Difficulty grading › Guide

Which Puzzle Steps Are Hardest? What Data Can Tell Us

By Cheng Xu · Published 2026-11-04

Completion rates and solve times cannot by themselves identify the hardest deduction. Solvedle currently records board starts, finishes and some interaction events, not a timestamped proof of each move. There is not yet enough evidence here to publish a player-based ranking of techniques.

This page explains the measurement problem and a study design. It is not a report claiming that a particular technique causes abandonment. The site launched on October 29, 2026; the observation window available for this November 4 publication is shorter than the proposed 2–4 week exploratory window.

What is observable today?

A defensible first report: board-level funnels

1. Choose a fixed date window and report its timezone. Separate the puzzle-set date from the UTC event timestamp.
2. Count distinct browser identifiers starting each board and match their solves to that same set and slot.
3. Report matched solvers / starters for each board, with the numerator and denominator beside the percentage.
4. Report median and other duration quantiles among solvers only, together with sample size and excluded or invalid timings.
5. Separate daily play from practice. Allow observation time for late completions; avoid comparing an unfinished current day with fully observed days.
6. For next-day return, use only cohorts whose following day has completed its observation window.

For the daily trio, preserve order in interpretation: slot 0 is 6×6 Tango, slot 1 is 8×8 medium one-star and slot 2 is 9×9 hard one-star. Users reaching slot 2 are already a selected group. A higher or lower completion rate cannot isolate the effect of the puzzle type or technique.

Why a pause is not automatically a difficult deduction

A person may leave to answer a message, switch devices, read the rules, stop the browser or simply run out of time. A long recorded session can include inactive time; a missing solve event can include lost connectivity. Solve durations are client reports, not independently verified proof replays.

Likewise, a board labelled hard may contain several advanced deductions plus visual scanning overhead. Board-level abandonment cannot tell us which one was the cause. Aggregating by grid size mixes puzzle construction, presentation, player selection and technique demands.

What a technique-level study would need

A separate, deliberately scoped study would need puzzle identifiers, valid intermediate states, the deduction available at each state, and time or action measurements between those states. It should distinguish an incorrect move, a request for help, inactivity and a valid alternate solve order.

Before collecting additional move data, define the research question, keep only the necessary signals, and state the privacy and retention choices. This article does not add that instrumentation. Even with such data, an observational association between a technique and stopping would not by itself prove causation.

When will a real ranking be justified?

Two to four weeks is a possible first review point, not a sample-size guarantee. Publish a board-level report only when each displayed comparison has meaningful denominators and stable observation windows. If samples remain small, keep collecting or report the uncertainty instead of creating a leaderboard of techniques.

For a ranking defined by the solver rather than player behaviour, read how the one-star grader assigns difficulty. For a concrete skill demonstration, use Queens checkpoints, Takuzu distinct-line examples, or Tango next moves.

Source contracts: event validation and stored fields, daily metrics definitions and engagement queries.