Courses, environments, and verification tooling for the estates that still run banks — COBOL, JCL, CICS, RPG, and the seams that hold them together.
Your mainframe, your AS/400 and your Oracle already exchange data through code nobody left understands. We make those handoffs explicit, tested and observable — and we prove equivalence, so you can change anything without guessing what breaks.
Every analysis vendor parses one platform and stops at its edge. We run both sides: execute the batch, produce the file, push it across the seam, and byte-compare what comes out the other end. Parse-and-diff tells you what you have. Run-and-compare tells you whether it's right.
Three facts define this market right now: the platforms are staying, the migrations keep failing, and the projects that succeed are the ones that can prove themselves. The numbers below are cited; the conclusions follow from them.
In the 20th annual BMC survey, 97% of practitioners see the mainframe as a long-term platform or a home for new workloads — the highest reading in the survey's history — and 72% forecast capacity growth.[1] The workforce flipped young: 66% of respondents are now millennial or Gen Z, up from 37% in 2018.[1] IBM shipped the z17 in June 2025 with AI inference designed to score every transaction in flight.[2] On the midrange, roughly three quarters of IBM i shops run half or more of their core applications on the platform.[3] And an estimated ~220 billion lines of COBOL remain in production.[4] This is not a dying estate. It is a permanent one with a staffing problem.
A 2020–2025 study of 29 mainframe-to-cloud migrations ($3M–$45M budgets) found 66% failed their stated objectives: 31% abandoned mid-project on data-conversion discoveries, 24% completed but reverted to the mainframe within 12 months.[5] The most common overrun cause: underestimating application complexity before starting.[6] Meanwhile the skills clock runs: IBM i shops now rank skills as their #1 concern at 69% — displacing security for the first time since 2017[7] — and the deadlines are no longer optional: SWIFT ended MT/ISO 20022 coexistence on 22 November 2025,[8] with domestic regulators cascading the same mandate. Generative AI is now writing mainframe code at scale — 65% of shops already use it[1] — which makes code cheaper and verification the bottleneck.
The canonical failure is TSB, 2018: a core migration that locked 1.9 million customers out, cost over £300M, drew £48.65M in regulatory fines — and a personal fine for the CIO. The root causes were testing causes: performance was tested on one data centre while production ran two.[9] Look again at the failure modes above — data conversion discovered mid-flight, systems reverted after a year in production. Those are not coding failures. They are absent-rehearsal and absent-equivalence failures: nobody ran the estate somewhere safe first, and nobody could prove the new system computed what the old one computed. That proof is precisely what this platform exists to manufacture: rehearse every change on an environment that behaves like the estate, and byte-compare the outcome of every cycle — before the customers find out for you.
| A permanent estate with a staffing problem — and no LPARs to learn or test on | Unlimited safe iron. SteelFrame puts a full mainframe- and IBM i-compatible environment in the browser for every student, developer and project — no MIPS, no queue, no IBM licence audit. The scarcity that starves both training and testing simply stops. |
| 66% of migrations miss their objectives — abandoned on data conversion, or quietly reverted within a year | Rehearsal before iron. Every failure mode in that study is discoverable in advance on a SteelFrame mirror of the estate: run the real batch, cross the real seam, find the data-conversion surprise in a sandbox where it costs an afternoon, not the project. |
| TSB-class cutovers fail on proof, not on code — and regulators now fine the individuals | Equivalence as evidence. SteelFrame executes the baseline and byte-compares every migrated outcome — posted state, control totals, reports — and files a divergence-explained evidence trail a board and a regulator can actually read. |
| GenAI writes mainframe code at scale now — verification is the new bottleneck | The proof layer for the AI era. Whoever — or whatever — writes the code, SteelFrame is where it runs, and the equivalence verdict is what decides. Verified, not autonomous: the agent proposes, the platform proves. |
Sources: [1] BMC 20th Annual Mainframe Survey, 2025 · [2] IBM z17 announcement, Apr 2025 · [3] Fortra IBM i Marketplace Survey 2026 · [4] Legacy modernization statistics compilation (Reuters-attributed COBOL estimate) · [5] 29-migration outcome study, 2020–2025 · [6] Kyndryl mainframe modernization report coverage, 2025 · [7] IT Jungle on the 2026 IBM i survey · [8] A-Team Insight on the Nov 2025 ISO 20022 cutover · [9] Charles Russell Speechlys, lessons from the TSB migration. External figures are cited as published by their sources; our own numbers elsewhere on this site come from our test harnesses.
These are the SteelFrame platform's own numbers — the engine behind every CobolStack environment, course lab and verification run — printed from its test harnesses, not estimated by its marketing.
Every SteelFrame number on this site is printed by a derivation script over the repository or by the suites themselves; none are estimates, and the static counts are deliberate undercounts. Ask us and we will run them in front of you.