Mechanism Design

1. Diagnose the two failures precisely — they have opposite fixes

Your symptoms come from one root cause: a lowest-price-wins rule selects on exactly the wrong variable. But the lowball problem is actually two distinct economic failures stacked on top of each other, and conflating them is why fixes backfire.

Adverse selection (hidden type, before award). Firms differ in private ways you can’t see: true cost efficiency, competence, solvency, and disposition to cut corners. A sealed low-bid auction doesn’t select the most efficient firm — it selects the firm with the lowest bid, which is a mixture of “genuinely cheap” and “most optimistic / most desperate / most willing to underbid and recover later.” You are running a reverse lemons market: the honest firm that prices the work correctly is systematically out-competed by the firm that mis-prices it.

Winner’s curse (a special case of the above, driven by common-value uncertainty). Much of a construction cost is common across bidders — soil conditions, material prices, weather exposure. Each bidder forms a noisy estimate. The firm that wins is disproportionately the one whose estimate erred low. A rational bidder shades upward to correct for “winning is bad news”; an inexperienced or aggressive one doesn’t, wins, and is then underwater on day one. Everything bad follows from a firm that is contractually locked into a price below its real cost.

Moral hazard (hidden action, after award). Once the price is fixed, quality is a hidden action. A firm that’s underwater — or simply one facing a high-powered cost incentive — recovers margin the only ways available: thinner slabs, cheaper aggregate, skipped cure time, and the claims-and-change-order game (bid low, then mine ambiguities in the spec for compensable extras).

The crucial asymmetry for design:

• Adverse selection / winner’s curse → fix by changing who can bid and how you score them (screening).
• Moral hazard → fix by changing what the contract pays for and what it withholds (incentive compatibility post-award).

A fix aimed at one can worsen the other. The classic error: pile risk onto the contractor at a hard fixed price to “discipline” them — which deepens the winner’s curse (they lowball-and-claim or price a huge risk premium) and sharpens the corner-cutting incentive.

2. The screening layer — make capable, honest firms self-select into winning

Replace lowest-price with a scoring auction (Che, 1993). Award on a published score, e.g. Score = Price − λ·QualityPoints, where quality points come from verifiable, pre-announced criteria (technical method, key-personnel experience, past defect/warranty performance, schedule credibility). A scoring rule that is announced in advance is itself incentive-compatible: a firm that knows quality is rewarded will invest in it, and a low-quality firm can’t win on price alone. Keep λ disciplined and transparent so it isn’t a discretion loophole.

Make a surety do your screening for you — this is the highest-leverage single mechanism. Require a bid bond to bid and a performance + payment bond to be awarded. A surety company underwrites the firm’s ability to finish and will not bond a firm it expects to fail. You are effectively renting a professional, financially-exposed third party’s private assessment of contractor solvency and competence — the exact hidden type you can’t see. The bond also defuses the winner’s curse: a firm that wins underwater can’t walk away cheaply; the surety completes the work and pursues the principal, so “win and default” stops being an escape hatch. Bonding is a self-selection device — only firms a surety will back can show up.

Prequalification gate (two-envelope). Open the technical/qualification envelope first; only the qualified have their price envelope opened. Screen on solvency (audited financials, bonding capacity), relevant completed work, and safety/defect record. This thins the pool to firms whose low bid is credible.

Abnormally-low-tender (ALT) challenge. Borrow the EU public-procurement device: any bid below a threshold (e.g. a set % under the engineer’s estimate or the field average) triggers a mandatory justification — the firm must show its price covers labor, materials, method, and bonding. Unjustified → rejected. This directly intercepts the lowball before it becomes a contract, and it’s procedurally defensible because the burden is on the bidder to prove the price is real, not on you to prove it’s fake.

A caution on “average-bid” awards. Awarding to the bid nearest the average (Italy, some US DOTs) does kill the incentive to lowball — but it replaces it with an incentive to collude to move the average and rewards mediocrity. Prefer ALT-justification (which removes outliers on the merits) over mechanical average-bid (which is gameable). If you use any statistical trim, trim before computing the reference and keep the rule hard to manipulate by a single bidder.

3. The incentive layer — make honest effort the firm’s best response after award

The governing idea here is the multitasking problem (Holmström–Milgrom, 1991), which is the precise economics of “cut corners”: when one dimension is measurable (cost) and another is hard to verify (quality/durability), a high-powered incentive on the measured dimension pulls effort away from the unmeasured one. So you have exactly two levers, and you should use both:

(a) Lower the power of the cost incentive where quality is unverifiable. A pure fixed-price contract is maximally high-powered and therefore maximally corner-cutting on anything an inspector can’t catch. Offer instead a menu of incentive contracts (Laffont–Tirole) that ranges from high-powered to low-powered via a cost-sharing fraction α (firm bears α of any overrun and keeps α of any saving):

Offering the menu is itself a screening device: efficient firms confident in their costs pick the fixed-price/high-α option (and accept the risk); firms facing real cost uncertainty pick more sharing rather than lowballing-and-praying. Truthful self-sorting replaces the bidder’s incentive to disguise type.

(b) Make quality measurable and contingent, so the high-powered cost incentive can’t reach it. Wherever you can convert “quality” into a verifiable outcome, do it — this is what lets you keep strong cost incentives without inviting corner-cutting:

• Retainage (retention): withhold ~5–10% of each payment, released only at acceptance and again after a defects-liability period. This is skin in the game that survives the firm’s cash-flow pressure.
• Warranty / defects-liability period: the firm is liable for latent defects for N years (construction’s worst failures surface late — exactly where a fixed price has no reach). Back the warranty with the surety or a maintenance bond. This re-internalizes the long-tail quality the contract price can’t see at acceptance.
• Performance specs with objective acceptance tests: density, compaction, concrete cylinder breaks, pavement smoothness/deflection. Pay-factor schedules (DOTs use these) scale payment to measured quality — turning the unverifiable into the verifiable and putting it inside the incentive.
• Liquidated damages + A+B bidding for time: bid both cost (A) and days (B); award on A + B × daily_road_user_cost. This prices schedule honestly and stops “cheap but endless.” Pair with liquidated damages so slippage is costed, not free.

(c) Discipline the claims-and-change-order channel, which is the lowball firm’s recovery route:

• Invest in complete design before tender — ambiguity is the raw material of claims.
• Score claims/dispute history into the scoring rule (§2), so a litigious recovery strategy is costly across contracts.
• Constrain compensable changes contractually and require pricing of change-order rates at bid time, so the firm can’t lowball the base and gouge the extras.

4. The information / risk layer — drain the winner’s curse at the source

The winner’s curse scales with the dispersion of bidders’ cost estimates. Narrow that dispersion and the curse shrinks (this is the linkage principle, Milgrom–Weber: the buyer does better by committing to release information):

• Publish a thorough Geotechnical Baseline Report and full site data. Don’t make bidders each guess the soil. Shared information collapses the estimate spread, so the winner is less likely to be the one who guessed too low.
• Don’t dump uncontrollable risk onto the contractor at fixed price. A differing-site-conditions clause that shares ground-condition risk against a published baseline is cheaper for you than a fixed price — because a contractor forced to swallow that risk either prices a fat premium (you pay for it regardless) or lowballs and claims (you pay for it through disputes). Allocate each risk to whoever can best control or absorb it; the contractor should bear execution risk, not geological lottery risk.
• Publish the engineer’s estimate or a reasonable reference, which anchors bids and supports the ALT challenge.

