Malleable Forms

Malleable Forms replaces upfront specification with lazy data space elicitation: you start with a sentence describing what you need, and the system helps you discover and refine your data schema through a structured conversational process.

The Problem

  • Traditional form builders: Require the full data space upfront
    Malleable Forms: Schema emerges through conversation
  • Traditional form builders: Assume data modeling skill
    Malleable Forms: Meets you where you are
  • Traditional form builders: No guidance for domain-specific decisions
    Malleable Forms: Surfaces considerations you hadn’t thought of

How It Works

You describe your intent in natural language. The system responds with design probes—targeted questions that surface tradeoffs and requirements you may not have considered. Each answer refines both the schema and the intent, converging on a form you could not have specified from scratch.

“I need a registration form for our youth soccer club’s new season”

Design Principles for Lazy Data Space Elicitation

Instead of front-loaded specification, the data space is elicited lazily: the creator begins with a rough intent and progressively refines the underlying schema through a structured conversational process. Four design principles govern this process.

Iterative Convergence

The data space emerges through successive refinement. Design probes act as cognitive forcing functions: each requires an active evaluative choice that channels domain knowledge into the specification. A novice triggers more follow-up questions; an expert skips intermediate rounds.

Intent–Schema Co-Persistence

The natural-language intent and structured schema are stored together as a single linked artifact—the intent portfolio. Domain context travels with the schema, so anyone can reopen the portfolio and understand not just what it contains but what it was intended to capture.

Decision Traceability

An append-only provenance log records each change together with its rationale. When a new collaborator encounters an unfamiliar field, the log explains why it was added—without requiring human mediation.

Scenario-Driven Derivation

New portfolio instances are derived from a shared base schema for specific scenarios. A derivation can be a sub-schema (subset of fields), a super-schema (extensions), or a mixed case combining both—without duplicating the data space.

Let's try it out