The 281 listed nodes in this graph now carry 963 broker ratings. Sort them into quartiles by this year's return and one rule falls out: average-target upside is largest where returns are worst. The bottom quartile (averaging -20% YTD) shows 56% upside; the top quartile (+159%) shows 27%. Buy-share runs the other way, higher on the winners (81%) than the losers (77%). Targets are written down after prices without fully catching up, while opinions follow momentum.
Large upside means ungraded, not cheap
So a "top upside" list is not an undervaluation list. Two kinds of names sit at the top. One is a residual like Pony.ai (-54% this year, +192% upside), where the price fell and the target has not yet followed. The other is the rare overlap like Eugene Technology (+2.7% this year, +434% upside), the one equipment name left behind while its whole layer ran, where chain logic and upside point the same way. The upside number alone cannot tell these apart; chain position can.
The flipped sign is the real information
The sharper signal is negative upside, nodes whose prices have overtaken the average target: SUMCO (+226% this year, -48%), Jusung Engineering (+364%, -39%), Tokyo Electron (+78%, -32%), Kokusai Electric (+70%, -28%), KC Tech (+108%, -51%). All of it is the equipment and materials layer. The list shows exactly where analyst pencils failed to keep pace with the chain: memory and foundry capex propagated into equipment faster than consensus refresh cycles.
At the opposite extreme sits Synopsys, negative this year with a standing buy majority: a laggard by chain position, a conviction by consensus. Divergence like that means one side is wrong, which is precisely why it is worth watching.
How to use it
Node pages and essays on this site place chain position (returns) next to consensus. Read the chain first, upside second. Where chain logic explains the lag and upside is large, you have a candidate; where only the upside is large, you have a residual. And the flipped-sign layer is a record that a bottleneck has already been confirmed, which makes it the right starting point for asking where the bottleneck moves next.