Template engines / document assembly

One template.
New data.
Exact output.

Five engines assemble the same documents. What does preparation cost—and what happens when the next render arrives?

ASSEMBLY / 001BYTE CONTRACT

01 / TEMPLATE<h1>$[d.title]</h1>
@input('footer')

02 / DATA{ title: "Template lab",
author: { name: "Nick" } }

03 / OUTPUT<h1>Template lab</h1>
<footer>Nick</footer>

SOURCE + BINDINGS → RENDERER → CHECKED BYTES

Nift 4.12.0Native embedded renderer

Jinja2 3.1.6Python compiled templates

Nunjucks 3.2.4JavaScript template runtime

Handlebars 4.7.10Compiled presentation functions

LiquidJS 10.30.0Constrained template model

17,950measured samples375supported cells20timed scenarios750retained warmups

Fresh article: Nift 4.695 ms per invocation. Persistent 10,000-record render: Nift 34.085 ms, Handlebars 2.249 ms with preparation included. Low invocation cost does not establish sustained rendering throughput.

01 / the contract

The document comes first.

Independent data and expected bytes define the task. Participant templates express the same literal text, values, branches, named records and include expansions. No tracked builds, generated sites or application servers are measured.

Fresh process

Start → load → render

Runtime startup, modules, source/data loading, complete rendering and stdout capture. Milliseconds per invocation.

Persistent runtime

Prepare → render

Preloaded data and sources. Timed engine calls, batched where needed. Milliseconds per template operation.

Prepared templates

Render → render again

Parsed/compiled artifacts reused where public APIs permit. Nift source reuse still parses; it has no exposed prepared-AST handle.

The corpus is intentionally raw text. Automatic HTML escaping, contextual safety and untrusted-template sandboxing are not certified here. A separate special-character gate verifies that template-looking values remain data.

02 / operational latency

A fresh process has a price.

These totals include host-language startup, both frozen data variants, all scenario templates and complete intended output. A single render still preloads an unused second variant in every adapter. This is the reproducible harness boundary, not a hand-tuned minimal request or engine-only throughput.

Fresh end-to-end invocation — Fresh invocation median ms; 30–50 measured samples per cell. SVG · PNG. On narrow screens, scroll the chart and table horizontally.

Fresh invocation median ms; 30–50 measured samples per cell.
Workload	Nift	Jinja2	Nunjucks	Handlebars	LiquidJS
Literal	4.426	78.642	50.702	63.797	63.970
Values	4.328	76.878	51.883	65.213	63.657
Article	4.695	78.686	54.578	69.893	69.959
1k records	11.024	80.401	58.704	69.804	120.795

03 / preparation

Before the first render.

Jinja produces code, Nunjucks eager-compiles, Handlebars precompiles JavaScript source and LiquidJS parses tokens. These preparation artifacts differ. Nift exposes no standalone preparation API; its column is unavailable.

Persistent preparation only — Preparation median ms per template; 50 samples; logarithmic axis. SVG · PNG. On narrow screens, scroll the chart and table horizontally.

Preparation median ms per template; 50 samples; logarithmic axis.
Workload	Nift	Jinja2	Nunjucks	Handlebars	LiquidJS
Literal	—	0.203	0.018	0.031	0.002
Values	—	0.781	0.084	0.252	0.015
Article	—	1.894	0.269	0.755	0.053

04 / records and output

More rows. More work.

Main source preparation plus rendering is timed on every operation. Nift uses its native render_text call. Preloaded contexts and output validation remain outside the timer.

Persistent source preparation + render — Median ms per render; 50 samples; logarithmic axis. SVG · PNG. On narrow screens, scroll the chart and table horizontally.

Median ms per render; 50 samples; logarithmic axis.
Workload	Nift	Jinja2	Nunjucks	Handlebars	LiquidJS
10 records	0.064	0.889	0.146	0.381	0.191
100 records	0.364	1.035	0.245	0.383	0.938
1,000 records	3.293	2.494	1.560	0.649	8.677
10,000 records	34.085	16.886	11.820	2.249	85.047

05 / prepared reuse

Preparation can be amortized.

These four engines reuse prepared templates with preloaded data. Nift is absent because it does not expose the same capability. Keeping a Nift Engine alive avoids process startup, but does not turn source reuse into parsed-template reuse.

Prepared-template rendering — Median ms per prepared render; 50 samples; logarithmic axis; Nift API unavailable. SVG · PNG. On narrow screens, scroll the chart and table horizontally.

Median ms per prepared render; 50 samples; logarithmic axis; Nift API unavailable.
Workload	Nift	Jinja2	Nunjucks	Handlebars	LiquidJS
10 records	—	0.027	0.012	0.005	0.096
100 records	—	0.172	0.117	0.021	0.836
1,000 records	—	1.593	1.038	0.294	8.584
10,000 records	—	15.194	10.200	1.884	84.503

Prepared template, alternating new data — Median total ms per repeated-render workload; 50 samples; logarithmic axis. SVG · PNG. On narrow screens, scroll the chart and table horizontally.

Median total ms per repeated-render workload; 50 samples; logarithmic axis.
Workload	Nift	Jinja2	Nunjucks	Handlebars	LiquidJS
100 renders	—	1.028	0.059	0.176	0.775
1,000 renders	—	10.298	0.638	1.703	7.691

06 / partials

Composition has two boundaries.

Memory-backed sources use normal warmed partial caches where available, while main preparation repeats. File-backed calls load from files with engine source/partial caches invalidated or disabled. OS caches stay warm. Nift includes its supported root-reset cost.

Memory-backed composition — Median ms per prepare + render; 50 samples. SVG · PNG. On narrow screens, scroll the chart and table horizontally.

Median ms per prepare + render; 50 samples.
Workload	Nift	Jinja2	Nunjucks	Handlebars	LiquidJS
Depth 1	0.043	0.800	0.102	0.317	0.097
Depth 4	0.089	0.814	0.102	0.267	0.086
Depth 8	0.163	0.822	0.131	0.375	0.157

File-backed composition — Median ms per file-backed prepare + render; 50 samples. SVG · PNG. On narrow screens, scroll the chart and table horizontally.

