Watch data-quality drift and recheck only dependent reports

The decision tree

5 typed decisions, each with an action for every answer

At fixed moments in a run, the engine puts one narrow question to a decision model. The decision model never writes text: it answers yes or no with a probability, picks from listed options, or gives a score, about a small slice of the material. The engine then does exactly what this tree says, which is what makes the run auditable. The thresholds are the template's design values, not measured results.

1. 1

   Planner, while planning

   Scope checkA choice among options

   Before work starts on a unit

   Does this column's profile differ between the two named runs by more than its documented seasonal or calendar pattern explains?

   Sees only: One column's profiles from two runs and its documented expected pattern

   Why: Filters routine variation so downstream rechecks are triggered only by real drift.

   • Unexplained shift→thenAccept
   • Explained by documented pattern→thenSkip this unit
   • No documented pattern to compare→thenEscalate to a strong model
2. 2

   After workers, the judge checks

   Evidence checkYes or no, with a probability

   After a worker answers

   Does the quoted line of this model's logic filter, join or aggregate on the drifting column, so that the shift can alter its output?

   Sees only: The drift description and the quoted lines of one dependent model

   Why: Lineage says a model touches a column; only the logic says whether the drift matters to it.

   00.50 and 0.851

   • Yes: 0.85 or higher→thenAccept
   • Unsure: 0.50 up to 0.85→thenEscalate to a strong model
   • No: below 0.50→thenReject and retry
3. 3

   Reconciler, while merging

   Conflict checkYes or no, with a probability

   While reconciling

   Do the column's current values contradict what its documentation says the column contains?

   Sees only: The column's documented meaning and a summary of its current profile

   Why: A change of meaning with an unchanged schema is the drift that tests never catch.

   00.30 and 0.701

   • Yes: 0.70 or higher→thenMark unresolved
   • Unsure: 0.30 up to 0.70→thenEscalate to a strong model
   • No: below 0.30→thenAccept
4. 4

   Run control, between rounds

   Another round?Yes or no, with a probability

   Between rounds

   Do the exposed models found in this round feed further models or reports that have not yet been checked?

   Sees only: The exposed models and their recorded downstream dependants

   Why: Walks the lineage only as far as exposure actually reaches.

   00.30 and 0.601

   • Yes: 0.60 or higher→thenContinue
   • Unsure: 0.30 up to 0.60→thenContinue
   • No: below 0.30→thenStop
5. 5

   Accountable person, before anything is settled

   Person decidesYes or no, with a probability

   Before anything is reported as settled

   Is any exposed dependant a report or metric that has already been published to its readers for the affected period?

   Sees only: One drift note and its list of exposed reports with publication dates

   Why: Whether a drift is a defect and whether to restate are calls for the platform and report owners.

   Accountable: The data platform owner decides whether a drift is a defect; report owners decide on restatement.

   00.15 and 0.401

   • Yes: 0.40 or higher→thenAsk a person
   • Unsure: 0.15 up to 0.40→thenAsk a person
   • No: below 0.15→thenAccept