Contract signals

Version 3.0. Score = competition maximum (up to 60) + execution/duration maximum (up to 40), with no renormalization by coverage, dataset or filter. Official findings are separate documentary elements, never points. No amount-based uplift. No target proportion of flagged contracts.

Coverage: eight checks, each flagged, evaluated without a threshold crossed, not assessable, or out of scope. The evaluated/established-applicable ratio does not count unknown applicabilities: those are reported in addition, not as negative cases. All conflicting groups, duplicated identifiers and unverified records are excluded. Aggregate audit dossiers receive no individual index. No evaluated check = “Not assessed”, never zero.

Weights: a single offer in an explicitly competitive procedure: 12; direct award: 18, all amounts. Bounded linear progressions rounded to the tenth: single-offer repetition and concentration, 12 at 60 % up to 40 at 100 %; direct repetition, 18 at 3 contracts up to 60 at 10; duration, 8 at 120 months up to 40 at 360 months; relative increase strictly > 20 %, 8 near the threshold up to 40 at +100 %; period < 15 days, 8 near the threshold up to 40 at 3 days. Entry/reliability thresholds remain: a graduated progression does not erase them. No points for an offer expected in a direct award.

Editorial choices, not calibrated to predict irregularity. Comparison v2.1/v3 and ranking sensitivity (JSON). Software verification is not legal or statistical validation. Flagged and unflagged cases must be reviewed. Full method and excerpt review · Colombia pilot method · Countries proposed for upcoming pilots.

SECOP II · Colombia pilot: same v3.0 framework (families, family maxima, “Not assessed” semantics; full method), jurisdiction-specific check set. Four checks: award declared without supplier plurality or under manifest urgency (18); repeated such awards to the same buyer and supplier document (18 at 3 contracts up to 60 at 10); concentration within buyer and contract type (12 at 60 % up to 40 at 100 %, minimum 10 identified contracts and 80 % coverage); declared duration ≥ 36 months (8 at 36 up to 40 at 120, parsed read-only from the published free text). The four offer/timetable/history checks of the French method are out of scope here: no offers table, no publication–deadline chronology, no amendment history. The bare direct-family modality (≈82 % of this cohort) and ordinary justifications — professional services, interadministrative agreements, minimum-amount rules, regime statutes — add no points. Amounts in COP stay visible and sort only. Thresholds are editorial, not Colombian legal ones; no CPI, no country coefficient, no cross-country ranking.

Tours full notices: the declared explanations have procedure-wide scope, possibly on other lots. The criteria displayed come only from the indicated lot; candidate-selection criteria and award criteria are distinct. Weighting codes and values are kept without converting an unknown unit into a percentage. No points for a legal reason or a price weight. Documents without an assessable period display “Not assessed”, not a zero score. Publications and dispatches are distinguished; no merging with DECP.

Six additional municipalities: Rennes, Nantes, Bordeaux, Grenoble, Dijon and Tours, chosen before the scores. Groups with ambiguous versions or identifiers are excluded from all calculations and remain visible. The enriched names are a current snapshot of the legal unit via the Annuaire des entreprises, not a historical verification: SIREN/SIRET and source are kept, with no effect on the score. Every typed SIREN of the six cities and of Paris & Ardèche is looked up; a name is kept only for an exact, fully public register entry, and none is inferred for the others.

Short bidding period: fewer than 15 calendar days, upper bound from midnight UTC on the day of the initial publication to the last explicit deadline, after extensions. Only explicitly open non-accelerated procedure, identified lot and procedure, verified chain without conflict. Reduction, reopening or unknown information: no calculation. Graduated weight from 8 to 40, competition family (maximum, not a sum). Editorial threshold, not legal nor statistically calibrated. Among the 10 imported notices, no period is assessable: corrections may be missing. Result dates are never used as the start.

Concentrated awards: same buyer/CPV to three digits, DECP cohorts 2024–2025 computed separately, at least 10 contracts to the single identified holder, coverage ≥ 80 %, same SIREN receiving ≥ 60 % of the known contracts. Repeated direct awards: at least three distinct contracts explicitly without competition, all amounts, same buyer/CPV/SIREN, same period. Groups in conflict, ambiguous identifiers and unmatched co-holders are excluded. A repetition may have a legal justification; the results describe only the published data of this cohort.

Groupings: CPV = purchase sector, not the company’s activity. Projects are linked only by an explicit documentary match; neither the scores nor the amounts of contracts are summed. The other groupings are navigation tools, not proof of a common project. Groups are ordered by their first contract in the chosen sort, then their contracts follow that sort. The block without a documented project comes after the documented projects. Unknown values come last in the sorts.

Declared amount increase: last newly published amount greater than the initial one by more than 20 %, with no euro threshold, price exclusively declared final and firm, with no conflicts of initial versions and no identified change of holder. A decrease or an ambiguous chronology triggers nothing. Modified amounts are never summed. An increase may fund additional services; it is not a measure of excess cost nor of final spending. The published history may be incomplete.

Repeated low competition: only for single-offer contracts of the DECP cohort. Same buyer, three-digit CPV group, notifications 2024–2025, explicitly competitive procedures: at least 10 contracts with a known number of offers, at least 80 % coverage and at least 60 % at a single offer. Undetermined adapted procedures and groups in conflict are excluded. The statistics are computed over the whole file before filtering; they describe the published purchases retrospectively, not the regularity of a supplier. No repetition computation on the BOAMP sample.

The former thresholds of €100k (direct award) and €50k (absolute increase) no longer enter the score. Unknown values trigger no indicator. Amounts: “≥” indicates a lower bound, “>” a strict bound and “≈” a rounded value; the sort and the filter use the cited numeric value, not an assumed exact amount. These rules do not take account of legal exceptions or the context of each purchase.