Source Research Lab / Applied Research Layer Qualification-Gated

Source Research Lab
Applied Systems Research

A selective applied research environment for capable AI systems researchers, benchmark operators, quant researchers, research assistants, technical builders, and advanced operators who can help test, document, benchmark, and improve the systems Source is building.

△ Apply for Research Review View Research Lanes →
02 Research Lanes
Benchmark Operators
AI Systems Research
Trading Systems Research
No Guarantees
Research Lab Definition

What Source
Research Lab Is.

Source Research Lab is the applied research and benchmark layer behind Source’s AI systems, Source University, market systems research, technical infrastructure, and operator workflows. Its purpose is to test whether the systems actually work: whether AI tutors improve learning, whether agent harnesses improve execution, whether RAG systems retrieve accurate source material, whether dashboards expose useful state, whether premium data workflows improve research quality, whether backtests are honest, and whether telemetry reveals failures early.

01

AI Systems Evaluation

01 / Infrastructure

Test AI tutors, agents, RAG, model workflows, local inference, hallucination mitigation, and dashboards.

02

Trading Systems Research

02 / Quantitative

Test data feeds, backtesting, simulations, broker/exchange maps, risk dashboards, and telemetry.

03

Benchmark Operations

03 / Measurement

Turn vague capability into measurable reports, scorecards, audits, and reproducible experiments.

04

Infrastructure Lab

04 / Sandbox

Work with controlled access to custom harnesses, local inference, servers, dashboards, and configurations.

05

Proof-of-Work

05 / Outputs

Produce structured reports, research memos, QA artifacts, benchmark datasets, and dashboards.

06

Gated Access

06 / Security

Access is scoped, project-based, monitored, permissioned, and strictly tied to useful research outputs.

Training Layer vs Research Layer

Not The Training Track.
The Research Layer Behind It.

Understand the distinction: Source University is built for capability development, while the Research Lab focuses on rigorous system evaluation, benchmarking, and failure testing.

SOURCE UNIVERSITY CAPABILITY

Source University

Capability Development Layer

Focuses on teaching, training, and building the operational competence of candidates within defined tracks.

  • Trains candidates to run active systems
  • Teaches AI infrastructure and market mechanisms
  • Focuses on learning and execution capability
  • Uses assignments and founder-led mentoring
SOURCE RESEARCH LAB EVALUATION

Source Research Lab

Evaluation & Benchmark Layer

Focuses on testing, auditing, validating, and documenting the boundaries, limits, and failures of active systems.

  • Evaluates systems under stress and live conditions
  • Runs rigorous benchmark experiments and audits
  • Identifies failure modes, hallucinations, and risks
  • Outputs structured research memos and metrics

Source University asks: Can this person learn the system?  •  Source Research Lab asks: Can this person help prove whether the system works?

Research Lanes

Two Applied Research
Lanes.

The Source Research Lab parent page routes qualified candidates into two major research lanes. Each lane has its own deeper page, research programs, infrastructure layer, application questions, and proof-of-work expectations.

LANE 01 // SYSTEM TESTING AI SYSTEMS

AI Systems Researcher /
Benchmark Operator

Evaluate Source’s AI tutors, agent harnesses, RAG corpora, model workflows, dashboards, local inference labs, and proprietary AI systems.

AI Tutors Agent Harnesses RAG / Retrieval Local Inference Model Routing Hallucination Audits BuildGraph

Built for AI systems researchers, research assistants, technical builders, LLMOps/MLOps operators, agent testers, and benchmark operators.

Explore AI Systems Research Lane  →
LANE 02 // QUANT TESTING MARKET SYSTEMS

Trading Systems Researcher /
Benchmark Operator

Evaluate premium data workflows, market data pipelines, backtesting integrity, simulations, broker/exchange maps, AI market research agents, and risk dashboards.

Premium Data Data Pipelines Backtests Simulations Broker APIs Risk Dashboards Telemetry

Built for quant researchers, market systems researchers, data scientists, AI-assisted research builders, risk analysts, and benchmark operators.

Explore Trading Systems Research  →

*Note: These pages represent research lanes and will be deployed as separate deep-dive pages after parent qualification.