5. The dynamic layer — let the repeated game do the cheap enforcement

Single-shot incentives are expensive; a credible relationship is cheap. Make the game repeated:

• Past-performance scoring feeds defect rates, warranty claims, schedule, and safety back into the §2 scoring rule. Corner-cutting today lowers tomorrow’s win probability — a discipline that costs you nothing to administer once the registry exists.
• Debarment / suspension for serious or fraudulent failures, with due process. The threat of losing the whole municipal pipeline dwarfs the margin on any one slab.
• Prequalified-contractor registry with periodic re-qualification, so the pool is continuously curated.

6. Recommended package, and why honest-capable firms self-select

For a typical municipal job with moderate scope uncertainty, I’d run:

1. Prequalification gate + mandatory bid/performance/payment bonds (surety does the solvency-and-competence screening).
2. Scoring auction — Price − λ·Quality, with quality drawn from objective, pre-published criteria including past defect/claim history.
3. ALT challenge on any bid materially under the engineer’s estimate; unjustified bids rejected.
4. A menu contract — default to cost-sharing (α ≈ 0.6–0.8) on uncertain scope rather than hard fixed price; let confident firms opt up to fixed price.
5. Objective acceptance tests with pay-factors, 5–10% retainage, and a multi-year warranty bonded by the surety.
6. A+B time bidding + liquidated damages where schedule matters; GBR + differing-site-conditions clause to keep geological risk off the contractor’s fixed price.
7. Past-performance feedback + debarment to make it a repeated game.

The incentive-compatibility argument, stated plainly: a corner-cutting lowballer must now (i) get past a surety that won’t bond it, (ii) survive an ALT challenge it can’t justify, (iii) overcome a competitor’s quality score built on a clean record, (iv) cut quality that is now measured and tied to pay-factors, (v) wait out retainage and a bonded warranty that claw back the savings, and (vi) damage its standing in every future municipal tender. Each layer raises the lowballer’s cost and lowers its win probability; none of them penalizes the honest, capable firm, whose true cost is credible, whose quality score is high, and whose warranty exposure is small. The honest firm’s best response is to bid its real cost and build to spec — that is now the winning strategy, not the losing one. That is what incentive-compatible means here.

7. Failure modes to watch

• Collusion / bid-rigging. Scoring and prequalification narrow the field, which makes a cartel easier. Counter with sealed bids, rotation monitoring, screening for suspicious bid patterns, and reserve prices.
• Over-screening kills competition. Bonding capacity and prequalification can exclude small/new-but-capable firms, thinning the pool until you lose price discipline (and invite collusion). Calibrate thresholds; consider unbundling large contracts to keep entry feasible.
• Scoring discretion as a corruption vector. Keep λ and all quality criteria objective, published, and audited; subjective scoring is where capture lives.
• Gaming the reference statistic. Prefer merits-based ALT justification over mechanical average-bid awards.
• Verification cost. Pay-factors and warranties only work if you can actually test and enforce; budget for inspection and for surety follow-through, or the incentives are paper.

If you tell me the typical contract size, how much scope uncertainty you face, and how deep your local bidder pool is, I can tune the specific levers — α on the cost-sharing menu, the ALT threshold, bonding levels, and whether unbundling is worth the loss of scale.

This is a classic mechanism design problem involving Information Asymmetry. In your current setup (standard low-bid reverse auction), you are encouraging bluffing (Winner’s Curse) and selecting for high-risk agents (Adverse Selection), which leaves the contract vulnerable to underperformance (Moral Hazard).

Here is an analysis of the core economic friction points and a redesigned mechanism using Screening and Transfer Functions to align incentives.

Part 1: Diagnostic Analysis

To design a solution, we must precisely map the failure modes to economic theory.

1. The Winner’s Curse (Common-Value Underestimation)

In typical construction auctions, the “value” of the project (the true cost to complete) is related to the project parameters (soil conditions, permit delays, material volatility) which are not fully known at bid time.

• The Problem: Because bidders compete aggressively, the winner is statistically the most likely to have underestimated the cost above average. If the lowest bidder wins by too much margin, they lose money unless they anticipate defaulting on quality to reclaim the margin.
• The Diagnosis: Bidders are rational but optimistic. The auction mechanism creates a Common Value equilibrium where the highest bidder (least aggressive) is the most rational, but the Lowest Bidder holds the best information on how much they can push the cost down without going bankrupt.

2. Adverse Selection (Lemons Problem)

In outbound/procurement markets, high-quality bidders (construction firms with high overheads, low litigation risk, and sustainable practices) have higher baseline costs than “speculators” (firms desperate for revenue, high overhead risk, or skilled labor shortages).

• The Problem: Since the contract is awarded purely on Price, the “Lemons” (low-capability firms) outbid the “cherries” (high-capability firms).
• The Diagnosis: The municipality cannot screen for quality ex-ante because the Quality Type is collinear with Cost Type in the bad equilibrium. Low-cost bids = High-Risk Selection.

3. Moral Hazard (Hidden Action)

Once the contract is signed, the firm knows the municipality cannot perfectly monitor inputs (hours logged, material quality).

• The Problem: The lowball winner bids “as low as technically feasible.” To realize that profit, they must reduce quality-saving resources (input) or cut corners post-award.
• The Diagnosis: The contract leaves the contractor with insufficient incentive to maintain quality. Their utility function shifts from Total Contract Value to Marginal Profit = (Revenue - Inputs). Minimum Inputs = Maximum Utility.

Part 2: The Mechanism Design Solution

We must shift from a Pure Price-Auction to a Hybrid Screening & Risk-Adjusted Mechanism.

Phase 1: Pre-Auction Screening (The “VET-Filter”)

Standard open bidding selects the lowest price. We must restrict the set of contenders to “Capable” types to resolve Adverse Selection.

• Rule: Pre-Qualification with Mandatory Disclosure.
  • Bidders must disclose historical defect rates, financial leverage ratios, and safety record.
  • Screening Parameter ($\theta$): A firm must prove $\theta_{min}$ (Quality Capability) exists.
  • Why: This transforms the auction. If firms may be rejected for non-compliance based on quality history, the “lemon” types (who lack quality history) opt out immediately. Only high-cost, high-quality firms enter the pool.

Phase 2: The Award Mechanism (Incentive-Compatibility)

To fix the Winner’s Curse, we must flatten the information asymmetry regarding the true project cost.

• Revelation Principle Implementation: We create a Risk-Adjusted Reverse Auction.
• The Two-Envelope Rule:
  1. Envelope A (Price): Sealed price bid.
  2. Envelope B (Risk Margin): A percentage “Absorption Capacity” bid. This defines the penalty the firm is willing to absorb if actual costs exceed estimates without defaulting.
• Auction Rule:
  • Do not select the absolute lowest bid.
  • Select the bid that satisfies the Municipal Target Cost ($T$) where $T > \text{Historical Average} \times \alpha$.
  • Cutoff Rule: If the winning bid is lower than $0.8 \times T$, the bid is invalidated.
  • Why: This protects against Winner’s Curse. It prevents the municipality from being forced into a contract where the price is physically impossible to maintain quality standards.

Phase 3: Payment Structure (Fixing Moral Hazard)

To fix Moral Hazard, the payment must be tied to Output Quality, not just Output Quantity/Budget.

