Problem

packages/slate-history/test/** had 17 deleted legacy fixtures, but they did not all mean the same thing anymore.

Some still matched the live slate-history contract. Others depended on wider legacy behavior that the current package no longer claims.

Without classifying that split first, the closure would either:

• resurrect dead behavior just to make the count look smaller
• or close the family with docs that outran the real proof

Root Cause

The deleted family mixed three different buckets:

1. live history semantics that still belong in the current contract
2. harness residue with no contributor-facing value
3. legacy-width behavior that exceeds the current package claim

The tricky part was that the live package already advertises a narrower history model:

• one outer transaction is one undo unit
• merge behavior is explicit through helper scopes
• fragment/delete proof is narrower than the old fixture corpus

If you treat the deleted file list as one undifferentiated “restore it all” queue, you lose that distinction immediately.

Solution

Close the deleted family against the live contract, not the raw legacy file count.

For slate-history, that meant:

1. freeze the deleted inventory first
2. group the deleted files by current claim cluster
3. add direct proof only for the live supported seams
4. mark wider legacy rows as explicit skips instead of pretending they were recovered
5. update the roadmap/ledger language in the same turn so the docs match the proof

The recovered proof landed in history-contract.ts and now directly covers:

• History.isHistory(...) across edit, undo, and redo lifecycle
• direct insertText(...) undo
• reverse block joins and reverse same-text deletes
• same-text delete undo with custom props and inline nodes
• delete-fragment selection restore after deselect/refocus
• insertBreak() undo
• current simple block fragment undo

The deleted rows that were wider than the live contract were cut explicitly:

• cross-block custom-prop delete fixture
• cross-inline multi-block delete fixture
• deep nested fragment fixture
• timing-based contiguous/non-contiguous auto-merge heuristics

The closure note for the family lives in 2026-04-09-slate-v2-slate-history-deleted-test-family-closure.md.

Why This Works

The closure becomes honest because it follows the same thing the release claim is supposed to follow: the current proved surface.

That avoids two bad failure modes:

• rebuilding dead legacy width just because deleted files existed
• quietly narrowing the contract without saying so

It also keeps parent-package truth clean.

In this batch, packages/slate-history/test/** could close, but packages/slate-history/** had to stay open because sibling deleted residue still exists in:

• packages/slate-history/src/history.ts
• packages/slate-history/CHANGELOG.md

Prevention

• For deleted-family closure work, freeze the inventory first and group it by live claim cluster before writing any closeout language.
• Recover direct tests only for behavior the current package actually claims.
• If a deleted fixture depends on wider legacy behavior than the current package advertises, mark it as an explicit skip and say why.
• Do not let roadmap or ledger docs say a child bucket is open after fresh proof closes it.
• Do not let a closed child bucket silently close the parent package when sibling deleted residue still exists.
• For slate-history specifically, treat these as the live contract anchors:
  • one outer transaction is one undo unit
  • merge semantics are explicit helper decisions, not timing heuristics
  • fragment/delete width must match the current proved core seams