Automated benchmark report

Exceptional performance

Node.js 24.16.0 V8 13.6.233.17-node.49 win32 x64 Intel Core i5-10210U at 1.60 GHz 2026-08-02T22:01:31.679Z

JavaScript/WASM performance

Function	Sequence length	Timed Sequencer workload	Sequencer ops/sec	Diamond Types ops/sec	Sequencer calls	Diamond Types calls	Sequencer avg µs/op	Diamond Types avg µs/op	Sequencer RME	Diamond Types RME
`__create`	100	Hydrate the retained Reel	53,359	31,036	256	256	18.741	32.221	±15.10%	±4.57%
`__read`	100	Read a rotating visible index	10,493,179	213,995	131,072	256	0.09530	4.673	±4.92%	±18.45%
`__length`	100	Read the visible length	36,429,872	20,699,648	131,072	131,072	0.02745	0.04831	±5.51%	±4.42%
`__recover`	100	Recover all retained values	392,927	222,668	256	256	2.545	4.491	±1.95%	±56.98%
`__update`	100	Insert one Frame at the midpoint	163,532	309,406	256	256	6.115	3.232	±10.45%	±9.03%
`__delete`	100	Soft-delete one midpoint Frame	259,538	290,698	256	256	3.853	3.440	±11.08%	±15.79%
`__merge`	100	Merge one new midpoint Frame	116,904	56,117	256	256	8.554	17.820	±33.51%	±21.39%
`__acknowledge`	100	Acknowledge materialized Realm progress	6,729,475	1,395,089	4,096	4,096	0.1486	0.7168	±4.23%	±7.00%
`__garbageCollect`	100	Release one soft-deleted Frame	270,856	—	256	—	3.692	—	±19.72%	—
`__snapshot`	100	Snapshot complete retained state	255,558	73,779	256	256	3.913	13.554	±19.95%	±3.56%
`__create`	1,000	Hydrate the retained Reel	79,853	39,769	128	128	12.523	25.145	±5.75%	±6.84%
`__read`	1,000	Read a rotating visible index	9,861,933	434,972	65,536	128	0.1014	2.299	±11.97%	±5.07%
`__length`	1,000	Read the visible length	19,988,007	26,567,481	65,536	65,536	0.05003	0.03764	±11.12%	±8.72%
`__recover`	1,000	Recover all retained values	164,663	302,755	128	128	6.073	3.303	±13.30%	±3.86%
`__update`	1,000	Insert one Frame at the midpoint	111,919	48,940	128	128	8.935	20.433	±11.75%	±9.17%
`__delete`	1,000	Soft-delete one midpoint Frame	302,663	93,853	128	128	3.304	10.655	±13.86%	±5.49%
`__merge`	1,000	Merge one new midpoint Frame	252,589	37,518	128	128	3.959	26.654	±9.56%	±9.98%
`__acknowledge`	1,000	Acknowledge materialized Realm progress	2,948,983	1,162,520	2,048	2,048	0.3391	0.8602	±3.71%	±5.96%
`__garbageCollect`	1,000	Release one soft-deleted Frame	527,704	—	128	—	1.895	—	±8.92%	—
`__snapshot`	1,000	Snapshot complete retained state	156,715	54,864	128	128	6.381	18.227	±16.82%	±4.00%
`__create`	10,000	Hydrate the retained Reel	2,573	17,225	64	64	388.702	58.055	±66.53%	±21.46%
`__read`	10,000	Read a rotating visible index	5,020,080	103,040	32,768	64	0.1992	9.705	±9.58%	±21.99%
`__length`	10,000	Read the visible length	16,350,556	12,603,983	32,768	32,768	0.06116	0.07934	±15.26%	±16.01%
`__recover`	10,000	Recover all retained values	16,344	106,826	64	64	61.183	9.361	±17.86%	±26.31%
`__update`	10,000	Insert one Frame at the midpoint	38,438	7,982	64	64	26.016	125.289	±39.17%	±82.21%
`__delete`	10,000	Soft-delete one midpoint Frame	172,891	13,819	64	64	5.784	72.364	±40.84%	±11.56%
`__merge`	10,000	Merge one new midpoint Frame	46,100	10,454	64	64	21.692	95.656	±55.93%	±20.84%
`__acknowledge`	10,000	Acknowledge materialized Realm progress	3,225,806	968,054	1,024	1,024	0.3100	1.033	±11.50%	±14.23%
`__garbageCollect`	10,000	Release one soft-deleted Frame	46,893	—	64	—	21.325	—	±107.46%	—
`__snapshot`	10,000	Snapshot complete retained state	43,301	22,213	64	64	23.094	45.019	±48.42%	±37.33%
`__create`	100,000	Hydrate the retained Reel	142	1,438	16	16	7,056.825	695.625	±30.00%	±56.66%
`__read`	100,000	Read a rotating visible index	4,083,299	10,018	8,192	16	0.2449	99.819	±15.38%	±31.71%
`__length`	100,000	Read the visible length	12,274,457	6,877,579	8,192	8,192	0.08147	0.1454	±6.06%	±26.14%
`__recover`	100,000	Recover all retained values	878	17,337	16	16	1,138.744	57.681	±32.91%	±14.07%
`__update`	100,000	Insert one Frame at the midpoint	98,580	1,812	16	16	10.144	551.969	±25.58%	±12.35%
`__delete`	100,000	Soft-delete one midpoint Frame	190,694	1,930	16	16	5.244	518.012	±20.29%	±8.73%
`__merge`	100,000	Merge one new midpoint Frame	89,485	1,372	16	16	11.175	728.825	±55.60%	±21.61%
`__acknowledge`	100,000	Acknowledge materialized Realm progress	4,182,350	1,050,089	256	256	0.2391	0.9523	±5.96%	±16.26%
`__garbageCollect`	100,000	Release one soft-deleted Frame	320,616	—	16	—	3.119	—	±5.37%	—
`__snapshot`	100,000	Snapshot complete retained state	1,559	4,558	16	16	641.431	219.381	±44.63%	±29.69%
`__create`	1,000,000	Hydrate the retained Reel	18	267	16	16	56,965.988	3,747.700	±14.14%	±10.55%
`__read`	1,000,000	Read a rotating visible index	2,768,549	754	8,192	16	0.3612	1,327.000	±57.49%	±25.38%
`__length`	1,000,000	Read the visible length	22,381,379	8,496,177	8,192	8,192	0.04468	0.1177	±7.55%	±2.88%
`__recover`	1,000,000	Recover all retained values	58	897	16	16	17,147.762	1,115.063	±39.12%	±24.01%
`__update`	1,000,000	Insert one Frame at the midpoint	99,443	246	16	16	10.056	4,072.900	±8.11%	±9.25%
`__delete`	1,000,000	Soft-delete one midpoint Frame	108,401	269	16	16	9.225	3,723.856	±14.51%	±11.97%
`__merge`	1,000,000	Merge one new midpoint Frame	59,018	262	16	16	16.944	3,811.575	±31.17%	±14.74%
`__acknowledge`	1,000,000	Acknowledge materialized Realm progress	6,385,696	1,419,849	256	256	0.1566	0.7043	±15.48%	±7.32%
`__garbageCollect`	1,000,000	Release one soft-deleted Frame	235,627	—	16	—	4.244	—	±7.87%	—
`__snapshot`	1,000,000	Snapshot complete retained state	156	912	16	16	6,398.656	1,096.031	±39.13%	±13.20%