Median ms per file-backed prepare + render; 50 samples.
Workload	Nift	Jinja2	Nunjucks	Handlebars	LiquidJS
Depth 1	0.078	1.315	0.187	0.466	0.218
Depth 4	0.168	2.766	0.403	0.997	0.526
Depth 8	0.300	4.746	0.622	1.692	0.852

07 / realistic assembly

A page is more than interpolation.

The article combines metadata, navigation, sections, conditional output and a footer. Nested loops and a 2,000-record collection exercise lookup and output growth. Uppercase is the same ASCII transformation in every engine.

Persistent preparation + rendering — Median ms per render; 50 samples; logarithmic axis. SVG · PNG. On narrow screens, scroll the chart and table horizontally.

Median ms per render; 50 samples; logarithmic axis.
Workload	Nift	Jinja2	Nunjucks	Handlebars	LiquidJS
Article	0.275	2.046	0.373	0.963	0.448
Nested loops	1.068	1.600	0.512	0.599	2.073
2k collection	7.149	4.042	2.615	0.867	17.205
Uppercase	0.021	0.498	0.036	0.179	0.037

08 / source diversity

A thousand distinct templates.

Unique main sources carry distinct output suffixes. Every source is prepared and rendered; this is not a cache-hit claim. Values below are complete workload totals rather than amortized individual template times.

Prepare + render distinct templates — Median total ms per workload; 50 samples; logarithmic axis. SVG · PNG. On narrow screens, scroll the chart and table horizontally.

Median total ms per workload; 50 samples; logarithmic axis.
Workload	Nift	Jinja2	Nunjucks	Handlebars	LiquidJS
10 templates	0.212	4.086	0.308	1.105	0.108
100 templates	2.142	39.918	2.380	6.235	1.023
1,000 templates	21.516	392.408	23.372	64.676	10.465

09 / resource scope

Memory belongs to a boundary.

These are unranked raw accounting observations, not memory-efficiency rankings. The final noop-child control rose from 16.75 to 80.75 MiB when its launch parent allocated 64 MiB: inherited wait4 floors can dominate fresh and whole-worker peaks. No constant is subtracted. Persistent-worker lifetime also includes preload, preparation, retained batch outputs and validation; it is not memory per render. Scopes remain separate.

Raw median maximum-child RSS MiB; inherited launch floor; unranked, 30–50 invocations per cell. SVG · PNG. On narrow screens, scroll the chart and table horizontally.

Raw median maximum-child RSS MiB; inherited launch floor; unranked, 30–50 invocations per cell.
Workload	Nift	Jinja2	Nunjucks	Handlebars	LiquidJS
Literal	27.875	27.875	50.125	52.633	54.625
Article	27.875	27.875	51.000	53.141	60.652
2k collection	27.875	27.875	52.562	53.902	67.766

Persistent-worker high-water observations

Unranked whole-worker maximum RSS, MiB; launch floor and lifetime scope, not memory per render
Workload	Engine	Operation	Peak MiB
literal	Nift	source-reuse	27.875
literal	Jinja2	render	27.875
literal	Nunjucks	render	59.504
literal	Handlebars	render	85.199
literal	LiquidJS	render	69.184
literal	Jinja2	reuse	27.875
literal	Nunjucks	reuse	71.754
literal	Handlebars	reuse	65.172
literal	LiquidJS	reuse	68.129
literal	Jinja2	parse	27.875
literal	Nunjucks	parse	63.391
literal	Handlebars	parse	66.246
literal	LiquidJS	parse	67.113
article	Nift	source-reuse	27.875
article	Jinja2	render	27.875
article	Nunjucks	render	73.113
article	Handlebars	render	81.598
article	LiquidJS	render	82.656
article	Jinja2	reuse	27.875
article	Nunjucks	reuse	94.820
article	Handlebars	reuse	94.762
article	LiquidJS	reuse	82.438
article	Jinja2	parse	27.875
article	Nunjucks	parse	63.020
article	Handlebars	parse	67.078
article	LiquidJS	parse	69.355
collection	Nift	source-reuse	27.875
collection	Jinja2	render	27.875
collection	Nunjucks	render	94.496
collection	Handlebars	render	98.910
collection	LiquidJS	render	103.289
collection	Jinja2	reuse	27.875
collection	Nunjucks	reuse	90.516
collection	Handlebars	reuse	152.941
collection	LiquidJS	reuse	103.789
collection	Jinja2	parse	27.875
collection	Nunjucks	parse	63.297
collection	Handlebars	parse	66.047
collection	LiquidJS	parse	70.125

10 / independent gates

A faster wrong document is rejected.

Cloud certification passed all 393 supported cells, including a separate raw-special-value fixture. Every timed render is compared against independent expected bytes. Fresh output hashes are retained; persistent workers check every batch output before emitting their timing row. Ordering, record counts, includes and changing-data variants follow the exact output contract.

The independent audit checks all 375 official cells, the complete sample index set, 17,950 measured observations, 750 separate warmups and every frozen corpus hash. No statistical outliers were removed. Thirty extra special-value/malformed/variant gates passed. Historical newline/Node-stdout preparation failures remain in the archived evidence; final certification and observations passed.

