Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Network Comparison Strategy · yogsoth-ai | DecimalAI

yogsoth-ai/network-comparison

SkillScore54Maintained

Compare N methods simultaneously including indirect evidence — network meta-analysis protocol design. Budget: 50 studies, 80 effect sizes, 60 web searches.

researchby yogsoth-ai

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Research pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/yogsoth-ai-network-comparison/SKILL.md

decimalai skills pull yogsoth-ai-network-comparison --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/yogsoth-ai-network-comparison/SKILL.md https://app.decimal.ai/s/yogsoth-ai-network-comparison/SKILL.md

Raw file: /s/yogsoth-ai-network-comparison/SKILL.md ↗ · machine-readable summary at /s/yogsoth-ai-network-comparison.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

24 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 18d ago

Yes — +38% lift · at higher cost+53% tokens0% turns54% of whole cases pass with skill (13/24)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-07	✗→✓	▲ Improved	316%	0%
case-09	✗→✓	▲ Improved	99%	0%
case-12	✗→✓	▲ Improved	44%	0%
case-13	✗→✓	▲ Improved	160%	0%
case-14	✗→✓	▲ Improved	135%	0%

Showing 5 of 24 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/10/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

yogsoth-ai / de-anthropocentric-research-engine

skills/network-comparison/SKILL.md

View on GitHub ↗

License Apache-2.0Imported 8/10/2026Version v1

Skill contents

SKILL.md

Network Comparison Strategy

Design a network meta-analysis (NMA) protocol comparing N>=3 methods simultaneously, leveraging both direct and indirect evidence.

Purpose

When multiple methods have been compared in various head-to-head studies but no single study compares all methods, network meta-analysis synthesizes direct and indirect evidence to rank all methods simultaneously. This strategy produces the complete NMA protocol.

Budget

| Resource | Floor | Target | |----------|-------|--------| | Studies identified | 35 | 50 | | Effect sizes extracted | 55 | 80 | | Web searches | 40 | 60 | | Network nodes (methods) | 3 | N | | Quality assessments | 25 | 50 |

Budget gate: cannot exit until 80% of floor met.

State Ledger

<HARD-GATE>
| Metric | Current | Floor | Target | Status |
|--------|---------|-------|--------|--------|
| Studies found | 0 | 35 | 50 | BLOCKED |
| Effect sizes planned | 0 | 55 | 80 | BLOCKED |
| Web searches done | 0 | 40 | 60 | BLOCKED |
| Network nodes | 0 | 3 | N | BLOCKED |
| Quality assessed | 0 | 25 | 50 | BLOCKED |
</HARD-GATE>

Available Tactics

| Tactic | When to Use | |--------|-------------| | effect-size-extraction | Extract effect sizes from each study arm | | quality-assessment-protocol | Assess RoB + CINeMA framework for NMA | | evidence-synthesis-planning | Plan NMA model (consistency, ranking) |

Available SOPs

| SOP | When to Use | |-----|-------------| | pico-formulation | Frame the multi-method comparison | | inclusion-criteria-design | Define eligibility across all comparisons | | effect-size-planning | Standardize effect sizes across study designs | | data-extraction-form | Multi-arm extraction template | | risk-of-bias-assessment | Per-study + network-level bias | | evidence-network-construction | Build the network geometry graph | | heterogeneity-source-analysis | Assess transitivity assumption | | sensitivity-analysis-design | Node-splitting, exclusion sensitivity | | publication-bias-assessment | Comparison-adjusted funnel plots | | meta-analysis-synthesis | Final NMA protocol assembly |

Execution Guidance

Frame — Run pico-formulation for the multi-method question
Scope — Run inclusion-criteria-design covering all treatment nodes
Search — Systematic search for all pairwise comparisons
Network — Run evidence-network-construction to map the geometry
Extract — Use effect-size-extraction tactic per study arm
Assess — Use quality-assessment-protocol (RoB2 + CINeMA)
Transitivity — Run heterogeneity-source-analysis to verify transitivity
Plan — Use evidence-synthesis-planning for NMA model selection
Synthesize — Run meta-analysis-synthesis for final protocol

Iterate steps 3-6 until budget floor met. Verify network connectivity after each batch.

