A Principal Variation Is a Search Snapshot
The PV is the engine's current preferred continuation from the root under the present search state. It is enormously useful for understanding why a move works, but it is not automatically a forced line. Quiet alternatives, transpositions or defensive resources may exist outside the displayed sequence.
Read the PV for the Mechanism
Instead of copying twenty moves, ask what the line demonstrates. Does it show a tactical refutation, a defensive resource, a favorable liquidation, a king-safety mechanism, a pawn break, or simply that one candidate reaches a healthier endpoint? Stop quoting the line once the lesson is established.
MultiPV Changes the Question
Single-PV asks, “What does the engine currently prefer?” MultiPV asks, “How do several root candidates compare under the allocated search?” That makes MultiPV useful for auditing human candidates and repertoire alternatives. It also divides search resources across lines, so a wide MultiPV is not “more truth for free.” Use the smallest number of lines that answers the analytical question.
Depth Is Only One Search Coordinate
Depth, selective depth, nodes, elapsed time, hardware, hash, threads, tablebase access, engine version and position type can all affect what you see. A reported depth is not a portable certificate of confidence. Two analyses labeled “depth 30” need not represent equivalent work.
Stability Is More Informative Than a Single Snapshot
Track three kinds of stability:
- best-move stability — does the preferred root move keep changing?
- evaluation stability — is the score settling or swinging materially?
- PV stability — does the explanatory mechanism persist even when move details change?
A position may have stable evaluation but unstable move order because several moves transpose. Conversely, a stable first move can hide a still-changing tactical assessment deeper in the line.
Stop When the Evidence Matches the Claim
You do not need maximal search for every annotation. A small practical lesson may require only enough analysis to establish the mechanism. A claim of forced mate, exact novelty refutation or critical theoretical evaluation deserves stronger verification. Match search effort to claim strength.
Transpositions Can Make PV Instability Harmless
A PV may change its move order while reaching essentially the same position and strategic conclusion. Before treating a line change as evidence of uncertainty, check whether the branches transpose. Conversely, identical first moves can lead to materially different deeper evaluations, so root stability alone is insufficient.
MultiPV Is Best Used at the Root Question
If your analytical question is “Which of these candidate moves works?”, MultiPV is valuable at the common root. If the question becomes “What is the best defense after candidate A?”, narrowing back to a focused search can be more efficient. Move between breadth and depth according to the question rather than leaving one setting fixed for the whole session.
Reference Framework
Engine search is iterative and selective. The current best move, evaluation and principal variation can change as the search invests more resources.
A useful audit therefore records not only an evaluation but also search context and stability.
Depth is metadata, not a universal confidence meter.
Principal Variation
The principal variation (PV) is the engine's current preferred continuation associated with the searched root choice.
PV Is Current Preferred Line
PV is the current preferred line under the current search state/settings; it can be the best explanatory sample the engine has, not a timeless truth.
PV Is Not Necessarily Forced
PV is not necessarily forced. Opponent alternatives may be nearly equal, and the line can change as search deepens.
PV Can Change
PV can change because iterative deepening/search discovers different moves, refutations, transpositions or evaluations.
Explain the PV
Explain what the PV demonstrates—tactical refutation, structural consequence, defender resource, endgame transition—instead of dumping moves.
What MultiPV Does
MultiPV asks the engine to analyze/output multiple leading root moves/variations.
Compare Human Candidates
MultiPV is valuable for comparing the player's actual candidates because it can show whether several moves are close or one candidate is clearly inferior.
Explore Repertoire Alternatives
MultiPV can explore repertoire alternatives, but opening choice still needs database/model-game/human plan context.
MultiPV Resource Cost
MultiPV consumes search resources. Current Stockfish documentation explicitly notes that computing more lines reduces resources available to the best line at a fixed resource budget.
Use Only Needed Number of Lines
Use only the number of MultiPV lines needed for the analytical question. More lines are not automatically better analysis.
Depth
Depth is a search-progress indicator tied to the engine's iterative/selective search, not literal uniform plies searched everywhere.
Selective Depth
Selective depth (seldepth) reports a deepest selective/PV-related search depth concept and is not interchangeable with nominal depth.
Nodes
Nodes counts positions/search nodes processed and is useful resource metadata, but comparisons across versions/settings/hardware can be misleading.
Search Time
Search time is a direct resource measure, but equal time on different hardware/configurations does not imply equal search work.
NPS
NPS measures nodes per second. Higher NPS alone does not prove stronger or more accurate analysis because engine version/search architecture/settings differ.
Tablebase Hits
Tablebase hits indicate tablebase probes occurred during search; they do not mean the entire root analysis is tablebase-exact unless the root falls within exact coverage/conditions.
Why Depth Alone Is Not Enough
Depth alone is insufficient for confidence. Consider best-move/eval/PV stability, resources, MultiPV, tablebase status, version and whether the search is still resolving.
Best-Move Stability
Best-move stability means the leading root move remains consistent across additional meaningful search.
Evaluation Stability
Evaluation stability means the score stops moving materially under additional search/resources.
PV Stability
PV stability means the explanatory continuation becomes less volatile in its important branches.
Search Still Resolving
Search still resolving means best move, evaluation or critical line remains volatile enough that strong conclusions should be provisional.
Engine Disagreement
Engine disagreement is a signal to deepen/check settings/versions/position rather than average the scores mechanically.
Preserve Uncertainty
Preserve uncertainty when the analysis is unstable or tools disagree. Unresolved under current resources is better than false certainty.
Search-Stability Practice
Try a search-stability drill: watch iterative output as the best move, evaluation and PV change, then decide when the evidence is stable enough for the analytical claim you want to make.