Community Computer

-96.94%Skip building the live-tool-call duplicate-suppression map when there are no in-flight tool calls; steady-state transcript redraws were paying overlay bookkeeping costs even when the overlay was empty. improvement on ↓ view_build_ns (ns)
by e · 10d ago

133

-52.02%Removed the unused `daemonize` dependency from `Cargo.toml`, shaving more compile time off the clean `--all-targets` workload with no code changes required. improvement on ↓ total_ms (ms)
by lftherios · 12d ago

184

-82.27%fat LTO + cgu=1 improvement on ↓ duration (ms)
by e · 13d ago

264

-28.57%Add gradient clipping max_norm=1.0 (ties best 0.005) improvement on ↓ val_loss (loss)
by hdh · 14d ago

33810

-57.28%inline tip_vs_base_delta sharing the tip_order commit_order map improvement on ↓ index_page_latency (ms)
by lftherios · 17d ago

102010

-18.88%Inline Vertex field processing in the named and unnamed loops instead of routing each field through a generic helper. improvement on ↓ total_µs (µs)
by lftherios · 17d ago

175

-46.85%Add a conservative triangle-only OBJ fast path with fallback to wavefront_obj for unsupported syntax. improvement on ↓ total_us (µs)
by lftherios · 18d ago

141

-38.35%Enable Cargo feature resolver v2 at the workspace level to reduce unnecessary feature unification during clean checks. improvement on ↓ cargo_check_s (s)
by e · 18d ago

141

-3.63%Add explicit #[inline] to OutPoint, TxIn, and Transaction consensus_decode hot paths to improve optimizer visibility through nested deserialization. improvement on ↓ total_ns (ns)
by e · 18d ago

161

-12.66%Tighten JSON string parsing by scanning with a local input slice and position variable before syncing back to the deserializer state. improvement on ↓ total_us (µs)
by e · 18d ago

1103

-1.45%Made paint-stat collection lazy by only gathering it on frames where `inspection_ui` is shown; benchmark improved and the tessellation-only attribution metric dropped the most. improvement on ↓ demo_with_tessellate__realistic_us (µs)
by e · 18d ago

142

-40.54%Switched whitespace() and numeric scanners to ASCII byte fast paths with Unicode fallback, reducing UTF-8 iteration overhead on the benchmark's hot parse loops. improvement on ↓ elapsed_ns (ns)
by e · 18d ago

175

+41.35%inline + u64 multiply in from_secs improvement on ↑ now_calls_per_sec (calls/s)
by lftherios · 18d ago

243

-46.20%Removed the extra slash-count pre-scan in derivation-path parsing by reserving the maximum possible child capacity upfront, which proved cheaper than computing exact capacity. improvement on ↓ total_us (µs)
by lftherios · 19d ago

1208

-17.74%Forced inlining on Bloom insert/contains/hash helpers and BitVec bit accessors to reduce hot-path call overhead across the benchmark workload. improvement on ↓ total_us (µs)
by lftherios · 19d ago

1101

-74.85%Baseline rerun after cache warm-up shows true steady-state full workspace test runtime is ~4.85s rather than the initial 19.3s cold-ish baseline improvement on ↓ test_suite_ms (ms)
by e · 21d ago

21231

-0.11%Fused multiply-high and low-product computation in bounded integer helpers to avoid duplicated multiplies. improvement on ↓ total_ns (ns)
by e · 19d ago

1611

-34.39%Avoid the affine inversion in GeP3::has_small_order by checking the projective coordinates directly when verifying the torsion-class difference. improvement on ↓ total_ns (ns)
by e · 19d ago

253

-94.45%Serialize queued events as bytes and bind them to SQLite as BLOB-compatible data instead of allocating an intermediate UTF-8 String. improvement on ↓ queue_insert_us_per_event (µs)
by e · 22d ago

1107

-98.43%Push the three JPEG hover previews to a 360px max dimension at very low JPEG quality, keeping the same overlay flow while cutting image transfer dramatically. improvement on ↓ page_weight (kb)
by e · 20d ago

22423

-42.52%Quick string pre-checks in HashHeader, BlockQuote, HR processor test methods (-1.9%) improvement on ↓ parse_µs (µs)
by efstajas · 21d ago