Output Format

yaml
protocol:
  question: [PICO with multiple interventions]
  network_geometry:
    nodes: [list of methods/interventions]
    edges: [direct comparisons available]
    connected: [yes/no]
  inclusion_criteria: [eligibility rules]
  studies_included: [list with arm-level metadata]
  effect_size_type: [standardized across network]
  model: [consistency/inconsistency NMA model]
  transitivity_assessment: [effect modifiers balanced?]
  ranking_method: [SUCRA/P-score/mean ranks]
  heterogeneity_plan: [global I2, tau2, between-design heterogeneity]
  inconsistency_plan: [node-splitting, design-by-treatment interaction]
  sensitivity_plan: [network reduction, node exclusion]
  bias_assessment_plan: [comparison-adjusted funnel, small-study effects]
  reporting: PRISMA-NMA extension

Available Tactics

Optional, no fixed order; the final leaf is always a sop.

| Tactic | When to use | | --- | --- | | effect-size-extraction | Systematically extract effect sizes and conditions from papers for meta-analytic synthesis | | evidence-synthesis-planning | Plan the statistical synthesis approach — model selection, heterogeneity strategy, and reporting | | quality-assessment-protocol | Methodological quality and bias risk assessment of included studies using validated tools |

Available SOPs

Optional, no fixed order; the final leaf is always a sop.

| SOP | When to use | | --- | --- | | data-extraction-form | Design structured data extraction form for systematic meta-analysis data collection | | effect-size-planning | Determine effect size types and calculation methods for meta-analytic synthesis | | evidence-network-construction | Build evidence network graph for network meta-analysis — nodes, edges, geometry assessment | | heterogeneity-source-analysis | Identify and classify sources of between-study heterogeneity (clinical, methodological, statistical) | | inclusion-criteria-design | Define inclusion/exclusion criteria for systematic study selection in meta-analysis | | meta-analysis-synthesis | Produce final meta-analysis protocol document assembling all planning outputs into PRISMA-compliant protocol | | pico-formulation | Construct PICO/PECO framework for the meta-analysis research question | | publication-bias-assessment | Plan funnel plots, Egger's test, trim-and-fill, p-curve, and selection model analyses for publication bias | | risk-of-bias-assessment | Assess methodological bias using RoB2, PROBAST, or QUADAS-2 validated tools | | sensitivity-analysis-design | Design leave-one-out, influence diagnostics, subgroup analyses, and robustness checks |

Illustrative Want to see what yogsoth-ai/network-comparison would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