• Mechanism: “Target Cost + Gain Sharing” Contract.
  • Standard Price: $P_{target}$ (Based on historical data, not lowest bid).
  • The Incentive:
    • If Final Cost < $P_{target}$: Contractor keeps 70% of savings (Gain Sharing).
    • If Final Cost > $P_{target}$: Contractor bears 50% of overage (up to $P_{target}$).
    • Quality Offset: Any final penalty for non-compliance with quality standards (code violations, rework) reduces the Gain Sharing, not just the profit.
  • Incentive Alignment:
    • Aggressive Corner Cutting: Increases the risk of Quality Penalties. If they cut corners, they trigger rework. Since they are sharing savings, they lose the 70% bonus.
    • Honest Costing: If they earn actual costs, they win the penalty buffer without quality risk.

Part 3: Detailed Contract & Rule Specifications

Here is the operational design to implement the above theory.

Rule 1: Mandatory Bonding & Insurance

• Action: Require a Performance Bond (90% + 10%) and Warranty Bond.
• Reasoning: If the winner cuts corners (Moral Hazard) and leaves, the municipality taps the bond. This adds significant cost to “skimmers,” driving them to drop out in favor of firms with higher operational capital.

Rule 2: The “Re-verification” Audit (Hard Audit)

• Action: Include a clause requiring a third-party independent engineer (City’s choice) to sign off on every phase payment.
• Incentive: The contractor knows their claims are being audited for “speculative” man-hours. Biddings will be higher, but bids will be more accurate.
• Mechanism: Introduce a Price Adjustment Factor (PAF). The final payout is: $$ \text{Payment} = \text{Billed Cost} \times (1 - \lambda \cdot \text{Discrepancy}) $$ Where $\lambda$ (Audit Shock) is a known number. This discourages padded invoices post-award.

Rule 3: Retainage for Defects (Long-Term Horizon)

• Action: Hold 10-15% Retainage for 12 months.
• Reasoning: Low-margin firms cannot afford the 10% hold. They will cut corners.
• Mechanism: Pay 50% upfront, 25% progress, 25% final.
  • Clayton Plan: Ensure that the “Change from Construction/Inspection” (CFI) cannot proceed unless payee has all necessary permits.

Rule 4: Disclosure of “False Economy”

• Action: Public announcement of Winning Contract Price is fine, but add a Public Disclosure of Quality Performance.
• Long-term Logic: This creates a reputational cost. If the municipality leaks specific project quality (e.g., “X firm won but failed inspection 2 years later”), other firms may refuse to bid in subsequent auctions, forcing previous winners to perform better.

Part 4: Why This Works (The Pitch)

The current approach fails because it treats Price as the only dimension of value. You must sell your new design as “Value-Based Competitive Screening.”

To present this to City Council:

1. Reduction of Risk, Not Cost: “We will slightly increase the upfront cost to 5-10%, but it secures a quality outcome with 0 probability of default.”
2. Safeguarding against Lowballing: The “Risk Margin” envelope ensures firms aren’t betting their solvency on a margin that isn’t in the budget.
3. Stopping the Corner-Cuting Race: The “Target Cost” contract means the firm earns money when they control waste. They don’t need to cut corners to make profit; the profit comes from the efficiency bonus, which protects their safety standards.
4. Incentive Compatibility: If firms are honest about costs:
   • High-quality firms bid their cost + risk margin. They win.
   • Low-quality firms (who need max margin to survive) bid high and lose (cannot win against high-quality firms if the threshold is too high) OR they bid low and risk losing the bond/audit.

Summary Checklist for Implementation

1. Auction Format: Two-envelope format (Price + Quality Bond).
2. Selection Criteria: Lowest price subject to a Minimum Performance Score (derived from historical defect data).
3. Contract Type: Unit Price Contract with Earned Value Tracking.
4. Key Metric: Ensure the Total Cost Ownership (TCO) metric (Price + Audit + Warranty) is maximized, not just Price.

By shifting to a mechanism that rewards Cost Certainty over Cost Aggression, you flip the Winner’s Curse. High-quality firms love price certainty; low-quality firms demand corner-cutting margins. By offering constant payment for constant quality, you select the former and exclude the latter.

Parties and the asymmetry

Parties: Contractor (Agent) and Municipality (Principal).

Private information / hidden action: Contract (Agent): Holds private information regarding true construction cost estimates ($C$), quality investment capacity, and risk tolerance; undertakes hidden action regarding material specifications and schedule compliance. (Principal): Holds private information regarding maximum project tolerance and budget constraints ($V_{max}$); observes only final bid price, milestone reports, and public performance history post-award.

Status: Asymmetry is explicit on both sides; prior full-information assumptions are rejected.

Selection vs. hazard

Adverse selection (hidden type): Pre-contract hidden type. Distortion is driven by risk-prone firms underbidding (lowballing) to enter pool; honest firms bid conservatively and exit; winner correlates with lower quality capability.

Moral hazard (hidden action): Post-contract hidden action. Distortion is driven by awarded bidder reducing effort/quality to recover margin gains ($\delta$) where inspection is imperfect and monitoring is costly.

The distortion

• Current Mechanism Failure: “Lowest Price Wins” rule incentivizes risk-prone firms to bid $C - \delta$ against honest bids $C$.
• Winner’s Curse: Aggressive partials win bids even with underestimated costs, increasing post-award run-rate risk and municipal funding shortfall.
• Corner-Cutting: Post-award, risk-prone bidder reduces quality to $\delta$ savings; cost saving exceeds verification cost, leading to municipal quality loss.
• Pool Degradation: Akerlof Lemons Spiral dynamics where honest firms exit or must lowball, leaving a pool dominated by risk-takers.
• Correlation: Municipal spending correlates inversely with project quality.

Named mechanisms in play

Screening Mechanism: 2-Stage Auction (Stage 1 Cost Transparency/Audit + Stage 2 Quality Tier/Competitive Bid). Stage 1 requires detailed cost breakdowns (labor, materials) and rejection of bids falling outside $\pm 10%$ benchmark tolerance.

Signaling Mechanism: Upfront resource commitment (e.g., advance non-returnable material deposits, parent corporate bonds) to separate high-quality bidders from those unable to afford risk.

Principal-Agent Alignment: Fixed % escrow bond held separate from operational funds; released only upon third-party verification at milestones.

Winner’s Curse Mitigation: Second-lowest bid threshold rule coupled with mandatory minimum viable bid ($C_{base} +$ Contingency Margin) to prevent rationally unsustainable prices.

Mechanism (for a design)

Proposal: 2-Stage Auction (Stage 1 Cost Transparency/Audit + Stage 2 Quality Tier/Competitive Bid). Stage 1 requires detailed cost breakdowns (labor, materials) and rejection of bids falling outside $\pm 10%$ benchmark tolerance. Upfront resource commitment (e.g., advance non-returnable material deposits, parent corporate bonds) to separate high-quality bidders from those unable to afford risk. Fixed % escrow bond held separate from operational funds; released only upon third-party verification at milestones. Second-lowest bid threshold rule coupled with mandatory minimum viable bid ($C_{base} +$ Contingency Margin) to prevent rationally unsustainable prices.

Participation constraint: Honest bidders’ expected profit must exceed outside-option ($\pi_{curr} > \pi_{outside}$); requires calibrating Quality Reserve/Margin to historical municipal profitability rates.

