubscribe preference. CRUD serializes account changes and commits their audit records atomically. The console controls are gated by the existing governance-review capability. Report delivery, plan defaults and comparison orchestration are provided by the commercial plugin.

The read-only report projections filter by account and optional attributed user before aggregation. Personal approvals count requests triggered through that user's keys or managed agents and requests on which they recorded a vote. Raw finding evidence is excluded from personal projections. Adjacent report windows are half-open, and replay-validation plus every governance_ purpose are excluded from customer usage totals. Governance cost appears only on its policy-section analysis line.

Deployment and API reference

Setting Default Meaning
GOVERNANCE_REVIEW_ENABLED / Helm governanceReview.enabled true Schedule deterministic weekly review and the account-filtered hourly sweep.
GOVERNANCE_REVIEW_MODEL_ANALYSIS_ENABLED / Helm governanceReview.modelAnalysisEnabled false Model analysis default; account override takes precedence.
COST_DIGEST_ENABLED true Monday factual exception-email schedule.
Account model_analysis_override null Inherit; true on; false off.

GET /api/v1/governance-review/settings returns model_analysis_override, model_analysis_default, and the effective model_analysis_enabled. It also returns model_analysis_override_available: true when the core schema supports storing account overrides; false on older core versions, where the console hides the selector and overrides cannot be written. Account admins write the override with PUT /api/v1/governance-review/model-analysis, body {"model_analysis_override": true} (or false / null). The platform-admin GET/PUT path is /api/v1/admin/accounts/{account_id}/governance-review/settings. Overrides are audited. Choosing inherit on an account without saved settings leaves its defaults unsaved. These routes are provided by Enterprise.

Most valuable work: recorded outcomes

The weekly review can show sessions and flow executions ranked by recorded repository outcomes per estimated dollar. This section appears when the server advertises weekly_work_value and the reader has View cost permission. The digest shows five entries; View all recorded outcomes opens the console for the same reporting window. Changing the Week selector switches the outcomes window too.

Each entry states its PRs/MRs, changed lines, linked issues closed or referenced, review comments addressed, and estimated cost. The method explains the score and shared credit. Unknown or partially priced work is Unranked, rather than counted as free. This measures observed repository activity, not business value. Review resolution, a later edit of the commented line, or an accepting author reply establishes addressed evidence; missing evidence stays unaddressed.

The console reads cached forge evidence. The feature makes no model calls and incurs no LLM spend. Forge observations refresh after seven days when available; failed refreshes can retain older observations. Missing attribution or incomplete forge coverage can leave work out of the ranking.

Forge coverage currently supports GitHub, GitLab and Bitbucket Cloud. Bitbucket Data Center is not supported, so its PRs have no observed evidence and are omitted. GitHub observes at most 100 review threads and 100 replies per thread. When those limits truncate a response, the email and console show a partial-review note; comment counts cover the observed threads only. Incomplete GitLab diffs show a line-coverage note and retain unknown line counts. Forge refresh runs only when at least one digest recipient may see the cost section.

Bitbucket Cloud merge credit uses an observed merge activity date, never the PR's last update time. The adapter reads at most 100 activity entries per merged PR; when the merge event is unavailable, the report discloses the unknown date and omits weekly merge and issue-closure credit. This adds at most one API call per merged Bitbucket Cloud PR to a cold cache fill.