SlimeRESCUE — Binary-First Legacy Recovery Family
Bit-exact conversion from load modules to Java/COBOL/Rust, for customers whose COBOL source code has been lost.
Source lost during vendor migration, vendor bankruptcy, M&A asset fragmentation, retiree-dependency — an industry-wide open secret. Gartner and Forrester touch on it: 15-30% of the world's 80 billion lines of COBOL (= 12-24 billion lines) are estimated to be source-lost. Still running, too risky to touch, blocked from migration — a domain with extremely high willingness to pay.
LLMs (GitHub Copilot / Claude / watsonx) cannot perform binary
semantic inference and are therefore physically incapable of
bit-exact conversion. Accenture / Capgemini / Micro Focus / Rocket /
mLogica all require source — this market is unreachable for them.
Javatel has consolidated its prior investment in machine-code → Slot IR
technology (SlimeASM-rev / SlimeELF-rev / SlimePE-rev) into the new
independent SKU family SlimeRESCUE.
A SlimeNENC-family tool that moves legacy code into a modern language without changing its meaning. A hand rewrite quietly drifts and causes incidents; SlimeNENC doesn't interpret meaning — it copies only the "skeleton" (structure), so the computed results stay identical to the original. It proves the behaviour first, so migration anxiety disappears. It copies only what it can, honestly, and isolates what it can't.
Hand rewrites drift on subtle numeric/exception differences (boundary conditions), and verifying that (old-vs-new testing) costs enormous labour. SlimeNENC faithfully mirrors language-specific traps, proves "zero divergence" via differential testing, backed by an independent reference implementation. The deliverable is a machine-checkable "certificate of behavioural invariance," not human UAT. No overstating; what it can't do isn't hidden.
Source is projected onto a Slot IR (language-independent structural intermediate form) and transcribed structure-preserving into the target. The statically-determined core is made bit-exact; dynamic, state-dependent parts are honestly isolated (isolate, don't confabulate). Backed by differential fuzzing plus formal methods where applicable, with deterministic verification a third party can reproduce locally.
Projection (π) of source as unique structure, not semantics. Primitives are modelled rigorously in bit-vector theory and formally verified over all inputs; composition/loops are covered by Csmith-style differential fuzzing — a two-tier guarantee. Non-reproducible computation (float/parallel/AI) goes to tier-③: meaning-equivalence + convergence + residual. "Where there is form, prove it in bytes; where there is none, in meaning. Lie on neither face."
📋 "Ask your AI at this level" copies this page's explanation with an instruction matched to the level you picked. Paste it into your own AI (Claude · GPT · Gemini · Grok) to dig deeper at that resolution.
SKU Structure (5 z/Arch SKUs + 1 IBM i SKU)
Combinations of machine-code ISA × load-module container format form 5 z/Arch SKUs. Each SKU has its own front-end (disassembler + container parser), while the Slot IR / S2-S5 / S6 emitter (5 languages) are fully shared. Same structural pattern as the SlimeCOBOL-MF / -GnuCOBOL split.
SlimeRESCUE-VOS3
Input: Hitachi VOS3 / Fujitsu MSP / GS21 / IBM z/OS LOADLIB (PDS) or s390x Linux ELF.
Output: Java / COBOL / Rust / C# / Kotlin / Go (bit-exact)
Target ISA: IBM S/370 / 390 / z-Architecture (PCM-compatible = identical processing across Hitachi/Fujitsu/IBM)
2026-05-20 achievements:
- z/Arch ISA 33 format families = 100% coverage (1,778 opcodes ingested from GNU binutils s390-opc.txt, vector and niche included)
- Both s390x ELF and z/OS PDS (LOADLIB) supported (incl. EBCDIC member name recovery)
- Slot IR mapping complete (67 mnemonic correspondence table, dialect_id=200 (Linux) / 201 (LM))
- Cross-validated against Hercules 3.13 official emulator — 6/6 disasm match + GR final state (GR1=350) exact match
- Python interpreter + Hercules-style trace output for behavioural verification
- HTTP service exposed on 127.0.0.1:8770 (4 endpoints) for external invocation
- Linux
file(1)recognises SlimeRESCUE-generated ELF as "IBM S/390" confirmed
Market: Domestic regional banks, municipalities, electric utilities, and large insurance companies retiring Hitachi VOS3 / Fujitsu MSP. Overseas: Amdahl / IBM-compatible PCM customers. 2026-2030 is the golden window (concentration of z/OS, GS21, AP8000 retirement schedules).
