Source files: hedge-fund-domain-model-agentic-job-audit-2026-raw.md · hedge-fund-ml-predictive-edge-prospectus-audit-2026-raw.md · hedge-fund-ai-podcast-practitioner-signals-2026-raw.md · hedge-fund-ai-linkedin-personnel-2026-raw.md · hedge-fund-ai-worker-layer-linkedin-2026-raw.md · buy-side-quant-ai-practitioner-signals-2026-raw.md · generic-finance-topical-2026-08-16-raw.md · future-alpha-vendor-customer-media-2026-08-18-raw.md Source status: first-party firm disclosures, technical papers, official recordings, and official listings prioritized; third-party job mirrors and podcast transcripts explicitly marked. Reconciled 2026-08-16.
Executive Summary
- Man AHL publicly describes AlphaGPT and AlphaTrend as workflows spanning hypothesis generation, production-code implementation, signal testing, and research gates. Man has not publicly identified a Man-trained finance language model.
- CFM publicly reports a narrow financial named-entity recognition fine-tune. AQR-affiliated research reports point-in-time language models trained from scratch, but public evidence does not establish AQR portfolio deployment.
- The recruiting record describes different build layers. G-Research names on-prem open-model inference, centralized MCP, and secure agent sandboxes. Point72 names fine-tuning pipelines, distributed training, and investment-team GenAI. CFM names paper-to-code research agents. WorldQuant names LLM agents for signal development. These are hiring or architecture signals, not a common measure of completed deployment.
- Numerai is not a “0 completed” model case. Its NumerCon 2026 recap publicly describes an 8-billion-parameter Predictive LLM trained on more than one million articles and used to generate features for the Faith dataset. The same public record describes agent Skills and MCP-based research and submission tools.
- Balyasny is not a “0 disclosed” model case. Balyasny-affiliated researchers published BAM financial embeddings and described a retrieval/RAG deployment. This is a domain-specific embedding artifact, not evidence of a general-purpose finance LLM or investment-return attribution.
- None of those listings, by itself, proves that a finance-specific language model has been trained and deployed. Hiring evidence measures technical direction and organizational capability. It does not measure completed systems, alpha, or changed weights.
- The public record contains multiple implementation patterns: frontier-model integration, domain-specific embeddings, narrow task fine-tuning, agent/tool surfaces, predictive ML, and academic model research. They should be tracked separately rather than collapsed into a maturity ladder.
Evidence register: the question changes the artifact
| Evidence question | Public examples | What the evidence supports | What remains unproven |
|---|---|---|---|
| Named model or changed weights | Numerai Predictive LLM; CFM financial NER; AQR-affiliated research | A model name, parameter count, training or fine-tuning claim can be recorded with its source class. | Weights, reproducibility, production scope, and portfolio attribution vary by artifact. |
| Embeddings and retrieval | Balyasny BAM embeddings; QRT-associated RAG artifacts; general finance embedding benchmarks | Domain-specific retrieval work and its reported evaluation can be recorded separately from generative models. | Current platform status, access permissions, and investment impact. |
| Agentic research workflow | Man AHL, Two Sigma, Bridgewater, CFM, WorldQuant | Tool use, research steps, code generation, evaluation, or submission surfaces when explicitly described. | Unrestricted autonomy, live-book permissions, and independent outcomes. |
| Investor or investment-process deployment | Citadel, Bridgewater, Man AHL, Schonfeld, Point72 | Firm- or vendor-reported adoption and bounded workflow descriptions. | Independent verification, exact model path, and return attribution. |
| Enabling infrastructure | G-Research, Point72, Millennium, QRT, Jump, Tower | Model serving, MCP, RAG, tools, observability, or secure execution intent. | A completed investment product or live signal path. |
| Academic or affiliated research | AQR-affiliated papers and Balyasny-affiliated paper | Research claims and affiliations, with authorship and publication date. | Current employer ownership, deployment, and investment authority. |
The Job Listings Reveal the Build Layer
Public product announcements tend to show the interface. Recruiting pages reveal the substrate.
