Before you begin
Before any customer-data learning, identify the exact source, purpose, rights holder and eligible task. Applicable terms require review and agreement, and customer learning requires explicit opt-in.
- Do not provide customer data or raw prompt archives to establish a task.
- Identify the owner who can withdraw or correct the source authorization.
Supported scope and current status
UMBRIQ currently has an executable local baseline, not a connected deployed result plane or a trained customer task. Its first-release signal is observational class/count and coverage, with no confidence, probability, span, tenant, policy or action authority.
- Treat a declared task or model as not run until current consumer evidence exists.
- Keep an observation separate from a policy decision.
Compare this path with the published coverage table and the coverage guide.
Set up the path
The planned customer workflow requires purpose-specific rights and separate task eligibility before any source is activated. A customer administrator would review the exact consent and recipient scope; no current public activation recipe is available.
- Ask for the current product and privacy terms for the specific task.
- Do not infer broad tenant permission from a single source consent.
Use the workflow
If a future release offers an approved task result, open the exact task and version, then review the signed result and typed degradation state. A correction or withdrawal is a new accountable action, not an edit to old history.
- Distinguish unknown, not run and input-limited states from a clean result.
- Request a correction against the exact source and result.
Verify the result
A future signed result must bind the trained model, runtime, rights and sequence to the exact task. A metadata card, sample output or passing local test is not proof that a customer consumer ran.
- Read back current task eligibility and result status.
- Confirm a withdrawn source is fenced from later use before treating deletion as complete.
Coverage references: 24 requirements, 0 currently passed
This guide is bound to 24 requirements in RedactWall's acceptance crosswalk, which is maintained internally and not published. A requirement counts as passed here only after current evidence for every bound scenario and authority has been published; none has been published yet.
See the public verification overview and the documentation coverage guide.
Denials and recovery
Missing rights, stale consent, absent labels or an unmeasured profile block activation. Do not turn an unavailable result into an allow decision or a reassuring zero.
- Record the exact blocking input without sharing source content.
- Retry only after an authorized owner supplies new rights and release proof.
For customer-side next steps, read Recovery and support.
Billing and integration handoffs
A future task may use a separate approved consumer and source connection. Neither a RedactWall workspace nor a paid device seat grants permission to pool, train on or export customer data.
- Review /privacy for the current learning-policy status.
- Confirm commercial and integration scope before any data handoff.
Limits to keep in view
Nine planned tasks remain blocked for customer activation pending task-specific privacy/legal approval, independent labels and signed measured numeric profiles. Synthetic-stage qualification would not authorize customer data.
- Do not describe local baseline behavior as a deployed task.
- Do not turn an observational signal into confidence, policy or action authority.
Review the current Disclosures before relying on a capability.