Research Program Map

Programs Built For
Systematic Measurement.

Source Research Lab organizes work into research programs. Below is a preview of the program map across the two lanes. Detailed dashboards are located inside each lane page.

AI Systems Research Programs

PROGRAM 01 AI Tutor / Learning System Evaluation

Tests whether AI tutors, founder guidance, structured assignments, and review loops improve real operator capability.

PROGRAM 02 Agent Harness Reliability Benchmarking

Tests whether custom Pi harnesses, Codex workflows, Claude Code, Hermes multi-terminal setups, and OpenClaw automation improve pipeline reliability.

PROGRAM 03 RAG / Retrieval Quality Evaluation

Tests retrieval precision, recall, citation faithfulness, hallucination behavior, chunking, embeddings, and rerankers.

PROGRAM 04 Vertical Intelligence Corpus Validation

Tests whether vertical corpora power useful specialist AI consultants, diagnostics, and assessments.

PROGRAM 05 AI Workflow QA and Human Control

Tests QA gates, validation scripts, human review queues, and failure taxonomies.

PROGRAM 06 Console / Dashboard Utility Testing

Tests whether consoles expose state, queue bottlenecks, costs, and next actions clearly.

PROGRAM 07 Acquisition System Benchmarking

Tests lead lifecycle systems, approval gates, reply classification, and source attribution.

PROGRAM 08 BuildGraph Multi-Agent Runtime Research

Tests branch variants, layer-locking, multi-agent critique loops, and convergence telemetry.

PROGRAM 09 Model Routing Cost-vs-Quality Evaluation

Tests which models handle which tasks, comparing cost, latency, quality, and routing policies.

PROGRAM 10 Documentation & Context Engineering

Tests whether SOPs, mistake logs, and decision logs improve AI/human execution consistency.

Trading Systems Research Programs

PROGRAM 01 Data Feed and Pipeline Validation

Tests Bloomberg-style, LSEG-style, alternative, and alternative on-chain data workflows.

PROGRAM 02 Backtesting Integrity and Bias Detection

Tests lookahead bias, survivorship bias, leakage, overfitting, slippage, and false robustness.

PROGRAM 03 Strategy Research Workflow Benchmarking

Tests how hypotheses become documented research, backtests, simulations, and review loops.

PROGRAM 04 AI-Assisted Quant Research Evaluation

Tests whether AI agents can summarize markets, review strategies, and detect assumptions.

PROGRAM 05 Risk Dashboard and Control System Design

Tests drawdowns, limits, leverage exposure, kill switches, and operator visibility controls.

PROGRAM 06 Paper-Trading / Simulation Environment Research

Tests broker disconnects, order rejections, partial fills, slippage, and readiness thresholds.

PROGRAM 07 Broker / Exchange Connectivity Mapping

Maps broker APIs, exchange APIs, FIX endpoints, order states, fills, and reconciliation.

PROGRAM 08 Trading Performance & Regime Analysis

Tests Sharpe, Sortino, drawdown expectancy, win rate, slippage, and strategy degradation.

PROGRAM 09 Trading Psychology & Operator Discipline

Tests journaling, rule adherence, override logs, review loops, and behavior drift metrics.

PROGRAM 10 Trading Telemetry & Observability

Tests logs, metrics, traces, data-feed health dashboards, and incident reviews.

Infrastructure Layer

Controlled Access To
Serious Research Infrastructure.

The appeal of Source Research Lab is the environment: controlled, qualification-based access to serious tools, custom systems, compute resources, and data workflows.

  • Proprietary Source configurations and confidential internal research tooling.
  • Internally developed harness configurations and private AI agent orchestration systems.
  • Custom Pi harness workflows and custom model-routing workflows.
  • Internal benchmark environments, private evaluation rubrics, and internal vertical intelligence corpora.
  • Internal Source University learning-system datasets operating-system artifacts.
  • Confidential prompt and context-engineering systems, and proprietary workflow compositions.

Selected researchers may work with proprietary Source configurations, confidential research workflows, custom harness environments, internal evaluation templates, private corpora, and experimental systems developed by the Source research team. Access is qualification-based and scoped by project.

