Empirical research and laboratory work.
We believe in verifiable engineering. Every claim is subject to rigorous hypothesis testing, transparent measurement harnesses, and public replication.
Methodological commitments.
How TIV designs, isolates, and documents experimental systems research.
Deterministic Reproducibility
All benchmark harnesses, synthetic workloads, and raw log files are version-controlled alongside research notes. Any researcher should be able to clone and run the experiment.
Real-World Load Profiling
We avoid synthetic micro-benchmarks that test ideal cache states. Our tests incorporate packet jitter, memory saturation, and concurrent I/O interference.
Vendor Neutrality
Experiments evaluate fundamental architecture rather than proprietary vendor features. Findings are independent of proprietary cloud platforms.
Preliminary & active experiments.
Current experimental initiatives validating core assumptions across TIV systems.
Zero-Copy Queue Pipeline Throughput
Replacing intermediate heap copies with memory-mapped buffers and io_uring submissions in OpenMail yields a 3.4x throughput increase on multi-core Linux hosts.
Synthetic benchmark harness generating 100,000 synthetic MIME payloads per second across 32 concurrent worker threads on bare-metal x86_64 nodes.
Achieved sustained 92,400 msgs/sec with p99.9 latency under 4.2ms. Memory consumption stabilized at 1.8GB with zero garbage-collection latency spikes.
Synthetic MIME payloads generated in memory without disk I/O persistence bottlenecks; real-world disk write sync and anti-spam scanning will lower aggregate message rates.
Sub-50ms Context Retrieval for Local AI Assistants
Hierarchical index caching of abstract syntax trees (ASTs) on NVMe storage allows local LLM agent loops to parse 500k-line codebases without cloud vector databases.
Continuous execution of real-world refactor workflows on 12 open-source repositories using Mercura running on consumer hardware (Apple M-series & Linux RTX 4090).
Initial benchmarks demonstrate cold-start context lookup of 38ms and warm query resolution under 12ms with strict deterministic reproducibility.
Evaluated on single-user local developer workstations. Multi-tenant concurrent indexing under contention has not yet been benchmarked.
Wavelength Attenuation in Urban Dark Fiber Conduits
Standard CWDM/DWDM optical transceivers operated across high-vibration municipal conduits can maintain Bit Error Rates (BER) below 10^-12 without active amplification.
Optical Time Domain Reflectometer (OTDR) profiling across 15km passive single-mode fiber links simulating temperature and traffic-induced conduit movement in laboratory spool racks.
Baseline calibration complete; physical carrier testing scheduled for deployment in Phase 2 field infrastructure tests.
Preliminary data derived from laboratory spools with simulated mechanical oscillation; real municipal duct environmental factors may introduce unmodeled dispersion.
Asymmetric Latency Tolerant Peer Consensus
A quorum-based gossip mesh with adaptive leader lease time reduces commit stall during trans-continental transit jitter spikes.
10-node cluster running Linux network emulation (tc/netem) injecting synthetic packet drops (2-8%) and simulated geo-distributed latency spikes up to 400ms.
Eliminated 94% of election churn during simulated trans-oceanic fiber degradation events.
Simulated via software packet drop and latency injection on a local test cluster, not trans-oceanic physical undersea cables.
Experimental boundaries.
Clear definitions of what we test, what we do not claim, and known experimental constraints.
Negative Results and Failures Are Published
Simulation vs Physical Deployment Distinction