G-Research’s Core AI role describes on-prem open-model inference, model serving, centralized MCP servers, governed access to tools and data, and isolated execution for autonomous agents. The listing does not prove an investment agent is producing signals. It describes a controlled runtime the firm may be building.
Point72’s current openings reach further into the training lifecycle. The firm names embedding and retrieval systems, fine-tuning pipelines, distributed model training, hyperparameter tuning, inference, observability, and direct integration with investment workflows. A separate L/S equity role places an AI engineer beside a portfolio manager and analysts. This is stronger evidence of organizational capability than the earlier synthetic-data listing, but it still stops short of a trained finance LM.
CFM’s hiring connects agents directly to research production. One role asks engineers to use LLMs and agent frameworks to turn academic papers into code. Another places code-generation and model-transformation agents inside the platform that deploys and monitors predictive models. Combined with CFM’s reported narrow fine-tuning case, this documents a hybrid pattern: frontier-model orchestration around research plus task-specific model adaptation.
WorldQuant’s AI Scientist description applies LLM agents to signal and algorithm development. Numerai’s older Quantitative LLM Researcher description names financial-corpus fine-tuning, Common Crawl data engineering, and LLM-architecture modification, but the newer firm-published NumerCon recap is the more direct model evidence: it identifies an 8B Predictive LLM and its Faith-dataset use. The job description remains hiring evidence; it is not needed to establish the released model claim.
Acadian was a material omission in the earlier hiring lane. June and April 2026 mirrors describe an Investment AI Engineer and an AI-driven research-platform lead. The roles cover model training, deployment, quantitative research, backtesting, and feature pipelines. They document predictive-ML and investment-platform hiring intent; they do not establish LLM agents or finance-LM fine-tuning.
Model and embedding corrections
Numerai Predictive LLM is a public model disclosure
Numerai’s NumerCon 2026 recap says the firm introduced a Predictive LLM with 8 billion parameters, trained on more than one million articles, to turn news into structured predictive signals. Numerai says those features power the Faith dataset. This is direct firm-published evidence of a named model artifact and a released feature use case. It corrects the earlier landscape label “0 completed.”
The public record does not include the model weights, full article corpus, point-in-time leakage controls, independent replication, or a model-specific return attribution. The correct classification is “firm-described 8B model used in dataset feature generation,” not “autonomous trading model.” Numerai’s homepage, MCP documentation, and example Skills repository are separate agent-facing platform evidence.
Balyasny BAM embeddings are a separate model lane
Balyasny-affiliated researchers published BAM embeddings, fine-tuned from Multilingual-E5-base on 14.3 million finance query-passage pairs. The paper reports held-out retrieval and FinanceBench results, and describes a retrieval service indexing millions of financial documents. This is model and deployment evidence for finance retrieval, not a general-purpose generative model. It belongs in the landscape’s embeddings/retrieval lane and in Balyasny’s artifact inventory.
The paper is a published research result with Balyasny affiliation. It does not disclose whether every described service remains current, expose the full production evaluation suite, or connect the embedding system to a specific investment decision or return stream. The repository’s BAM/Greenback synthesis records the architecture, dataset construction, and caveats.
Adjacent vendor model and platform surfaces
The Future Alpha sponsor graph also led to first-party vendor material that should remain adjacent to, rather than inside, the hedge-fund-owned model inventory. Clearwater Analytics’ GenAI architecture brief describes CWIC digital-specialist personas for research, quantitative analysis, risk, compliance, and operations. Its architecture diagram names RAG/vector databases, internal and external APIs, authentication and authorization, a LangChain-plus-custom workflow framework, several LLM options, and cloud AI platforms. This is vendor architecture evidence; it does not establish a named hedge-fund customer, that every listed model is used in production, or a return result.