Incentive-compatibility: For risk-prone bidders, expected financial loss ($E[\text{Forfeiture}]$) must exceed potential savings from evasion ($\delta$). Non-linear forfeiture function ($Q \approx 0.7-1.0$) with bounded max cap (100%) ensures enforceability. Third-party inspection schedule tied to escrow releases (Mobilization, Materials, Structure, Final). Calibration Requirement: Specific thresholds (bond %, tolerance $\pm$%) depend on internal historical cost/loss data; public benchmarks (e.g., Chicago DPS) serve as proxy if internal data unavailable.

Residual gaming surface: Contract terms locked in Stage 1 to prevent post-audit offers for different service levels. Verified subcontractor list required to prevent fly-by-night mid-project swaps. Default thresholds set to industry-wide standards (if internal data unavailable) to reduce risk of arbitrary floor-setting. Escrow re-insurer/third-party oversight required to counter administrative bypass. Change order caps and upfront escrow set to prevent profitable scope creep.

Read

Conclusion: The current “Lowest Price Wins” arrangement misfires because it incentivizes risk-prone firms to bid $C - \delta$ against honest bids $C$, resulting in Municipal spending that correlates inversely with project quality and a Pool Degradation where honest firms exit. The proposed design achieves Incentive Compatibility by ensuring risk-prone bidders face expected financial loss ($E[\text{Forfeiture}]$) exceeding potential savings from evasion ($\delta$), and satisfies Participation Constraints by calibrating Quality Reserve/Margin to historical municipal profitability rates.

Posture: Analysis (Distortion/Selection vs Hazard) + Design (Mechanism/Constraints). Confidence is High on mechanism design principles mapping to standard categories (Adverse Selection + Moral Hazard), while specific parameter values (bond %, forfeiture curve steepness) are Medium Confidence expectations pending empirical calibration of internal historical data.

Confidence and assumptions

High Confidence: Mechanism design principles (Screening + IC) are established; mapping problem to standard categories (Adverse Selection + Moral Hazard) is robust.

Medium Confidence: Specific parameter values (bond %, forfeiture curve steepness) are plausible proxies pending empirical calibration.

Missing Data (Gaps): Municipal specific historical loss rates ($\delta$) and project cost medians are not in vault (public data only provides general proxies). Honest bidder’s specific outside-option profit rates ($\pi_{outside}$) unverified.

Assumptions: Honest bidders exist in current bidder pool; screening cost is less than quality improvement value.

Parties and the asymmetry

Parties: The municipal procurement authority (Principal) and construction firms (Agents).

Private information / hidden action: Firms hold private pre-contract information across three dimensions: cost type (true cost structure, overhead loading, backlog, equipment availability), quality type (technical capacity, workforce skill, management depth, supply-chain reliability), and strategy type (intent to recover margin through honest execution versus post-award extraction via change orders, corner-cutting, or opportunistic default). Post-contract hidden actions include material substitution, supervision/effort levels, defect concealment, and adherence to safety/timeline protocols, which are unobservable or prohibitively costly to monitor continuously.

When: Before contracting (type); after contracting (action). The observed “burning” is the combined output of both asymmetries.

Selection vs. hazard

Adverse selection (hidden type): A lowest-bid auction applies a single price signal across a heterogeneous population. Honest, capable (H-type) firms will not bid below the true cost of quality delivery plus a normal risk premium. Lowballer (L-type) firms—those who either misestimate costs or intend to recover via change orders, corner-cutting, or default—bid below cost because they possess a profitable post-award exit path. The contract is awarded to the L-type. This operates as a lemons dynamic: H-types learn that participation is unprofitable and exit, self-reinforcing the pool toward the exact firms the municipality wishes to avoid.

Moral hazard (hidden action): Once an L-type wins, the bid is unprofitable if executed to specification. Because effort and material quality are partially unobservable, the firm cuts corners to avoid loss, transferring the cost (rework, delay, default, municipal remediation) to the public. The firm is partially shielded from the downside, as bonds are rarely drawn, defects often emerge after the warranty period, and subsequent contracts are pursued in different jurisdictions. Authority over execution is separated from cost-bearing.

Both asymmetries are load-bearing. Treating pre-contract type sorting and post-contract effort as a single problem misses half the mechanism; they require distinct, layered countermeasures.

The distortion

• Degenerating-pool trajectory: In Round 1, a lowballer bids ~20% below honest cost, wins, recovers margin via change orders and corner-cutting, and delivers before a defect surfaces in year two. In Round 2, the municipality tightens the specification. The honest firm bids true cost plus margin, while the lowballer underbids by ~20% again and wins. The municipality pays more in change orders than it saved on the nominal bid. By Round 3, the honest firm, observing that winning at an honest price is impossible, either exits, lowers quality to compete, or adopts an aggressive change-order strategy. The degenerating pool is the equilibrium outcome of the auction as designed.
• Who overpays and who exits: The public overpays through remediation and change-order leakage that exceeds nominal bid savings. Honest firms exit the market, and high-risk firms concentrate. The outcome is a pooling failure that the mechanism must convert into a separating equilibrium.

Named mechanisms in play

• Winner’s curse: Operation here: The classic common-value winner’s curse occurs when bidders estimate an unknown true project value and the most optimistically mistaken bidder wins, ex-post experiencing loss despite honest intent, cured by better ex-ante information aggregation. The problem described is distinct: asymmetric-information lowballing, where the bidder privately knows their intent to cut corners or exploit change orders. This is an adverse-selection problem cured by screening and bonding, not information sharing. In a competitive lowest-bid auction, winning is itself a bad-news signal about type, reflecting either the most optimistic estimate, the most aggressive extraction assumption, or the most desperate need for work.
• Screening: Operation here: A substantive prequalification gate set before any price is seen, requiring verified bonding capacity from a rated surety, audited financials showing adequate working capital and liquidity, documented relevant experience with named personnel, safety records below threshold, and beneficial-ownership disclosure. The cost of qualifying is asymmetric. H-types accept it because it documents capacity they already hold. L-types either cannot meet the gate (lacking financials or bonding capacity) or find meeting it uneconomical, causing their expected value of participation to turn negative and driving self-selection out.
• Signaling: Operation here: Requiring an itemized bid breakdown (labor, materials, equipment, overhead, profit separated) makes revealing actual cost cheap and truthful for H-types, but costly and risky for L-types, who must construct a cost structure that survives auditable scrutiny (e.g., CPA certification for bids exceeding an anomaly threshold). Past performance and bonding capacity also function as credible signals of hidden type that low-quality firms cannot costlessly mimic.
• Principal-agent (skin-in-the-game): Operation here: Post-contract mechanisms re-couple the contractor’s downside to its effort and quality choices. Performance bonds, liquidated damages tied to observable milestones, substantial holdback/retention, extended-warranty bonds, and strict change-order discipline internalize the cost of delay, shirking, or corner-cutting, making extraction a dominated strategy.

Mechanism (for a design)

Proposal: Replace lowest-bid-wins with a layered, two-stage best-value multi-attribute procurement anchored to statutory bonding regimes (e.g., Miller Act or state Little Miller Act thresholds).

