Skip to content

Anti-Collusion Monitoring

Behavioral monitor — instantiates Bounded Rivalry Governance

Reads the pattern of bids, prices, and moves for the statistical fingerprints of secret coordination, so a field that looks competitive isn't quietly rigged.

Anti-Collusion Monitoring is the surveillance function that treats a suspiciously well-behaved field as a red flag. It continuously reads the observable record of a rivalry — bid sequences, price moves, market shares, timing — for the signatures that competition is being faked: bids that rotate turns, margins that move in lockstep, entrants that never undercut one another. Its defining move is inference from pattern rather than confession: collusion is normally secret, so the mechanism detects it indirectly, from the shape of the data, not from a smoking-gun agreement. It works the anti-collusion half of the archetype's guardrail; the sabotage-and-attack half, and the punishment, belong to other mechanisms.

Example

A regional roads agency runs dozens of paving tenders a year, each of which looks competitive on its own. Its monitoring unit screens the whole bid record and finds a pattern: four firms appear to take turns winning, the losing bids sit just above the winner's by a near-constant margin, and no firm ever bids in another's home county. None of that proves an agreement — but bid rotation, cover bidding, and geographic division are exactly what a procurement cartel produces.

The pattern trips a threshold and triggers a deeper investigation and referral. The point is that no single tender revealed anything; only the shape across many of them exposed the coordination that individual, apparently-competitive auctions had hidden.

How it works

What distinguishes this mechanism is that it detects and refers — it does not judge or punish — and it lives on observability:

  • Structural screens — flag markets whose conditions (few players, homogeneous product, repeated interaction) make collusion easy, so attention is aimed where it is likely.
  • Behavioral screens — hunt the fingerprints in the data: price parallelism, bid rotation, suspiciously stable market shares, cover bids clustered just above the winner.
  • Cross-round pooling — pattern only emerges across many contests, so evidence is aggregated over time rather than judged one auction at a time.
  • Anomaly-to-investigation handoff — a flagged pattern opens an inquiry; the monitor produces suspicion, and the verdict is made elsewhere.

Tuning parameters

  • Sensitivity vs. false-positive rate — how aggressive the screen is; tight screens catch more cartels but flag innocent coincidences.
  • Observability depth — how much of the record is visible (bids only, or costs, timing, communications); more data sharpens detection but costs privacy and effort.
  • Screen mix — how much weight on structural (market-ripe) versus behavioral (pattern) markers.
  • Reaction threshold — whether a flag triggers investigation or an automatic penalty; monitoring is safest when it stops at referral.
  • Retrospective window — how many past rounds are pooled before a pattern is called.

When it helps, and when it misleads

Its strength is catching the failure a leaderboard cannot see — a field that is rigged rather than merely gamed, where every entrant looks compliant because they have quietly agreed not to compete. Its failure mode is that pattern evidence is circumstantial: parallel behavior can be perfectly innocent, since rivals reacting to the same cost shock look much like colluders, so screens throw false positives and cannot by themselves establish intent.[1] The classic misuse is running it selectively — screening only disfavored bidders, or waving a "clean" screen to exonerate a favored one. The discipline that keeps it honest is to use screens to target investigation, never as a verdict, and to apply the same fixed criteria to the whole field.

How it implements the components

Anti-Collusion Monitoring fills the detection side of the guardrail, and only that:

  • anti_sabotage_and_anti_collusion_guardrail — it operates the anti-collusion half, surfacing the secret coordination that softens a field.
  • information_disclosure_and_observability_rule — it depends on, and shapes, what must be observable (bids, timing, shares) so patterns can be read at all.

It does not penalize the coordination it finds, nor the sabotage half of the guardrail — that is Sabotage or Foul Penalty Schedule; it does not surface the personal ties that create conflicts up front — that is Conflict-of-Interest Disclosure; and it scores nothing itself, reading the record that Contest Rulebook and Ranked Leaderboard With Audit define and publish.

Notes

A clean screen is not proof of innocence and a flagged pattern is not proof of guilt — the monitor produces suspicion, not judgment. Wiring it directly to penalties inverts due process; the referral must pass to an investigation and a fair-process layer before any sanction lands.

References

[1] In competition economics, screening uses structural and behavioral markers (bid rotation, price parallelism, abnormally stable shares) to flag likely collusion for investigation. It is a targeting tool, not proof — the same markers can arise from firms independently responding to shared conditions. ↩