Context & efficiencyjackin❯ Context Engine

09 — Current reuse decision matrix

Decision matrix for reusing, embedding, adapting, or declining current context-tool mechanisms and dependencies.

Summary

jackin should reuse proven mechanisms selectively—vendor RTK's reviewed corpus, reimplement documented recovery patterns, and avoid adopting any tool as a wholesale runtime. Tool-state cutoff: 2026-08-12. This page uses only current stable releases and current license boundaries. Version pins are design candidates, not authorization to edit the workspace dependency graph; implementation must re-run source, API, license, supply-chain, and benchmark checks.

Question and scope

Which existing tools and repository components should the engine reuse, wrap, evaluate, or reject at each layer?

Method

The matrix compares current repository primitives and external tools by boundary coverage, determinism, recovery, licensing, maintenance, and measured outcome evidence.

Findings

Decision table

ToolCurrent version / licenseCurrent decisionCurrent reusable surface
RTKv0.45.0 / Apache-2.0VENDOR reviewed declarative corpus; PORT selected kernelsCurrent command/filter corpus, classification shape, shorter-only guard, exit preservation, raw recovery. Skip hook installers, telemetry, approximate gain ledger, and current-outcome assumptions.
Headroomv0.34.0 / Apache-2.0REIMPLEMENT documented mechanismsTyped content routing, CCR recovery, lossless seam, cache-aware request policy, holdout design. Skip global proxy, Python runtime, and Kompress dependency in the deterministic core. No stable published Rust library boundary was verified.
lean-ctxv3.9.18 / Apache-2.0PORT selected kernels; optional sidecar experimentRead modes, cached handles, deterministic crushers, tree-sitter views, recovery doctrine, graph scoring, self-overhead gates. Skip daemon/dashboard/global installers/proxy/default cognitive behavior in the embedded path.
semv0.21.0 / MIT OR Apache-2.0DEPEND on reviewed sem-core + native ledgerEntity extraction/graph, structural diff, bounded context surfaces. Generalize the attention-ledger concept at the jackin boundary. Disable cloud, telemetry, self-update, and global git/hook mutation.
fffv0.10.3 / MITDEPEND after current API reviewfff-search and fff-query-parser for resident lexical search. Disable unneeded native/egress surfaces; verify current dependency tree and features before pinning.
codedbv0.2.5838 / BSD-3-ClauseDESIGN DONORQuery/output grammar, budget packing, pagination/escalation patterns, warm-index economics. Reimplement in Rust; do not adopt broad installer or MCP policy.
mqv0.8.2 / MITDEPEND after current API reviewmq-lang and mq-markdown for structural Markdown selection. Keep HTTP/process features off unless a concrete requirement passes review.
Cavemanv2.0.0 / MIT interfaces + BSL-1.1 engine-side componentsNATIVE PROMPT PACK; design review only for engineReuse/adapt current MIT output-skill register concepts and clarity/verbatim guards. Do not copy BSL engine/proxy/MCP code into Apache-licensed jackin; treat deterministic routing/recovery patterns as architectural evidence only.
pxpipev0.13.1 / MITMETHOD DONORCounterfactual accounting and cache-alignment invariants. Keep text-to-image representation behind a model-specific harness; never make pixels the only source for exact text.
Serenav1.7.0 / MITOPTIONAL SIDECAR / protocol donorLanguage-server symbol retrieval and exact editing verbs. Prefer established language servers directly where jackin already owns the runtime boundary.

Current source of truth: tool comparison, rising tools, evidence, and release register.

Current candidate dependency set

# Design candidates only; confirm again at implementation time.
fff-search       = { version = "=0.10.3", default-features = false }
fff-query-parser = { version = "=0.10.3", default-features = false }
sem-core         = { version = "=0.21.0", default-features = false }
mq-lang          = { version = "=0.8.2", features = ["sync"] }
mq-markdown      = { version = "=0.8.2", features = ["json"] }

No dependency change is implied by this research update. Exact feature names and transitive licenses must be verified against the current crates before implementation.

License boundary

  • Apache/MIT-compatible code may be depended on or vendored only with current notices and repository policy checks.
  • Caveman v2's current engine/proxy/MCP path is BSL-1.1. Its readable source does not make it an Apache-compatible code donor today.
  • Context Mode uses Elastic License 2.0 and is not a dependency candidate for the embedded engine.
  • Projects whose GitHub API exposes no SPDX identifier require direct license review before any reuse.

Current acceptance bar

A borrowed mechanism ships only if a current implementation-specific A/B shows:

  1. accepted-task quality is non-inferior;
  2. total tokens or subscription quota per accepted task falls at least 20%, net of self-overhead;
  3. cache-read continuity does not regress unexpectedly;
  4. raw recovery succeeds for every lossy transform;
  5. hook/config behavior stays inside /jackin/ and preserves jackin host-write rules;
  6. dependency/license/supply-chain checks pass.

Current upstream percentages are not implementation evidence for jackin.

Implications for jackin

These findings define the current reuse decision matrix contract for jackin; any implementation must preserve the acceptance gates and boundaries above.

Limitations and unknowns

Revalidation triggers

Re-run this page when any candidate changes stable version, license, published crate boundary, default telemetry, installer behavior, or independent evidence. Replace old conclusions; do not append a release history.

Sources

Evidence and repository references are cited inline beside the claims they support.

On this page