9/3/2026, 08:14 UTC · 24 cases

Lift

+38%

54% vs 17%

Agent turns

1.0 vs 1.0

Tokens

+53%

8,110.42 vs 5,300.75

fail → pass · 9pass → pass · 4fail → fail · 11

Per-case results (24)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-07 I have a single randomized trial paper comparing Drug A vs Placebo in heart failure. I need you to assess this specific paper using the Cochrane Risk of Bias tool (RoB 2) across its five domains: randomization, deviations from intervention, missing outcome data, outcome measurement, and selection of reported result.	fail→pass	30,607	142,726	+366%	1	1	0%	1,553	6,460	+316%	0	0	—
case-08 We are preparing a network meta-analysis protocol comparing six non-pharmacological insomnia interventions (CBT-I, sleep restriction, stimulus control, relaxation therapy, mindfulness, and sleep hygiene education). We want to output a YAML protocol file. Some team members suggested structuring the top-level keys as meta_analysis_plan: or study_outline:. Please generate the complete structured YAML protocol for this study.	fail→fail	64,979	44,560	-31%	1	1	0%	8,270	9,667	+17%	0	0	—
case-01 I am preparing a research study to evaluate four distinct frontline hypertension medications (amlodipine, lisinopril, losartan, and hydrochlorothiazide). While several head-to-head clinical trials exist between pairs of these drugs, no single study has evaluated all four together. Please draft a complete network meta-analysis protocol in structured YAML format. The protocol needs to clearly delineate the PICO framework, define the evidence network geometry (nodes and direct connections), specify criteria for study inclusion and effect size standardization, outline how transitivity and consistency across direct/indirect links will be assessed, detail the statistical modeling and ranking methods (such as SUCRA), and describe plans for heterogeneity and publication bias evaluation.	fail→fail	32,841	46,497	+42%	1	1	0%	5,231	9,728	+86%	0	0	—
case-09 Our psychiatric research team is synthesizing evidence across seven psychotherapy modalities for major depression. Modelers often default to ranking interventions solely by unweighted point estimates or raw p-values from pairwise comparisons. Please produce a structured YAML network meta-analysis protocol for ranking these seven modalities.	fail→pass	27,935	67,297	+141%	1	1	0%	4,587	9,116	+99%	0	0	—
case-02 We are conducting a systematic benchmark synthesis comparing five optimizer algorithms (AdamW, SGD, Lion, Sophia, and Shampoo) used in neural network training across disparate machine learning literature. We need a formal network meta-analysis protocol formatted as a YAML document. Please layout the multi-intervention research question, the graph structure of treatment nodes and pairwise edges, the criteria for study eligibility, how effect sizes will be extracted across different benchmark tasks, the method for evaluating transitivity assumptions, the consistency modeling approach, sensitivity checks, and the reporting guideline framework.	fail→fail	48,548	38,487	-21%	1	1	0%	8,012	7,936	-1%	0	0	—
case-03 Our research team wants to conduct a multi-treatment evidence synthesis comparing three primary reading intervention programs implemented in primary schools. Because existing studies only perform pairwise comparisons, we need a network meta-analysis protocol. Produce a YAML-formatted output that details our study question using a multi-arm PICO, maps the network geometry and connectivity, outlines study inclusion rules, specifies quality assessment approaches, and lays out analytical plans for statistical consistency, node ranking, sensitivity analyses, and small-study bias detection.	fail→fail	59,390	57,670	-3%	1	1	0%	7,732	9,684	+25%	0	0	—
case-04 I am writing a protocol for a traditional pairwise meta-analysis comparing Cognitive Behavioral Therapy (CBT) directly against Selective Serotonin Reuptake Inhibitors (SSRIs) for major depressive disorder. We are only interested in direct head-to-head trials between these two interventions. Please generate a pairwise systematic review and meta-analysis protocol outline.	pass→pass	30,925	46,759	+51%	1	1	0%	4,382	9,656	+120%	0	0	—
case-05 Our team is designing a new primary multi-center 4-arm randomized controlled trial to test four novel immunotherapies in patients with stage III melanoma. We need an IRB-ready clinical trial protocol covering patient recruitment, randomization procedures, dosage administration schedules, primary safety endpoints, and informed consent protocols.	pass→pass	65,238	47,221	-28%	1	1	0%	8,249	9,646	+17%	0	0	—
case-06 I want to run an empirical benchmark experiment comparing AdamW, SGD, and Lion on a ResNet-50 image classification task. Please write a Python script using PyTorch that trains ResNet-50 with each optimizer, logs training loss and top-1 accuracy per epoch, and plots the learning curves.	pass→pass	43,051	55,154	+28%	1	1	0%	6,994	9,026	+29%	0	0	—
case-10 We are developing an evidence synthesis protocol comparing eight biologic therapies for plaque psoriasis. For quality and risk of bias assessment, reviewers typically rely exclusively on standard single-study RoB 2 assessments. Please generate the network meta-analysis protocol in structured YAML, ensuring the risk of bias and quality framework accounts for network-level confidence across indirect comparisons.	pass→pass	39,099	42,420	+8%	1	1	0%	7,162	8,611	+20%	0	0	—
case-11 We are drafting an NMA protocol comparing five classes of topical glaucoma drops (prostaglandin analogs, beta-blockers, alpha-agonists, carbonic anhydrase inhibitors, and rho kinase inhibitors). To evaluate whether direct and indirect evidence agree, analysts often suggest simple meta-regression on study year. Generate the complete YAML network meta-analysis protocol specifying appropriate statistical techniques for assessing network inconsistency.	fail→fail	24,457	44,313	+81%	1	1	0%	4,290	8,381	+95%	0	0	—
case-12 Our endocrinology group is designing a network meta-analysis comparing six classes of type 2 diabetes medications on HbA1c reduction. Traditional systematic reviews use standard symmetric funnel plots, but multi-treatment networks require specialized methods because interventions are not compared against a single placebo baseline. Draft the NMA protocol in YAML format detailing publication bias detection.	fail→pass	32,244	33,802	+5%	1	1	0%	4,820	6,931	+44%	0	0	—
