Reread shared artifacts before incremental writes

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

While coauthoring, paragraph A becomes author-edited A2 after your read; you have approved addition B. Rewriting cached A+B would erase A2, so inspect current disk state before writing.

Mechanism

Reread before append/targeted edits, checking target/version. Merge agreed scope only, preserving human edits; compare unexpected target changes rather than blindly applying cached text. Keep raw material separate. Simultaneous writers also need revision checks, locking or coordination.

Bad example

Overwrite the article with cached A+B, losing A2, and edit raw interview material to match the synthesized draft.

Good example

Reread the teaching article before adding B. Preserve current A2 and append approved B only. For targeted rewrites inspect current target/differences and stay within scope; coordinate unexpected changes instead of overwriting others. Keep raw input unchanged and read back A2/B, with revision conditions/locking where needed.

Why the change matters

Disk is shared current state; cache is an earlier observation. Rereading/scoped merge preserves human intervention while conditional writes address subsequent races. Raw separation prevents interpretations from rewriting evidence.

Observable expectation

The fixture yields A2+B, not A+B. Intervening edits in the target produce explicit reconciliation.

Inspect readback, scope and unchanged raw inputs. Write success alone does not prove preservation.

Limits

Rereading does not remove the read/write race; real concurrent work needs atomic or collaborative controls and explicit ownership. This is not guaranteed automatic conflict merging.