Methodology
How the review signals are computed, what they can and can’t show, and the public data behind them.
What this is
A concept demo built only on public records from data.ok.gov (OMES): P-card transactions, vendor payments, payroll, competitive-bid exemptions, purchase-order exceptions, and the general ledger. Signals are heuristic review prompts, not findings of wrongdoing. The named-entity review console is access-gated, mirroring how the production integrity system would stay internal while only an aggregate subset is public.
A full implementation would ingest the State’s source accounting and purchasing-card systems — the approval chains, receipts, purchase-order lines and per-card limits that the highest-value checks require.
Integrity check coverage
What public data supports today — and the high-value checks that need the production backbone.
- Competitive-bid exemption review Live nowPublic bid-exemption dataset (source-stated decision).
- Vendor / merchant concentration Live nowPublic vendor-payment + P-card spend.
- Repeated-merchant clustering PartialNo cardholder key in public data — cannot dedupe per card.
- Statutory-threshold proximity PartialAmounts present; contract/exemption context missing.
- Missing-receipt detection Requires backboneThe State’s purchasing-card system (per-transaction receipts).
- Configured per-card / per-agency limit breach Requires backboneThe State’s card-limit configuration.
- Three-way match (PO ↔ receipt ↔ invoice) & encumbrance Requires backboneThe State’s accounting system (PO/voucher/approval tables).
Signal specifications
Each computed signal carries a rule, rationale, false-positive control and benign caveat.
Competitive-bid exemption review
Live nowsource-stated · may name- Rule
- Group bid-exemption requests by reason and decision; surface approved exemptions with weak/blank stated reasons, or high cost relative to low/no stated savings vs. the comparable item.
- Why
- Exemptions bypass competitive bidding. The bid-exemption dataset is the only public source that carries the State’s own Approval/Denial decision and stated reason, so this restates a published fact rather than inferring one.
- Statutory hook
- Central Purchasing Act competitive-bid requirements; Strom HB3413–HB3420 transparency mandate (eff. Nov 1 2026).
- False-positive control
- Uses only the source-stated decision and reason; no derived "should have been bid" judgement.
- Benign caveat
- Many exemptions are legitimate (emergency, genuinely sole-source, grant deadlines). Shown verbatim for review.
Vendor / merchant concentration
Live nowsource-stated · may name- Rule
- Per agency and fiscal year, compute Herfindahl–Hirschman Index (HHI) and top-5 vendor share over vendor payments and P-card spend.
- Why
- Spend concentration is a standard procurement-health indicator and a starting point for competition review.
- Statutory hook
- Open Records Act 24A.4 (lawful disclosure of public-fund expenditure).
- False-positive control
- Aggregate only; deterministic vendor grouping only (no fuzzy merges).
- Benign caveat
- High concentration is normal for specialized categories (utilities, a single statewide contract).
Repeated-merchant clustering
Partial- Rule
- Same agency + same normalized merchant + same/near transaction date with multiple P-card charges. Descriptive grouping ONLY — presented as a pattern to review, not as a threshold-circumvention finding.
- Why
- A pattern worth a human glance; clustering can accompany routine recurring buys.
- Statutory hook
- Procurement-control review (descriptive).
- False-positive control
- Count-based and neutral. Public data has no unique cardholder key, so charges cannot be attributed to one buyer.
- Benign caveat
- Routine recurring purchases (fuel, supplies, subscriptions) cluster legitimately. Cardholder identity in public data is name+initial only and may collide.
Statutory-threshold proximity
Partialsource-stated · may name- Rule
- Count, per agency, P-card charges at or over the $5,000 single-purchase guideline. (Distinct from the $50,000 Central Purchasing Act competitive-bid threshold.) An aggregate count labeled as proximity to a published limit — NOT a confirmed breach.
- Why
- Charges at/over a published guideline are a starting point for confirming the correct purchasing path was used.
- Statutory hook
- Oklahoma statewide P-card single-purchase guideline ($5,000); Central Purchasing Act competitive-bid threshold ($50,000).
- False-positive control
- The guideline is a published number, not a judgement; contract/exemption context is absent from public data. Aggregated to the agency, not the cardholder.
- Benign caveat
- Crossing the guideline is not a violation — the purchase may be exempt, on a statewide contract, or properly authorized.
Oklahoma context
The statewide spending-transparency mandate is the Strom package (HB3413–HB3420), effective Nov 1 2026, requiring public contractor lists, contract values, project status, subcontractor disclosure and a public contract database. The integrity emphasis aligns with the Division of Government Efficiency (DOGE-OK, EO 2025-04, housed in OMES), the Legislative Office of Fiscal Transparency’s finding of ~$3B spent without oversight, and State Auditor Cindy Byrd’s identification of $93.4M in misspent expenditures. Publishing public-fund expenditure is supported by the Open Records Act (24A.4), with personal-financial and competitive-advantage exemptions (24A.10) respected.
Data notes & caveats
- Detail (P-card, vendor payments, payroll) is loaded for the most recent fiscal years; headline figures use the latest complete fiscal year for comparability.
- Payroll FY2022 is excluded from the cross-year trend: the State changed payroll reporting that year to a finer, lump-sum-inclusive basis (~4 pay lines/employee vs ~2 after), so its totals are not comparable. Every record remains searchable in the payroll explorer.
- General Ledger totals are computed directly against data.ok.gov; a fiscal year is shown only when every quarter loaded cleanly (data.ok.gov rate-limits bulk ledger reads, so some years may be omitted rather than shown partial).
- Vendor names are deterministically normalized from public records; cross-dataset matches are name-based and approximate. Authoritative vendor identity is a source-system (backbone) asset.