REPOBOUND PROOF LAB

Real context compilations.

Inspectable evidence. Reproducible state. Three deterministic cases from the actual RepoBound product path.

01 / TASK CONTEXT

Fix session race condition

A task becomes a bounded Context Receipt: source, store, repository instructions, and architecture evidence.

Open case →

CONTEXT SNAPSHOT5a92ee4b4567

Fix session race condition

4SELECTED

11DROPPED

669 / 800ESTIMATOR TOKENS

src/session.ts
src/session-store.ts
docs/SESSION.md

02 / REVIEW CONTEXT

Review a tracked ledger change

A real tracked change produces bounded surrounding context and explicit structural evidence.

Open case →

REVIEW RECEIPTf88df456f0d0

src/ledger.ts

4SELECTED

7DROPPED

823 / 1kESTIMATOR TOKENS

src/caller7.ts
└─ imports → src/ledger.ts
STRUCTURAL FACT

03 / CONTEXT DEBUGGING

Why was the integration test dropped?

Read the recorded budget decision, apply Include, and compare the immutable rebuild.

Open case →

SEMANTIC DIFFREAL CONTROL

integration/session.integration.test.ts

DROPPEDSELECTED

Recorded reason: section allocation could not fit this candidate.
Control: Include.

Fixture evidence only. RepoBound compiled context; no coding model executed. These cases do not establish coding success, review accuracy, productivity, completeness, or token savings.