Readiness model

Quorlo measures how ready a table is for an AI agent, as scores between 0% and 100%: one overall and one for each of the five questions.

Dimensions

Every check answers exactly one of the five questions. In the code these are the members of quorlo.readiness.Dimension:

Dimension	Question	Shown as	Checks
meaning	What is it?	Meaning	descriptions, primary key, readable names
certification	Is it the right one?	Certified	near-duplicate tables
trust	Can I trust it?	Trust	freshness column
lineage	Where did it come from?	Lineage	planned: needs dbt or OpenLineage input
governance	Am I allowed to use it?	Governed	unmarked personal data

A dimension with no checks is reported as not scored yet, never as 0% or 100%. Scores are always shown per dimension, because one opaque number can't tell a data owner what kind of gap they have.

Checks and findings

A check reports only what is wrong, as findings. There are two kinds:

• a table check judges one table on its own. All but one of the built-in checks are this kind;
• an estate check judges tables against the rest of the estate, which is how near-duplicates are found. It observes each table as schemas stream past, keeping only a small signature, and reports once every table has been seen.

Each check declares:

• a scope: table (one verdict for the whole table) or column (one per column). The personal-data check is table-scoped on purpose: one unmarked column is enough to make a table unsafe to use, though its findings still name each column;
• a severity: low, medium or high;
• a weight: how much it counts towards the score;
• which tables it applies to. For example, views are not expected to have a primary key.

All built-in checks are listed in the checks reference.

Scoring

Checks never compute scores; the engine does, in three steps.

1. Each check gets a pass rate. A table-scope check evaluates one unit per table, a column-scope check one per column:

pass rate = 1 - findings / units evaluated

2. Each table gets a weighted average of its checks' pass rates: per dimension using only that dimension's checks, and overall using all of them.

3. A scan's scores are the average of its tables' scores.

Example

The Meaning score of a table with 20 columns, all documented and readably named, with a primary key but no table description:

Check	Weight	Pass rate
table.description.missing	2	0 / 1 passed → 0%
column.description.missing	1	20 / 20 → 100%
table.primary_key.missing	1	1 / 1 → 100%
column.name.cryptic	1	20 / 20 → 100%

Meaning = (2 × 0 + 1 × 1 + 1 × 1 + 1 × 1) / 5 = 60%.

Averaging per check, rather than counting every column as a unit, keeps a wide table's column checks from drowning out its table-level checks. Counted per unit, this table would score about 95% despite having no description at all.

Nothing evaluated is not zero

If no check applies to a table (say, a view with no columns), its score is n/a, not 0%.