17. Entropy

Definition

Entropy is the natural tendency of a data model to lose order, clarity, and analytical meaning over time, as changes, exceptions, patches, and unstructured decisions accumulate.

A model does not degrade all at once.
It degrades gradually.

In Power BI, entropy appears when the model continues to function technically, but becomes increasingly difficult to understand, maintain, and extend.

Nature

Temporal, structural, and cumulative.

Entropy is not a one-time error.
It is a continuous process.

A model can be:

• correct today,
• useful this quarter,
• functional this year,

and still be entering entropy if its evolution is not controlled.

Function

Recognizing and managing entropy allows teams to:

• preserve model quality,
• maintain long-term coherence,
• and avoid silent degradation.

In Power BI, managing entropy:

• protects analytical sustainability,
• reduces technical debt,
• and keeps the model readable over time.

Consequence

When entropy is controlled:

• the model remains clean,
• changes are understandable,
• analytics retains its value.

When entropy dominates:

• the model becomes confusing,
• each adjustment costs more effort,
• analytics is abandoned or replaced.

Entropy does not break the model.
It wears it down.

Signals of Low Entropy (✔️)

• Clear and stable structure.
• Well-named and organized measures.
• Reusable logic.
• Documented or self-explanatory changes.
• Ease of onboarding new analysts.

Signals of High Entropy (❌)

• Duplicated and contradictory DAX.
• “Temporary” measures that become permanent.
• Accumulated exceptions without refactoring.
• Difficulty understanding the full model.
• Phrases like: “better not touch that”.

🔧 Samples — Entropy in Power BI

🔹 Sample 1 — Patch accumulation

❌ High entropy:
Each new requirement is solved with an additional measure without reviewing existing ones.

✅ Low entropy:
The model is periodically refactored to integrate changes structurally.

👉 Complexity is managed, not accumulated.

🔹 Sample 2 — DAX duplication

❌ High entropy:
Multiple measures perform the same logic with slight variations.

✅ Low entropy:
Base measures are reused and extended.

👉 Less code, more meaning.

🔹 Sample 3 — Unabsorbed exceptions

❌ High entropy:
Special rules hardcoded for isolated cases.

✅ Low entropy:
The model absorbs exceptions through design, not patches.

👉 Order prevails.

🔹 Sample 4 — Disordered growth

❌ High entropy:
New tables and columns added without a common criterion.

✅ Low entropy:
Growth guided by a clear semantic framework.

👉 The model grows without losing shape.

🔹 Sample 5 — Anti-pattern vs Pattern

❌ Anti-pattern — Entropic model

• Accumulated complexity.
• Costly maintenance.
• High fragility.

✅ Pattern — Model with controlled entropy

• Conscious complexity.
• Continuous refactoring.
• Analytical longevity.

📌 Practical rule:
If the model works but nobody understands it,
entropy has already taken hold.

Interactions with Other Properties

• Plasticity enables adaptation without increasing entropy.
• Consistency slows semantic degradation.
• Transmissibility counters loss of understanding.
• Entropy measures model wear over time.

Synthesis

Entropy cannot be eliminated.
It must be managed.

A mature model:

• accepts change,
• controls complexity,
• and preserves meaning over the long term.