  • Bloomberg Terminal access, Bloomberg B-PIPE, and Bloomberg Market Data Feeds.
  • Bloomberg Enterprise Data workflows and LSEG Workspace (Refinitiv) platforms.
  • LSEG Real-Time, LSEG Real-Time Direct, and LSEG Ultra Direct feeds.
  • FactSet Workstation, FactSet Real-Time Data Suite, and S&P Capital IQ.
  • S&P Global Market Intelligence, ICE Data Services, Morningstar Direct, MSCI, Barra, Axioma, PitchBook, and Preqin.

Access to premium data, terminals, paid datasets, or institutional systems is not guaranteed and may depend on project scope, licensing, availability, qualification, and approval.

  • Databento high-performance market data APIs.
  • Polygon.io and dxFeed real-time tick and historical APIs.
  • Intrinio, Tiingo, IEX Cloud, Financial Modeling Prep, and Alpha Vantage.
  • Twelve Data, Finnhub, EOD Historical Data, and QuantConnect data libraries.
  • AlgoSeek-style historical tick datasets, broker-provided market data APIs, and exchange WebSocket feeds.
  • SEC / EDGAR filings and earnings call transcripts parsing workflows.
  • Analyst estimates and economic calendar feeds.
  • FRED macroeconomic data, central bank data, interest rate, and inflation datasets.
  • News sentiment feeds, RavenPack-style news analytics, and social sentiment data.
  • App usage proxies, web traffic datasets, job posting datasets, and supply-chain/shipping data.
  • Weather datasets for commodities, prediction market data (Polymarket), and web-scraped structured datasets.
  • Kaiko, Coin Metrics, Glassnode, CryptoCompare, and Amberdata pipelines.
  • CoinGecko and CoinMarketCap developer APIs.
  • Binance, Coinbase, Kraken, OKX, and Bybit historical and WebSocket APIs.
  • On-chain wallet/entity datasets, DEX liquidity data, and DeFi protocol data.
  • Stablecoin flow metrics, perpetual futures order-book data, and exchange reserves.
  • Blockchain explorer APIs, token holder concentration data, and smart-contract event logs.
  • Dedicated private virtual servers (VPS) and isolated research VMs.
  • Windows research desktops and Linux servers with high-core CPU and high-RAM.
  • GPU-enabled workstations and multi-node VM compute clusters.
  • Distributed parallel processing engines (Ray, Dask, Spark) and job environments.
  • Isolated sandboxes for running reproducible experiments.
  • Local LLM inference servers and private model endpoints.
  • Ollama-style local inference and vLLM-style serving layers.
  • Hugging Face TGI-style serving and llama.cpp environments.
  • Quantized GGUF and TensorRT-LLM model testing setups.
  • ONNX Runtime, Triton Inference Server, and Hugging Face Transformers.
  • Local embedding models, rerankers, and privacy-sensitive RAG model deployments.

Researchers may work with local inference labs and private model deployments rather than only using consumer chat interfaces.

  • OpenAI (ChatGPT/Codex), Claude (Claude Code), Gemini, and DeepSeek workflows.
  • Qwen, Kimi, and MiniMax specialized model routing.
  • Local open-source models and Hugging Face model hub workflows.
  • Autonomous coding agents, research agents, QA agents, and document extraction agents.
  • Reasoning models, multimodal setups, embedding models, and judge models.

Selected researchers may work with frontier and open-source model workflows, custom routing setups, local inference systems, coding agents, and evaluation protocols. Access is controlled, project-scoped, and subject to availability, cost, licensing, and approval.

  • Custom fine-tuned LLMs, LoRA adapters, and QLoRA configurations.
  • Domain-specialized assistants for vertical business diagnostics and market research.
  • Source University tutor models and retrieval-grounded assistant agents.
  • Synthetic training data generation and supervised fine-tuning (SFT) datasets.
  • Model distillation experiments, preference-tuning, and model evaluation cards.

Researchers may help evaluate whether custom-tuned, retrieval-grounded, or domain-specialized models outperform generic frontier models on Source-specific tasks.