All measured medians, p95 and ranges

Fresh ms/invocation; all persistent modes ms/template operation; no boundary pooling
Workload	Engine	Mode	Median	p95	Min	Max	Samples
literal	Nift	fresh	4.4261	5.6212	3.6464	9.0663	50
literal	Nift	source-reuse	0.0152	0.0286	0.0146	0.0298	50
literal	Jinja2	fresh	78.6419	85.0563	73.3372	100.9441	50
literal	Jinja2	render	0.2390	0.3255	0.2163	0.4039	50
literal	Jinja2	reuse	0.0090	0.0160	0.0087	0.0198	50
literal	Jinja2	parse	0.2033	0.3897	0.1877	0.9995	50
literal	Nunjucks	fresh	50.7022	60.1301	47.4936	61.7883	50
literal	Nunjucks	render	0.0227	0.1042	0.0140	0.1274	50
literal	Nunjucks	reuse	0.0005	0.0015	0.0004	0.0017	50
literal	Nunjucks	parse	0.0183	0.0798	0.0109	0.1263	50
literal	Handlebars	fresh	63.7970	72.5479	60.2171	77.9489	50
literal	Handlebars	render	0.0437	0.2081	0.0189	0.2860	50
literal	Handlebars	reuse	0.0014	0.0084	0.0013	0.0116	50
literal	Handlebars	parse	0.0313	0.1095	0.0115	0.1744	50
literal	LiquidJS	fresh	63.9702	77.2523	60.7756	80.5271	50
literal	LiquidJS	render	0.0036	0.0219	0.0022	0.0265	50
literal	LiquidJS	reuse	0.0010	0.0118	0.0009	0.0252	50
literal	LiquidJS	parse	0.0016	0.0058	0.0013	0.0212	50
interpolation	Nift	fresh	4.3276	5.6980	3.6496	8.9494	50
interpolation	Nift	source-reuse	0.0280	0.0436	0.0271	0.0545	50
interpolation	Jinja2	fresh	76.8781	86.7747	73.8807	94.0831	50
interpolation	Jinja2	render	0.6766	1.0704	0.6323	1.3550	50
interpolation	Jinja2	reuse	0.0120	0.0241	0.0117	0.0480	50
interpolation	Jinja2	parse	0.7808	1.7127	0.6073	2.4405	50
interpolation	Nunjucks	fresh	51.8829	58.7476	49.1177	62.1285	50
interpolation	Nunjucks	render	0.0712	0.1792	0.0476	0.3035	50
interpolation	Nunjucks	reuse	0.0007	0.0030	0.0006	0.0090	50
interpolation	Nunjucks	parse	0.0844	0.2407	0.0470	0.2964	50
interpolation	Handlebars	fresh	65.2132	78.2362	62.0791	81.6657	50
interpolation	Handlebars	render	0.2692	0.5914	0.1379	0.7077	50
interpolation	Handlebars	reuse	0.0033	0.0247	0.0016	0.0288	50
interpolation	Handlebars	parse	0.2519	0.6989	0.1369	1.4230	50
interpolation	LiquidJS	fresh	63.6574	71.2344	60.9407	86.4300	50
interpolation	LiquidJS	render	0.0372	0.1618	0.0242	0.2351	50
interpolation	LiquidJS	reuse	0.0138	0.0613	0.0110	0.0952	50
interpolation	LiquidJS	parse	0.0153	0.0946	0.0105	0.1078	50
condition	Nift	fresh	4.1892	4.8143	3.7012	5.3136	50
condition	Nift	source-reuse	0.0229	0.0390	0.0213	0.0515	50
condition	Jinja2	fresh	77.8444	88.0658	73.0504	90.9726	50
condition	Jinja2	render	0.5321	0.7621	0.4945	1.0677	50
condition	Jinja2	reuse	0.0098	0.0169	0.0094	0.0263	50
condition	Jinja2	parse	0.4801	0.8312	0.4604	1.1934	50
condition	Nunjucks	fresh	51.0503	56.4302	47.9723	58.0936	50
condition	Nunjucks	render	0.0569	0.1990	0.0371	0.5495	50
condition	Nunjucks	reuse	0.0005	0.0019	0.0004	0.0042	50
condition	Nunjucks	parse	0.0467	0.2082	0.0371	0.3145	50
condition	Handlebars	fresh	66.1081	72.5365	62.6521	75.7466	50
condition	Handlebars	render	0.3229	0.7497	0.1489	1.0832	50
condition	Handlebars	reuse	0.0024	0.0151	0.0016	0.0338	50
condition	Handlebars	parse	0.2060	0.5296	0.1273	0.6510	50
condition	LiquidJS	fresh	63.9807	69.8989	61.7739	74.1567	50
condition	LiquidJS	render	0.0397	0.1554	0.0203	0.2685	50
condition	LiquidJS	reuse	0.0095	0.0474	0.0047	0.0701	50
condition	LiquidJS	parse	0.0099	0.0722	0.0072	0.0802	50
nested-object	Nift	fresh	4.1875	4.7946	3.6676	8.5395	50
nested-object	Nift	source-reuse	0.0229	0.0330	0.0221	0.0340	50
nested-object	Jinja2	fresh	76.9953	83.1994	72.8081	88.0045	50
nested-object	Jinja2	render	0.5582	0.6347	0.5261	0.7449	50
nested-object	Jinja2	reuse	0.0113	0.0218	0.0107	0.0293	50
nested-object	Jinja2	parse	0.4989	0.7886	0.4760	2.1077	50
nested-object	Nunjucks	fresh	51.1950	58.1469	47.7610	63.5917	50
nested-object	Nunjucks	render	0.0524	0.2149	0.0386	0.5389	50
nested-object	Nunjucks	reuse	0.0006	0.0094	0.0005	0.0164	50
nested-object	Nunjucks	parse	0.0442	0.2568	0.0350	0.3042	50
nested-object	Handlebars	fresh	65.1363	78.2033	61.7493	81.0628	50
nested-object	Handlebars	render	0.1932	0.5483	0.1082	0.7188	50
nested-object	Handlebars	reuse	0.0019	0.0064	0.0014	0.0171	50
nested-object	Handlebars	parse	0.1586	0.4076	0.0948	1.0385	50
nested-object	LiquidJS	fresh	63.9874	68.4241	60.3253	72.1575	50
nested-object	LiquidJS	render	0.0445	0.1589	0.0188	0.1791	50
nested-object	LiquidJS	reuse	0.0108	0.0351	0.0077	0.0674	50
nested-object	LiquidJS	parse	0.0098	0.0862	0.0065	0.1645	50
loop-10	Nift	fresh	4.4065	6.0660	3.7697	7.8079	50
loop-10	Nift	source-reuse	0.0637	0.1321	0.0606	0.2026	50