SlimeRESCUE-MSP (Fujitsu MSP detailed support)
Fujitsu MSP runs on the IBM-compatible ISA, so SlimeRESCUE-VOS3 is usable as-is. This SKU adds MSP-specific runtime / NetCOBOL integration. Derivative of VOS3, +2-4 weeks effort.
SlimeRESCUE-GCOS
Bull GCOS 7 / 8 (European mainframe), Honeywell GCOS, with NEC ACOS-4 (= GCOS 7 derivative) as a secondary path. Implementation via bitsavers' Honeywell architecture spec + reverse engineering.
SlimeRESCUE-ACU
ACUCOBOL-GT load modules. Targeting manufacturing / distribution AcuCorp customers.
SlimeRESCUE-MCP
Burroughs MCP (Unisys ClearPath base) load module support. For Unisys mainframe retirees.
※ NEC ACOS-4 as a standalone SKU is intentionally excluded — see Why Javatel for the rationale. Strategic partnership with NEC is the preferred path.
Note: IBM i (AS/400) is a separate architecture branch — it carries TIMI / MI, not machine code — covered by SlimeRESCUE-AS400.
SlimeRESCUE-AS400 (IBM i / separate branch)
Input: IBM i (AS/400) *PGM objects + physical files.
Output: SlimeCL → Rust / bash (Java later)
Nature: unlike z/Arch, a *PGM encapsulates TIMI/MI —
not machine-code disassembly, but the observability / recoverable-information path
Validated on a live IBM i 7.5:
- CL rescue = world-first: from a
*PGMwhose source-file slot is empty,RTVCLSRCrecovered the real source → SlimeCL → Rust, byte-exact == the liveCALL(425) - Data rescue: every physical-file record read byte-exact as EBCDIC/zoned/packed COMP-3/binary/float/DECFLOAT (P10-P19)
- RPG/COBOL rescue = in development: retrieval API (
QTEDBGS) reached and bound; completion needs a debug-mode context
Market: AS/400 sites that ran unattended for 20-30 years and lost their source. SlimeRESCUE-AS400 details →
Why this Binary-First market is monopolisable (3 structural advantages)
- Maximised willingness-to-pay: With source, "rewrite ourselves" is an option. Without source, Javatel is the only option.
- Physical impossibility of competition: LLM-based tools (Copilot / Claude / watsonx) cannot perform binary semantic inference — a hard physical limit. Accenture / Capgemini decline binary-start engagements as "unestimable". Micro Focus / Rocket require source.
- Temporal window: IBM z/OS, Fujitsu GS21, Hitachi AP8000, and Unisys ClearPath retirement schedules are concentrated in 2026-2030 — a 4-year golden window.
Technical Stack (SlimeRESCUE-VOS3 example)
| Phase A | z/Arch instruction disassembler (Python 757 lines, 33 format = 100%, 1,778 mnemonics ingested from GNU binutils s390-opc.txt) |
|---|---|
| Phase B | s390x ELF64 BE parser (260 lines, Linux file(1) recognises "IBM S/390") |
| Phase C | Slot IR mapping (406 lines, 67 mnemonic table, Core64+Ext32 compatible with existing SlimeNENC backend) |
| Phase D | z/OS LOADLIB / PDS parser (275 lines, EBCDIC member name recovery + byte-exact member extraction) |
| Phase E | Python z/Arch interpreter (342 lines, 25 instructions + packed BCD decimal, Hercules-style trace) |
| Hercules cross-val | 6/6 disasm + GR state exact match against Hercules 3.13 official emulator |
| Phase F | unified CLI + HTTP service (127.0.0.1:8770, 4 endpoints, ELF/PDS → slot/java/trace) |
| Total implementation | 3,588 lines (src 2,719 + samples 869) + reference data, all 8 tests PASS, regression 0 |
| WASI port | Phase F-2 achieved (2026-05-20) — ~580 lines of C ported to wasm32-wasi: slimerescue.wasm = 84 KB (about 1/6 of the <500 KB estimate). Python ↔ native ↔ WASM produce byte-exact slot streams (audit_hash match confirmed) |
Related products / pricing
SlimeRESCUE is positioned as a mainframe binary-origin premium SKU, listed at 1.5x = $14.95 per LOC over source-origin SlimeCOBOL (reflects binary-recovery effort + no competing tooling). PSDP bundling, volume discount, and SIer partner margin: see Partners. Details disclosed under NDA.
Related: SlimeCOBOL (source-first) · SlimeASM-rev (x86_64 binary, current) · PSDP (in-language parallelisation)