⚠ Reconciled at canonical schema v1.0.0. This narrative was authored against a bespoke minimal schema (Agent D, PR #1) that is now superseded. Any claim below that a construct is “UNREPRESENTABLE in schema v1.0.0” or “pending-v1.1” refers to that bespoke schema, not the canonical one — the canonical schema expresses these constructs. The executable truth for this domain is the canonical scene fixture conformance/case-studies/government-procurement/government-procurement.json, which validates + audits clean via node conformance/runner/run.mjs. See case-studies/README.md for the reconciled status table and schema/BACKLOG-v1.1.md for the real findings.

Case Study: Government Procurement #

Adversarial test corpus — executable. This is one of the domains the Commerce Model Formal Sufficiency Test claims to have passed. The JSON below is a real fixture under conformance/case-studies/government-procurement/ that validates against schema v1.0.0 and passes auditCommerce. Run it: node conformance/audit.mjs conformance/case-studies/government-procurement.

Reference platforms: Morocco Marchés Publics (Portail des Marchés de l’État), EU TED (Tenders Electronic Daily), GCC government procurement portals. Fixture: public-tender-with-protest.json

The domain and the hard cases it stresses #

Government procurement is not a price auction. A public tender selects among suppliers on weighted multi-criteria scoring — technical capability, price, local content, past performance — not on who bid lowest. This is the most important structural difference from the auction domains already modelled.

The hard cases this domain puts on the model:

• Scored selection, not price maximisation. Bids are evaluated by a committee against a published scoring matrix. The “winner” is the highest composite score, which may not be the lowest or highest price.
• Minimum technical threshold. A bid that scores below a minimum on technical criteria is disqualified regardless of price. This is a gate, not a criterion.
• Award protest. A losing supplier can formally challenge the award decision before contract signature. If upheld, the initial award is rescinded and a re-evaluation or re-award follows. This creates a fork in the commitment lifecycle: an accepted commitment that later transitions to cancelled.
• Compliance documentation. Suppliers must submit statutory documents (tax clearance, criminal record, professional qualification certificates) before the contract can be activated. Missing documents block execution even if the bid was winning.
• Phased payment against service milestones. Government contracts typically release payment tranches tied to delivery milestones (mobilisation, interim acceptance, final acceptance), not as a single lump sum.

Extensions relied upon (schema gaps — the central finding of this domain) #

This domain exposes three concrete gaps between the v0.3 prose spec and the v1.0.0 executable schema. These are not modelling choices — they are cases where the spec defines vocabulary that does not exist in the schema.

Extension 1 — ScoredSelection: UNREPRESENTABLE in schema v1.0.0 #

The v0.3 spec defines:

AuctionMechanism::ScoredSelection {
  criteria: Vec<ScoringCriterion> { name, weight, max_points }
  minimum_threshold: Option<u32>
  evaluation_committee: Vec<PartyID>
  publication_required: bool
}

The schema v1.0.0 AuctionMechanism is a closed oneOf with exactly four variants: english, dutch, sealed_bid, vickrey. There is no scored_selection variant. Any fixture that sets "mechanism": {"scored_selection": {...}} fails schema validation.

Resolution in this fixture: The tender is encoded with "mechanism": {"sealed_bid": {...}} — the closest available variant (bids are submitted privately and evaluated before revelation, which is structurally similar). The scoring criteria, weights, minimum threshold, and evaluation committee exist only in the subject_description prose field.

Impact: The spec’s claim that “government procurement passed” the Formal Sufficiency Test cannot be demonstrated with an executable schema fixture. The prose model is richer than the schema.

The scoring for this tender was:

Supplier B (BID-B) is disqualified for failing the minimum technical threshold (61 < 65). This threshold logic has no schema representation.

Extension 2 — AwardProtest: no schema object #

The v0.3 spec defines an AwardProtest auxiliary record:

AwardProtest {
  id: ProtestID
  filed_by: PartyID
  against: CommitmentID
  auction_process: AuctionProcessID
  grounds: Vec<String>
  filed_at: Timestamp
  deadline_for_response: Timestamp
  reviewing_body: Option<PartyID>
  state: ProtestState { Filed | UnderReview | Upheld { remedy } | Dismissed }
}

Schema v1.0.0 has no award_protests array in Fixture and no AwardProtest schema object. The protest cannot be represented as a first-class object.

Resolution in this fixture: The upheld protest (AP-001) is modelled through its effect on commitments:

BID-C: tendered → accepted → cancelled
  (reason: "award protest upheld — AwardProtest AP-001 — original award rescinded")

BID-A-REAWARDED: draft → tendered → accepted → ... → fulfilled
  (re-tender of Supplier A's terms following protest)

The protest narrative, grounds, reviewer, and remedy are in prose. The commitment lifecycle correctly captures the outcome (accepted → cancelled), but the regulatory record of why — the protest document itself — has no schema home.

