Indexed literal code-search alternatives — 2026-08-10

Status: INCOMPLETE. Issue #6184 tracks real search-system runs and independent resource/cost witnesses.

Capability and workload

internal/trigram.Index builds distinct-rune trigram postings and verifies candidate lines exactly; it also falls back safely for short queries. This contract covers indexed literal search only. Regex candidate extraction and lexical similarity remain separate capability debt.

Every arm loads/indexes the same six Go files and executes four queries: a literal in three files, a short literal, a Unicode literal, and an absent literal. Correctness requires all exact ordered paths and line numbers. Build/load and query phases are measured separately.

Arms

Arm	Class	Local status
fak native trigram indexed literal search	native	available
optimized in-memory linear scan	tuned no-index baseline	available
ripgrep	external	unavailable
git grep	external	unavailable
Zoekt	external	unavailable
livegrep	external	unavailable
Sourcegraph Search	external	unavailable

No equivalent first-class fak integration was found. A future equivalent must appear as its own fak + integration arm. External rows remain measurement-zero until real processes/services run the identical corpus; wrappers and reimplementations are not witnesses.

Completion evidence

Complete arms report exact queries, false positives/negatives, location errors, build/load and query latency, throughput, CPU/RSS, corpus/index/storage/network bytes, setup/operator time, and total cost. Versions, index configuration, commands, raw output, and independent read-back must be pinned.

TestCompareLocalKeepsIndexedSearchAlternativesExplicit locks inventory, native/baseline correctness, and unavailable zeros. BenchmarkBuildAndQueryLiteralCorpus rebuilds the real index and runs all four queries per iteration. Local timing is not a cross-system claim; no system is ranked until #6184 has complete real runs.