Skip to content

VERDICT.md

Latest commit

 

History

History
161 lines (119 loc) · 7.84 KB

VERDICT.md

File metadata and controls

161 lines (119 loc) · 7.84 KB
Raw

Verdict — Rynix vs End (who is ahead?)

Date: 2026-08-30
Peer: IrMaho/End main @ bdc8732
(local clone D:\0\End-peer, git fetch + pull --ff-only — no peer source edits)
This repo: Rynix Golden Remaining closed (30–38); Lead path: GOLDEN_LEAD.md;
Niche-10 certified (NICHE10.md).

This document answers one question under audit rules:

Is continuing Rynix a waste of time relative to End, or is Rynix already ahead on what actually ships?

One-line answer

Rynix is ahead on the shipping agent/systems core End cannot match today: MCP-19, contracts/scope, NDJSON diags, VS Code LanguageClient depth path, Suite5 checksum CI, honest deferred ADRs. End @ bdc8732 now ships real rustls TLS and h2 HTTP/2 in host Rust — stop claiming “End TLS is simulated.” End still leads spectacle (README domain wallpaper, ~60 CLI names, agent DSL keywords, C11/UI narrative) — surfaces we refuse to fake (ADR-0007, 0008, 0009).

You are not behind on real language value. Time spent on Rynix built auditable depth; End’s green tables often describe stubs.

Judgment rules (how we score)

Rule Meaning
Code > README A 🟢 row without tests/RT is theater
Same harness Suite5 ms only vs Suite5; never Suite5 vs End suite12
Strength reduction disclosed Faster checksum-same binary ≠ identical asm
Deferred ≠ loss Refusing fake UI/C11/CDN is honesty, not defeat
Peer untouched End clone is read-only for fair audit

Full methodology: END_PEER_GAP.md. Positioning: COMPARE.md.

Scorecard — shipping axes

Axis End @ bdc8732 Rynix Winner
Native binaries Real C11 → host CC; host Rust runtimes Real LLVM IR → clang ThinLTO Tie product paths; Rynix size gates
Async / I/O runtime Fibers + host tokio net Fibers + portable / uring / IOCP Rynix
TLS (product path) rustls in host; C/std partial Real SChannel / OpenSSL product Parity — not “Rynix wins fake TLS”
HTTP/2 h2 in host Rust HTTP/1 + WS product End host on H2; Rynix refuses brochure chase
Crypto SHA-256 + HMAC real; some AES/Argon theater SHA + HMAC + AES-GCM KATs Rynix (KAT discipline)
HTTP Host HTTP/2 + thin .end One-shot + bounded loop + JSON Rynix product ergonomics
WebSocket — / incomplete RFC 6455 + wire smokes Rynix
Agent CLI ~60 names, uneven depth graph/slice/impact/eval/patch/verify… Tie overlap; End broader names
MCP Absent 19 tools (mcp-serve, includes rynix_slice) Rynix
Packages Local path; PubGrub theater Path + local index + unity/semver/lock/workspace + local digest attest Rynix
WASM Text / toy surface emit-ll --target=wasm32 + emit-wasm + Node run — no WASI Rynix (honest subset)
Memory / escape Real region/borrow subset Escape + move + #^ effect: pure + explain-alloc Rynix (transparency)
Editor LSP LSP server; VS Code without LanguageClient lsp-serve + LanguageClient + CodeLens + workspace goto Rynix
Microbench fairness suite12 (different programs) Suite5 C↔Rynix CI + optional End slot Rynix
Docs honesty 62 files; STATUS often green on stubs Evidence-gated maturity + ADRs Rynix
Spectacle / domain matrix Strong marketing Refuses wallpaper End (spectacle only)

Tally (shipping): Rynix wins the axes that decide “usable systems + agent language.”
Spectacle: End wins the brochure. Brochures do not compile TLS.