loop-10	Jinja2	fresh	77.4983	105.2507	73.9972	116.9527	50
loop-10	Jinja2	render	0.8888	1.0565	0.8421	1.3105	50
loop-10	Jinja2	reuse	0.0265	0.0475	0.0259	0.0579	50
loop-10	Jinja2	parse	0.8160	1.7805	0.7841	2.5044	50
loop-10	Nunjucks	fresh	52.6948	60.5095	50.7471	67.6170	50
loop-10	Nunjucks	render	0.1461	0.3823	0.0850	0.6140	50
loop-10	Nunjucks	reuse	0.0117	0.0210	0.0102	0.0264	50
loop-10	Nunjucks	parse	0.1018	0.3503	0.0775	0.5189	50
loop-10	Handlebars	fresh	67.3288	84.6068	63.8084	108.4824	50
loop-10	Handlebars	render	0.3809	0.9908	0.1969	1.2816	50
loop-10	Handlebars	reuse	0.0051	0.0313	0.0029	0.0910	50
loop-10	Handlebars	parse	0.2728	0.7089	0.1837	0.9146	50
loop-10	LiquidJS	fresh	65.7343	84.3587	63.1271	90.1482	50
loop-10	LiquidJS	render	0.1907	0.7412	0.1123	1.0468	50
loop-10	LiquidJS	reuse	0.0965	0.1975	0.0868	0.5068	50
loop-10	LiquidJS	parse	0.0164	0.1165	0.0128	0.1649	50
loop-100	Nift	fresh	4.8396	5.3569	4.3307	5.8283	50
loop-100	Nift	source-reuse	0.3643	0.6632	0.3543	1.1559	50
loop-100	Jinja2	fresh	77.1476	84.2377	73.3244	91.3559	50
loop-100	Jinja2	render	1.0348	1.1685	0.9927	1.4267	50
loop-100	Jinja2	reuse	0.1718	0.2041	0.1669	0.3858	50
loop-100	Jinja2	parse	0.8295	1.7697	0.7859	2.3293	50
loop-100	Nunjucks	fresh	53.6097	58.6037	51.1621	62.7372	50
loop-100	Nunjucks	render	0.2449	0.9213	0.1829	1.2520	50
loop-100	Nunjucks	reuse	0.1168	0.1795	0.0985	0.3540	50
loop-100	Nunjucks	parse	0.1150	0.3636	0.0817	0.4534	50
loop-100	Handlebars	fresh	66.9474	72.4665	64.4685	86.9737	50
loop-100	Handlebars	render	0.3834	0.9395	0.2241	1.5553	50
loop-100	Handlebars	reuse	0.0211	0.0689	0.0140	0.2634	50
loop-100	Handlebars	parse	0.2949	0.7853	0.1823	1.1169	50
loop-100	LiquidJS	fresh	74.4833	82.8043	69.9790	85.5465	50
loop-100	LiquidJS	render	0.9377	1.9441	0.8237	2.8882	50
loop-100	LiquidJS	reuse	0.8363	1.2428	0.7764	2.2382	50
loop-100	LiquidJS	parse	0.0175	0.1352	0.0130	0.2082	50
loop-1000	Nift	fresh	11.0243	12.4321	10.5550	13.8082	30
loop-1000	Nift	source-reuse	3.2934	3.6550	3.2192	7.2242	50
loop-1000	Jinja2	fresh	80.4006	89.9420	76.5213	94.6888	30
loop-1000	Jinja2	render	2.4941	2.9182	2.4119	3.1289	50
loop-1000	Jinja2	reuse	1.5927	2.7383	1.5417	3.0826	50
loop-1000	Jinja2	parse	0.8104	1.6755	0.7728	1.8544	50
loop-1000	Nunjucks	fresh	58.7043	64.9430	55.6729	70.4583	30
loop-1000	Nunjucks	render	1.5602	2.8825	1.1271	4.7745	50
loop-1000	Nunjucks	reuse	1.0381	1.8836	0.9505	2.4397	50
loop-1000	Nunjucks	parse	0.0980	0.4565	0.0781	0.4839	50
loop-1000	Handlebars	fresh	69.8036	76.8036	66.9621	78.9887	30
loop-1000	Handlebars	render	0.6493	1.4998	0.3717	3.5819	50
loop-1000	Handlebars	reuse	0.2937	0.5694	0.1293	0.9390	50
loop-1000	Handlebars	parse	0.3177	1.8361	0.2086	1.9618	50
loop-1000	LiquidJS	fresh	120.7950	133.3768	116.0802	138.0007	30
loop-1000	LiquidJS	render	8.6769	14.3764	7.7187	16.6323	50
loop-1000	LiquidJS	reuse	8.5839	16.4251	7.7400	25.9177	50
loop-1000	LiquidJS	parse	0.0178	0.1260	0.0130	0.1532	50
loop-10000	Nift	fresh	76.7503	84.0528	71.4375	84.6973	30
loop-10000	Nift	source-reuse	34.0847	37.8297	33.2001	69.5894	50
loop-10000	Jinja2	fresh	113.5029	122.7730	106.7649	124.6484	30
loop-10000	Jinja2	render	16.8858	18.7210	16.2922	21.8873	50
loop-10000	Jinja2	reuse	15.1943	16.0843	15.0442	24.6687	50
loop-10000	Jinja2	parse	0.8217	1.8227	0.7883	3.0012	50
loop-10000	Nunjucks	fresh	98.8274	109.0909	93.4323	115.3841	30
loop-10000	Nunjucks	render	11.8195	17.7200	10.0070	27.0111	50
loop-10000	Nunjucks	reuse	10.2000	13.7770	9.1717	21.6150	50
loop-10000	Nunjucks	parse	0.1124	0.3668	0.0738	0.5397	50
loop-10000	Handlebars	fresh	99.3733	109.8616	93.0798	112.2132	30
loop-10000	Handlebars	render	2.2486	5.3166	2.1086	8.4113	50
loop-10000	Handlebars	reuse	1.8837	4.2244	1.6953	7.1547	50
loop-10000	Handlebars	parse	0.2578	0.7967	0.1837	0.9490	50
loop-10000	LiquidJS	fresh	240.1026	260.8873	227.0205	262.9817	30
loop-10000	LiquidJS	render	85.0467	97.7849	81.6179	110.6598	50
loop-10000	LiquidJS	reuse	84.5030	97.5672	81.7115	103.0087	50
loop-10000	LiquidJS	parse	0.0202	0.1674	0.0141	0.1963	50
nested-loops	Nift	fresh	6.0909	7.1461	5.4670	7.9292	50
nested-loops	Nift	source-reuse	1.0676	1.7010	1.0376	2.9206	50
nested-loops	Jinja2	fresh	77.2261	85.1275	73.9228	95.4614	50
nested-loops	Jinja2	render	1.6002	1.7590	1.5489	2.3589	50
nested-loops	Jinja2	reuse	0.3447	0.4343	0.3364	0.7418	50
nested-loops	Jinja2	parse	1.2087	1.9913	1.1555	2.3859	50
nested-loops	Nunjucks	fresh	53.6076	55.8791	51.3149	62.6151	50
nested-loops	Nunjucks	render	0.5115	1.3899	0.3448	2.3594	50
nested-loops	Nunjucks	reuse	0.2479	0.3373	0.2061	0.3941	50
