SC LABS · PROOF & EVIDENCEPUBLIC EVIDENCE CENTRE
SC Agent Lab
This is where SC LABS separates product claims from demos. We publish what was tested, the environment, the result and the limitation — while proprietary ARKTOR control, recovery and trust mechanisms stay private.
Proof should make a claim smaller and stronger at the same time.
“It worked” is not enough. Useful evidence says what ran, where it ran, what counted as success and what the result does not prove. When a failure changes the product, that failure belongs here too.
Five current proofs worth knowing about.
ARKTOR RESCUEPhysical V0.1 recovery proof accepted
Boot, controlled remote diagnosis, read-only Windows inspection, persistent reconnect and real hash-verified data rescue all passed on physical hardware.
LIMIT · Continuous multi-reboot repair handoff is still the next milestone. ARKTOR ANDROIDPhone workbench retained; automotive LTE transport accepted0.8.7 remains the stable Android workbench reference. 0.8.8 physically closed the automotive transport regression over independent LTE.
LIMIT · The separate automotive chat/model UI issue remains open. ARKTOR VAULT0.3.0-alpha.1 durable-memory proofThe current local alpha passed 21/21 retained end-to-end memory tests and demonstrated cross-model reuse after a fresh process.
LIMIT · Public release, encryption-at-rest and multi-user authorization remain future gates. ARKTOR CONTROLAccepted control-layer candidate remains greenThe accepted V0.3 reliability baseline remains intact after a reproduced selection defect was fixed and the deterministic quality gates were rerun.
LIMIT · Internal state, routing, recovery and completion mechanisms are not public specifications. ARKTOR GORetained real-Windows proof, fresh release gate still openThe portable Windows path has real external E2E evidence, while the latest clean-PC one-click release criterion still requires fresh verification.
LIMIT · Historical functional proof is not the same as current public-release readiness.Physical devices matter.
Windows PCs, Android endpoints and recovery media are tested as real systems rather than reduced to screenshots. Device-brand names appear only when they are relevant to the evidence record.
Benchmarks stay tied to their question.
Local model and runtime measurements remain useful when the task, machine and caveat are explicit. A bounded benchmark is not promoted into a universal model ranking or product claim.
A regression is evidence when it changes the product.
Transport failures, lifecycle regressions and incomplete model results are retained when they materially change what SC LABS can claim. A safe incomplete result is better evidence than a polished false success.
Older proof stays available without crowding the current story.
SC Node remains the historical open-source proof of an earlier controlled-runtime direction. Older ARKTOR checkpoints remain in the Journal and Guides where chronology is useful, but they no longer compete with current product evidence on this page.
Enough detail to prove the work. Not enough to hand over the proprietary design.
Agent Lab publishes capabilities, outcomes, test context and limitations. It deliberately does not publish ARKTOR's private control state, internal recovery decisions, trust contracts, routing rules or production infrastructure. The goal is credible proof, not public engineering replication.
AGENT LAB
Back to ProductsThe evidence centre is public and continuously updated. It is not a separate SC LABS product.