Performance (Suite5 — same algorithms)

Harness: benchmarks/suite5/ — identical integer kernels, opaque trip counts, checksum parity required. End’s suite12 is a different program set — do not cross-compare ms.

vs C / Rust / Go / Zig (2026-08-25)

Rynix fastest on 11/12 (Zig edged nested). Rynix/C ratios ~0.19–0.98. Large wins use disclosed strength reduction (same checksum). See END_PEER_GAP.md §2 and benchmarks/suite5/README.md.

vs End on Suite5 (2026-08-25 Phase 16-A, this machine)

Peer endc built release-only from untouched D:\0\End-peer @ bdc8732 (ENDC_PATH=…/endc/target/release/endc.exe).

Peer compiler regressions discovered during prior audit (not Rynix edits):

  1. Statement if cond { … } fails to parse — End’s own benchmarks/suite12/suite12_end.end errors with Expected token Else, found LBrace.
  2. Expression a if cond else b always evaluates to 0 in probed binaries.
  3. Rynix-owned Suite5 .end ports use only for / while / getenv as i64 (see benchmarks/suite5/regen_end_ports.py + END_INTEGRATION.md).

Live head-to-head: benchmarks/suite5/suite5_summary_2026-08-25_phase16.txt (Rynix 11 · End 1 on matrix) and table in END_PEER_GAP.md §2. Checksums must match C on every counted End row.

Binary size (hello full RT gate): <300 KiB (~13 KiB measured 2026-08-25). Suite5 --bench Rynix bins were smaller than End --strip on prior Windows run.

Doubt checklist (anxiety → evidence)

Worry Evidence response
“End README looks finished; maybe I’m late” Host Rust @ bdc8732 is real on TLS/H2/DB; brochure rows still overclaim C/UI/registry
“End has more CLI commands” Volume ≠ depth; Rynix has MCP (End has zero)
“End has C11 / UI / frameworks” Real C transpile + host runtimes exist; copying wallpaper wastes months. We deferred C11/UI on purpose (ADRs)
“Maybe Suite12 proves End faster” Different workloads; Rynix ports MATCH ids with checksum locks, not End ms tables
“Am I wasting time?” Rynix ships MCP-19, fibers+IOCP/uring, Suite5 CI, package unity — Lead path deepens agent niche
“Should I copy End domains?” No — FFI/Refuse per GOLDEN_LEAD.md Fate matrix

What we still refuse (and why that is leadership)

Copying End’s green wallpaper would make Rynix look equal and become less trustworthy. Leadership here means:

  1. Real I/O and crypto over labeled stubs
  2. MCP + contracts evidence over feature/skill keyword theater
  3. Checksum CI over medal tables with divergent oracles
  4. ADRs for deferred work instead of fake Stable rows

How to re-verify (reproducible)

# Peer: update only (never edit End sources)
cd /path/to/End-peer && git fetch && git pull --ff-only
cargo build --release --manifest-path endc/Cargo.toml

# Rynix Suite5 including End
set ENDC_PATH=/path/to/End-peer/endc/target/release/endc.exe   # Windows
python benchmarks/suite5/run_suite5.py --langs c,rust,go,zig,rynix,end --summary

# Named product gates (sample)
cargo test -p rynixc --test size_echo_gates
cargo test -p rynixc --test agent_cli

Final verdict

Question Answer
Who is ahead on real shipping systems + agent toolchain? You (Rynix)
Who is ahead on Suite5 same-algorithm wall-clock vs End@bdc8732? You — 11 wins, 1 loss (matrix) (Phase 16-A; checksums OK)
Who is ahead on brochure / domain spectacle? Friend (End) — by design of their README
Is Rynix a waste of time vs End? No — Rynix leads MCP/agent toolchain; End leads wallpaper
What would reverse this? End ships MCP + drops green-on-stub STATUS — then re-audit

Canonical Lead SoT: GOLDEN_LEAD.md.
Execute history: LEAD_AHEAD.md.
Living gap log: END_PEER_GAP.md.