1. Stage 1 (Prequalification): Mandatory verified single-project and aggregate bonding capacity from a rated/T-listed surety, audited financials, documented experience, strict safety thresholds, and beneficial-ownership disclosure.
2. Stage 2 (Evaluation): Score technical proposals (methodology, schedule logic, personnel, QA/QC plan, risk register). Proposals below a quality threshold are eliminated. Technically qualified bidders proceed to price evaluation, awarded on a quality-adjusted score: Score = w_T × Technical + w_P × normalized(Price), with w_T typically in the 0.4–0.6 range. Utilize a Vickrey-style (second-score) payment rule within this framework to restore dominant-strategy incentive compatibility.
3. Post-award controls: Mandate 100% performance and payment bonds; liquidated damages; substantial holdback/retention (~10–15%) released only after defects-liability periods; flow-down bonding and retainage for major subcontractors; and strict change-order discipline (owner-initiated at cost + fixed fee, contractor-initiated requiring independent verification, with aggregate caps). Incorporate positive incentives such as quality bonuses, early-completion bonuses, and shared savings to preserve H-type margins.

Participation constraint: H-type expected utility from bidding must be ≥ 0, covering true cost, risk premium, and normal profit, beating the outside option of private work or other jurisdictions. Prequalification excludes L-types, raising H-type win probability at honest bids. The technical gate allows H-types to win on quality, not just price. The contract’s positive incentives (bonuses, shared savings) and the long-game return of a relational past-performance record ensure the inside option of honest bidding remains profitable.

Incentive-compatibility: H-type’s best response must be to bid and execute honestly; L-type’s best response must be to not bid or to bid honestly and accept losing. For an L-type with a true cost higher than the bid price, the expected utility of winning and cheating must be negative. This is enforced by sizing bond premiums, holdbacks, and penalties to exceed the short-term corner-cutting savings. If the L-type slips through, extraction is constrained, downside is large, and future contracts are jeopardized, collapsing their expected value across the relevant horizon.

Residual gaming surface:

• Burn and walk away (bankruptcy): Mitigated by phased bonding with progressive release tied to milestones, joint-and-several liability for principals of small-firm bidders, and escrow/trust accounts for progress payments.
• Shell-firm re-entry: Mitigated by beneficial-ownership disclosure and exclusion of affiliated entities.
• Bait-and-switch subcontracting: Mitigated by strict key-personnel and subcontractor approval clauses with flow-down bonding/retainage.
• Collusion on technical scoring: Mitigated by open published criteria, independent technical panels, scoring-variance monitoring, and whistleblower protection.
• Evaluator weight manipulation: Mitigated by pre-committed, published weights and audit trails, though vulnerability to weight manipulation remains a known limit of multi-attribute auctions.
• Political override: Officials may waive bonds or award to non-qualified bidders. The mechanism cannot supply political will; this requires transparency, published criteria, and audit trails to deter.

Read

Posture: This is a mechanism design proposal resting on an explicit diagnostic foundation. The transition from explaining why the current arrangement misfires to prescribing the fix is marked, not blurred.

The current arrangement misfires because it treats a two-part information problem as a simple price-minimization problem. By applying a single price signal to a heterogeneous pool without screening, the municipality systematically selects L-types (adverse selection) and leaves them unmonitored post-award (moral hazard). The proposed layered mechanism converts this pooling failure into a separating equilibrium. It forces L-types to either reveal their true cost (making them uncompetitive) or exit the bidding process entirely, while simultaneously rewarding H-types for technical quality and honest execution.

Confidence and assumptions

Confidence: High on the classification of the asymmetries (adverse selection pre-contract, moral hazard post-contract) and the applicability of standard procurement mechanisms (screening, signaling, principal-agent). Medium on specific parameter calibration (e.g., exact w_T/w_P weighting, holdback percentages, defects-liability periods). These are conventional ranges requiring calibration to local market conditions, project scale, and jurisdictional norms.

Load-bearing assumptions:

1. The municipality has the administrative capacity to enforce multi-attribute evaluation, manage retainage, and consult past-performance records.
2. The municipality can credibly commit to enforcement—bonds are actually drawn, and disqualifications are actually enforced.
3. A rigorous, competitive surety market exists to serve as the delegated monitor.

If the institution is captured or politically volatile, the design fails from its environment, not its logic. This is the single assumption requiring validation before implementation.

Interpretive assumptions: “Honest” is interpreted as truthful cost and capacity revelation; “capable” as solvent and technically qualified; “self-select” as screening-driven exit by low-quality bidders; “burning” as post-award financial or quality loss.

Scope boundaries: This design does not draft specific bid documents, scoring rubrics, or contract clause language, nor does it resolve every statutory requirement in every jurisdiction (e.g., specific Little Miller Act thresholds, prevailing-wage rules, or local-content mandates). It identifies the mechanism; successful implementation requires downstream legal review, technical advisory input, and stakeholder engagement.

Parties and the Asymmetry

A note on posture before the substance: Parts 1–2 below are analysis — explaining why the current arrangement misfires, tracing a distortion. Parts 3–5 are design — proposing a fix, where the obligation is to satisfy participation and incentive-compatibility and to survive gaming. The switch is marked explicitly at the section-3 boundary; the two postures are not blurred.

Parties:

• Principal: the municipal authority (buyer). It wants spec-compliant construction at a defensible price; it sees only the bid number.
• Agents: bidding firms of two types — H (capable, intends to deliver to spec, prices compliant delivery) and L (will cut corners and/or is thinly capitalized, prices the corner-cut version).

Private information / hidden action:

• Hidden information (pre-contract): each firm privately knows its own type — its true cost of compliant delivery and its intent/capacity to deliver it. The authority cannot read type from a sealed bid. → adverse selection.
• Hidden action (post-contract): the winner privately chooses a quality/effort level (concrete mix, rebar cover/spacing, compaction, curing time, waterproofing), much of which is buried and surfaces only as latent defects years later. The authority cannot cheaply observe it in real time. → moral hazard.

When: type is known before contracting (→ selection); action is exercised after contracting (→ hazard). The two are distinct and must be kept separate: selection is about which firm wins; hazard is about what the winner then does. Screening cannot cure hazard; incentives cannot cure selection. Conflating them is the main analytical trap and would misdirect the fix.

Selection vs. Hazard

Adverse selection (§1 — pre-contract, hidden type)

Two compounding forces both push bids down:

• Intent/cost-driven lowballing. A firm pricing against its cost-of-cutting-corners (or planning to recover margin through change-order claims) has a genuinely lower expected performance cost, so at any headline bid its margin is higher; it can profitably bid below H’s honest compliant cost. The low bid is strategic understatement, not estimation error.
• Winner’s curse proper. With a partly common cost component (unforeseen conditions, material moves), each bidder’s estimate is true-cost-plus-noise; under “lowest price wins” the winning bid is mechanically the most optimistic draw. Even among honest firms the winner is the one who most underestimated cost — the auction format itself selects for the error.

These compound: the winner combines the most optimism and the most willingness to absorb the gap by cutting unobservable quality. A low bid is therefore a negative signal, and the lowest-price rule rewards exactly that signal — winning is bad news about the winner.

Why H exits: H prices full compliant cost plus normal margin, so its number structurally sits above L’s lowball; under lowest-price-wins it loses repeatedly. Bid preparation is costly; once expected win probability approaches zero, expected value of bidding goes negative and H stops entering. This is Akerlof unraveling — the degenerative spiral running on the bid distribution — where the price rule drives out the type the buyer wants, accelerating each round “the low bid won again.” The change-order ecosystem widens the gap: L bids low knowing variations recover margin; H won’t play that game.