  • Hallucination-associated neuron detection and suppression/amplification vector experiments.
  • Model internals analysis, activation steering, and sparse autoencoder interpretability.
  • Internal-state probing and compliance/refusal behavior testing.
  • Factuality/truthfulness benchmarks, hallucination audits, and groundedness scorecards.
  • Mechanistic interpretability pipelines, steering vector analysis, and citation faithfulness testing.

Source is interested in neuron-level hallucination mitigation research, including hallucination-associated neuron detection, activation steering, local model intervention experiments, groundedness evaluation, and reliability benchmarking.

  • Custom Pi harnesses and Oh-My-Pi style coding harness workflows.
  • Pi.dev agent orchestration layers and Claude Code desktop setups.
  • Codex task execution, Hermes Agent-style workflows, and OpenClaw automation.
  • DeepSeek TUI and coding agent execution logs.
  • Task, run, and artifact ledgers with automated validation gates and BuildGraph telemetry.

Researchers may work inside custom Pi-style harness environments where AI agents are assigned, observed, logged, benchmarked, corrected, and compared across controlled execution runs.

  • Model Context Protocol (MCP) server endpoints and WebSocket bridges.
  • Custom tool registries, tool-calling workflows, and function-calling schemas.
  • API-connected agents, local tools, cloud integrations, and browser automation.
  • Permission-gated tool access, tool audit logs, and tool safety boundary testing.
  • Human-in-the-loop approval systems and event verification.
  • Frontier model comparison matrix data and model routing policies.
  • LLM-as-judge workflows, multi-model consensus, and disagreement telemetry.
  • Model debate setups, cost-vs-quality curves, and latency-vs-quality trade-offs.
  • Long-context vs. RAG retrieval evaluations.
  • Model fallback chains, committees, and ensemble configurations.
  • Source University curriculum, Research Lab, and Lead Engine corpora.
  • Ingested vertical intelligence datasets, transcript databases, and SOP libraries.
  • LangChain, LangGraph, LlamaIndex, pgvector, Qdrant, Pinecone, and Weaviate.
  • BM25, semantic search, vector indexing, rerankers, and chunking experiments.
  • Knowledge graph concepts, source citation retrieval, and contextual compression.
  • Experiment tracking setups, prompt versioning, and dataset registries.
  • MLflow tracking servers, LangSmith trace endpoints, and Weights & Biases integrations.
  • LLM-as-judge grading runs, pairwise model comparisons, and golden-answer sets.
  • Regression test suites, prompt regression tests, and synthetic evaluation data.
  • Evaluation reports, cost-quality dashboards, and evaluator agreement scoring.
  • Grafana dashboards, Loki logging, Prometheus monitoring, and OpenTelemetry integrations.
  • Tempo distributed trace visualizers.
  • Model-call metrics, token cost logs, latency stats, and agent run traces.
  • API failure logs, server health stats, risk dashboards, and queue monitors.
  • Broker connectivity and data feed latency displays.
  • Market data pipeline sandboxes, tick data validation, and OHLCV checking.
  • Order book replay simulations and paper trading environments.
  • Event-driven and vectorized backtesting systems.
  • Slippage modeling, transaction fee calculations, and execution latency buffers.
  • Broker/exchange API simulations and order lifecycle state machines.
  • Risk dashboards, drawdown limits, kill switches, and max exposure gates.

Market systems research is simulation-first and paper-trading-first. This is not investment advice, a signals service, a profit guarantee, a funding guarantee, or a live-trading promise.

  • Benchmark reports, technical research memos, and experiment ledger tables.
  • Failure-mode registries, system diagram files, and model comparison matrices.
  • Reproducibility test packages, evaluation rubrics, data cards, and model cards.
  • Strategy research briefs, case studies, and approved public proof-of-work pages.
  • Bloomberg/LSEG premium feeds connected to local inference servers.
  • Custom tuned models integrated with RAG vector search parameters.
  • Neuron-level hallucination mitigation vectors aligned with retrieval verification mechanisms.
  • Codex and Claude Code loops monitored by Pi-harness validation engines.
  • Simulated paper-trading brokers linked to risk dashboards and telemetry logs.

