Current research snapshot
The methodology version attached to the current DDI dataset.
The generation date recorded with the current dataset.
The complete original Microsoft FreeCell deal range analysed by the current pipeline.
Deals with a numeric DDI score and comparative ranking.
1 deal is kept as unsolvable and 3 deals are currently left unresolved rather than being forced into the numeric ranking.
What the DDI measures
The FreeCell Deal Difficulty Index is a comparative model. It applies the same versioned analysis framework across the classic 32,000-deal set and combines search-based evidence with structural properties of each deal to place rated boards on a common difficulty scale.
A higher DDI means that a deal ranked as more difficult within this model. The score is useful for comparing deals against one another; it is not a claim that the model captures every element of human difficulty.
Signals used by the model
The current public methodology groups the model inputs into three broad areas:
- Search effort: evidence about how much work different search approaches require before they can resolve a deal.
- Opening structure: signals describing the starting position and how restrictive or permissive the opening is under the analysis model.
- Search structure: signals describing properties of the explored state space that help distinguish easier and harder deals.
FreeCell Now does not publish the exact internal weighting formula, search-profile configuration or operating thresholds. The methodology version is exposed so material model changes can be identified rather than silently mixed with earlier results.
Search robustness in v0.4.2
Earlier versions could be overly sensitive to the behavior of a particular search strategy on an individual deal. Version 0.4.2 reduces that sensitivity by comparing evidence from multiple independent search approaches rather than allowing one search path to dominate the difficulty estimate.
The model is also designed so machine speed is not itself a difficulty signal. Solvability certification is handled separately from the comparative difficulty score, which helps prevent extra work used to establish solvability from automatically inflating a deal's DDI.
Why first-solution length is not weighted
Solution-path length is intentionally not used as a weighted feature in the current DDI. A search algorithm's first discovered solution is not guaranteed to be move-optimal, so treating that path length as a direct difficulty measure could reward or penalise deals for properties of the search order rather than the board itself.
This is also why DDI should not be interpreted as a minimum-move benchmark.
How ranking and percentiles work
Only deals with a supported numeric result enter the ranked population. Rated deals are ordered by DDI score, and the percentile indicates where a deal sits relative to the other rated deals. A higher percentile means the deal ranks among the more difficult boards in the current model.
Unsolvable and unresolved results are kept outside that numeric ranking. This avoids inventing a score where the pipeline does not support one.
Known limitations
- DDI is a model of comparative deal difficulty, not a direct measurement of every human player's experience.
- The public methodology intentionally does not disclose the exact weighting formula, search-profile configuration or internal operating thresholds.
- The first solution found by a solver is not assumed to be move-optimal.
- Unresolved deals are not treated as unsolvable and are not assigned a numeric score without enough evidence.
- Future methodology versions may change individual scores or relative rankings; material changes should be versioned and logged.
Dataset provenance
The current DDI dataset is created and maintained by FreeCell Now and covers the original Microsoft FreeCell deal range from #1 through #32,000. The dataset record on the main Deal Difficulty Index page exposes the methodology version and generation date from the live dataset metadata.
FreeCell Now is the research creator and publisher identity for this work.
Updates and corrections
Material methodology changes should receive a new version and be reflected in the Corrections & Updates log. If you believe a deal result or methodological statement is wrong, email [email protected] and include the deal number or affected URL together with evidence that helps reproduce the issue.