dillylang-decompose

name	dillylang-decompose
description	Decomposes a problem into axioms, derivations, and assumptions. Trigger= /dillylang-decompose PROBLEM

Dillylang decompose

[[THIS is_grounded_by: urn:unique_reference:dillylang::spec-primer]]

Decompose the given problem statement into its structural foundations.

Separate what's load-bearing from what's assumed. Not a generic outline — a search for axioms.

Prompt constraints

• Axioms are foundational — not derivable from other statements in the decomposition.
• Derivations follow from axioms with an explicit dependency list.
• Assumptions are treated as true without justification. Every assumption must be assessed for load-bearing status and testability.
• Be ruthless. Three sharp axioms beat ten soft ones.

Calibration example:

Rejected: "Axiom: The system should be reliable. Justification: Reliability is important." (Not foundational — restates a vague desideratum with no structural content.)

Accepted: "Axiom: Every pipeline run must emit an inspectable trace. Justification: Debugging requires per-step visibility; without traces, prompt iteration is guess-and-check against end-to-end output only."

Steering

If a focus is provided (e.g. /dillylang-decompose @file.md "requirements and constraints"), narrow the decomposition to those structural aspects. Otherwise, perform general structural decomposition over the full input.

Output template

Axioms (AX-n)

Foundational statements that are not derivable from other statements in this decomposition. The bedrock.

Each entry must include:

• Statement: the axiom
• Justification: why this is foundational, not derived

Derivations (DV-n)

Claims that follow from axioms. Each must trace its logical ancestry.

Each entry must include:

• Claim: what follows
• Depends on: which AX-n or DV-n entries this derives from

Calibration example:

Rejected (circular): "DV-1. The system must be tested. Depends on: AX-1 (traces are required), AX-2 (operators produce structured output)." (The claim doesn't follow from those axioms — it follows from general engineering practice. The "Depends on" launders an unstated premise.)

Rejected (restatement): "DV-1. The pipeline must be staged. Depends on: AX-1, AX-2, AX-3, AX-4." (Cites every axiom without showing how any of them entails staging. A derivation that restates an axiom's implication in different words, or hedges by citing everything, has not derived — it has summarized.)

Accepted: "DV-1. Budget exhaustion must halt mid-pipeline, not retry. Depends on: AX-1 (fixed call ceiling), AX-3 (drift dominates past saturation). If calls are capped (AX-1) and extra calls degrade quality (AX-3), retrying on exhaustion makes the output worse, not better."

Assumptions (AS-n)

Statements treated as true without justification. The hidden load-bearing walls — identify them before someone removes one.

Each entry must include:

• Statement: what's being assumed
• Load-bearing: yes | no — would the argument collapse if this were false?
• Testable: how you'd check whether this assumption actually holds

Calibration example:

Rejected: "AS-1. The approach will work. Load-bearing: yes. Testable: we'll know when we try it." (Unfalsifiable — "we'll know when we try it" is not a test, it's a deferral. Name the observable that distinguishes success from failure.)

Accepted: "AS-1. LLMs can reliably distinguish structural transformation from restatement when prompted. Load-bearing: yes. Testable: run decompose on three inputs with and without the self-review instruction; compare axiom sets against human judgment of which items add structure vs. restate the input."

n is sequential starting from 1 in each section.

Self-review

After generating your output, review each axiom: state what structural relationship it reveals that the input text didn't make explicit. Axioms that restate the input in different words have not done work — strengthen or replace them.