PFP V: A Practitioner's Guide to Checked Proposals

PFP V: A Practitioner's Guide to Checked Proposals, is an initial instructional preprint by Jeremy H. Carroll on the Proposal Fidelity Protocol (PFP), formerly the Symbol Game. The deposit contains the guide, claim ledger, editorial history, a 32-question Landau quiz with a worked study guide, and self-contained Python companion code. Its separately configured native calculator is not deposited. The guide takes one small inconsistent graph through a complete session: declare the instance, propose a change, interpret a refusal, admit a repair, verify the result, and restore an earlier state. Three events with unit offsets around a directed triangle give claims (1, 1, 1) that no vertex potential can satisfy. With edge weights (1, 2, 3), a two-line lower-bound argument shows that the weighted minimum is 3, attained by one explicit potential. The guide then separates choosing a potential, which leaves the cycle sum unchanged, from replacing the claims by the consistent (−2, 1, 1), which changes the problem’s data. A scripted session shows the host refusing a proposal that raises the minimum from 3 to 5, admitting the exact repair, verifying it, and retracting to a saved snapshot. The manuscript and scripts use weights (1,2,3) throughout. An offline worksheet checks the arithmetic without the native calculator. The two copied modules record their provenance and local patches. Worked-session scripts print their selected calculator path: exact Python enumeration when no native executable is configured, or the configured native route whose refusals propagate. Regression checks cover declared inputs, exact values, scoped edge verification, and rollback after a projection error. Eight examination levels progress from instance declaration and a hand-checkable optimality bound to stationary moves, verification scope, exact input parsing, bounded-search policy, evidence custody, and an executable capstone. The fresh publication review records offline tests and explicit native skips; earlier validation records remain historical. This guide adds no new theorem, live provider experiment, learner study, or measured model-performance result.