P02 · 15 rules · 15 package cases

Prompt remasterer

Paste a long, messy prompt and it reorganises it, flags what is vague and counts the tokens.

Verified calculationThe result depends only on your data and on formulas covered by the tests.

The interface is in English, but what the engine writes is not: rule messages, error messages, audit-log entries and the catalogue of nodes and patterns come from the specification packages, which are written in Spanish and are the source of truth. Their codes are always visible.

Paste a long, messy prompt. It reorganises it into context, task, input, format and constraints, flags what cannot be measured, sets the contradictions against each other and tells you what is missing. It does not change a single word of yours: it only reorders them.

Your prompt

Reorganised

CONTEXTO
I need help with something at the university, it is about students dropping out and honestly we do not really know what is going on.
Analyse the information and tell me which are the best students to reach first, I want something professional and fast.
Also compare the programmes against each other.
Be brief but include all the detail you can on every case.
Do not make up data.

TAREA
[FALTA: TAREA]

ENTRADA
[FALTA: ENTRADA]

FORMATO DE SALIDA
[FALTA: FORMATO DE SALIDA]

RESTRICCIONES
[SIN RESTRICCIONES DEFINIDAS]

What cannot be measured

No vague terms detected.

What is missing

[FALTA: ENTRADA]
[FALTA: FORMATO DE SALIDA]
[SIN RESTRICCIONES DEFINIDAS]

¿Sobre qué material debe trabajar? No se encontró la entrada.
¿En qué formato debe entregar el resultado?

Where each part came from

Every fragment with the section it went to and what triggered it. This is what makes the reorganisation auditable.

Every fragment of the prompt, the section it went to, which rule moved it and where it ended up.
Fragment	Section	Why	Position
I need help with something at the university, it is about students dropping out and honestly we do not really know what is going on.	Context	precede a la primera tarea	0–132
Analyse the information and tell me which are the best students to reach first, I want something professional and fast.	Context	precede a la primera tarea	133–252
Also compare the programmes against each other.	Context	precede a la primera tarea	253–300
Be brief but include all the detail you can on every case.	Context	precede a la primera tarea	301–359
Do not make up data.	Context	precede a la primera tarea	360–380

Tokens

Before: 95
After: 132
Change: +38.9 %

APPROXIMATE · With no tokeniser declared, the count estimates four bytes per token. Good for comparing, not for billing.

Reorganising usually adds tokens rather than removing them: headings and markers for what is missing get added. Fewer tokens does not mean a better prompt either, which is why the tool shows the figure without celebrating it.

Audit log

8 decisions

P02-R-001
Se separó en fragmentos conservando el offset de cada uno; ninguna palabra se alteró.
P02-R-002
Cada fragmento se copió a su sección del orden canónico, con el indicio que la disparó.
P02-R-008
No se detectó entrada: se marca el vacío y se pregunta.
P02-R-009
No se detectó formato de salida: se marca el vacío y se pregunta.
P02-R-010
No se inventa ninguna restricción: sólo se hace constar que no se definieron.
P02-R-011
Sin tokenizador declarado: el conteo es una estimación por caracteres y se marca como aproximado.
P02-R-014
Se muestra la diferencia absoluta y porcentual. Reducir tokens no significa mejorar el prompt.
P02-R-015
La herramienta reorganiza y diagnostica; no realiza la tarea que el prompt pide.

engine 1.0.0 · rule package c1aeceac109c1ec1…