100
Scripts in scope
Act 1: Workload Reality
The scripts move like sand until you ask the right question
Each particle is one MSSQL script. Watch the same workload re-organize by purpose, table family, risk, and execution pressure.
Input Dataset
Non-obvious challenges
Act 2: Why This Matters
Performance debt becomes migration risk
What goes wrong in manual-only migration
Teams hand-translate syntax quickly, then spend weeks chasing hidden semantics: lock hints, transaction behavior, null handling in joins, and collation differences. Costs rise in the review cycle, not the first draft.
Why business leaders care
In this dataset, a small set of scripts carries most runtime load. If those scripts regress, checkout latency, ETL freshness, and finance reporting all absorb the hit.
The migration thesis
LLMs are strongest when paired with routing rules and verification loops: accelerate the easy 80%, isolate the dangerous 20%, and produce an audit trail everyone can inspect.
Act 3: The Six-Step LLM Pipeline
Click a step to inspect the mechanism
Act 4: Script-Level Proof
Three real conversions, with the hidden decisions exposed
Act 5: Business Simulator
Dial your migration scenario
Move the controls to simulate volume, workload intensity, model quality, and verification depth. This is a planning model, not a quote.
Manual Budget
$0
0 weeks
LLM-Assisted Budget
$0
0 weeks
Operational Risk Delta
0 fewer incidents/mo
0%
Estimated Business Lift
$0
ROI 0x
Act 6: Caveats and Controls
Where automation can fail, and how we contain it
Semantic drift risk
Equivalent syntax can still produce different results under null, collation, timezone, or transaction isolation differences.
Performance inversion risk
A query that runs in SQL Server can degrade in MySQL if join strategy, indexes, or grouping plans are not re-validated.
Verification scope risk
Unit checks are not enough for ETL and financial scripts. You need row-level invariants, reconciliation totals, and synthetic edge cases.
Governance requirement
Every conversion should ship with traceability: prompt lineage, diff artifacts, test evidence, and explicit manual-review decisions.