Limits

Evidenced

• Containment of a prompt-injected agentTested in CI·environment:Lab
• Hardened container arguments for agent executionTested in CI·environment:Lab
• Fail-closed secret handlingTested in CI·environment:Lab
• Task-branch isolation and git integrity checksTested in CI·environment:Lab

Boundary

• Not verifiedExternal penetration test or security audit of anything listed here.

  Security claims are backed by automated tests of specific scenarios, not by independent review.

• External penetration test or security audit: NOT VERIFIED. (boundedcode-prompt-injection-containment)
• Container escape resistance verified in CI: NOT VERIFIED (runtime probes run only locally). (boundedcode-sandbox-hardening)

Evidenced

• Token-budgeted context packsTested in CI·environment:Lab
• Deterministic, budgeted model-escalation policyTested in CI·environment:Lab

Boundary

• PlannedAn LLM gateway with streaming, tenant-aware rate limiting across replicas, failover and cost metering.

  Planned as inference-gateway. No claim until its tests and load-test reports are public.

• Not verifiedGPU serving, vLLM, KServe or Triton in production.

  I have no GPU-serving evidence to publish. BoundedCode runs a local model on one laptop GPU, which is not production serving.

• That frontier escalation improves task outcomes: NOT VERIFIED. (boundedcode-escalation-policy)

Evidenced

• Crash-safe task ledger with lease takeoverTested in CI·environment:Lab
• Safe event processing for money-moving workflowsImplemented·environment:Production

Boundary

• Multi-node coordination or replicated state: NOT CLAIMED. (boundedcode-crash-resume)
• Public, inspectable evidence: NOT AVAILABLE (proprietary). (employment-event-processing)

Evidenced

• Fail-before / pass-after verification gateTested in CI·environment:Lab
• Specification-first, AI-assisted delivery with test gatesImplemented·environment:Production

Boundary

• Not claimedThat AI-written code is correct because tests pass.

  A passing test proves only what it asserts. That is why claims here name the tests that support them.

• Correctness of agent-written fixes in general: NOT CLAIMED. The gate only withholds TASK_VERIFIED when evidence is missing. (boundedcode-fail-before-pass-after)
• Public, inspectable evidence: NOT AVAILABLE (proprietary). (employment-ai-native-delivery)

Evidenced

• Streaming and historical time-series servicesImplemented·environment:Production

Boundary

• Public, inspectable evidence: NOT AVAILABLE (proprietary). (employment-time-series)

Evidenced

• Real-time market-data path in productionImplemented·environment:Production

Boundary

• Not verifiedInternet-scale or multi-region behaviour.

  No public evidence covers multi-region operation.

• Not claimedPublicly verifiable scale and latency figures from employment.

  Employment metrics are proprietary and are not published here; employment evidence is labelled as such.

• Public, inspectable evidence: NOT AVAILABLE (proprietary). A reference is available on request. (employment-realtime-data-path)

Evidenced

• Observability stack with customer-impact alertingImplemented·environment:Production

Boundary

• Published SLOs or postmortems from employment: NOT AVAILABLE (proprietary). (employment-observability)

Evidenced

• Production operations for a live trading platformImplemented·environment:Production

Boundary

• PlannedSLOs with multi-window burn-rate alerts, a chaos experiment and a published postmortem.

  Planned for platform-lab. My production incident work is described only as employment history, without proprietary details.

• Public, inspectable evidence: NOT AVAILABLE (proprietary). (employment-production-operations)

Evidenced

No published evidence.

Boundary

• Not claimedAWS or Azure deployment experience.

  I have no public evidence of AWS or Azure work. A Terraform AWS reference architecture is planned and will be labelled REFERENCE, never DEPLOYED.

Evidenced

No published evidence.

Boundary

• PlannedGitOps delivery with Argo CD, signed images and admission policy.

  Planned for platform-lab; no public evidence yet.

Evidenced

No published evidence.

Boundary

• PlannedKubernetes cluster administration (kubeadm upgrades, etcd backup and restore, troubleshooting drills).

  Planned as public, CI-verified lab drills in platform-lab. Nothing is claimed until those drills are published.

• Not claimedOwnership of an employer's production Kubernetes clusters.

  I have not administered Kubernetes in production, so it is not claimed.