Neuralk’s research page and financial-services solutions page describe tabular foundation models for classification, regression/forecasting, risk scoring, fraud detection, and financial time-series signal generation, alongside the TabBench benchmark. The company page names Alexandre Pasquiou and Antoine Moissenot as co-founders and records the company’s public model and benchmark milestones. These pages add a model/vendor route for structured financial data and title-blind discovery, not evidence that a hedge fund uses Neuralk or that its benchmark claims have been independently reproduced.
The same sponsor-graph expansion adds three more adjacent routes. Zerve’s quant/enterprise role page describes AI agents that execute data workflows, stateful and reproducible environments, deployable outputs, and a vendor claim that early users include Tier 1 quantitative firms using the platform for research workflows and agentic trading systems. The page does not name a customer or disclose a permission map. OpenBB’s first-party AI-leadership article names Michael Struwig, identifies his prior Hudson & Thames Quantitative Research leadership, and describes finance-data and generative-AI workflows; current Enterprise documentation adds role-based AI/data permissions, private deployment, and internal-system integrations. YipitData’s investor pages explicitly target hedge funds and quant/systematic funds with point-in-time, research-ready alternative data, while its feed catalogue describes transaction, web, app, cloud-spend, and other data sources. These are vendor and platform surfaces; they do not establish named-fund adoption or investment outcomes.
Registered-fund filings add a strategy-and-model layer
The prospectus sweep adds evidence that is easy to miss when the search is limited to firm websites, conference talks, and job listings. The filings are legal product disclosures, so they should be read as stated process and risk language—not as an audit of a manager’s private research stack.
| Filing signal | What is stated publicly | Evidence boundary |
|---|---|---|
| AQR Funds | ML and NLP for certain strategies; alternative-data examples include consumer behavior, social sentiment, and internet-search/traffic data; testing and model-modification risks are acknowledged. | Current registered-fund disclosure, not a public inventory of AQR’s private models or GenAI deployment. |
| AQR Delphi | Long/short design using beta, quality, value, and proprietary quantitative models. | Preliminary filing for a fund with no operating history; no performance or production evidence. |
| Counterpoint Quantitative Equity ETF | Gradient-boosted trees and neural networks rank companies using value, reversal, momentum, profitability, sentiment, and stability variables. | Registered-product methodology; no claim about another manager’s implementation. |
| QRAFT AI-Enhanced U.S. Large Cap ETF | Deep learning predicts relative four-week price appreciation; Bayesian neural networks estimate forecast uncertainty. | Licensed process in an ETF prospectus; no independent validation or portfolio attribution in the filing. |
| Sparkline | Founder/CIO Kai Wu’s filing says Sparkline uses ML and computing to seek alpha in large, unstructured datasets; the filing also records a Boston quant-hedge-fund and GMO background. | Adviser/product disclosure and biography; not proof of a general-purpose finance LM. |
| Bluerock AI200+ | A multi-model ensemble combines gradient boosting, deep neural networks, and a fine-tuned financial-text LLM for classification, sentiment, earnings guidance, and AI-opportunity assessment, with PM review and override. | New registered product disclosure; no independent evaluation, filled-role proof, or AI-attributed return evidence. |
| Goldman Sachs Asset Management | NLP and ML may extract information from textual or audio data for quantitative portfolios, with portfolio construction and execution-cost controls. | Large asset-manager prospectus language; not evidence about hedge-fund deployment. |
| Rayliant NxtGen Multifactor ETFs | ML models estimate risk-adjusted returns and determine constituent weights; the filing describes purchased and public-web data, proprietary cleaning, automated/manual checks, and human trade discretion. | Registered ETF disclosure; no architecture, model weights, or independent evaluation. |
The full source-by-source map is in the prospectus audit. The most useful research implication is not a league table. It is a better taxonomy: predictive price models, event-impact models, text/audio extraction, embeddings and retrieval, regime models, and agentic research workflows should be tracked as separate artifacts.
