I now have a complete per-batch picture. Here's the answer.

## Which letter-suffix batches are blends vs. real synthesis variants

The rule (user-confirmed scientist walkthrough): a letter suffix is either a solvent/additive blend of the parent batch (no synthesis data — application tests only) or a synthesis variant (has its own reaction rows) [context/30-entity-links.md, Confirmed links, "Letter-suffix batches can be blends"]. Applying that test to each letter-suffix record:

Blends (no reaction rows — only an application sample + tests):

• 48 K, 48 BC, 48 PA50, 48 B99 (samples S 9076 #48K/#48BC/#48PA50/#48B99)
• 50 A, 50 B
• 51 A, 51 B, 51 C
• 52 A
• 55 D

These carry 0 process steps but a linked sample with 2–4 application tests.

Real synthesis variants (own reaction rows with reactants/intermediates + reaction results):

• 24 A, 24 B, 24 C, 24 CI, 24 CR, 24 CR2
• 25 A, 25 B, 25 I, 25 II
• 26 A, 26 B
• 27 A, 27 B, 27 C, 27 D, 27 E
• 28 A, 28 B, 28 C
• 29 A, 29 B, 29 D
• 34 B, 34 C, 37 A, 38 C, 38 D
• 48 Q (own ELN; 3-step synthesis) and 54 A

Re-interpret the 24 CR / CR2 / 24 CRREWORK trio: the 24 CRREWORK record is a sample-level rework ("reactor workup carried out on the above parent batch product; not one of the parent's three synthesis steps" — exp-9076-048Q-type note), so the "CR" line is a rework, not a fresh synthesis.

Gaps / caveats:

• Batch 24 CI and 28 C records errored on my step read (one result row has a missing parameter); both show 3+ steps in the list, so they fall in the synthesis-variant class, but I could not enumerate their full material roles.
• The "blend vs variant" split is a per-batch judgement per the rule above, not a field stored in the records — I inferred it from presence/absence of reaction rows.