nested-loops	Nunjucks	parse	0.1769	0.6521	0.1262	1.2929	50
nested-loops	Handlebars	fresh	67.7136	73.2727	64.8081	74.2442	50
nested-loops	Handlebars	render	0.5988	1.6230	0.3311	3.3512	50
nested-loops	Handlebars	reuse	0.0583	0.1159	0.0374	0.2571	50
nested-loops	Handlebars	parse	0.4171	0.9623	0.2792	1.7729	50
nested-loops	LiquidJS	fresh	78.5052	87.4217	75.1915	90.6321	50
nested-loops	LiquidJS	render	2.0726	3.5077	1.7807	4.1771	50
nested-loops	LiquidJS	reuse	1.8678	2.8290	1.6837	5.0831	50
nested-loops	LiquidJS	parse	0.0269	0.1572	0.0212	0.2005	50
includes-1	Nift	fresh	4.3987	4.7569	3.7388	5.4986	50
includes-1	Nift	source-reuse	0.0434	0.0826	0.0411	0.0934	50
includes-1	Nift	files	0.0781	0.1173	0.0758	0.1233	50
includes-1	Jinja2	fresh	77.1416	86.7477	74.2007	100.4816	50
includes-1	Jinja2	render	0.7999	1.1783	0.7559	1.5032	50
includes-1	Jinja2	reuse	0.0164	0.0350	0.0161	0.0513	50
includes-1	Jinja2	parse	0.7347	1.3711	0.6935	1.4904	50
includes-1	Jinja2	files	1.3153	1.5385	1.2518	1.6506	50
includes-1	Nunjucks	fresh	52.8592	63.2610	51.0474	71.1261	50
includes-1	Nunjucks	render	0.1019	0.3546	0.0610	0.5805	50
includes-1	Nunjucks	reuse	0.0021	0.0327	0.0019	0.0405	50
includes-1	Nunjucks	parse	0.0767	0.3610	0.0576	0.6221	50
includes-1	Nunjucks	files	0.1871	0.5383	0.1214	0.8829	50
includes-1	Handlebars	fresh	66.7962	71.7125	64.0374	77.0327	50
includes-1	Handlebars	render	0.3170	0.8359	0.1578	1.0441	50
includes-1	Handlebars	reuse	0.0030	0.0207	0.0018	0.0466	50
includes-1	Handlebars	parse	0.2341	0.5615	0.1476	0.8923	50
includes-1	Handlebars	files	0.4661	1.0953	0.2755	1.4993	50
includes-1	LiquidJS	fresh	65.2035	73.1481	62.8935	76.9329	50
includes-1	LiquidJS	render	0.0974	0.3522	0.0416	0.4198	50
includes-1	LiquidJS	reuse	0.0245	0.0746	0.0181	0.1521	50
includes-1	LiquidJS	parse	0.0155	0.0862	0.0114	0.1186	50
includes-1	LiquidJS	files	0.2184	0.5916	0.1316	0.8224	50
includes-4	Nift	fresh	4.4392	5.1673	3.8126	5.3932	50
includes-4	Nift	source-reuse	0.0890	0.1890	0.0857	0.3005	50
includes-4	Nift	files	0.1678	0.2377	0.1616	0.2467	50
includes-4	Jinja2	fresh	78.4327	85.4734	74.4581	88.9807	50
includes-4	Jinja2	render	0.8139	1.0179	0.7734	1.0637	50
includes-4	Jinja2	reuse	0.0307	0.0502	0.0297	0.0845	50
includes-4	Jinja2	parse	0.8988	1.8381	0.6971	2.9960	50
includes-4	Jinja2	files	2.7660	3.2677	2.6420	4.3760	50
includes-4	Nunjucks	fresh	54.1569	63.1533	51.9838	67.1992	50
includes-4	Nunjucks	render	0.1019	0.4834	0.0669	0.6503	50
includes-4	Nunjucks	reuse	0.0088	0.0401	0.0062	0.0748	50
includes-4	Nunjucks	parse	0.1114	0.3572	0.0652	0.4456	50
includes-4	Nunjucks	files	0.4034	1.1382	0.2250	1.6808	50
includes-4	Handlebars	fresh	67.7791	75.5850	64.6565	80.9506	50
includes-4	Handlebars	render	0.2671	0.8209	0.1548	1.0699	50
includes-4	Handlebars	reuse	0.0037	0.0331	0.0024	0.0391	50
includes-4	Handlebars	parse	0.2807	1.0162	0.1667	1.5083	50
includes-4	Handlebars	files	0.9969	2.3695	0.4996	2.6027	50
includes-4	LiquidJS	fresh	65.4844	74.7830	63.1063	78.7590	50
includes-4	LiquidJS	render	0.0859	0.3575	0.0451	0.5023	50
includes-4	LiquidJS	reuse	0.0321	0.1346	0.0246	0.1408	50
includes-4	LiquidJS	parse	0.0189	0.0852	0.0122	0.1372	50
includes-4	LiquidJS	files	0.5256	1.4685	0.3030	1.6216	50
includes-8	Nift	fresh	4.8741	5.5560	4.1811	7.3288	50
includes-8	Nift	source-reuse	0.1629	0.2382	0.1580	0.2776	50
includes-8	Nift	files	0.3004	0.4662	0.2882	0.6218	50
includes-8	Jinja2	fresh	83.4359	96.1542	78.0738	108.7327	50
includes-8	Jinja2	render	0.8216	1.1068	0.7799	1.3260	50
includes-8	Jinja2	reuse	0.0488	0.0854	0.0469	0.1383	50
includes-8	Jinja2	parse	0.7505	1.7882	0.6981	2.1688	50
includes-8	Jinja2	files	4.7458	5.5628	4.5185	7.9408	50
includes-8	Nunjucks	fresh	58.1122	62.5263	54.2429	66.1784	50
includes-8	Nunjucks	render	0.1315	0.5304	0.0800	1.1033	50
includes-8	Nunjucks	reuse	0.0198	0.0836	0.0136	0.1117	50
includes-8	Nunjucks	parse	0.0947	0.2977	0.0597	0.5292	50
includes-8	Nunjucks	files	0.6224	1.6066	0.3818	3.4187	50
includes-8	Handlebars	fresh	74.6415	82.7258	68.8397	85.2903	50
includes-8	Handlebars	render	0.3754	0.7309	0.1575	1.7701	50
includes-8	Handlebars	reuse	0.0051	0.0176	0.0033	0.0207	50
includes-8	Handlebars	parse	0.2546	0.9141	0.1455	1.0704	50
includes-8	Handlebars	files	1.6924	3.8504	0.7983	4.9524	50
includes-8	LiquidJS	fresh	68.3819	78.8433	64.1259	88.4977	50
includes-8	LiquidJS	render	0.1569	0.3923	0.0632	0.5770	50
includes-8	LiquidJS	reuse	0.0505	0.1760	0.0403	0.1855	50
includes-8	LiquidJS	parse	0.0162	0.0927	0.0112	0.1253	50
includes-8	LiquidJS	files	0.8524	2.9029	0.5571	3.5492	50
