Running the engine…
The question every HR leader asks
How many active employees do we have in ?
Four systems, four answers. Numbers below are the corrected view: source data plus the fixes a steward has approved ( so far). Source files are never changed.
Approve them one at a time and watch the numbers converge.
MONTHLY PAYROLL COST
| Payroll pays | |
| Core HR contracts (excl. new joiners) |
ACCESS RISK
enabled logins with no active employee behind them
STEWARD PROGRESS
approved · rejected · matches to review
Ranked by fixed severity tier first (access risk, then money, then report counts, then cosmetic), then by the impact number inside each tier. Each card shows the rule and the number.
trusted system may be wrong
steward to enter value
No fixes match these filters.
Loading fix…
Evidence, system by system
Red border = record this fix would change. Highlighted row = the field in question.
Suggested fix
The trusted system is the odd one out here. Pick the value you believe is right.
Or enter:
Why we think these records are one person
Single record: no cross-system match involved (see evidence).
↔
Impact check, before you approve
TOUCHES
BEFORE → AFTER
| → |
No headline number changes.
WHAT CHANGES
| → |
DEPENDENCY CHAIN (dependencies.md)
→
Decision
Match review
Middle-band matches the engine will not decide alone. Confirm or reject each one; the engine re-runs and the fix list updates. Scores:
Nothing to review.
Audit log
Every approve, reject and match decision: who, when (UTC), what, before, after, reason. Append-only.
| When (UTC) | Who | Action | Target | What | Before → after | Reason |
|---|---|---|---|---|---|---|
No decisions yet.
Accuracy on the holdout set
Synthetic holdout set (seed , people, records) with planted errors. Thresholds were tuned on a separate dev set and frozen before this set was evaluated.
Match precision
record pairs, automatic
Match recall
Issue precision
Issue recall
Per error type (automatic, no steward input)
| Type | Planted | Flagged | Precision | Recall | False-positive rate | Suggested value right |
|---|---|---|---|---|---|---|
False-positive rate = flagged issues that match no planted issue ÷ all flagged issues of that type. Synthetic data from our own generator: a real-data number needs the pilot.
Dev (tuning) set for comparison
No holdout report yet. Run make eval.