case-13 We are synthesizing trial data for four endovascular thrombectomy devices for acute ischemic stroke. Some authors publish using standard PRISMA 2020 guidelines intended for pairwise reviews. Draft the YAML protocol for this network meta-analysis, specifying the exact reporting guideline extension required for multi-treatment network evidence.	fail→pass	21,777	36,331	+67%	1	1	0%	3,163	8,228	+160%	0	0	—
case-14 Our team is comparing five disease-modifying anti-rheumatic drugs (DMARDs) in rheumatoid arthritis. When planning indirect comparisons, researchers sometimes assume transitivity holds automatically as long as all trials include adult patients. Generate the NMA protocol in YAML format detailing how transitivity must be formally assessed.	fail→pass	29,470	30,856	+5%	1	1	0%	3,028	7,111	+135%	0	0	—
case-15 We are designing a network evidence synthesis for seven second-generation anticonvulsant drugs for focal epilepsy. Data extraction templates often extract study-level aggregate outcome differences between intervention and control. Generate the YAML NMA protocol specifying the required structure for multi-arm trial data extraction.	fail→pass	35,257	26,723	-24%	1	1	0%	5,128	5,524	+8%	0	0	—
case-16 We are outlining an NMA protocol evaluating five biologic agents for severe asthma. Standard sensitivity analysis plans only propose excluding high risk of bias studies. Draft a YAML network meta-analysis protocol that includes network-specific structural robustness and sensitivity checks.	fail→fail	29,637	40,387	+36%	1	1	0%	4,356	6,686	+53%	0	0	—
case-17 Our neurology department is synthesizing direct and indirect evidence for six oral and injectable preventive migraine medications. Researchers often describe network structure only as text narrative in the introduction. Produce a structured YAML network meta-analysis protocol defining the evidence network's graph properties.	fail→fail	23,182	45,257	+95%	1	1	0%	4,370	9,635	+120%	0	0	—
case-18 We are preparing a network synthesis comparing four heart failure medication classes (ARNI, ACEi/ARB, beta-blockers, and SGLT2i). Biostatisticians frequently start with independent pairwise random-effects models. Draft the NMA protocol in YAML format specifying the statistical model selection approach for network-wide synthesis.	fail→fail	25,960	34,288	+32%	1	1	0%	3,994	9,653	+142%	0	0	—
case-19 We are building a multi-treatment comparison protocol for four surgical mesh techniques in ventral hernia repair. A common mistake is planning study-level binary outcome tables without arm-specific tracking. Produce the YAML NMA protocol specifying how included studies and arm-level details will be structured.	fail→fail	37,217	36,314	-2%	1	1	0%	6,673	7,815	+17%	0	0	—
case-20 Our anesthesiology team is writing a protocol for a network meta-analysis evaluating five antiemetic drug combinations for preventing postoperative nausea. Standard PICO definitions only allow a single intervention vs single comparator. Provide the NMA protocol in YAML format with multi-intervention PICO framing.	fail→fail	35,315	44,396	+26%	1	1	0%	5,589	9,105	+63%	0	0	—
case-21 We are setting up an operational evidence synthesis workflow for eight systemic treatments for atopic dermatitis. Before exiting the preliminary search and extraction phase, team leaders want to establish hard quality and volume thresholds. Draft the YAML protocol for this multi-treatment synthesis including study and extraction volume requirements.	fail→fail	20,436	21,052	+3%	1	1	0%	3,759	5,297	+41%	0	0	—
case-22 Our orthopedics group is conducting a network meta-analysis of five intra-articular injection therapies for knee osteoarthritis (hyaluronic acid, PRP, corticosteroids, stem cells, and saline). When mapping the network, researchers sometimes aggregate all biological injections into a single category. Draft the NMA protocol in YAML mapping each distinct therapy as an independent node.	fail→pass	19,731	23,449	+19%	1	1	0%	4,031	5,710	+42%	0	0	—
case-23 We are planning an NMA across four chemotherapy regimens for metastatic colorectal cancer incorporating both hazard ratios and odds ratios from various trial publications. Draft the YAML protocol detailing how effect sizes will be standardized across different study designs and reporting formats.	fail→pass	33,759	21,838	-35%	1	1	0%	5,879	5,712	-3%	0	0	—
case-24 We are organizing a network meta-analysis of four disease-modifying therapies for early Alzheimer's disease. Analysts frequently report only total Cochrane Q without separating between-study and between-design variance. Generate the NMA protocol in YAML format detailing heterogeneity assessment.	fail→pass	33,681	51,978	+54%	1	1	0%	5,966	9,332	+56%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 24 cases were attempted. The headline lift of +38 percentage points is the difference between those two pass rates over the 24 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 16.7%
Pass rate with the skill: 54.2%
Cases attempted: 24
Cases counted toward lift: 24
Cases that regressed: 0
Skill version measured: v1
Run completed: 9/3/2026, 08:14 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-07

Prompt

I have a single randomized trial paper comparing Drug A vs Placebo in heart failure. I need you to assess this specific paper using the Cochrane Risk of Bias tool (RoB 2) across its five domains: randomization, deviations from intervention, missing outcome data, outcome measurement, and selection of reported result.

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run54

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+38% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 17 days ago

Other measured skills in the registry, with their headline benchmark lift.

Ready to supercharge your AI agents?

yogsoth-ai/network-comparison

SkillSafety

Source

Network Comparison Strategy

Purpose

Budget

State Ledger

Available Tactics

Available SOPs

Execution Guidance

Output Format

Available Tactics

Available SOPs

How this skill was measured

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-07

Related skills