Skip to main content

Foundation transfer audit

Use 60–90 minutes of the existing three-hour evidence defense for this cross-semester transfer. Prepare by indexing existing artifacts during the existing project-assembly allocation; do not rebuild the earlier labs. Semester 10 remains 530–785 hours. This audit makes G7's unseen-change requirement concrete and supports G3/G4; it adds no new gate or product-feature quota.

Entry evidence and honest status​

Carry the original prerequisite ledger, including S00 design/diagnosis and retention, S01 mathematical connections, S02 CA/AC computational-limits gates, S03 DS construction clinic, S05 PC native parallel work, S06 DB engine-boundary work and S08 SL statistical-learning bridge. Preserve S07 proposal-review and S09 artifact-selection records. Required earlier gates remain required even when the capstone product does not use their technique.

For each entry, record artifact revision, prediction, actual observation, independent explanation, changed case, assistance, reviewer status and unresolved limit. Use implemented, measured, modeled, proposed or pending precisely. A supplied reference passing locally is author evidence; it is not your independent implementation or review.

Select change cards appropriate to the project​

The reviewer selects two cards: one affecting correctness and one affecting evidence or operations. Give the learner a fresh variant, not the public worked case. If no independent reviewer participates, record a self-assessed rehearsal and leave the original external-review requirement pending.

  • Representation change: replace a nullable status with explicit draft/published/archived states. Identify valid combinations, API observations and migration/rollback behavior. Reuse S03's representation reasoning; a new type alone does not repair invalid stored data.
  • Performance pressure: a stakeholder requests more workers after seeing a faster kernel. State end-to-end measurement boundaries, partition ownership, serial work and failure cleanup. Reject unsupported speed claims. A justified sequential design can pass; no parallel product feature is required.
  • Uncertain outcome: the service writes state and loses the reply, then receives a retry. Trace commit, event publication, deduplication and response separately. State what actual engine evidence establishes and why process-exit checks do not prove power-loss durability.
  • Misleading score: a proposal reports excellent prediction or retrieval quality while repeated users cross the split. Identify the deployment question, independent unit, leakage and a defensible baseline. A no-model decision is acceptable. The capstone need not add machine learning merely to demonstrate literacy.
  • Changed release: the reviewed candidate passes, but the selected artifact or builder identity changes. Identify stale tests/attestations and verify the exact object reaching the running boundary. A valid signature with an unexpected identity is not sufficient.
  • Impossible promise: a requested checker claims always to decide whether arbitrary user programs terminate. Explain the scope of the impossibility argument, then propose a bounded execution or restricted-language contract with explicit unresolved outcomes. Do not confuse NP-hardness with undecidability.

What an adequate defense shows​

For each card: state the invariant and changed assumption; identify which old evidence survives; make a falsifiable prediction; reproduce a relevant existing check or explain exactly what new check is needed; and preserve the original decision before revising it. Critical missing runtime evidence stays pending until executed. A persuasive plan cannot close a failed authorization, restore or artifact-verification requirement.

Apply the existing G7 0–3 scale. Level 2 needs independent explanation, inspectable evidence and honest limitations; repeating the reference narrative is level 1. A changed case with a repaired claim supports level 3 only where the existing rubric permits it. All other module, strand, cloud-deployment, accessibility, peer and exam requirements remain conjunctive.

Advanced studios: credit evidence, preserve new work​

Carry matching proof, language, numerical, native performance, storage and evaluation artifacts into the advanced studios. Attempt each entry diagnostic before rereading. A passing prior artifact can satisfy the same diagnostic or setup step; count that work once. It cannot replace a studio's fresh transfer, expanded mechanism, research reproduction or maintenance period.

The computational-limits studio still requires its independent solver/pruning investigation; the S02 bridge is preparation. The performance studio still needs its workload-sensitive repeated experiment; the S05 histogram is a starting point. The residency still requires releases separated by observed use, migration, recovery and handover. Neither a capstone completion claim nor this audit fabricates elapsed maintenance or external participation.