article	Nift	fresh	4.6949	5.5039	4.0456	5.8725	50
article	Nift	source-reuse	0.2748	0.5307	0.2606	0.7297	50
article	Jinja2	fresh	78.6855	87.7151	75.1414	90.6292	50
article	Jinja2	render	2.0464	3.2086	1.9244	3.9432	50
article	Jinja2	reuse	0.0456	0.0601	0.0418	0.1948	50
article	Jinja2	parse	1.8944	3.5091	1.7972	3.7629	50
article	Nunjucks	fresh	54.5781	60.5419	51.6308	71.6696	50
article	Nunjucks	render	0.3730	1.3140	0.2212	1.6773	50
article	Nunjucks	reuse	0.0219	0.0472	0.0188	0.0760	50
article	Nunjucks	parse	0.2685	1.0763	0.1949	1.5687	50
article	Handlebars	fresh	69.8928	77.1008	67.3761	110.0634	50
article	Handlebars	render	0.9627	2.3016	0.5174	4.2980	50
article	Handlebars	reuse	0.0078	0.0344	0.0052	0.0778	50
article	Handlebars	parse	0.7549	1.5534	0.4703	2.4844	50
article	LiquidJS	fresh	69.9595	79.9298	66.8822	85.3770	50
article	LiquidJS	render	0.4484	1.0092	0.2225	2.8583	50
article	LiquidJS	reuse	0.2091	0.4850	0.1618	0.7786	50
article	LiquidJS	parse	0.0531	0.3570	0.0400	0.7922	50
collection	Nift	fresh	17.6440	24.5978	16.8772	26.7375	30
collection	Nift	source-reuse	7.1487	12.0038	6.9096	13.0758	50
collection	Jinja2	fresh	84.3397	91.8468	79.6775	93.3692	30
collection	Jinja2	render	4.0420	4.9196	3.9427	8.2429	50
collection	Jinja2	reuse	3.1240	3.4487	3.0724	7.0688	50
collection	Jinja2	parse	0.7736	1.2153	0.7268	1.7465	50
collection	Nunjucks	fresh	63.1874	70.9357	60.0178	73.2733	30
collection	Nunjucks	render	2.6148	4.4671	2.0631	5.0435	50
collection	Nunjucks	reuse	2.0918	2.9263	1.8775	3.0353	50
collection	Nunjucks	parse	0.1024	0.3802	0.0721	0.6498	50
collection	Handlebars	fresh	72.2749	84.2659	68.3395	84.8205	30
collection	Handlebars	render	0.8673	2.6272	0.5297	3.3048	50
collection	Handlebars	reuse	0.7486	1.2162	0.4499	2.3485	50
collection	Handlebars	parse	0.2548	1.0548	0.1859	1.2533	50
collection	LiquidJS	fresh	151.2657	164.8805	142.5834	167.6684	30
collection	LiquidJS	render	17.2048	29.6182	15.7468	33.4031	50
collection	LiquidJS	reuse	17.4451	31.5081	15.5456	35.7015	50
collection	LiquidJS	parse	0.0166	0.1273	0.0130	0.1569	50
many-renders-100	Nift	fresh	6.2441	6.7601	5.8210	8.4631	30
many-renders-100	Nift	source-reuse	0.0218	0.0282	0.0212	0.0435	50
many-renders-100	Jinja2	fresh	126.7340	141.3064	121.0456	142.5641	30
many-renders-100	Jinja2	render	0.4873	0.5361	0.4626	0.5616	50
many-renders-100	Jinja2	reuse	0.0103	0.0179	0.0100	0.0307	50
many-renders-100	Jinja2	parse	0.4551	0.5207	0.4252	0.6501	50
many-renders-100	Nunjucks	fresh	88.3816	94.8619	82.4448	95.2227	30
many-renders-100	Nunjucks	render	0.0262	0.1269	0.0240	0.2180	50
many-renders-100	Nunjucks	reuse	0.0006	0.0013	0.0004	0.0021	50
many-renders-100	Nunjucks	parse	0.0244	0.1310	0.0229	0.2267	50
many-renders-100	Handlebars	fresh	135.3628	147.9425	121.7949	149.6789	30
many-renders-100	Handlebars	render	0.1000	0.2742	0.0869	0.3533	50
many-renders-100	Handlebars	reuse	0.0018	0.0039	0.0014	0.0051	50
many-renders-100	Handlebars	parse	0.0736	0.2309	0.0660	0.2545	50
many-renders-100	LiquidJS	fresh	79.2000	89.7198	74.4932	110.1837	30
many-renders-100	LiquidJS	render	0.0161	0.0447	0.0148	0.1242	50
many-renders-100	LiquidJS	reuse	0.0077	0.0176	0.0070	0.0306	50
many-renders-100	LiquidJS	parse	0.0066	0.0173	0.0063	0.0403	50
many-renders-1000	Nift	fresh	26.7723	29.7733	25.7046	40.8920	30
many-renders-1000	Nift	source-reuse	0.0225	0.0315	0.0216	0.0513	50
many-renders-1000	Jinja2	fresh	556.2439	593.8707	532.9274	599.1695	30
many-renders-1000	Jinja2	render	0.4741	0.5020	0.4552	0.5056	50
many-renders-1000	Jinja2	reuse	0.0103	0.0150	0.0100	0.0170	50
many-renders-1000	Jinja2	parse	0.4430	0.4748	0.4203	0.5874	50
many-renders-1000	Nunjucks	fresh	264.5821	299.5823	253.6532	355.8594	30
many-renders-1000	Nunjucks	render	0.0249	0.0611	0.0228	0.1002	50
many-renders-1000	Nunjucks	reuse	0.0006	0.0008	0.0004	0.0010	50
many-renders-1000	Nunjucks	parse	0.0230	0.0395	0.0218	0.1021	50
many-renders-1000	Handlebars	fresh	318.1022	348.4469	304.2679	439.4891	30
many-renders-1000	Handlebars	render	0.0842	0.1111	0.0747	0.1863	50
many-renders-1000	Handlebars	reuse	0.0017	0.0032	0.0016	0.0047	50
many-renders-1000	Handlebars	parse	0.0693	0.0908	0.0651	0.1229	50
many-renders-1000	LiquidJS	fresh	172.7177	192.2219	164.9564	271.7213	30
many-renders-1000	LiquidJS	render	0.0151	0.0253	0.0130	0.0646	50
many-renders-1000	LiquidJS	reuse	0.0077	0.0123	0.0069	0.0153	50
many-renders-1000	LiquidJS	parse	0.0059	0.0089	0.0057	0.0141	50
many-templates-10	Nift	fresh	4.7371	5.1056	4.1490	5.1384	30
many-templates-10	Nift	source-reuse	0.0212	0.0432	0.0206	0.0491	50