Moral hazard (§2 — post-contract, hidden action)

Type is settled; the hidden variable is the winner’s quality effort. A fixed-price contract fails to compel quality because each enforcement link leaks:

• Residual-claimant / timing asymmetry. Under fixed price, every dollar saved by cutting an unobservable corner is immediate certain profit; the countervailing defect downside is probabilistic (may not be detected), delayed (latent defects surface in years), and escapable (a judgment-proof or dissolved firm). Large certain upside vs. small delayed probabilistic downside makes cutting rational regardless of firm character.
• Non-verifiability. Quality is multidimensional and buried; courts need verifiable, causally-attributable evidence, which latent defects lack.
• Enforcement-cost asymmetry. Expected penalty = P(detect) × P(prove | detect) × sanction × P(collectable). Thin, scheduled, visual (not destructive) inspection makes P(detect) low; causation proof is slow and costly so P(prove) is low. Expected penalty sits far below cost saved by cutting. The cost of proof — not the absence of a clause — is the enforcement gap.
• Limited liability / judgment-proofing. A thin-cap L can cut, collect, and dissolve; the sanction is uncollectable, capping the downside near zero.
• Payment timing. Milestone/completion payment against visible completion rather than verified quality banks the gain before defects are testable.
• Monitoring gaps specifically: under-resourced inspectorate; announced/gameable inspections; visual sign-off instead of core samples/compaction tests at the moment work is about to be covered; retention too small and released too early; no skin in the game past handover.

The IC condition for quality fails: private cut-saving > private expected penalty.

The Distortion

• Good actors (H) exit; the winner is systematically L (or an over-optimistic H that runs out of money).
• The buyer’s true cost (lowball + change orders + defect/remediation + litigation) exceeds what an honest compliant bid would have cost.
• The pool degrades toward the firms the buyer least wants — a separating-by-default into the wrong type.
• Scope boundary: supply-chain/material-price escalation is an exogenous common shock, not a bidder-type problem — it is handled separately by indexation clauses, not by screening. If most overrun history is escalation rather than bidder behavior, indexation and complete documents do more work than screening.

Named Mechanisms in Play

• Principal–agent frame (ruled IN, post-award): the city pays against observable proxies (milestones, visible finish); the agent optimizes the proxy, not the unobservable outcome (durable quality); decision authority over build choices is separated from cost-bearing for defects — the defining condition. The cure is to re-couple them.
• Skin in the game (Taleb, Skin in the Game/Incerto): the party making build decisions must bear their consequences. Concept attribution is Taleb’s; “re-coupling” is the analyst’s own descriptor for applying it here — converting the defect downside from delayed, probabilistic, escapable into immediate, certain, inescapable.
• Signaling — ruled OUT as the firms’-side problem: firms aren’t trying and failing to signal quality; the lowest-price rule gives them no channel and no reward to do so. The fix must come from the uninformed side (screening).
• Screening — ruled IN as the load-bearing pre-award frame: the uninformed principal sets a menu/requirement and lets agents sort themselves through it.
• Buyer-required surety bond — classified as screening, not signaling: the principal imposes the requirement; the surety market does type-discrimination, underwriting only firms it judges able to complete and pricing premium by risk (L pays more or can’t be bonded). It exploits a signaling-style single-crossing condition (the bond is costlier for L than H) but is principal-initiated, so the term is “screening,” reserving “signal” for agent-initiated moves.
• Average-bid / closest-to-mean auction — ruled OUT. The honest discriminator is not collusion-proneness — the chosen design (prequalification + scoring + bonding) shrinks the pool and carries the same collusion exposure, so collusion-proneness cannot be the ground without indicting the chosen design. The real reason: average-bid rewards clustering at mediocrity and gives quality no channel — it punishes every deviation from the herd, deterring the wanted firm as hard as the unwanted one. The scoring + abnormally-low-justification combination disarms the lowball while still rewarding quality.
• Engineer’s-estimate disclosure — ruled IN (as a winner’s-curse/common-value anchor): publishing the authority’s cost estimate range collapses the common-value optimism gap so bidders update toward a shared cost basis and the lowest bid is less likely to be merely the most over-optimistic; it pairs with the ALB rule (the disclosed estimate is the trigger anchor). Caveat: can become a focal point for collusive convergence — combine with bid-pattern monitoring.
• A+B (cost-plus-time) bidding — ruled IN conditionally: bidders compete on price (A) plus monetized schedule (B = days × daily road-user/occupancy cost); use only where schedule is itself a corner-cutting margin (e.g., curing time skipped to finish faster), making the time dimension priced and visible. Where schedule is not a quality lever it adds complexity without addressing the type problem — leave it out.
• Vickrey/second-price sealed bid — ruled OUT for this setting: second-price truthful-bidding logic addresses valuation honesty, not quality-type honesty; it does nothing about the corner-cutting margin and is rarely used in public works for transparency/collusion reasons.

Mechanism (the Design)

This is the design posture. Both constraints must hold throughout: participation (IR) — a high-quality firm must expect non-negative profit from bidding and from winning at an honest price; incentive-compatibility (IC) — a low-quality firm must find mimicking a high-quality bid unprofitable.

Proposal — pre-award screening (§3)

(a) Scoring (best-value / multi-attribute) auction replacing lowest-price. Award on a published formula combining price with verifiable quality dimensions (delivered past performance, key-personnel credentials, financial capacity, technical method, schedule realism). This pulls the hidden quality dimension into the award rule so an honest higher price can win. Score-IC refinement: split inputs into cost-asymmetric ones genuinely cheaper for H to produce (verifiable multi-year defect-free record, named senior staff with attributable history, in-house QA capacity, bonding capacity) vs. cheaply-faked ones (paper qualifications, borrowed references, optimistic method statements). Only the first class screens; the second is decorative. Weighting cuts both ways: price-dominant collapses to lowest-price-with-extra-steps; quality-dominant rewards gold-plating. Treat the score as a coarse filter built on cost-asymmetric inputs; the real separating load sits on the bond and warranty. The score must be objective/auditable (defined rubrics, documented references) or it becomes a corruption surface.

(b) Prequalification gate. Hard floors: licensing, bonding capacity, minimum portfolio of projects delivered to spec, audited financials. Screens out the judgment-proof, thin-cap firms whose limited liability drives both the lowball and the corner-cutting; ensures a counterparty with collectable assets.

(c) Surety bonding — outsourced screening / cost-asymmetric instrument. Require performance (and payment) bonds from a rated surety: it screens (a firm that can’t be bonded can’t bid) and produces a cost-asymmetric instrument (cheap for genuine H, expensive/unobtainable for L) — renting the surety’s private information. Genuineness condition: this is a real screen only if the surety underwrites on information the city cannot cheaply access — audited financials, bonding-capacity history, prior claims experience, completed-work record under the surety’s own monitoring. If the surety reads only the same public prequalification packet, the bond is a relabel, not a screen — it merely relocates the hidden-type problem to the surety. Confirm sureties underwrite on private information before leaning on the bond.