Extension 3 — ComplianceDocumentation: no CommitmentCondition schema object #

The v0.3 spec defines:

CommitmentCondition::ComplianceDocumentation {
  required_documents: Vec<String>
  submission_deadline: Timestamp
  verified_by: PartyID
  if_not_submitted: CommitmentState
}

Schema v1.0.0 has no conditions array on Commitment and no CommitmentCondition schema at all. The pre-award compliance check — tax clearance certificate, company registration, professional licence — is a regulatory requirement that blocks contract execution if not satisfied.

Resolution in this fixture: Compliance verification is noted in the re-award acceptance transition’s reason field: "compliance documents verified". There is no structured enforcement.

The scenario #

The Ministry of Digital Transformation (Morocco) issues a public tender for IT infrastructure deployment and 24-month managed services. Reference: MDT-2026-007. Budget ceiling: 12,000,000 MAD. Reserve price: 7,000,000 MAD.

Timeline:

Total paid: 9,000,000 MAD = 4,500,000 (T1) + 4,500,000 (T2). All MAD.

The model objects #

Parties (6) #

Values (3) #

Intent (1) #

INT-TENDER-001 — the Ministry’s procurement intent, active from tender publication, converted to BID-A-REAWARDED once the re-award decision is final.

Commitments (5) #

AuctionProcess (1) #

TENDER-MDT-2026-007 — the public tender coordination record.

• Mechanism: sealed_bid (schema representation) — see Extension 1 above
• State: closed, winning_commitment: "BID-C" (initial award before protest; the auction record is not updated for the re-award since no re-opening occurred)
• All four bids are in tendered_commitments

Note: The auction’s winning_commitment reflects the initial award to BID-C, which is the state at normal_close. The protest and re-award are a post-close administrative process. The actual contract (BID-A-REAWARDED) is correct, but the auction object cannot express the protest mechanism or the re-award — another consequence of the AwardProtest gap.

Fulfillments (4) #

Lifecycle as a transition sequence #

Intent INT-TENDER-001:       active → converted(BID-A-REAWARDED)

AuctionProcess TENDER-MDT-2026-007:
  scheduled → open → closed(winning: BID-C, normal_close)
  [protest and re-award are post-close; not representable in AuctionState]

Commitment BID-A:            draft → tendered → cancelled (not selected initially)
Commitment BID-B:            draft → tendered → cancelled (disqualified)
Commitment BID-C:            draft → tendered → accepted → cancelled (protest upheld)
Commitment BID-D:            draft → tendered → cancelled (not selected)

Commitment BID-A-REAWARDED:  draft → tendered → accepted
                             → partially_fulfilled → fulfilled

  Fulfillment FUL-MOBILISATION (service_performance):
    planned → in_progress → completed

  Fulfillment FUL-PAYMENT-TRANCHE-1 (money_transfer):
    planned → in_progress → completed

  Fulfillment FUL-IT-SERVICES (service_performance):
    planned → in_progress → completed

  Fulfillment FUL-PAYMENT-TRANCHE-2 (money_transfer):
    planned → in_progress → completed

The critical transition: BID-C: accepted → cancelled. This is a valid transition in the Commitment table (accepted > cancelled) and is how the protest outcome is expressed. State monotonicity (I-2) is preserved because cancelled is a forward transition, not a reversal.

Invariants exercised #

What the model can express without extensions #

The five base primitives handle the procurement lifecycle correctly:

• Multiple competing bids as parallel Tendered commitments
• Loser bids transitioning to cancelled with reason carrying the evaluation rationale
• An initial winner’s commitment going accepted → cancelled (protest effect)
• A re-award modelled as a fresh Tendered → accepted lifecycle
• Phased service delivery through partially_fulfilled
• Milestone-linked payments as separate Fulfillments with money_transfer method
• service_completion Evidence confirming acceptance at each milestone

What the model cannot express — schema gaps #

The following v0.3 spec constructs are not representable in schema v1.0.0:

These are not minor gaps. ScoredSelection is the defining mechanism of government procurement — every public tender in Morocco, the EU, and the GCC uses weighted multi-criteria scoring, not a price auction. Without it, any schema-valid government procurement fixture is structurally misrepresented as a price-only sealed bid. The spec’s claim to have “passed” government procurement cannot be reproduced as an executable fixture that correctly names the mechanism.

The AwardProtest gap means the regulatory audit trail — which procurement authorities are legally required to maintain — has no first-class schema object. The protest can be inferred from commitment state transitions but cannot be queried, validated, or tracked as a structured record.

The ComplianceDocumentation gap means the gate that protects against non-compliant suppliers (those who did not submit tax clearance, professional licences, etc.) has no enforcement path in the schema. A fixture could model a non-compliant supplier reaching accepted without any validator catching it.

Run it #

node conformance/audit.mjs conformance/case-studies/government-procurement
# ✓ conformance/case-studies/government-procurement/public-tender-with-protest.json
# auditCommerce: 1 passed, 0 failed, 0 warnings, 1 fixtures

Full suite (all domains):

node conformance/audit.mjs
# auditCommerce: 22 passed, 0 failed, 0 warnings, 22 fixtures