The most valuable research environment is not any single tool. It is the composition of proprietary Source configurations, premium data, local inference, frontier models, custom harnesses, RAG corpora, telemetry, dashboards, and human review into controlled experimental systems.

“The most valuable research environment is not any single model, terminal, tool, server, or dashboard. It is the composition of proprietary Source configurations, local inference, frontier models, custom harnesses, premium data workflows, RAG corpora, telemetry, dashboards, benchmark protocols, and human review into controlled experimental systems.”

Proprietary Assets

Proprietary And
Confidential Source Systems.

Not all Source Research Lab infrastructure is publicly described. Certain technologies, configurations, internal benchmarks, private datasets, custom harnesses, model workflows, dashboard concepts, research templates, and experimental systems are intentionally held back from the public page. Selected researchers may be introduced to additional systems only after qualification, project assignment, confidentiality review, and access approval.

01

Custom Harness Systems

01 / Execution
  • • Custom Pi harness workflows
  • • Codex / Claude Code execution
  • • Hermes-style orchestration
  • • OpenClaw-style automation
  • • Agent run ledgers & gates
02

Private Corpora

02 / Intelligence
  • • Source University corpora
  • • Source Research Lab corpora
  • • Vertical business corpora
  • • Market & trading research
03

Model Workflows

03 / Routing
  • • Custom model-routing setups
  • • Local inference experiments
  • • Custom fine-tuned LLMs
  • • Domain-specialized agents
  • • Hallucination mitigation
04

Research Dashboards

04 / Telemetry
  • • Benchmark & cost dashboards
  • • Retrieval quality analytics
  • • Agent reliability charts
  • • Market telemetry displays
  • • Drawdown & risk boards
05

Research Protocols

05 / Rubrics
  • • Evaluation scoring rubrics
  • • Failure-mode taxonomies
  • • Experiment logging rules
  • • Access & approval boundaries
  • • Proof-of-work guidelines
Source may provide selected researchers with controlled access to proprietary configurations, confidential research workflows, internally developed harness systems, private corpora, custom model-routing setups, experimental dashboards, benchmark protocols, and Source-built evaluation environments. Some systems may be disclosed only after qualification, project assignment, and confidentiality review.
Research Work

What Selected Researchers
May Actually Do.

This is not passive reading. Selected researchers are assigned targeted testing tasks designed to validate the reliability, cost, and efficiency of Source's stack.

01

Run Experiments

Compare models, custom tool setups, workflows, data feeds, and RAG architectures under identical benchmark criteria.

02

Audit Failure Modes

Document hallucinations, retrieval failures, broken assumptions, data feed gaps, and strategy drawdowns.

03

Produce Reports

Write structured research memos, scorecard evaluations, and audit logs detailing system limits.

04

Test Dashboards

Assess whether dashboards communicate real-time system state, exposure levels, and failure alerts clearly.

05

Evaluate Tutors

Monitor tutor explanation metrics, agent automation pass rates, and human verification queues.

06

Validate Market Tech

Map broker WebSocket streams, test exchange order latency, and verify data-feed alignment.

07

Improve Docs

Identify gaps in SOPs, mistake logs, and runbooks, organizing them into retrieval-ready formats.

08

Build Proof-of-Work

Compile verified code sandboxes, benchmark datasets, and visual systems maps.

Output Standard

Research Output
Comes First.

We measure value strictly by output. Researchers demonstrate progress by delivering clear, reproducible reports and data scorecards that improve the Source codebase.

AI Tutor Scorecards

Evaluation matrices comparing AI mentor explanation accuracy.

Agent Run Memos

Logs detailing trace paths, API cost thresholds, and loop failures.

RAG Quality Audits

Metrics testing chunking methods, vector lookups, and recall gaps.

Data Feed Audits

Verifications of pipeline logs against authoritative market datasets.

Backtest Audits

Assessments targeting survivorship bias, leakage, and slippage fees.

