Building the Timeline

Opening Scene

Knowing which witnesses generally matter to cases of this type doesn’t tell you what happened in this specific case. A real investigation needs a timeline — a precise account of how each individual piece of evidence contributed to this one outcome. That’s the shift from feature importance to something far more specific: explaining one decision, for one person, in full detail.

SHAP values are the most widely used tool for building that timeline.

In Plain English

SHAP (SHapley Additive exPlanations) assigns each input feature a precise contribution value for one specific prediction, showing exactly how much that feature pushed the outcome up or down relative to a baseline. Unlike feature importance, which describes overall model behavior, SHAP answers “why did this applicant get this specific score” — with contributions that add up exactly to the difference between the baseline and the actual prediction.

The Old Way

Before SHAP formalized this, people attributed outcomes to specific factors much more informally:

A teacher explaining one student’s specific grade by pointing to individual assignments and their relative weight, without a rigorous, additive accounting.
An insurance adjuster explaining one specific claim payout by listing contributing factors, without a mathematically guaranteed breakdown that sums to the total.
A sports analyst attributing one game’s outcome to specific plays, based on judgment rather than a rigorous, provably fair allocation method.

These explanations were plausible but not provably complete — nothing guaranteed the listed factors actually accounted for the whole outcome.

What’s Changing (and Why AI Is the Reason)

SHAP is built on Shapley values, a concept from cooperative game theory with a rigorous mathematical guarantee: contributions are fairly allocated and always sum exactly to the actual prediction, closing the gap that informal explanations always had.
Efficient algorithms, particularly for tree-based models, have made SHAP practical to compute at scale, turning a theoretically elegant idea into something usable on real production models.
SHAP has become close to a default standard specifically because it works consistently across many model types, giving investigators one common language instead of a different bespoke method for every model.

The Metaphor, Fully Extended

The Investigation	SHAP Concept
The full timeline for one specific case	The SHAP explanation for one specific prediction
Each piece of evidence’s precise contribution to the outcome	Each feature’s SHAP value for that prediction
The evidence adding up to exactly explain the verdict	SHAP values summing exactly to the prediction
A rigorous, provably fair allocation of credit	The game-theoretic guarantee behind Shapley values

For Beginners: What to Actually Do

Start with a single prediction and its SHAP explanation before trying to interpret SHAP values across an entire dataset — the individual case builds real intuition first.
Understand that a positive SHAP value pushes a prediction up and a negative value pushes it down, relative to a baseline, not in absolute terms.
Practice explaining one SHAP result out loud to someone non-technical — this is exactly the skill Article 16 later builds on directly.

For Practitioners and Leaders: The Deeper Layer

Use SHAP explanations as the default response when a specific stakeholder asks about one specific decision, not feature importance, which answers a different question.
Be aware that SHAP computation cost scales with model complexity — a concern this series revisits directly in Article 17.
Combine SHAP explanations for individual cases with feature importance for overall model behavior — they answer complementary, not competing, questions.

Quick Recap

SHAP values explain one specific prediction by assigning each feature a precise contribution amount.
Contributions are guaranteed, by game theory, to sum exactly to the actual prediction.
SHAP works across many model types, making it close to a common standard for individual-decision explanations.
SHAP and feature importance answer different questions — one specific, one general — and are genuinely complementary.

Where This Fits in the Series

This article covered building a precise timeline for one case using SHAP. Article 6 covers a related but distinct technique for the same underlying question.