• Degraded mode — thin or non-discriminating surety market. This screen rests entirely on the surety-discrimination assumption; the developing-market case (e.g., the Ghana lowballing context) is exactly where surety depth can’t be assumed. Substitute a self-funded bond surrogate: a refundable cash/escrow performance deposit scaled to the bid (or bid-bond-plus-escrow), held in interest-bearing escrow, released against the same defects horizon as the holdback. Separating logic survives because the deposit ties up capital L can less afford to risk against its higher P(forfeit). A mutual-guarantee pool (members co-underwrite and bear losses) is a second option. Calibration constraint (governs over the robustness gain): the surrogate is a real working-capital cost — size it to bind on L without walling out thinly-capitalized H; where small-H IR would break, lean harder on the verifiable-record score (a) and the registry (§4e) and lighter on the cash deposit.

(c-warranty) Bonded multi-year latent-defect warranty — the core separating device. Require each bidder to commit to a warranty (e.g., 5–10 years on structural elements) backed by retention or a warranty bond. Cleanest cost-asymmetric signal: for a firm that will deliver, expected warranty cost is low so the commitment is cheap; for a corner-cutter, expected claims are high so it is expensive — only H can afford to offer it (separating equilibrium). It reaches through the award into the moral-hazard problem (§4). Load-bearing caveat: this separates only the solvent, going-concern low type; a judgment-proof firm discounts future warranty/bond liability toward zero, so the warranty does not separate the phoenix type by itself — the officer/principal-attached liability and rated-surety conditions are part of the separating argument, not residual cleanup.

(d) Winner’s-curse direct attacks. (i) Reduce common-value noise: publish the engineer’s estimate range, geotechnical data, and complete construction documents before tender — incomplete drawings are a named owner mistake driving overruns and the change-order channel lowballers exploit. Completing documents is a no-cost-to-bidders curse reducer. (ii) Abnormally-low-bid (ALB) justification rule: any bid below a trigger (a set % under the engineer’s estimate, or a statistical outlier against the field) must be justified — demonstrate the cost basis for compliant delivery; fail → rejected. This converts “lowest wins” into “lowest defensible wins,” removing the payoff to strategic understatement and defusing winner’s-curse selection. Anchor the trigger to the engineer’s estimate (not only field average) and treat it as a soft prompt for scrutiny, not a hard cliff.

Proposal — post-award incentives (§4)

Principle: re-couple the firm’s defect downside (skin in the game) — convert it from delayed/probabilistic/escapable to immediate/certain/inescapable.

(a) Holdback/retainage tied to the latent-defect horizon. Withhold retention (commonly 5–10%; 10% typical/often statutorily required on public contracts, e.g. DC, often reducible at 50% completion) released only after a defects-liability period long enough for latent defects to surface (12–24 months is a conventional general span, locally calibrated; longer for structural). Release milestone payments against inspected quality gates (including buried/structural work before it is covered), not visible completion. Aligns payment timing with quality realization, countering the “paid before defects surface” gap.

(b) Warranty/maintenance bond callable on defect; bonds as private enforcement. A rated surety stands behind remediation and then pursues the firm — a collectable third party backs quality even if the firm dissolves; size to remediation cost, not just completion. The surety, protecting its own exposure, polices the contractor during construction — converting the city’s costly monitoring into the surety’s self-interested monitoring. (Where surety markets are thin, the §3c escrow surrogate carries this load.)

(c) Independent destructive testing at burial points, not visual sign-off. Pay milestones against third-party test results (core samples, compaction tests) taken before work is covered. Raises P(detect) at the moment detection is cheapest and the firm’s gain from cutting is largest.

(d) Pre-defined liquidated damages + remediation protocol. Pre-agreed damages for specified defect classes (and delay) so the authority needn’t prove consequential loss — lifts the P(prove) term that currently kills enforcement; remediation at firm’s cost is the default, bond-backstopped. Decennial latent-defect liability/insurance — a civil-law statutory device (France, UAE, much of the Middle East) with no direct common-law parallel — extends collectable structural liability to ten years; consider only where the jurisdiction actually carries it.

(e) Past-performance registry / reputation — the hinge to pre-award screening and strongest long-run lever. Record verified delivered quality in a registry feeding future scoring (3a) and prequalification, converting a one-shot game (where defection dominates) into a repeated one: cutting corners today lowers future win probability, and for any firm intending to keep bidding the discounted future revenue loss from a bad record exceeds the one-shot cut-saving. Condition: works only if records are credible and tied to the firm’s principals/officers, not just the corporate shell (else phoenix firms shed history). Entry-barrier tension: the registry and prequalification gate jointly build an entry barrier — a genuinely capable new H with no track record scores poorly and risks failing the very screen meant to protect H (re-shrinking the pool and easing cartel risk). Counter with a new-entrant on-ramp: (i) a probationary score band admitting unproven firms to a defined slice of work with full bonding/testing but neutral (not penalized) reputation weight; (ii) weighted credit for verified key-personnel records where the firm has none (capability travels with people); (iii) the surety/escrow position as substitute assurance for missing firm history. The on-ramp must admit capable entrants without re-admitting L through the back door — probationary entrants still face the full post-award incentive stack.

(f) Material-cost indexation clause for verifiable external shocks. Supply-chain/material-price escalation is a cost driver independent of bidder behavior; a rigid fixed price forces honest bidders either to pad (and lose) or get squeezed into corner-cutting when prices spike. A narrow indexation clause for documented, indexed material moves removes a genuine corner-cutting pressure on honest firms without rewarding inefficiency — separating “bidder lied” from “the world moved.”

Participation constraint (why H joins) — IC verification §5

Participation (IR) holds for H. Scoring (3a) + ALB (3d) restore H’s win probability at a compensatory price; complete documents and published estimate shrink the curse so H isn’t punished for bidding realistically; its warranty/bond cost is low because it will deliver; indexation protects against external shocks. Expected profit from bidding turns positive, H re-enters, and the spiral reverses. Binding tension: bonds (3c, 4b), the escrow surrogate, and holdback (4a) are real working-capital costs; piled too high they violate H’s IR and screen out small-but-capable H, re-shrinking the pool. Holdback %, bond/deposit size, and defects-liability length must be calibrated so H’s expected profit stays positive — this is the binding design constraint governing the calibration of every cost-adding screen, including the §3c fallback and the §4e on-ramp.

Incentive-compatibility (why honesty/effort is the best response) — IC verification §5

IC pre-award (self-selection) holds against the solvent low type. Bond premium/escrow forfeiture-risk and prequalification cost are higher for L (single-crossing). Under the ALB rule an L bidding the corner-cut price must either (i) justify it — and can’t, since compliant delivery genuinely costs more — and is rejected, or (ii) bid the compliant price — surrendering its only edge and losing on the quality score. Either branch yields L no profit from masquerading. L’s remaining options are don’t bid (self-selects out — success) or bid honest spec-cost (becomes a higher-cost honest firm; distortion gone — success). Separating equilibrium.

IC post-award (effort) holds. Compliance is the best response when: cut-saving < P(detect via independent test) × (holdback/escrow + LD + bond consequences) + discounted future revenue loss from the registry. Independent destructive testing (4c) raises the P(detect) term; the registry (4e) raises the future-loss term; both are tunable upward until the inequality flips.

The separating engine is the bond-and-warranty cost asymmetry — cheap for the type that will deliver, expensive for the type that won’t. The equilibrium is stable as long as that differential exceeds the lowball profit for L and stays below it for H — the single-crossing condition the whole design rests on.

