All earlier work ↗
Historical engagement record
About this record
This is an earlier engagement or reference design, preserved as authored. Reported metrics and delivery status have not been independently revalidated for this preview; source dates are not established here. Modeled figures and design targets are not realized results. This record does not substantiate outcomes for the proposed product.
Explore Tensile today ↗
This is a capability and reference-architecture case study: how we scope and architect this build, not a delivered reference. The figures below are acceptance gates and design targets.
ECC end-of-support turns ABAP remediation into a deadline, and the deadline hides the real risk: you can rewrite thousands of custom programs and still not be able to prove they do the same thing. The compounding asset here is an engine plus a parity discipline. Deterministic transforms do the heavy lifting; the LLM handles the genuinely ambiguous edge; and nothing ships unless it replays historical production data to parity and clears a branch-coverage gate. The proposed design would return validated code through the customer’s transport and record the decision trail.
The challenge
Every SAP customer must finish the ECC→S/4HANA migration before support ends, and custom-ABAP remediation is the most heavily-staffed, highest-variance part of it. The real risk is proving the migrated code behaves exactly as the old code did. Brittle, manual remediation can’t prove equivalence at scale.
The approach
Static anti-pattern detection across the ABAP estate, then a deterministic codemod core with a scoped LLM edge for the cases that genuinely need it: deterministic first, LLM second.
Prove equivalence by replaying historical production data: reconciliation parity plus branch-coverage acceptance gates (“no green, no ship”).
Hand validated code back through the customer’s own transport, with a tamper-evident, attributable decision trail for audit.
Artifacts and handoff
A reference architecture and engine design: anti-pattern detection, deterministic codemod core + scoped LLM edge, the replay-based parity and coverage gates, customer-transport handback, and the tamper-evident decision trail.
Reported measures and design targets
Parity acceptance gate: reconciliation parity ≥99% on in-scope objects (design gate)
“No green, no ship”: branch-coverage gate before any code is accepted
Customer-owned handback: validated code returned via the customer’s own transport
Tamper-evident trail: append-only, attributable decision log
Explore the failure mechanisms, the tests they need and the evidence that establishes the customer outcome.
AI reliability, from production to customer outcome.