IndustriesProcurement and grants integrity
Across eleven tenders, in rotation. No screen looks there, because award data and contract data have never been read together. Nexyron reads them as one model.
Too few bidders. Round prices. The same firm winning one region. Bids minutes apart. Cover bids too high to win.
Every classic indicator is documented, taught and freely available. A group operating for a decade knows all of them, so the bids arrive from a plausible number of firms at plausible intervals with plausible spreads.
A cartel that allocates work has to actually allocate it. That allocation is in your contract data, and nothing you own reads award data and contract data together.
There is a deeper reason the classic screens have stopped working. Every one of them is a statistic computed over bids in a table. Bids are the part of this the participants control most completely, and a table has no way to express a relationship, a sequence, or what happened after the contract was signed.
The screen returns nothing, and that is not a failure of the screen. It is what happens when a detection method becomes public and the subject can read.
What Nexyron does instead
Nexyron builds one model from the tender notices, the bid records, the award decisions, the evaluation scores, the contract performance data, the subcontractor filings and the company register, and then asks it questions a bid table cannot hold.
Common officers, shared addresses, shared agents, shared history. A firm controls what its bid looks like far more easily than it controls what the registry says about its directors. In Nexyron that is a traversal, not a manual chart somebody maintains.
The losing firm has to be compensated, and it usually happens as a subcontractor to the winner. It is sitting in the contract data and nobody joins it to the bids.
One tender is close to impossible to judge. Eleven across four years is a pattern with a shape, and time is a property of the model rather than a separate export.
The signal moved to where the screens are not, which is exactly the place a connected model can reach and a table cannot.
A simulated screen
Constructed to show the method. No authority, no supplier, no claim about detection performance.
Eight firms, 47 tenders, four years. Participation averages 4.2 bidders, winning margin 6%, no firm above 22% of awards. Every classic screen has been run and every one is clean.
Joined to the company register. Six are unrelated. Two share a director who resigned from one four months before the other was incorporated. It changes the effective number of independent bidders in every tender both entered.
Counted, the award distribution is unremarkable. In sequence, within one category, the winner rotates through five firms in an order that repeats twice across four years.
In eight of eleven tenders, at least one losing bidder later appears as a subcontractor, consortium member or supplier to the winner of the same contract.
Most score consistently. One does not. Across seven tenders their scores for one firm drift steadily upward relative to their own scoring of everyone else, while colleagues' scores stay flat. In two of the seven, the technical specification carried requirements only one firm met.
Three comparable categories did not receive it, so the question is answerable. The direct subcontracting largely stops, and a new pattern appears in which the compensation flows one step further removed.
The behaviour did not stop. It adapted, in a way the original screen would never catch. That is the finding that tells you what to do next.
Every conventional concentration measure is a count, which is exactly why sophisticated rotation survives all of them.
Did the losing bidder work on the contract they lost. Computable today from data most authorities already publish and nobody joins.
Invisible in one tender. Unmistakable across seven, once the evaluator is a subject with a history rather than a column.
Interventions here are announced, evaluated once, and almost never re-examined for displacement.
The dimension nobody in this category occupies
Integrity pacts, e-procurement mandates, lot-splitting rules, mandatory disclosure. Each is introduced, each is credited with an improvement, and the improvement is almost always measured as a before-and-after on a market that was changing anyway.
Nexyron carries 42 causal and decision procedures on the same graph the tenders live in. Comparable categories that did not receive the reform are identified rather than assumed. Uplift is estimated rather than asserted, and where the comparison cannot honestly be built, the engine declines instead of publishing a number that will not survive contest.
This is the question every oversight body is asked by its legislature and no procurement analytics product has ever answered.
The evidence trail is the product
Findings here become public reports and are challenged by well-resourced parties. That is the design constraint, and Nexyron is built to it.
Every conclusion carries the records behind it, the time basis, the transformations applied, and a clear separation between what was measured, what was extracted from a document and what was inferred. The methods underneath, community detection, centrality, shortest path and shared-neighbour weighting, are peer-reviewed with decades of literature behind them.
What you hand an investigator is a structure, its evidence, and an explicit account of what the data does not establish. That is what makes it usable when somebody expensive starts arguing with it.
Running it
For a national audit body this is frequently not a preference but a legal constraint. Procurement records, evaluation records and supplier data may not leave the jurisdiction, which closes most of the analytics market outright.
Nexyron installs like an application, starts a full engine on hardware you already own, and grows into a cluster inside your own infrastructure when the caseload demands it.
Take your last completed contract category. List the losing bidders, then read the performance records and count how many went on to work on the contract they lost.