Every formal research memo submitted to the lab is structured to include:

1. Research Question
8. Assumptions Tested
15. Confidence Score
2. Target Hypothesis
9. Comparative Results
16. Recommendations
3. Systems Tested
10. Edge Case Failures
17. Future Experiments
4. Tools & Hardware Used
11. System Limitations
18. Internal Review Notes
5. Data Corpus Source
12. Risk Factors
19. Access / License Details
6. Methodology Steps
13. Interactive Logs / Images
7. Benchmark Metrics
14. Reproducibility Guide
Access Model

Access Is
Qualification-Gated.

Not every researcher receives access to every system. Source Research Lab access is qualification-based, project-scoped, permissioned, staged, revocable, and confidentiality-bound.

01

Public Review

Candidate reviews research lanes, program summaries, and boundaries.

02

Application

Candidate submits background details, desired lane, and proof-of-work.

03

Screening

Source evaluates technical fit, logical clarity, and confidentiality readiness.

04

Program Match

Candidate matched to specific, scoped RAG, model, or database projects.

05

Controlled Access

Researcher receives scoped access keys for target compute tools.

06

Report Output

Researcher submits structured memos, code runs, and scorecards.

07

Review

Strong candidates evaluated for paid collaboration or deeper tasks.

Access Bounds Rationale: Selected researchers may receive controlled, qualification-based access to Source research environments, configurations, custom harnesses, private corpora, model workflows, and dashboards depending on project fit, approval, licensing, confidentiality, and availability.

Certain technologies, internal benchmarks, and experimental setups are held back from public visibility and may only be discussed with qualified candidates after review. Access to compute, terminals, paid APIs, local inference setups, or frontier models is not guaranteed.

Candidate Fit

Who This Is For.
Seriousness Parameters.

Source Research Lab requires rigorous research discipline and structured reports. Read the requirements below to verify if your background matches our fit parameters.

FIT CRITERIA STRONG FIT

Strong Fit Indicators

Candidates who demonstrate technical curiosity, systematic documentation habits, and a willingness to verify assumptions.

  • Detail-oriented, data-aware, and heavy on written documentation
  • Capable of drafting structured, logical research reports and memos
  • Comfortable with staged database access and protecting client confidentiality
  • Interested in local model fine-tuning, RAG parameters, and quant simulations
NON-FIT INDICATORS WEAK FIT

Weak Fit Indicators

We screen out candidates looking for quick solutions, passive tutorials, or signals trading systems.

  • Wanting compute access without submitting detailed research outputs
  • Looking for guaranteed job placement, salaries, or trading signals
  • Unwilling to document experiments or write reproducible code sandboxes
  • Interested only in surface-level prompting or get-rich strategies
Compliance Boundaries

What This Is Not.
Strict Non-Guarantees.

Source Research Lab maintains rigorous compliance gates to ensure data security, legal safety, and realistic outcome targets.

Boundary 01

No Guaranteed Jobs

This is an applied research layer. Participating in benchmark testing or submitting memos does not guarantee employment, paid fellowships, or contract engagements.

Boundary 02

No Open compute

Compute resources, alternative data feeds, and frontier API access are project-scoped and qualification-gated. We do not provide free compute for general use.

Boundary 03

Not a Beginner Course

We provide no introductory tutorials on python syntax or prompting basics. Researchers must have prior technical context or high capacity to self-learn.

No Signals. No Guarantees. Systems Research Only.

Trading involves substantial risk. AI does not remove risk. Machine learning does not guarantee performance. Backtesting does not guarantee live results. Simulation does not guarantee execution quality. Premium data does not guarantee profitable decisions. No strategy is guaranteed to work.

The market systems research lane is research-focused, benchmark-focused, simulation-focused, paper-trading-first, and systems-focused. It is not investment advice, not a signals service, not a guaranteed-profit program, not a funding guarantee, and not a live-trading promise.

Application Preview

What Applicants Should Be
Ready To Answer.

We evaluate candidates based on structured technical parameters. Prepare your responses for these core questions before submitting an application.

General Fit Preview

  • • Why do you want to work with Source Research Lab?
  • • Which research lane matches your target expertise?
  • • What existing technical proof-of-work can you show?
  • • Are you comfortable producing detailed, structured reports?
  • • Are you comfortable with confidentiality boundaries?