many-templates-10	Jinja2	fresh	80.8272	86.5768	77.3650	100.4631	30
many-templates-10	Jinja2	render	0.4086	0.4769	0.3807	0.5802	50
many-templates-10	Jinja2	reuse	0.0095	0.0099	0.0093	0.0254	50
many-templates-10	Jinja2	parse	0.3805	0.4812	0.3558	0.8468	50
many-templates-10	Nunjucks	fresh	54.6383	61.5916	52.2397	61.7499	30
many-templates-10	Nunjucks	render	0.0308	0.1523	0.0203	0.1713	50
many-templates-10	Nunjucks	reuse	0.0005	0.0018	0.0004	0.0053	50
many-templates-10	Nunjucks	parse	0.0240	0.0940	0.0194	0.1895	50
many-templates-10	Handlebars	fresh	70.5466	80.0497	67.3158	82.0017	30
many-templates-10	Handlebars	render	0.1105	0.2987	0.0500	0.3945	50
many-templates-10	Handlebars	reuse	0.0017	0.0035	0.0013	0.0043	50
many-templates-10	Handlebars	parse	0.0772	0.2754	0.0522	0.3502	50
many-templates-10	LiquidJS	fresh	66.3754	70.3344	63.0563	70.9917	30
many-templates-10	LiquidJS	render	0.0108	0.0620	0.0089	0.0923	50
many-templates-10	LiquidJS	reuse	0.0050	0.0151	0.0041	0.0216	50
many-templates-10	LiquidJS	parse	0.0046	0.0163	0.0044	0.0207	50
many-templates-100	Nift	fresh	7.4420	8.0554	6.8669	8.1494	30
many-templates-100	Nift	source-reuse	0.0214	0.0235	0.0209	0.0372	50
many-templates-100	Jinja2	fresh	120.9770	125.9972	113.7080	136.9911	30
many-templates-100	Jinja2	render	0.3992	0.4329	0.3776	0.4507	50
many-templates-100	Jinja2	reuse	0.0097	0.0099	0.0094	0.0193	50
many-templates-100	Jinja2	parse	0.3653	0.4283	0.3439	0.4959	50
many-templates-100	Nunjucks	fresh	87.3222	91.3841	84.4883	93.9330	30
many-templates-100	Nunjucks	render	0.0238	0.1078	0.0213	0.1887	50
many-templates-100	Nunjucks	reuse	0.0007	0.0014	0.0006	0.0024	50
many-templates-100	Nunjucks	parse	0.0216	0.1214	0.0197	0.2175	50
many-templates-100	Handlebars	fresh	132.5488	142.7682	127.4625	143.9316	30
many-templates-100	Handlebars	render	0.0623	0.1967	0.0496	0.3331	50
many-templates-100	Handlebars	reuse	0.0020	0.0032	0.0015	0.0067	50
many-templates-100	Handlebars	parse	0.0579	0.1673	0.0472	0.2512	50
many-templates-100	LiquidJS	fresh	82.7361	90.8531	78.6093	93.3431	30
many-templates-100	LiquidJS	render	0.0102	0.0614	0.0089	0.0997	50
many-templates-100	LiquidJS	reuse	0.0048	0.0100	0.0043	0.0284	50
many-templates-100	LiquidJS	parse	0.0048	0.0207	0.0045	0.0260	50
many-templates-1000	Nift	fresh	35.8258	38.3633	34.0162	58.6698	30
many-templates-1000	Nift	source-reuse	0.0215	0.0251	0.0211	0.0369	50
many-templates-1000	Jinja2	fresh	498.1425	570.2601	482.9707	593.3529	30
many-templates-1000	Jinja2	render	0.3924	0.4624	0.3737	0.4874	50
many-templates-1000	Jinja2	reuse	0.0103	0.0166	0.0098	0.0174	50
many-templates-1000	Jinja2	parse	0.3596	0.3857	0.3447	0.3938	50
many-templates-1000	Nunjucks	fresh	341.0884	378.6479	328.4072	447.4057	30
many-templates-1000	Nunjucks	render	0.0234	0.0679	0.0206	0.1101	50
many-templates-1000	Nunjucks	reuse	0.0009	0.0017	0.0008	0.0021	50
many-templates-1000	Nunjucks	parse	0.0206	0.0476	0.0194	0.1181	50
many-templates-1000	Handlebars	fresh	371.0627	407.8127	355.6176	407.9452	30
many-templates-1000	Handlebars	render	0.0647	0.1368	0.0565	0.1548	50
many-templates-1000	Handlebars	reuse	0.0039	0.0088	0.0022	0.0100	50
many-templates-1000	Handlebars	parse	0.0499	0.0806	0.0473	0.1134	50
many-templates-1000	LiquidJS	fresh	163.7064	174.3214	151.3854	234.3372	30
many-templates-1000	LiquidJS	render	0.0105	0.0174	0.0093	0.0187	50
many-templates-1000	LiquidJS	reuse	0.0054	0.0106	0.0048	0.0120	50
many-templates-1000	LiquidJS	parse	0.0045	0.0071	0.0043	0.0108	50
transform	Nift	fresh	3.9872	4.7867	3.5669	5.1503	50
transform	Nift	source-reuse	0.0211	0.0362	0.0206	0.0468	50
transform	Jinja2	fresh	76.7184	86.9897	73.2419	94.5503	50
transform	Jinja2	render	0.4980	0.7247	0.4729	1.0988	50
transform	Jinja2	reuse	0.0100	0.0197	0.0096	0.0217	50
transform	Jinja2	parse	0.4812	1.0587	0.4577	1.4084	50
transform	Nunjucks	fresh	50.4048	57.1680	48.7282	58.6833	50
transform	Nunjucks	render	0.0357	0.2093	0.0254	0.2133	50
transform	Nunjucks	reuse	0.0006	0.0017	0.0005	0.0035	50
transform	Nunjucks	parse	0.0330	0.1825	0.0211	0.2738	50
transform	Handlebars	fresh	64.8234	73.4004	62.0512	79.8703	50
transform	Handlebars	render	0.1789	0.4224	0.0915	0.4362	50
transform	Handlebars	reuse	0.0022	0.0118	0.0016	0.0379	50
transform	Handlebars	parse	0.1398	0.4050	0.0939	0.5711	50
transform	LiquidJS	fresh	63.5418	68.2374	61.3649	79.6120	50
transform	LiquidJS	render	0.0370	0.1581	0.0270	0.2034	50
transform	LiquidJS	reuse	0.0090	0.0490	0.0046	0.0655	50
transform	LiquidJS	parse	0.0128	0.0719	0.0057	0.0922	50

