Foray 445 · Essai 01 · 18 October 2026
A useful answer sold by the piece. Six small commissions test the boundary between a checkable result and an open-ended professional promise.
Choose a case, inspect the submitted records and follow the receipt. A report that finds defects can be a successful delivery.
Submitted input
| Row / ID | Activity | Start → finish | After |
|---|
Demand: two. Capacity: one. This authored resource fact lives in a separate scenario note. Neither audit algorithm received it. The same valid dependency extract fits both a feasible two-bay world and an infeasible one-bay world.
Delivered report + verification
A separate implementation recomputes the predicates from the submitted records. It was written by the same agent: this is a machine-verification result, not independent professional or buyer acceptance.
Same supported findings and holds on every submitted revision. Implemented as row scans and a reachability table.
Same result, with indexed relations and cycle detection. No demonstrated decision advantage over the ordinary method.
Deliberate false claims, omissions and broken evidence were rejected. This is a bounded challenge set, not a reliability estimate.
The controls below describe possible terms for the selected case. They do not create a sale or change the executed evidence above.
Every minute and rate below is an editable scenario assumption. Human delivery effort was not timed. The four-minute production allowance is an illustration, not a measurement.
The defaults assume equal recurring effort. Only setup differs. Neither has a measured speed advantage.
The correction allowance is an assumed 20% chance of 20 extra minutes, charged to the supplier. It is neither an observed error rate nor a limit on responsibility. Supplier mistakes do not become new buyer-paid units. Loaded cost should include ordinary overhead; taxes, financing, acquisition cost and exceptional harm are outside this small calculation.
Cost sensitivity to the assumed hourly rate
A positive arithmetic margin does not establish willingness to pay, a fair market price, an agreed commission or a viable business.
Nine reports and a stale-reuse challenge expose what a narrow acceptance basis can check. Two inputs needed supplied clarification before a complete audit could be delivered.
Both implementations reach the same supported conclusions. The reusable machinery has earned reproducibility here, not a demonstrated reduction in labour or better professional judgement.
The shared-bay counterexample defeats a release-readiness claim. Resource discovery, scope completeness and authority may require a wider engagement. This study has no buyer, payment, repeat-use or benefit evidence.
Used: dependency logic and resource availability are separate parts of a credible schedule.
Built: a narrow finish-to-start check plus a resource counterexample.
Limit: six predicates are not a GAO-compliant schedule assessment.
Used: a result is derived from a particular input; revision changes that relationship.
Built: source-bound reports and stale-receipt rejection.
Limit: a hash binds bytes; it does not authenticate an owner or establish truth.
Used: distinguish what parties observe from what an acceptance mechanism can verify.
Built: explicit result, correction and changed-input boundaries.
Limit: no optimal-contract result, legal analysis or market validation is inferred.
All cases are deliberately authored and small. The producer and verifier share a written specification and author, but have separate implementations. The verifier uses pairwise row scans and a transitive-closure matrix; the producer uses indexed adjacency and strongly connected components. Agreement can still leave common specification errors. Narrative findings and professional usefulness have not been assessed by an independent practitioner.
Dates represent exclusive boundaries in a calendar-day model. Every activity must have positive duration; all supported links are finish-to-start with zero lag. This chosen convention makes equality valid. It does not reproduce working-calendar software or GAO's inclusive-day examples. No claim is made about absent work, resource feasibility, durations, critical path, risk or actual project performance.
The ordinary workflow is an executable table-check procedure, not an observed manual time trial. Cost inputs are scenarios. No execution-runtime-to-human-labour conversion is used. Data and scripts reproduce the reports, verdicts, finite challenge tests and calculations.
This is the first bounded test within A useful answer sold by the piece. Other services, acceptance arrangements and forms of professional exchange remain open.