Podcasts Expose Research Method More Often Than Model Ownership
The podcast record adds useful operating detail without resolving model ownership. Two Sigma’s Ben Wellington discusses LLMs as timestamp-safe feature-discovery tools. A named Acadian practitioner describes a path from text and relationship data through expected-return forecasts, portfolio construction, and execution. HRT’s Iain Dunning discusses deep-learning constraints in short-horizon markets. Man Group CTO Gary Collier exposes the firmwide data and platform layer beneath agentic research.
Several additional appearances fill firm-specific gaps. Schonfeld CTO Tom DeBow discussed its in-house AI tool, unified data systems, and human-plus-machine operating model on the Wharton FinTech Podcast in April 2025, before the firm published the FE AI Lab. CFM chief scientist Jean-Philippe Bouchaud appeared on Bloomberg’s The Big Picture in April 2026 and reinforced CFM’s empirical, model-driven research culture. WorldQuant founder Igor Tulchinsky appeared on Voices of Impact in June 2026; the episode is a high-priority acquisition target because Tulchinsky also leads the firm’s AI initiatives. Citadel founder Ken Griffin discussed data, AI, research, and independent judgment on S&P Global’s The Leaders.
The medium has a consistent limitation. Podcasts reveal how leaders think about data, research discipline, human judgment, and platform design. They rarely disclose training corpora, checkpoints, fine-tuning methods, evaluation results, or production authority. The claims therefore remain practitioner signals until a paper, job artifact, system description, or first-party deployment disclosure supplies the missing contract.
New topical recordings add an infrastructure and model-use layer
The expanded title-blind sweep surfaced several recordings that materially sharpen the evidence register. The full capture ledger is Generic Finance Podcast Discovery — 2026-08-16.
- Jane Street: the firm-linked GPU conversation describes LLM training and custom architectures adapted to trading data, while the companion tour shows the physical training environment. Jane Street’s first-party page reports 4,032 GPUs across 56 liquid-cooled racks and about 8,000 km of fiber in the Texas facility. Its ML page separately reports neural-network models driving trading strategies, distributed training/inference work, LLM/RL/CUDA research, and custom trading architectures. This is direct predictive-ML and infrastructure evidence; it is not a disclosed finance-language-model weight inventory or return attribution. (conversation, data-center page, ML page)
- HRT: the new Odd Lots recording adds a current practitioner account of compute, power, data-center capacity, model-training scale, and research-cycle constraints. The episode page describes token spending, memory prices, compute bottlenecks, and possible chip development. It supports the existing HRT infrastructure/process classification, not a public model-weight or autonomous-trading claim. (episode)
- Numerai: the NumerCon closing recording corroborates the written account of Numerai Skills, MCP-based model creation/diagnostics/submission, and the Predictive LLM/Faith workflow. The controlling source remains Numerai’s written recap, which identifies the 8B model and its article-trained feature generation. (recap, closing recording)
- Bridgewater: the topical discussion led back to Bridgewater’s first-party AIA Labs page, which now lists the June 30, 2026 Learning to Replicate Expert Judgment in Financial Tasks work and describes models tuned for specific organizational tasks. The page also states Bridgewater’s own live-capital and Pure Alpha integration claims. Those are important public disclosures, but they remain firm-reported until independently replicated. (AIA Labs, recording)
- Method signal: the CFA Institute and RavenPack recordings add context on causal inference, factor construction, alternative-data transformation, and NLP signal extraction. They are useful for the research-method map, but do not establish that a named hedge fund operates the described systems. (CFA recording, RavenPack recording)
Personnel evidence reveals dedicated AI ownership—and corrects stale attributions
A July 25 LinkedIn audit identifies three unusually direct organizational signals: Aaron Linsky is CTO of Bridgewater’s AIA Labs, Iain Dunning is Head of AI at Hudson River Trading, and Gideon Mann is Global Head of AI in Millennium’s technology organization. These titles do not prove model training or investment impact. They do show that AI has a named senior owner rather than being absorbed into a generic data-science function.
