Fragment approaches are long-established in target-based ligand discovery yet their full transformative potential lies dormant, because progressing hits to potency remains underserved by methodological work. The only credible progression paradigm is multiple cycles of costly conventional design-make-test-analyse (DMTA) medicinal chemistry, necessitating picking winners early and discarding others. It is effective to cheaply parallelize large numbers of non-uniform multi-step reactions, because, even without compound purification, a high-quality readout of binding is available, viz. crystallography. This can detect low-level binding of slightly active compounds, which the targeted binding site extracts directly from crude reaction mixtures (CRMs). In this proof-of-concept study, we expand a fragment hit from a crystal-based screen of the bromodomain PHIP2, using array synthesis on low-cost robotics to implement 6 independent multi-step reaction routes of up to 5 steps, attempting the synthesis of 1876 diverse expansions, designs entirely driven by synthetic tractability. The expected product was present in 1108 (59%) CRMs, detected by automated mass spectrometry, 22 individual products were resolved in crystal structures of CRMs added to crystals, providing an initial SAR map, pose stability in 19 and instability in 3 products and resolved stereochemical preference. One compound showed biochemical potency (IC50=34 μM) and affinity (Kd=50 μM) after resynthesis.