First-party measurements behind our engineering decisions — each with its method, its numbers, and its limitations in the same breath. None of these are independent benchmarks, and every page says so.
Across five repository-analysis tasks on one codebase (91 files, 4,374 symbols), a conventional file-by-file agent workflow opened 43 files, read 28,963 lines, and consumed 115,852 tokens over 65 tool calls. The same investigation answered through Context Zero Engine's code graph opened zero files directly and used 22,056 tokens over 6 MCP calls — roughly an 81% token reduction on this workload.
We measured whisper-large-v3 and distil-large-v3 against the production engine Lven Instant ships, running on CPU, on one consumer laptop (Intel i5-10300H, 4 cores/8 threads; NVIDIA GTX 1650 4 GB) over 73 clean-speech clips. whisper-large-v3 was about one point more accurate on this set, and 7–40× slower depending on configuration. Realtime dictation needs a real-time factor well below 1; on this hardware only our engine delivers it.
Lven Instant runs a deterministic pass over each transcript before it types it: hyphen-fragment and stutter pruning, n-gram repeat collapse, filler purge, contraction and proper-noun restoration, sentence casing, and terminal punctuation. No language model, no network, no regular expressions on the hot path; the same input always yields the same output. We compiled the polisher's golden regression corpus — 31 input/expected pairs that CI holds every platform's implementation to — against the shipped Windows source and ran it on the publication date: 31 of 31 pass. Across the corpus, 154 words go in and 134 come out; zero words are inserted; 20 are removed and every one is a stutter fragment, a filler, or a repeat; one word is spelling-restored (accellent → excellent); 3 gain an apostrophe; the remaining differences are casing.