27  Dead Ends

Things that were tried and abandoned, with a pointer back to the journal entry — kept so they don’t get quietly re-attempted later.

• Delete-then-insert factor-value lift (reverted) — replacing the INSERT … ON CONFLICT DO UPDATE upsert into gold.fact_factor_value with a per-factor DELETE + plain INSERT to dodge an operator-backfill OOM. It silently dropped the UPSERT’s intra-statement de-duplication (breaking ~20 annual factors on a duplicate-key error) and still OOM’d at full scale because memory accumulates across factors on one connection. A single-factor test validated it and missed both. See 2026-08-09.