Managing Implementation Slices

Keep engineering work in small batches that can be understood, reviewed, verified, and integrated without waiting for the entire initiative to finish.

Define a slice

A good implementation slice:

• serves one coherent purpose;
• leaves the repository in a working state after it lands;
• includes the tests or verification needed for its behavior;
• has a reviewable diff with minimal unrelated churn;
• makes dependencies on earlier or later slices explicit;
• has a clear rollback or containment story when the change is risky.

Small means conceptually focused, not an arbitrary line count.

Prefer vertical progress

When possible, cut through the smallest set of layers needed to demonstrate a useful behavior. A vertical slice is usually easier to validate than separate “frontend”, “backend”, and “tests” branches that are individually incomplete.

Horizontal enabling slices are appropriate when they create a stable seam, compatibility layer, schema expansion, test harness, or shared contract that later work can safely build on.

Keep changes independently safe

Each merged slice should preserve system correctness. Do not rely on a later pull request to repair a deliberately broken intermediate state.

Use compatibility techniques when needed:

• additive schema or contract changes before removals;
• feature flags or dormant code paths for incomplete rollout;
• adapters/seams before replacing implementations;
• tests before high-risk refactors when behavior is not already protected.

Separate concerns that review differently

Split work when combining concerns makes evidence harder to interpret. Common candidates:

• mechanical refactor vs. behavior change;
• dependency upgrade vs. feature adoption;
• schema expansion vs. data migration vs. contract cleanup;
• generated output vs. hand-authored logic;
• infrastructure enablement vs. product behavior.

Do not split so aggressively that reviewers cannot understand the purpose or validate the slice in isolation.

Handle dependencies explicitly

If slice B depends on slice A, say so. Prefer merging A first. When continued work must proceed before A merges, use stacked pull requests or another explicit dependency mechanism rather than hiding both changes inside one oversized branch.

Control scope creep

When new work appears, classify it:

• required for this slice — without it the accepted behavior is incorrect or unverifiable;
• follow-up — useful, but the current slice can safely land without it;
• unrelated — move it out immediately.

Do not expand a slice because the nearby code “could also be cleaned up.”

Completion check

Before requesting review, confirm:

• the slice has one explainable purpose;
• the diff contains no accidental neighboring work;
• relevant tests/checks pass;
• the system remains usable after this slice alone;
• dependencies and follow-ups are documented;
• the next slice can start from a known state.

References

• DORA, Working in small batches: https://dora.dev/capabilities/working-in-small-batches/
• Google Engineering Practices, Small CLs: https://google.github.io/eng-practices/review/developer/small-cls.html
• GitHub Docs, About stacked pull requests: https://docs.github.com/en/pull-requests/get-started/about-stacked-prs