# Suited connectors facing a three-bet: selected study data

Retrieved September 13, 2026. Six-handed NLHE cash, equal starting stacks of 50 or 100bb, no ante. Hero opens to2.5bb; villain three-bets to7.5bb from BTN or12bb from a blind; all other players fold. Catalogued as5% rake with4bb cap, as displayed by the original app. The underlying settings for these legacy cash artifacts were not independently available. Full solver version, postflop abstraction and achieved convergence are unavailable.

Primary hypothesis: calling every positive-opening-reach hand in54s65s76s87s98sT9sJTs is within0.1bb of the best stored action for at least95% of their marginal combo mass in every chosen node. Four discovery nodes: BTNvsBB and COvsBTN, both depths. Eight held out: HJvsBTN, UTGvsBTN, COvsBB, BTNvsSB, both depths. These are twelve related configurations, not eighty independent solves.

Local call loss = max(all stored action EVs) minus call EV. An action amount denotes total preflop contribution, not an extra payment. Fold EV is0 at this decision; the opening raise is already committed. Frequencies are decoded placeholder probabilities. Weights preserve actual opening reach. Within-node means and quantiles use4*hand-class reach weight. Range composition uses6 pair,4 suited,12 offsuit combinations. Ranges are marginal, before mutual card removal and folded-player card bunching; these are not deal-frequency estimates. Board/hand occurrences are not simulated. No complete changed policy or adaptive opponent was evaluated.

80 of84 requested hand/node rows have positive reach. Four absent rows stay explicitly excluded. Tiny reach is not replaced with a convenience floor. The0.02bb near-tie label and0.05/0.1/0.2bb sensitivity screens describe this study, not universal error definitions. Unknown convergence limits tiny differences. Public frequencies/EVs retain more digits for arithmetic than the article displays.

The two range figures show one selected50bbBTNvsBB example. Hero's grid shows its conditional response mix. Villain's single-color grid shows the reached three-bet range membership and entry weights only; it is not a next-action policy. Both entering ranges were verified by multiplying previous range weights by the player's chosen action frequency and comparing a current exported response node.

Files: tested-hands.json (selected classes only); node-summary.json (aggregates and sensitivity); range-views.json (two selected explanatory grids); range-composition.json (defined class partitions); call-loss-map.svg (plot derived directly from node-summary.json). Full proprietary exports, catalog and access information remain private.

Interpretations are hypotheses consistent with these outputs. Price, position, ranges and native continuations change together across configurations. No specific blocker, equity-realization or causal range-advantage mechanism was measured.

Independent audit: no >0.1bb regret with>=10% use in eligible connectors or positive-reach hero response ranges. Six such flags occur in villain responses to four-bets in five other configurations; these later response strategies were not published as a policy recommendation. Dropping opening weights below0.001,0.01 or0.05 preserves all twelve main threshold verdicts. The flags and unknown achieved convergence limit exact equilibrium claims; the article compares the stored values.