Single-crossing satisfiability heuristic. The type-cost differential scales with the unobservable fraction of the work (buried, structural, latent-defect-prone, where cutting pays and detection lags). It is most likely to hold on heavy-civil and structural projects (deep foundations, water/wastewater, bridges) and most likely to fail on cosmetic finish-out, where little is concealed and the warranty-cost gap is small. Project-type heuristic: deploy the full apparatus where the hidden fraction is large; for finish-dominated work a lighter screen suffices and the bonding/warranty overhead may not pay for itself.

Cost-effectiveness ordering of the IC knobs. The registry future-loss term is near-zero marginal cost once built — turn it up first by widening future-scoring weight on a clean/bad record. Independent destructive testing has recurring per-project cost but highest leverage on P(detect) — fund it second, concentrated at burial points. Holdback/escrow size is the most expensive knob in IR terms — it ties up the firm’s capital and trades directly against small-H participation, so raise it last and only as far as the IR constraint allows.

Residual gaming surface (cleverest defections still permitted)

• Cartel formation / pool-shrinkage. Prequalification and scoring shrink the pool, raising bid-rigging/rotation risk and possibly prices (the Akerlof-screening cost trap, a corrective overshoot); the registry’s entry barrier and any disclosed engineer’s estimate compound it. This is a genuine cost the chosen design carries — not a solved problem — which is why §3 does not lean on collusion-proneness to reject average-bid. Mitigate: reserve price, periodic re-opening of prequalification, the new-entrant on-ramp, statistical bid-pattern/clustering monitoring; calibrate screening to the asymmetry’s actual cost.
• Scoring-rubric capture. Subjective quality scores are a corruption/favoritism surface — the scoring official becomes the chokepoint (a cui bono shift, not elimination). Mitigate with objective, documented, auditable criteria and separation of scorer from award authority.
• Phoenix firms / entity-shifting. Dissolve and re-incorporate to shed bad registry record and warranty/bond obligations. Per the IC analysis this is not merely a residual leak — it is the case against which the separating equilibrium itself fails unless liability and performance records attach to principals/beneficial owners, with ownership disclosure required at prequalification and bonds that outlive the entity. The new-entrant on-ramp is the obvious channel for a re-formed L to re-enter clean — guard it accordingly.
• Key-personnel bait-and-switch. Name star staff to win the score, swap them post-award. Counter: contractually bind named personnel with substitution penalties — sharper now that the on-ramp credits personnel records, so the credit must follow the people who actually show up.
• Surety fronting / weak sureties. A warranty/bond is worth only its backer’s solvency at claim time, and the screen fails (per §3c) if the surety underwrites on nothing beyond the city’s own public prequalification. Counter: require rated sureties; in the escrow-surrogate regime, require the deposit be genuinely the firm’s capital, not a re-lent third-party float.
• Defect timing beyond the warranty window. Cut in ways that surface just after the defects-liability period. Counter: longer horizons for structural elements; decennial latent-defect liability/insurance where the jurisdiction carries it.
• Change-order rent recovery. The lowball-then-claim play migrates from bid to change-order process. Counter: complete design documents before tender, owner-controlled scope, a priced/pre-agreed variation mechanism, change-order discipline.
• ALB / justification theater. Bidding just above the trigger, or passing the gate with optimistic paper. Counter: anchor to the engineer’s estimate, treat the trigger as a soft scrutiny prompt; require a competent reviewer empowered to reject, not a checkbox.

Read

• Posture (analysis, §§1–2): the current arrangement misfires because the lowest-price rule rewards a negative signal. Two distinct asymmetries operate — hidden type before the award (adverse selection + winner’s curse, which together drive H out and leave L or a doomed-optimistic H to win) and hidden action after it (fixed-price moral hazard, where cutting is rational because the upside is certain/immediate and the downside is probabilistic/delayed/escapable). Screening cannot cure the second; incentives cannot cure the first.
• Posture (design, §§3–5): the fix is screening from the uninformed side, not waiting for firms to signal. A scoring auction built on cost-asymmetric inputs, a prequalification gate, and — load-bearing — a bonded multi-year latent-defect warranty form a separating equilibrium where only H can cheaply offer the warranty. Post-award, re-couple the defect downside via horizon-tied holdback, callable warranty bonds, destructive testing at burial points, pre-defined liquidated damages, and a principal-attached past-performance registry. The separating engine is the bond-and-warranty cost asymmetry; the equilibrium holds exactly when the single-crossing condition holds.
• The judgment-proof low type is the load-bearing exception. A firm intending to dissolve discounts warranty claims, retention loss, and bond recourse toward zero, so the bond/warranty cost asymmetry does not separate it on its own. Separation against this type depends entirely on liability and performance history attached to principals/officers (so dissolution sheds neither obligation nor record) and rated, solvent sureties (so the bond is collectable). Stated precisely: IC holds given the warranty PLUS officer-attached liability AND surety solvency — not given the warranty alone. If those conditions fail, the separating engine leaks against exactly the worst type.

Confidence and Assumptions

• Confidence: moderate-to-high on the classification and the direction of every mechanism (adverse selection + winner’s curse pre-award; fixed-price moral hazard post-award — well-established results shown operating on this situation, not name-drops). Moderate on parameters (holdback/deposit %, ALB trigger, bond sizing, warranty length, on-ramp band width) — calibrate to local cost data and the actual H–L cost gap.
• The design rests on four information assumptions; if any fails, confidence drops sharply:
  1. Quality has some verifiable ex-ante signal (track record, certifications). Pure unverifiability collapses the scoring screen.
  2. A functioning, rating-discriminating surety market exists that underwrites on private information the city lacks. This is the main pre-award screen (3c); if sureties don’t price type (or read only public prequalification), that lever is dead and the §3c escrow/mutual-guarantee surrogate becomes the primary screen instead — the fallback is what lets the separating equilibrium degrade gracefully rather than collapse.
  3. Quality is testable during construction even if latent at handover (core samples, compaction). This is the main post-award lever (4c); if untestable until years later, raising P(detect) fails.
  4. Procurement is repeated. The registry (4e), the strongest long-run device, needs a future to discipline the present.
• With the §3c fallback in place, the most fragile remaining assumption is #3 (construction-time testability) — it has no substitute lever in this design; if it fails, the design needs rework, not just recalibration. Assumption #2 now degrades gracefully. First fact to confirm before building: #3 (testability); second: local surety-market depth (which selects between the §3c primary and fallback). Tighten weights with the authority’s past-bid-vs-final-cost data, which also reveals how much overrun history is bidder behavior vs. external escalation — if mostly the latter, indexation and complete documents do more work than screening.

Genuine, still-unresolved tensions:

• Participation: whether the parameter envelope (5–10% holdback + rated-surety bond/escrow surrogate + 12–24mo liability) leaves positive IR for genuinely small-but-capable H is domain-empirical — resolves with a construction-procurement domain reviewer plus local H–L cost-gap data.
• Surety as screen vs. relabel: whether real surety markets price type finely enough to produce a separating (not pooling) bond-stage outcome, and whether they hold type-distinguishing private information beyond prequalification — resolves with surety-underwriting data on premium dispersion across risk tiers / local underwriting practice. The §3c escrow fallback partially de-risks this.
• New-entrant on-ramp calibration: whether the probationary on-ramp admits enough capable entrants to offset registry/prequalification entrenchment without re-admitting L — resolves with entrant-cohort performance data once the registry runs.