UVP Prompt Lab
See what changes when a prompt becomes an operating system.
Enter a rough instruction, choose a task family, and move the depth slider from Basic to Master. The public lab teaches the architecture behind stronger prompting while the full Prompt Revolution Engine continues through semantic recomposition and validation testing.
Prompt Revolution Engine
Prompt Optimizer, rebuilt as a transparent learning surface.
The long-term engine is designed to understand rough language, classify the real task, preserve constraints, choose the right prompt architecture, recompose the instruction, validate semantic fidelity, and repair weak output. This public beta exposes the depth architecture without falsely claiming that a browser template performs the whole semantic job.
Your rough instruction
Adds evidence/source rules, process, assumptions, and review.
Architecture preview
Advanced prompt scaffold
MISSION Create a small customer intake app for a local service business. TASK FAMILY Product Design CONTEXT [Relevant background, user/audience, current state, and why this matters] CONSTRAINTS - Preserve explicit user requirements. - Do not invent unsupported facts or requirements. OUTPUT [Define the final deliverable, format, and intended use] SOURCE / EVIDENCE RULES - Use supplied sources before relying on general knowledge when the task depends on them. - Separate confirmed facts, assumptions, and recommendations. - Mark missing evidence rather than filling gaps. PROCESS 1. Confirm the objective and available inputs. 2. Identify material ambiguities and risks. 3. Execute the task in a sensible order. 4. Review the result against the requested outcome.
Privacy note: this public preview is generated in your browser from a structural scaffold. It does not send this text to UVP or pretend to perform the full semantic recomposition of the Prompt Revolution Engine.
UVP master-prompt standard
Longer is not automatically better.
Mission clarity
State what the system must actually accomplish.
Context preservation
Keep the facts, names, constraints, and decisions that matter.
Boundary control
Say what must not be invented, changed, exposed, or assumed.
Execution logic
Give complex work an order rather than a pile of instructions.
Verification
Define how the result will be checked instead of trusting confident output.
Honest uncertainty
Separate facts, assumptions, estimates, and unresolved questions.
What comes next
The full optimizer has a stricter acceptance bar than this demo.
Semantic recomposition
The production engine must normalize rough language, infer the actual task without changing intent, choose a domain-aware schema, and produce directly usable output instead of wrapping raw input in generic adjectives.
Golden tests before production claims
Known failures become regression tests: image requests must remain image requests; app-creation requests must not become maintenance jobs; constraints must survive; unsupported details must not appear; and copied output must stay clean.