The same audit changes how two older disclosures should be read. Umesh Subramanian now identifies as an ex-Citadel CTO; Citadel’s official page names Andrew Janian interim CTO. Thomas DeBow’s profile still carries a CTO headline but lists a notice period at Schonfeld. Evidence tied to Subramanian and DeBow remains relevant to the systems built during their tenures, but it should not be presented as current executive ownership.
Named practitioners also reveal a distinct research-leadership layer. Giuseppe Paleologo is Global Head of Quantitative Research at Balyasny, Igor Tulchinsky identifies as WorldQuant’s Head of Research as well as CEO, and Philip Seager is CFM’s Head of Portfolio Management. These are signals of research depth. They are not evidence that the firms fine-tune language models.
The worker layer adds infrastructure detail
Leadership titles show sponsorship. Worker roles reveal the functions that can operate a system. The July 25 worker audit records AI engineering, ML/HPC architecture, principal AI/ML engineering, GenAI software, quant-software roles, ML research, quant-ML research, and MLOps across the reviewed firms. Bridgewater’s public record separately names AIA Labs staffing and technical roles.
Millennium and Point72 expose geographically distributed AI-engineering roles. Schonfeld exposes a Generative AI Engineer whose profile names LLMs, RAG, Bedrock, Azure OpenAI, Claude, LangChain, and Python, plus an engineer supporting the internal SchonGPT platform. This strengthens the integration thesis: the visible work is provider orchestration and internal-system connection, not proprietary foundation-model training.
Acadian provides a worker-level connection to the investment process. Siddhartha Pant’s public profile describes AI-enabled workflows spanning portfolio construction, transaction-cost analysis, optimization, and systematic trading. It remains self-described profile evidence and does not establish language-model fine-tuning.
These personnel signals belong to different evidence classes: workflow descriptions, infrastructure hiring, dedicated AI-lab staffing, and investment-process profiles. None of them by itself establishes finance-LM weight training or changes the interpretation of the model artifacts above.
Five Capability Layers Should Not Be Collapsed
| Layer | Public examples | What the evidence permits |
|---|---|---|
| Language-model weight modification | CFM, AQR-affiliated research; intended at Numerai | Claim changed LM weights only when the source identifies pretraining or fine-tuning. |
| Agentic research workflow | Man AHL, Two Sigma, CFM, WorldQuant; Point72 in development | Claim tool-using research automation when the workflow reaches hypotheses, code, features, tests, or signals. |
| Investor or investment-process deployment | Citadel, Bridgewater, Man AHL, Schonfeld, Point72 | Claim bounded use, not autonomous authority or alpha, unless independently demonstrated. |
| Enabling infrastructure | G-Research, Point72, Millennium, QRT, Jump, Tower | Claim organizational capability and architecture, not a completed investment product. |
| Predictive quantitative ML | HRT, Jane Street, XTX, Voleon, PDT, Renaissance, Winton | Claim trained financial forecasting models, not finance language models. |
The distinction prevents the most common category error in this market. HRT can train proprietary deep-learning models that directly influence trading without having trained a finance language model. G-Research can operate an advanced open-model and agent runtime without disclosing an investment-research agent. Citadel can deploy an investor assistant without disclosing whether any language-model weights changed.
