5PP
A five-phase discipline for deciding what must be understood, bounded, planned, executed and verified.
The Five-Point Protocol for bounded, reviewable AI-assisted work
5PP separates clarification, scope, planning, execution and verification so that work does not silently move from an unclear request into irreversible action. This page distinguishes the method from 5pp-gate, the later deterministic compliance checker that validates recorded protocol traces.
Method, record contract and enforcement tool are related, but not identical.
A five-phase discipline for deciding what must be understood, bounded, planned, executed and verified.
A Markdown work record containing phase evidence, an approval checkpoint, self-audit material and one declared verdict.
A stdlib-only Python linter that performs deterministic structural, ordering, enumeration and count checks. It uses no LLM and no network.
Drag the phase nodes, select a profile and click any phase to inspect its role, evidence and common failure modes.
Explore a bounded model of the gate rules. This browser simulation is educational and is not the Python gate.
Change the evidence controls and run the synthetic evaluation.
pending
Declared verdict
Advance
A representative, source-grounded network. Drag nodes, filter layers and inspect relationships. It is intentionally not a complete estate graph.
Two canvas-based graphs separate measured repository receipts from analyst-coded evidence coverage.
Counts are taken from the pinned changelog: 107, 131, 135 and 139. They are source-reported and were not rerun for this publication.
A zero under external validation means no independent effect study was identified in the inspected source set. It does not prove that no such study can exist.
Clarification through Verification.
Standard, constraint-hardened and silent.
Advance, Narrow, Revise, Park and Reject.
Unanimously confirmed findings described as fixed before v1.0.
H1-H6 described and regression-pinned in v1.1.
Source-reported in the pinned changelog.
The page separates mechanism, evidence, hypotheses and limits rather than presenting adoption as proof of efficacy.
AI-assisted work can cross from ambiguous intent into action without a stable scope, observable plan or explicit verification gate. 5PP creates named transition points where drift can be noticed before or after execution.
Making phase boundaries and evidence explicit should increase inspectability and make some classes of omission or ordering error easier to detect. This is a design hypothesis, not a measured universal effect.
5pp-gate converts part of the protocol into deterministic checks over Markdown records. It can validate presence, ordering, enumerations, counts and configured markers. It cannot determine whether the underlying reasoning is true.
Useful evaluation would compare matched tasks with and without the protocol, measure correction cost, missed constraints, false confidence, operator workload and completion quality, and publish negative as well as positive outcomes.
Examples of where the method or its records appear. The list is intentionally incomplete.
Claims are bounded to the inspected sources and pins below.
fbratten/5pp-gate @ e351dae2f39ca67af556037be99c9996140df2c5README.md, CHANGELOG.md, src/fivepp_gate/checks.pyadaptivearts.ai/blog/from-giant-meta-prompt-to-5pp/agentspool-showcase/demos/capabilities-radar @ Chart.js 4.4.7fbratten.github.io/methods/5pp/This page is a public-safe explanatory artefact. It performs no repository writes, LLM calls, networked evaluation, local file access or real gate execution.