H:award-bid-complementarity — Award-layer and bid-layer information are complementary

Sequencing matters under costly observability. The hypothesis is that award-layer triage adds information to bid-distribution screens on the same adjudication-anchored target — not because awards dominate bids, but because the two layers operate at different evidentiary stages. A joint classifier should outperform each layer individually.

Intuition (plain-language)

There are two layers of procurement data: cheap administrative award records (who participated, who won) and expensive bid-level microdata (every bid amount in every tender). The hypothesis: they carry complementary, non-redundant information for cartel detection. The data strongly support this — joint scoring (using both layers) gains +0.10 AUC over either layer alone with p = 10⁻²⁶. The two layers are not measuring the same thing, and this matters for the architecture of enforcement.

Evidence strength: Partial (strongly supported). The complementarity claim is established at the same-sample level with formal statistical significance and an operational deployment envelope: (i) AUC bands (AN-010): Imhof full 0.888 [0.865, 0.911]; FL14 alone 0.921 [0.914, 0.928]; joint 0.955 [0.943, 0.967]. (ii) Formal DeLong incremental tests (AN-033): Imhof + FL14 vs Imhof full Δ = +0.096, p = 1.2 × 10⁻²⁶; FL14 vs Imhof Δ = +0.035, p = 0.014 (the award-layer signal alone is at least as discriminating as the full Imhof pipeline, at lower information cost — a complementarity diagnostic, not an outperformance claim). (iii) Feature decomposition (AN-033): Imhof-base AUC 0.785 → +participation features +0.154 → +FL binary +0.003. Continuous participation is the load-bearing complement to Imhof; FL14 binary is the auditable simplification of that signal. (iv) Same-sample horse race (AN-011, AN-015): continuous AUC 0.939 dominates the binary FL14 flag (0.924); the gap is 0.015 under the corrected FL14 ≥ 14 definition (DeLong Z = −4.38, p = 1.2 × 10⁻⁵). (v) Operational sequential envelope (AN-034): sequential award → bid gatekeeping at Stage-1 K=2,000 recovers 131 of 193 cobidders (recall 0.679) using 17% of the bid-microdata footprint (2,000 of 11,676 firms). The architecture approximates the full-observability upper bound at much lower forensic cost. Imhof CV-only is chance-level (0.585) — the bid-distribution pipeline needs award-side features to reach its headline AUC. Promotion to 🟢 (Confirmed) requires non-BEC replication of the within-data complementarity pattern — see the H6 page section on why within-data DeLong significance does not satisfy the bar.

Theory

Bid-distribution screens \citep{imhof2018screening,imhof2019detecting, wallimann2023machine} evaluate suspicious bidding once bid-level data are recovered. Award-layer signals are visible earlier. If the two carry different information margins, the joint signal should beat either alone. This is the cost-of-evidence framing in \citet{chassang2022robust,harrington2008detecting}.

Prediction

On the cobidder target:

• AUC(award ∪ bid) > AUC(bid only);
• AUC(award ∪ bid) > AUC(award only);
• the increment over each is positive at the relevant operating points.

Competing prediction

Award is redundant. If the bid layer already encodes the loser-side information, the increment from adding the award layer would be statistically zero. The hypothesis predicts a positive increment; the null predicts none.

Case evidence

Imhof-style bid-distribution screens have been shown to discriminate cartelized auctions in Swiss procurement \citep{imhof2018screening,imhof2019detecting}. The Brazilian setting has both layers available — the test asks whether they substitute or complement.

Empirical test

• Sample: BEC firms with both award and bid features available.
• Outcome: cobidder indicator.
• Specifications:
• award-only: FL14 and log(1+tenders_count);
• bid-only: Imhof-style features (within-tender bid moments, CV, relative rank);
• joint: union of features in a single classifier.
• Identification: same-sample horse race; DeLong test for AUC difference.

Data requirements and limitations

Requires the LANCES bid-level export. Limitation: the bid layer is more expensive to recover, so the joint score is a full-observability upper bound rather than an operational benchmark. The complementarity result therefore supports — but does not require — the gatekeeping deployment of H:gatekeeping-cost-of-evidence.

Evidence

Analysis	Bearing	Status	Key takeaway
AN-010 (Imhof benchmark)	Direct	done	Imhof 0.888 vs FL14 0.921 vs joint 0.955
AN-011 (horse race)	Direct	done	Continuous dominates binary, DeLong p = 2e-5
AN-015 (D1 harmonized)	Supports	done	D1 passes; price coefficients align in single-score specs
AN-033 (formal DeLong incremental)	Direct	done	Imhof + FL Δ = +0.096, p = 1.2e-26; FL alone vs Imhof Δ = +0.035, p = 0.014; FL binary marginal beyond TC = +0.003
AN-034 (sequential envelope)	Direct	done	Sequential K=2,000: 74% of joint recall at 17% microdata footprint

Open tests

• Decomposition of complementarity into modality (Convite vs Pregão).
• Marginal AUC contribution of each Imhof feature in the joint model.
• Sequential envelope under temporal holdout (some data already in AN-013).
• Cross-jurisdiction replication of the same incremental DeLong tests on ComprasNet federal or a non-BR procurement panel.

Why not 🟢 Confirmed?

H6's within-data evidence is uniquely strong:

• DeLong p = 1.2 × 10⁻²⁶ on the joint complementarity gain (Imhof + FL vs Imhof) is at the level where statistical artifact essentially cannot explain the result. The two feature sets carry independent signal in an information-theoretic sense.
• The sequential envelope demonstrates the complementarity at the operational level: 74% of joint recall at 17% bid-microdata cost.
• The Shapley-like decomposition pinpoints that continuous participation is the load-bearing complement (+0.154 over Imhof base), while FL14 binary's marginal beyond continuous is only +0.003.

Two artifact families nonetheless remain untested by the within-data evidence:

1. BEC-specific bid-data structure. The Imhof features (cv_mean, cv_sd, skew_mean, kurt_mean, spread_mean, minmax_mean, second_low_mean) are computed from BEC's specific bid-recording convention. If BEC bid microdata has a structured noise pattern that interacts with award-layer participation in a particular way, the complementarity could be partially a feature-construction artifact. Imhof's original work uses Swiss data with different bid conventions; the complementarity might look different there.

2. Same-sample CADE label structure. Both Imhof and FL are evaluated against the same 193-cobidder positive class. If CADE adjudications systematically select cartels where loser-side participation differs from bid distribution (e.g., cases driven by tip-offs about bid patterns rather than participation patterns), the complementarity could be specific to CADE's adjudication selection.

Both can only be ruled out by replicating the incremental DeLong tests on a non-BEC panel with an independent cartel anchor. Until that exists, H6 stays at Partial (strongly supported), consistent with the project-wide rule documented in findings/index.md.