Technical Limitations and Responsible Use Statement
Document version: 2026-10-17-v2. English translation. Operator: TYRENX TECHNOLOGIES, 202603272412 (AS0523259-D); contact itgtrackercs@gmail.com. This supplements the Terms, order, Privacy Policy and DPA; it does not wholly exclude our own illegality, breach or fault.
1. QA and AI
Language, context, rule versions, model state, evidence scope and collection gaps can cause mistakes. An automatic pass is not supervisor confirmation, insufficient evidence is not a proven issue and not applicable is not 100% quality. Settings may exclude conversations, queue tasks or require human verification.
Review complete available conversations, original evidence and published rules. Company exceptions are not universal rules. Retaining company corrections does not guarantee models never repeat mistakes or authorise cross-company reuse.
2. Response times and statistics
Statistics use captured verifiable messages and applicable settings. Automated-reply classification, consecutive-message pairing, cross-day handling, scope, late replies, missing timestamps and valid company exceptions affect them. Unanswered waits are not zero-second averages. Charts may use request dates, reply dates, eligible waits, Episodes or QA records: check units, denominators and definitions, not total working time.
Displayed/scan scope does not prove every historical message was collected, scored or deleted. Updates and human review can change reference averages and coverage, without falsifying original model results.
3. Attribution and operational evidence
Handle, Sessions and Handover assist historical responsibility. Shared accounts, current owners, last responders, device names or one Handle do not prove all monthly authorship. Unverifiable attribution must remain unconfirmed.
Done alone does not prove the latest request resolved. Exit, heartbeat gaps and offline states alone do not prove deletion, evasion or misconduct. Deletion evidence requires collection status, object, time and context checks.
4. Collection and dependencies
Authority, rate limits, client updates, networks and devices affect collection. Every message/image, recovery of every deletion and reconstruction of uncaptured history are not guaranteed. Telegram, Google, Billplz, bank and infrastructure failures are assessed by cause and contract, not automatic exclusions of our own fault.
5. Local operation and resources
Local QA requires compatible hardware, model, program, authority and workspace settings; power-off or model failure may pause it. Speed varies with content length, model and backlog; low-load mode may be slower. Power control supports only implemented hardware, not every NVIDIA, AMD or laptop device, automatic capping everywhere or complete fan silence.
6. Security, exports and personnel decisions
Encryption, device approvals, signatures, checksums and permissions reduce some risks but do not guarantee no copying, attack, reverse engineering or breach. Authorised users may download/screenshot; customers must protect devices and exports. Exports reflect saved scope, not necessarily all facts or guaranteed court acceptance.
QA, statistics and online status cannot alone determine discipline/dismissal or significant personnel decisions. Employers must investigate lawfully and permit explanations/corrections. We retain duties of reasonable care, correction, confidentiality, incident handling and other law/contract obligations.
7. Payments and mandatory rights
Initiating bank/FPX payment, redirects and actual receipt differ. Refunds may need provider processing time. Use only methods actually open on the invoice; unenabled digital assets are not offered. This does not remove liability for wrong/duplicate receipts or non-delivery.
Liability arrangements apply only where lawful, without excluding fraud, wilful misconduct or mandatory consumer, data-subject or other rights. Malaysian law and lawfully competent dispute channels apply.