# Sources and construction choices

Read on 18 October 2026. These primary sources provide selected concepts, not a certification of the service, an optimal commercial contract or a claim of original scheduling theory. The synthetic cases, offer boundaries, acceptance rules and cost assumptions are authored for this essai.

## GAO — Schedule Assessment Guide (GAO-16-89G, 2015)

Source: [official report](https://www.gao.gov/assets/gao-16-89g.pdf), especially Best Practice 2, printed pp 27–30 (dependency types), and Best Practice 3, printed pp 49–61 (resource assignment and feasibility); see the correctly sequenced but resource-overallocated example at printed pp 59–60.

**Concept used:** dependency sequencing and resource feasibility require different information. A schedule with consistent supplied links can still demand unavailable resources.

**Construction:** the six predicates check a narrow declared extract. A second-world test preserves the same passing extract while changing test-bay availability outside it. This prevents widening a report into a release-readiness assurance claim.

**Limitation:** we have not applied the guide's ten best practices, checked scope completeness, developed a critical path or conducted schedule risk analysis. Our exclusive-boundary calendar-day convention differs from the guide's inclusive-day worked examples; it is explicitly supplied for this experiment.

## W3C — PROV-DM Recommendation (2013)

Source: [PROV-DM](https://www.w3.org/TR/prov-dm/), §2.1.2 and §5.2 on derivation, and §2.2.1.1/§5.2.2 on revision.

**Concept used:** a result derives from a particular input entity; a revised input is distinguishable from the original.

**Construction:** each report names its source revision and canonical-JSON SHA-256. Acceptance against changed input fails, prompting a fresh report. Evidence references retain the submitted filename and row number.

**Limitation:** this is a small JSON implementation informed by those concepts, not a PROV conformance claim. Hash equality binds the report to supplied bytes, not to a truthful source or authenticated owner. Canonicalisation here is the specified Python sorted-key compact JSON encoding, not a general cross-language canonical-JSON standard.

## Oliver Hart — Incomplete Contracts and Control (Nobel lecture, 2016)

Source: [official lecture](https://www.nobelprize.org/uploads/2018/06/hart-lecture.pdf). Selected reading: the distinction between observable and verifiable costs in the discussion of contractual implementation. The relevant passage was accessible through the indexed primary-source extract; direct full-PDF retrieval failed in this run. No claims rely on unseen portions.

**Concept used:** what a party may recognise is not automatically what a specified acceptance mechanism can establish.

**Construction:** separate an observable concern (“are we ready?”), machine-verifiable extract predicates, owner decisions and buyer acceptance. Offer controls expose correction and changed-input obligations.

**Limitation:** no ownership, equilibrium or optimal-contract theorem is applied. Our verifier is not a court, independent practitioner or buyer. Economic and legal enforceability, strategic conduct and willingness to pay remain outside the executed evidence.

## Lineage and scope

The work follows Foray 445's question about an independently acceptable unit of professional purchase. It does not reopen the general agent-capability enquiry, demonstrate the practical value of every management intervention, or pool outcome risk across clients. All project data and buyer context are fictional; no private backlog, commercial correspondence or client information is reproduced.