 Solution

Use a longitudinally segmented electromagnetic calorimeter with fine transverse position readout. The electromagnetic shower centroids in two separated depth layers define a line, providing photon pointing without assuming the longitudinal collision position. Combine this information with the reconstructed primary collision vertex from charged-particle track reconstruction. The two lines from that vertex to the shower positions then determine the opening angle.

For an order-of-magnitude requirement, a single-direction precision of $10\mathrm{mrad}$ corresponds to a transverse position uncertainty of $15\mathrm{mm}$ over a $1.5\mathrm{m}$ vertex-to-shower lever arm, provided the vertex is known sufficiently well. Centroid reconstruction across finely segmented cells can reach this scale. At high collision pileup, identifying the correct primary vertex is essential: using the wrong vertex can dominate the angular error even if the shower position is accurately measured. Converted photons offer an additional pointing constraint from the reconstructed $e^{+}{e}^{-}$ conversion tracks, but conversion is not required by the method described here.