The delivered artifact

The review queue

Not a Slack message. The decision this informs is which transactions a person opens this week, so the output is the worklist that person works, ordered, with the evidence and the lines attached.

Transaction screening, 2026-05-30 to 2026-08-273 for review19,800 lines · $215,593,342 · 16 units · BH at 5% FDR

Approval limit tested: $5,000, with a $400 window either side. Population conformity: close (MAD 0.0012). 149 lines on the pull list.

1Meridian Facilities GroupVendorClusters below the approval limitq = 2.5e-10
unit            Meridian Facilities Group (1,250 lines this quarter)
digit test      chi-square p = 0.0263, MAD 0.0104, acceptable
limit test      66 below / 8 above $5,000, p = 1.6e-11
q after BH      2.5e-10 across 16 units
under the limit $139,733 in excess lines, 66 to pull
Pull first, highest amount under the limit
INV-1113-00312026-06-03Emergency callout$4,974.58
INV-1113-04432026-06-13Fit-out labour$4,972.11
INV-1113-01622026-07-16Fit-out labour$4,970.55
INV-1113-01712026-07-10Building services$4,967.66
INV-1113-00732026-06-12Emergency callout$4,966.64
INV-1113-11302026-06-26Building services$4,964.52

Check whether any of these were raised against the same purchase order, the same requester, or the same week. That is what would distinguish a split purchase from a vendor whose work happens to price near the limit, and the data here cannot make that call for you.

2Regional BuildoutCost centreClusters below the approval limitq = 5.2e-5
unit            Regional Buildout (1,600 lines this quarter)
digit test      chi-square p = 0.3186, MAD 0.0073, acceptable
limit test      49 below / 13 above $5,000, p = 4.8e-6
q after BH      5.2e-5 across 16 units
under the limit $86,085 in excess lines, 49 to pull
Pull first, highest amount under the limit
EXP-1114-03182026-07-15Temporary power$4,968.81
EXP-1114-12152026-07-03Contract labour$4,967.46
EXP-1114-05282026-06-02Site prep$4,966.80
EXP-1114-02532026-06-30Site prep$4,960.53
EXP-1114-00692026-07-23Temporary power$4,955.13
EXP-1114-05092026-07-21Site prep$4,945.49

Check whether any of these were raised against the same purchase order, the same requester, or the same week. That is what would distinguish a split purchase from a vendor whose work happens to price near the limit, and the data here cannot make that call for you.

3Northwind LicensingVendorDigit distribution onlyq = 2.1e-11
unit            Northwind Licensing (600 lines this quarter)
digit test      chi-square p = 6.4e-13, MAD 0.0364, nonconforming
limit test      34 below / 23 above $5,000, p = 0.1451
q after BH      2.1e-11 across 16 units
under the limit no excess, spend is smooth across $5,000

Flagged on digits alone. The spend is smooth across the approval limit, so nothing here points at splitting. Most likely explanation is fixed price points, which produce this exact digit shape legitimately. Confirm against the contract and suppress the unit if it matches.

Cleared this run, no action

Atlas Freight Systems, Corvus Print & Mail, Helio Facilities, Northgate Legal, Pinewood Catering, Quarry Road Hardware, Sable Analytics, Field Operations, Marketing, R&D Prototyping, Facilities, Travel, IT Support.

What this document is not
  • This is not a finding of fraud, an accusation, or a determination that anything is wrong. It is a ranking of where to spend the next forty audit hours.
  • A Benford deviation is evidence of an anomaly and nothing more. Fixed price lists, capped ranges, narrow spend bands, and a vendor that bills one number all year all violate Benford while being entirely legitimate. Item 3 below is exactly that case.
  • Clustering under an approval limit is a statistical pattern, not an intent. It has innocent causes, including a vendor whose own pricing sits near the limit and a policy that produces natural rounding.
  • No item here should be raised with an employee or a vendor before a human has opened the underlying documents and formed their own view.
Why the ordering is by dollars and not by p-value

Item 3 has the smallest p-value on the page and belongs at the bottom of it, because no money is sitting under the limit and the digit shape has an obvious explanation. Sorting a review queue by statistical strength sends the reviewer to the most surprising thing rather than the most expensive one. Those are different questions.