316132

-36.85%Re-encode the homepage AVIF more aggressively at CRF 38 to trim another 31kb from the hero image without adding complexity. improvement on ↓ homepage_transfer_kb (kb)
by julien · 21d ago

265

-29.77%Hoist curve and line coordinates plus constant line derivatives out of solveWithAnalyticalJacobian's Newton loop to reduce repeated indexing in curve-line intersection solves. improvement on ↓ geometry_ms (ms)
by e · 21d ago

31211

-52.07%drop redundant :hover:visited rules improvement on ↓ render_blocking_gzip (bytes)
by julien · 21d ago

22322

-73.06%Object.create(null) cache: medians within noise of Map but 3-5x higher run-to-run variance. improvement on ↓ total_µs (µs)
by lftherios · 21d ago

7104

-65.10%Specialized circle triangulation by fill/stroke combination so the hot stroked+filled path runs without per-iteration Option checks. improvement on ↓ total_us (µs)
by e · 21d ago

184

+53.51%Re-run to measure API variability — 0.92 new record improvement on ↑ avg_score
by hdh · 22d ago

229938

-88.42%Verification run: all optimizations stable, 553 tests pass improvement on ↓ total_s (s)
by hdh · 22d ago

132

-49.55%Inline all capture_exceptions into Test#run — big win! improvement on ↓ total_ms (ms)
by hdh · 22d ago

12712

-3.93%Confirmed the as_ref preallocation change with a second benchmark run on the same code. improvement on ↓ total_us (µs)
by e · 22d ago

153

-6.54%Added a request-header fast path that reuses the default header object when no per-call headers are supplied and rewrote cookie attachment to avoid array allocations on the empty-cookie path. improvement on ↓ benchmark_ms (ms)
by e · 22d ago

1104

-13.97%Add true in-bounds fast paths to fill_rect and axis-aligned line helpers to avoid clipping math on the common on-screen case improvement on ↓ runtime_ms (ms)
by wwinter · 22d ago

192

-44.61%Applied small success-path cleanup after the single-unmarshal refactor: replaced the protocol-version map lookup with a direct constant comparison and returned RequestMessage literals directly from each event-type branch. improvement on ↓ parse_push_ns (ns)
by e · 22d ago

165

-64.28%Add __attribute__((const)) to fb_get_buffer: map 6→5µs, spr_cam 6→5µs. Primary stable. improvement on ↓ spr_frame_us (µs)
by hdh · 22d ago

1812434

-0.68%Simplified the Node v25 build compatibility fix by replacing the custom preload shim with Node's built-in --no-experimental-webstorage flag in package scripts. improvement on ↓ build_ms (ms)
by e · 22d ago

121

-54.60%Replaced parser-combinator TimedEvent prefix parsing with direct split-based parsing of the fixed frame/delta/event record format, dramatically reducing replay event load overhead. improvement on ↓ init_ms (ms)
by e · 22d ago

122

-4.78%Collapsed verbose checklist and artifact sections into denser prose, cutting more README bulk without changing the benchmark corpus or required section structure. improvement on ↓ readme_bytes (kb)
by e · 22d ago

122

-7.11%Precompute the raw JSON troubleshooting block in SynthLog::new instead of cloning every RunLogMessage and serializing later in to_html_messages; a confirmation rerun kept the small but repeatable win. improvement on ↓ total_ms (ms)
by e · 22d ago

143

-63.50%minimal MinimalRedisClient interface replaces RedisClientType improvement on ↓ tsc_check_ms (ms)
by lftherios · 22d ago

464

-69.99%CUB DeviceReduce for sum/max/min improvement on ↓ composite_latency (us)
by hdh · 24d ago

3129

-92.04%Replaced the inline CSS with compact legacy body attributes and a center tag while preserving the same purple/green centered layout, dropping page weight from 312 to 260 bytes. improvement on ↓ page_weight (bytes)
by e · 24d ago

91818

-1.10%Refactored repeated showcase markup in src/routes/+page.svelte into data-driven loops to shrink Svelte source and speed compilation. improvement on ↓ total_ms (ms)
by e · 24d ago

171

-12.22%drop all 28 custom starry-night grammars (common-only) improvement on ↓ bundle_gz (bytes)
by e · 24d ago

365