Answer Shape
Defines the expected shape: text, table, status, Markdown or structured data.
Answer Engineering works on output shape, admissible answer space and result extraction. Kerkito extends this logic with a deterministic answer contract and local structural validation.
Prompt-based learning literature has distinguished input design from expected output design since 2021. Shape, admissible values and result extraction become engineering objects in their own right.
Defines the expected shape: text, table, status, Markdown or structured data.
Defines allowed values or states when the task can explicitly bound them.
Defines how the useful result is recovered from model or system output.
Checks observable rules such as sections, fields, dates or statuses without claiming to verify factual truth by itself.
After P.O.N.T. and the four stages, Kerkito can compile a portable representation of the expected output. The contract describes what must be present, what may remain unknown and which rules should prevent false certainty.
An answer can follow a schema exactly and still be factually wrong. Kerkito therefore separates locally checkable properties from those requiring actual source retrieval or verification.
Required sections, formats, allowed values, requested dates, explicit unknowns and other deterministic contract rules.
Remote source authenticity, scientific validity of a conclusion or whether a citation actually supports a claim require a separate verification process.
This page relies on publications documenting Answer Engineering, prompting techniques and structured outputs. Results remain bound to their protocols and do not constitute general guarantees.