Data quality and record readiness
Understand what blocks trustworthy analysis, and define when a record is ready for a process.
Blocchi separates three questions that are often mixed up:
| Concept | Question |
|---|---|
| Data quality | Is the underlying data correct, complete, consistent, fresh and trustworthy? |
| Metric readiness | Does Blocchi have enough mapped evidence to compute this metric? |
| Record readiness | Is this specific record fit to move forward in a business process? |
Data quality (/app/data/quality)
The page answers one question for the whole workspace: can you trust your data? It covers every connected source (external CRMs, Blocchi CRM, files, finance and others), not one product. There is no blended quality score: every number is something you can act on.
- Across all sources — open issues, how many of them block a business action, records ready against those needing attention, and sources that are failing, stale (no successful sync in 24 hours) or never synced.
- Sources — one row per source with its sync state, mapping coverage (mapped fields out of the fields the entity has), readiness of the records it feeds and the open issues that name it. Each row links to the one fix that matters most for it: fix a failing sync, run a first sync, finish a mapping, review issues or see readiness. Blocchi CRM is written in place, so it is always live and has no mapping; a file is as current as its last import. Readiness is evaluated per record, so two sources feeding the same records show the same counts. Issues that span sources (duplicates, conflicting values) are counted in the fix queue rather than on one row.
- Fix queue — with Data+, the top issues from the stewardship inbox and suggested corrections you can accept or reject right there; each decision keeps its evidence and can be rolled back in Quality resolution. Without Data+, the page shows how many issues were detected and what Data+ adds. It never shows an empty "all good".
- Who depends on this data — the products that read these sources. Revenue comes first: which CRM is counted (Blocchi never mixes two; an admin chooses one under Commercial model,
/app/commercial-model), when it last synced, and whether each Revenue number is complete, with a link to the fix. Legal links to Legal operations when Legal is enabled.
When an external CRM is the one counted, the Revenue section also links to its hygiene checks (activity, close dates, contacts). Those describe how the CRM is kept, not whether the numbers are complete; the Revenue page Can we trust the CRM data? covers them.
The same read is available to integrations as GET /api/v1/data-plus/quality and the read-only MCP tool blocchi.get_data_quality. Its counts are workspace-wide, so people whose data access is limited to their own or their team's records are not shown it.
AI Cleanup Studio
Data+ stewards can open /app/data-plus/cleanup-studio to profile a stored current-state, retained-history or live tenant-authorized dataset, propose typed cleanup rules manually or with AI, and compare changes before approval. AI receives aggregate schema metadata, not raw rows. Uncertain values are quarantined; source records stay unchanged.
Approved recipes are versioned and reversible. They are replayed in the Studio and in Postgres metric totals, groups and distributions. Retained-history datasets use immutable recipe-pinned derived Parquet snapshots in Supabase Storage or GCS, preserving the historical original. Product limits: retained-history materialization supports at most 100,000 records per immutable snapshot; historical grouped metrics, buckets and record drill-downs support at most 10,000 records per snapshot. Above these limits, Blocchi blocks publication or querying instead of showing partial results. Increasing these limits is outside the current scope. Source CRMs, native operational tables and canonical Party identities are not modified automatically. Counts are sample-based (up to 500 records).
Tenant-authorized live connectors use the same approved rules before filtering and calculating scalar, grouped and bucketed KPIs. Product limit: Live cleanup metrics and drill-downs support at most 10,000 records per dataset, with a complete count-verified read. Above this limit or when the provider read is incomplete, Blocchi blocks the query rather than reporting partial totals; use persisted sync instead. Increasing this limit is not in the current scope.
The read-only REST endpoint GET /api/v1/data-plus/cleanup-profile and MCP tool blocchi.get_cleanup_profile return only schema/quality summaries and the latest full-validation counts to an authorized tenant steward. They do not expose row values or accept changes.
Viewing approved changes elsewhere: open Data Explorer and choose Show approved cleanup to inspect derived values next to authorized source records, or request cleaned=approved in the Data Explorer REST/MCP read. Record drill-down also supports a derived annotation. Explorer search and sorting stay source-based; persistent Postgres metric aggregations use the published recipe consistently. No cleanup overwrites original source records, original historical Parquet objects, live-only connections or Golden Party profiles.
Record readiness (/app/data/readiness)
Define what it means for a record to be ready — for example Opportunity qualification readiness or Customer onboarding readiness.
- Create zero, one or many checklists per semantic entity. Checklists attach to Blocchi entities, not provider objects, so they stay portable.
- Each check has a condition, optional applicability criteria, a severity (blocking, required or recommended), a weight, remediation guidance and an order.
- Start from a suggested template; applying it creates your own copy that you can customize.
A record is evaluated to one of four states:
| Status | Meaning |
|---|---|
| Ready | No applicable check failed and no required or blocking evidence is unknown. |
| Needs attention | A non-blocking check failed, or required evidence is unavailable. |
| Blocked | At least one blocking check failed. |
| Not applicable | No check applies to the record. |
The percentage score is secondary: a blocking failure is never hidden by a high score. Readiness is used, for example, when moving an opportunity between stages in the Pipeline.
Related
Last updated on