11 / practitioner view

Would I actually use Nift as a template engine?

Qualitative practitioner assessment / not benchmark evidence

Human + agent: Nift for Nift-owned composition, Jinja2 for an existing Python application, Handlebars for logic-light JS prepared reuse. Agent-primary: the same direct conditional choices; named bindings, source identity and output contracts help automated maintenance, but they do not erase the large source-render gap or missing prepared API. Nift is an attractive text-template language in its own right. Its value expressions, named loop locals and explicit includes make realistic composition inspectable. These are language advantages, independent of the website generator around it.

The API matters as much as syntax.

Nift’s source-versus-path distinction and per-render Context separate data, source policy and execution cleanly. Jinja/Nunjucks remain familiar and expressive in their existing hosts. Handlebars and Liquid deliberately restrict logic; that can make a presentation layer easier to maintain, rather than weaker.

The implementation exposed concrete conventions: Jinja’s dot lookup collided with a dictionary key named items, and its default newline policy changed bytes. Handlebars compile is lazy, so prepared functions must be forced before a reuse timer. Exact fixtures caught these differences. Nift source caching also needed explicit filesystem treatment.

What would I want an agent to maintain?

Named bindings, predictable scopes, explicit composition and a declared output/escaping contract matter more than minimum character count. Nift’s richer functions can remove awkward helper plumbing, but business policy and side effects should not drift into templates accidentally. Formatted Nift, Jinja and Nunjucks are all good representations; constrained Handlebars/Liquid can be preferable when the project wants a smaller editing surface.

Current preferences at equal performance: requirements, not a universal ranking
Requirement	Preference	Reason
Small native text generation	Nift is a credible first choice	Direct embedded renderer; named scopes, expressive text syntax, explicit source/path APIs.
Existing Python application	Jinja2	Native Python values and concrete framework integration already in the application.
Expressive JS application views	Nunjucks	Native JS data and view composition without a native library deployment dependency.
Logic-light JS views with heavy reuse	Handlebars	Compiled functions and application-owned helpers match the desired presentation boundary.
Product standardized on Liquid syntax	LiquidJS	Its constrained filters/blocks are the required authoring contract.
Nift-owned or mixed-generation project	Nift	Shared expression/composition model plus tracked project ownership.
Agent-primary Nift-owned composition	Nift	Explicit scopes/includes and checked output contract within the existing project; standalone prepared workloads remain a host-engine choice.
Repeated fixed complex server templates today	A prepared-template engine	A concrete public preparation/reuse capability Nift does not yet expose.

Present-day versus mature Nift.

A public prepared-template artifact is the major standalone gap. I want explicit source identity, include invalidation, lifetime and concurrency contracts—not an undocumented benchmark fast path. If those become reliable and repeated-render performance is comparable, I see no architectural requirement to retain another renderer underneath Nift.

Existing Python/JS projects may still choose their host engine because of concrete value/framework/deployment requirements. Nift-owned projects already have a compelling single-model advantage. AI-DX authoring safety is qualitative: autonomous edit success, partial-error recovery and product-native structured feedback were not measured. Harness JSON is not a product feature. Context-sensitive HTML escaping, sandboxing, async filters, localization and streaming were not certified by this text-rendering campaign.

Full practitioner review ↗

12 / frozen evidence

Every number has a source.

One fresh g6-standard-4 shared-CPU Linode, us-east, Ubuntu 24.04. AMD EPYC 7642 host CPU, 4 shared vCPUs / 8 GiB RAM, Linux 6.8.0-134-generic, Python 3.12.3 and Node 22.22.1. All participants use one CPU affinity; order rotates within each round. Two separate warmup rounds, workload-sensitive fresh counts and 50 persistent samples. Batches are frozen from local feasibility. There is no machine-cold cache claim.

Nift source: c14d06835bca515a5281e41da0e40219c2d60ef0
Measured suite: 3c0dcdd1aad3cbb0a0c638c9041233b1e329dc6f
Canonical evidence: 19e390183b9bcb039abe6aa553ad88a2305b49ee
Experiment assessment/audit: 3c0dcdd1aad3cbb0a0c638c9041233b1e329dc6f

Raw observations, warmups, environment, hashes and audit ↗ · Adapter boundaries ↗ · Protocol ↗

Publication contains compact summary statistics and these displayed figures. Raw samples, corpora, traces and lifecycle receipts remain in the evidence repository. Existing reports and graph designs are preserved.

One template.New data.Exact output.