AI Systems Fit Preview

  • • Have you deployed local open-source models (Llama, Qwen)?
  • • Have you built vector embeddings or hybrid RAG pipelines?
  • • How would you test whether an AI tutor improves learning?
  • • How would you systematically audit LLM hallucination rate?
  • • How would you compare two frontier models fairly?

Trading Systems Fit Preview

  • • Have you processed tick data, OHLCV data, or order books?
  • • Have you utilized Python, pandas, Polars, or Jupyter?
  • • Have you configured Bloomberg, LSEG, dxFeed, or Polygon APIs?
  • • What are the most common causes of backtesting bias?
  • • How would you evaluate strategy regime degradation?
Support Desk

Frequently Asked
Questions.

Clarifying operational details, access stages, and researcher guidelines for Source Research Lab.

No. This is a selective research and benchmark layer inside Source. Admitted candidates are researchers, not employees. Strong candidates who produce high-quality work may be considered for paid collaboration, case studies, or future Source tasks, but nothing is guaranteed.
No. Source University is the training and capability-development layer designed to teach candidates how to run systems. Source Research Lab is the evaluation, benchmark, and research layer behind the systems that tests whether they actually work under pressure.
No. While we welcome PhD-level researchers, a degree is not a requirement. We value real proof-of-work, documentation discipline, logical clarity, and the capacity to write structured research reports more than academic titles.
No. The Research Lab is not a training course. While self-taught individuals are welcome, applicants must possess prior technical context (in AI systems or financial data) and a high capacity to self-direct and document their findings.
Active programs include AI tutor evaluation, agent harness benchmarking, RAG chunking and vector recall tests, model cost-vs-quality routing, console utility tests, backtesting integrity, risk dashboard configurations, and trading telemetry observability.
Possibly. Access is qualification-gated, staged, project-scoped, and subject to approval and confidentiality agreements. Selected candidates receive only the specific access required for their matched research tasks. Access is not guaranteed.
Access to compute VMs, GPU servers, alternative data feeds, paid APIs, and frontier models is project-dependent and subject to licensing, availability, and approval. Resource allocation is gated and not guaranteed.
Possibly. Strong researchers who consistently deliver high-quality, reproducible scorecards and memos may be considered for paid tasks, contract database setups, or future roles on our operator bench. This is not guaranteed.
Possibly. Public proof-of-work, case studies, or research summaries may be published only by mutual agreement. Public exposure is optional and not guaranteed.
No. The market systems lane is strictly research-focused, benchmark-focused, and simulation-first. We do not provide trading signals, investment advice, profit promises, or live account funding.
Expected outputs may include benchmark reports, research memos, scorecards, failure-mode reports, data-quality audits, retrieval evaluations, dashboard reviews, model comparison reports, risk reports, experiment ledgers, and proof-of-work artifacts.
Possibly. Strong candidates who consistently deliver high-quality, reproducible scorecards, memos, and audits may be considered for paid collaboration, contract database setups, or future roles on our operator bench. This is not guaranteed.
Possibly. Public proof-of-work, case studies, testimonials, or public research summaries may happen only where mutually approved and appropriate. Public exposure is not guaranteed.
No. Access is controlled, project-scoped, and subject to availability, cost, licensing, approval, qualification, and confidentiality boundaries.
Click the "Apply for Research Review" button below to submit your interest, technical background, desired research lane, availability, and links to your existing proof-of-work.
Research Review Gate

Apply For
Research Review.

If you are serious about helping test, benchmark, document, and improve advanced AI systems, agent harnesses, RAG corpora, model workflows, dashboards, market data pipelines, premium data workflows, backtesting systems, simulations, risk dashboards, and telemetry systems, you may apply for Source Research Lab review.

Qualification-Gated No Guarantees Outputs Matter
△ Submit Research Application View Source University → View AI Systems Program View Trading Systems Program
Access, paid collaboration, public exposure, compute, premium data, proprietary systems, model workflows, local inference labs, custom harnesses, broker/exchange workflows, live-risk systems, and future Source involvement are not guaranteed. They are qualification-based, project-scoped, and subject to approval, availability, licensing, confidentiality, risk boundaries, and research fit.