This is a model, not a result
Everything in this article is an estimate built from stated assumptions. It is not measured output from your system, and we would rather label it than let you discover that in a procurement review. For what we have measured, see the reference run on /platform.
What we publish
For a Delphi system between 50,000 and 200,000 lines of code, the price is on the pricing page: $75,000 – $180,000, fixed per project, assuming standard complexity (the assessment sets the final level). Above 200K lines we do not print a number: a validated discovery sets the waves and each wave gets its own fixed price and its own date.
We do not publish a timeline either. This article used to say "3–4 months with 2–3 reviewers". Nothing measured supported it, and a duration printed next to a price reads as a promise.
The manual side is your number, not ours
The comparison people want is against rewriting by hand. That side of the equation depends on your team's cost, your developers' availability and how much of the system is still worth keeping, so the honest thing is to hand you the shape of the model and let you fill it in: developer-months times your loaded rate, plus the parallel maintenance you keep paying while the rewrite is in flight. A rewrite of a 200K-line system is typically quoted in developer-years, not developer-months, and that is the order of magnitude the published price should be read against.
This article used to print the manual side as "$880,000 – $1,660,000" and a "55–72% lower cost". Both were our assumptions about your team. They are gone; the saving is something you compute with your own inputs during the discovery, and it goes into the proposal as a scenario.
What we deliberately do not claim
You will not find a success-rate percentage here. This article used to carry one — "98%+ logic preservation" — and it came from no measurement at all. What exists instead is a real run on a 3,885-file Delphi corpus, with its residue published alongside it. A number with a run behind it can be checked, argued with, and improved on the next project. A number without one can only be repeated.
The rest of the return is real but harder to price: less parallel maintenance, a system your current developers can staff, and the option to keep changing it.