SFT Part VI

Case Studies

Part VI uses concrete cases to show how SFT reads artifacts, migrations, incidents, AI-generated shortcuts, and end-of-life decisions as changes to operation support, policy, observations, and field memory.

Coupon PRD

The coupon PRD is the running example. It targets checkout and payment as a feature addition artifact action, but an underspecified PRD can expose several operation families.

ConsequenceEnvelope path classes
lawful policy insertion path
  + hidden PaymentAdapter dependency path
  + rounding semantic obstruction path
  + UI-only discount drift path
  + unknown / unmodeled remainder

When the specification states rounding order, discount composition, payment authorization boundaries, and refund semantics, support and policy can be shaped toward the lawful path.

Coupon PRD descriptor
RawArtifactAction
  source artifact = PRD
  target field components = requirements, constraints, operation support
  target architecture regions = Checkout / Payment
  action boundary = coupon feature addition only
  non-conclusions = pricing strategy, marketing success, fraud model
  • Missing invariants Rounding order, discount composition law, payment authorization boundary, and refund / cancellation semantics.
  • Claim boundary Specification tightening narrows selected witness families; it does not prove global risk reduction.

Migration

Migration is a field transformation with old-new projection mismatch and partial migration risk. SFT reads migration as reconfiguration of field memory and operation support, not just replacement work.

  • Bridge path A temporary relation between old and new projections.
  • Dual-run path A field state where two implementations coexist under a comparison boundary.
  • Rollback path A supported path back from selected migration states.
MigrationEnvelope
bridge path
  + dual-run path
  + replacement path
  + rollback path
  + old-new projection mismatch
  + partial migration risk
  + consumer compatibility boundary

Incident response

Incident response forces a field update through runtime observation. The incident can classify root-cause witnesses, reveal missing invariants, update review or CI governance, and revise the forecast boundary.

A local post-incident fix does not prove future trajectory safety. SFT keeps the missing invariant and observation boundary visible.

IncidentFeedback
incident observation
  + root-cause witness classification
  + missing invariant discovery
  + review / CI governance update
  + runtime observation update
  + forecast boundary revision

AI-generated shortcut

AI-generated proposals can amplify local patterns and make shortcut paths look cheap. SFT evaluates them against both AAT theorem boundaries and SFT consequence envelopes.

AIShortcutCase
prompt boundary
  + generated operation support
  + selected shortcut path
  + missed theorem boundary
  + review / CI mediation
  + observed obstruction witness
  + posterior field update

End-of-life decision

End-of-life is a lifecycle governance choice. SFT compares architecture signature, repair cost, migration support, runtime risk, ownership boundary, staffing boundary, and the consequence envelope.

It does not predict human intention or market success. It describes how repair, migration, contraction, or deletion changes architecture futures and field capacity.

EndOfLifeDecision
current architecture signature
  + repair cost
  + migration support
  + runtime risk
  + ownership / staffing boundary
  + ConsequenceEnvelope
  + non-conclusions

Case-study boundary and non-conclusions

These cases show how the same formal vocabulary is used across feature work, migration, incidents, AI proposals, and lifecycle decisions. They are examples of bounded diagnosis, not claims that the artifact caused every observed outcome or that the repository has been completely reconstructed.

  • Coupon PRD Demonstrates artifact action, missing invariants, and `ConsequenceEnvelope` path classes.
  • Migration and incident response Demonstrate field transformation and runtime feedback as updates to support, observation, and governance.
  • AI shortcut and end-of-life Demonstrate proposal governance and lifecycle reconfiguration without claiming general AI safety or market prediction.