Snapshot the current state before each pipeline stage transition — recoverable checkpoints throughout production prevent catastrophic rework
Before an output advances from one pipeline stage to the next (draft to review, review to polish), snapshot the current state as a version to create recoverable checkpoints throughout the production process.
Why This Is a Rule
Pipeline stage transitions are the natural checkpointing moments in output production. The draft is complete → snapshot → begin review. Review is complete → snapshot → begin polish. Each snapshot captures the state at a known-good boundary, creating recovery points if something goes wrong in the next stage.
Without snapshots, a bad review pass (accidentally restructuring in a way that loses the original argument) or a bad polish pass (over-editing that strips the voice) has no rollback capability. You'd need to reconstruct from memory or start over. With snapshots, you revert to the pre-review or pre-polish state and try again. This is Preserve intermediate outputs as concrete artifacts at every workflow boundary — enables resumption, failure recovery, and reuse's intermediate artifact preservation applied specifically to single-output production.
The snapshot is also a decision record: it preserves what the output looked like at each stage boundary, enabling retrospective learning. You can compare the draft snapshot to the final output and identify which changes added value and which were unnecessary rework. This feedback loop improves your production process over time.
When This Fires
- At every pipeline stage transition for outputs that justify version tracking (Scale versioning to stakes: full (named versions + changelogs) for high-stakes, date-stamped for medium, none for low — match overhead to value)
- Before any destructive editing operation (major restructuring, significant cuts, tone overhaul)
- When you're about to make changes you might want to undo
- Complements Move outputs forward through pipeline stages (Draft → Review → Polish → Deliver) — no skipping, no backward oscillation without explicit regression decisions (pipeline stages) and Scale versioning to stakes: full (named versions + changelogs) for high-stakes, date-stamped for medium, none for low — match overhead to value (versioning tiers) with the specific timing of version captures
Common Failure Mode
"I'll remember what it looked like before": relying on memory for the pre-edit state. Memory of the original text degrades within minutes of editing — you can't un-see the changes. The snapshot is the only reliable way to compare before and after or to revert if the changes made things worse.
The Protocol
(1) Before advancing an output from one stage to the next, save a snapshot: "Draft-final.md," "Post-review.md," or use your version control system's snapshot capability. (2) The snapshot should be a complete copy, not a diff. You need to be able to open it and see exactly what the output looked like at that boundary. (3) Label the snapshot with the stage boundary: "Draft-complete-2026-03-30" or "Pre-polish-v2." (4) If the next stage goes poorly (bad restructuring, over-editing, lost coherence), you can revert to the snapshot and retry from the stage boundary rather than starting over. (5) For short outputs (under 500 words), a quick copy-paste into a "previous versions" section at the bottom of the document may suffice. For longer outputs, use separate files or version control.