Diamond Types 1.0.2: “The world's fastest CRDT. WIP.” Both implementations use one-character strings, the same logical Sequence lengths, midpoint mutations, warm-up sample counts, and measured sample counts. Each uses its public native snapshot and patch format. Diamond Types has no direct indexed-read API, so its read result calls get() before selecting one character; its clean-state get() is the closest recovery equivalent. Diamond Types exposes no public garbage-collection operation, shown as —.

Methodology: avg µs/op is the arithmetic mean of the measured per-call latencies. Ops/sec is its reciprocal; displayed values are rounded as one reciprocal pair. Setup, cleanup, and warmup calls are excluded. Mutating operations receive a fresh prepared state for every sample; read-only operations use stable state. Batch sizes only amortize timer overhead and are reported through call counts. These Node.js results do not represent browser runtimes.

Small bundle size

Format	Raw	Minified	Minified + gzip
ESM	91.3 kB	53.8 kB	19.0 kB
CommonJS	91.6 kB	56.3 kB	19.3 kB

Compact data model

Reel workload	Average bytes per operation	MessagePack	MessagePack + gzip	Gzip reduction
1,000 one-frame updates	39.6 B	39.6 kB	7.8 kB	80.2%
1,000 one-frame masks	36.0 B	36.0 kB	5.1 kB	85.9%
Snapshot containing 1,000 one-frame strips	39.6 B	39.6 kB	7.8 kB	80.2%

Reel sizes use MessagePack and gzip level 9 over the complete 1,000-operation collection. Gzip reduction is measured against the MessagePack payload.