Record Principles, Not Prompts
Record reusable production principles rather than preserving exact prompts as the public knowledge artifact.
Capture
Successful prompts aged quickly.
Model updates, moderation changes, new reference assets, and different scene goals altered their behavior. The exact wording lost value while the reasoning that produced it remained useful.
Prompt libraries preserved many artifacts without always preserving why they worked.
Why
A working prompt appears to be the most concrete form of knowledge.
It can be copied, stored, and reused, so it seems like the natural unit of production memory.
Why-Not
Prompts encode an implementation for a particular system state.
Principles such as reducing decisions, separating motion layers, aligning geometry, and changing one variable at a time remained useful even when the wording changed.
The scar was learning that archiving yesterday's syntax is not the same as preserving tomorrow's method.
Commit
Decision: Record durable production knowledge as decisions and principles. Preserve prompts as examples, not doctrine.
For each successful technique, capture the assumption, the evidence that contradicted it, and the rule adopted in response. Let implementation evolve without erasing the accumulated scars.
After all repeatable execution decisions have been externalized, one decision remains deliberately unassigned: whether the resulting work is right.
Timestamp
2026-07-24