Mental Models

Transient cognitive filters applied directly to the generation pipeline to focus analytical execution.

In addition to formal Methods and multi-step Recipes, the Brainstorm Reactor supports transient Mental Models (labeled as "Tägliche Tipps" / "Daily Tips" in the System Config and displayed via the collapsible "Biases · Gesetze · Paradoxien · Prinzipien" drawer on the canvas). This feature is restricted strictly to Direct Mode to prevent interference with structured multi-step workflows. When a user is in Workflow Mode (Step or Auto Mode), the collapsible drawer is hidden, active badges/pills under the input field are suppressed, and no prompt filter is sent to the LLM.

1. System Integration

When a Mental Model (Tipp) is enabled, its configuration is retrieved from the reactor_mental_models database table by ID. The API routes dynamically query locale-aware keywords (keywords_en or keywords fallback) depending on the active user locale (DE/EN) to provide tailored context to both the interface and the LLM.

Outbound Prompt Injection

Before sending the payload to the language model, the system prepends an additive epistemic instruction block to the final assembled system prompt:

COGNITIVE LENS ACTIVE: Analyze the input through the lens of "[Model Name]" (Category: [Category]).
Core concept to apply: [Keywords].
You MUST structure your reasoning to reflect this model's logic. Explain your synthesis or evaluation steps by explicitly referencing the principles of [Model Name].

This injects a targeted cognitive constraint into the Epistemic Architecture (Layer 2 - Context/Model injection) without requiring the user to manually engineer prompts or change the active Method.

2. Catalog & Taxonomy

The engine maintains a static catalog of 500 curated models spanning multiple disciplines, including:

Biases (e.g., Confirmation Bias, Hindsight Bias)
Laws (e.g., Campbell's Law, Goodhart's Law)
Principles (e.g., Occam's Razor, First Principles)
Paradoxes (e.g., Ship of Theseus, Abilene Paradox)
Fallacies (e.g., Sunk Cost Fallacy, Straw Man Fallacy)

UI Controls & Selection Patterns

The active model is displayed via a dedicated badge ("Tipp: [Name]" or "Tip: [Name]") on the Smart Input Pill and dynamically synchronized with the Ambient Watermark on the empty canvas. The collapsible drawer at the bottom of the canvas provides these controls:

Apply / Deactivate ("Anwenden" / "Deaktivieren"): Toggles whether the selected model acts as an active prompt filter on outbound queries.
Next Model ("Weiter"): Pulls an arbitrary model from the catalog for exploratory sessions, switching the mode from "daily" to "random".
Turn off feature ("Aus"): Completely disables and hides the collapsible drawer.
Daily selection ("täglich"): Automatically determined by a deterministic modulo selector across the 500 catalog entries, exposing users to a new framework every day.

3. Intellectual Property Guardrails

To comply with the strict clean-room policy defined in IP Protection Guidelines, all mental models in the catalog are audited to prevent trademark infringement or copyrighted text duplication.

Trademarked concepts (such as the term IKEA-Effect) are replaced with functional, descriptive equivalents (e.g., Self-Assembly Effect).
Math and logic concepts are expressed via raw structural descriptions rather than proprietary formulations.