Firm-by-Firm Public Evidence Register
| Firm | Public evidence as of 2026-08-12 | Boundary |
|---|---|---|
| Man AHL | AlphaGPT and AlphaTrend are described as research workflows connecting hypotheses, code, testing, and research gates. | No Man-trained finance LM or independent return attribution is public. |
| CFM | Financial-NER fine-tune; paper-to-code and predictive-platform agent hiring. | Narrow task result and hiring evidence; no general finance LM is disclosed. |
| AQR | Point-in-time LM research with AQR-affiliated authors. | Affiliation and paper evidence do not establish AQR deployment. |
| Citadel | Equity-research assistant, central AI/ML research, and LLM/fine-tuning hiring. | Exact architecture and weight status are unresolved. |
| Bridgewater | AIA Labs, PAT research tool, and human-machine investment-process disclosures. | Firm/vendor-reported workflow; no independent attribution. |
| Point72 / Cubist | Fine-tuning, distributed training, RAG, agents, and direct investment-team roles in current listings. | Hiring intent and infrastructure evidence; no completed domain LM is public. |
| G-Research | Open-model serving, MCP, agent sandboxing, and applied-AI platform language. | Role and engineering evidence; no investment-agent deployment is public. |
| Two Sigma | Firm-aware model research, semantic-type tooling, and agentic feature-development roles. | Workflow and code evidence; no public model weights or trading authority. |
| Schonfeld | FE AI Lab, internal tools, model partnerships, and pilot/evaluation controls. | Firm-reported workflow; exact deployment and authority are not public. |
| Millennium | AI leadership, federated agent platform, routing, internal search, and end-user tools. | Firm-reported operating model; model inventory and authority are not public. |
| WorldQuant | LLM-agent signal-research role and public AI leadership statements. | Hiring and firm-reported direction; production authority is unverified. |
| Numerai | 8B Predictive LLM for Faith features; AI-scientist workflow; Skills and MCP research/submission tools. | Model weights, leakage controls, and portfolio attribution are not public. |
| Balyasny | BAM financial embeddings paper, retrieval/RAG deployment description, Applied AI case study, and agent evaluation claims. | Paper and firm/vendor claims; no general finance LLM or return attribution is public. |
| Acadian | Investment AI and research-platform roles; systematic ML process disclosures. | LLM-specific model ownership is not public. |
| Jane Street | Proprietary trading-model research, LLM research, training loops, and AI-assistant signals. | No firm-owned finance LM is identified. |
| Jump Trading | LLMs, generative models, agents, HPC serving, tools, and data integration. | No disclosed finance-LM fine-tune or investment authority. |
| QRT | AI-platform roles, QRT Labs research, and firm-associated RAG/embedding artifacts. | quant-mind is a fork; authorship and deployment are not established. |
| Tower Research | Agent-harness governance, ML hiring, and public agentic-research discussion. | No specific model, signal path, or investment authority is public. |
| HRT | Deep-learning trading models; generative-AI research assistance; agent-versus-human signal-ideation evaluation; reported token consumption; compute-site and power procurement; hardware-team investigation. | Predictive ML is not evidence of a finance language model. Team-level spend and productivity remarks are anecdotal; no public model weights, returns, or autonomous trading authority are disclosed. |
| XTX | ML price forecasts, deep-learning researchers, AI internship, and ML performance engineering. | No public LLM-agent claim found in the reviewed sources. |
| PDT | Applied ML scientists and large-scale research infrastructure. | No public language-model evidence found in the reviewed sources. |
| Voleon | AI/ML research and production trading-data signals. | No firm-controlled language-model program was located. |
| D. E. Shaw | Investment-side GenAI hiring signals; separate molecular-science LM research. | Do not transfer research-business LM work to the investment business. |
| Winton | ML research with interpretability and selection-bias discipline. | No current LLM-agent role found. |
| Renaissance | Statistical-model and compute/research infrastructure signals. | No public finance-LM or agent evidence found in the reviewed sources. |
| Systematica | No official LLM/agent role retrieved in this pass. | Unknown, not “no AI.” |
| Marshall Wace | Technology and quant-research hiring; one public recursive-LM workshop artifact. | The workshop does not establish a production RLM program or investment use. |
| Brevan Howard | Macro-research role describing LLM-assisted ingestion, RAG, embeddings, and guardrails. | Employer-described workflow; exact title and deployment require qualification. |
| Caxton Associates | Hedge-fund and quantitative talent material without a retrieved firm-owned LLM/agent program. | Unknown, not “no AI.” |
| Aspect Capital | Systematic-investment and quant-research material discussing ML/LLM context. | No current firm-owned LLM-agent deployment was located. |
| Squarepoint Capital | ML-qualified investment hiring and integrated research/trading systems. | No retrieved firm-owned LLM-agent role in this pass. |
Evidence Boundaries
- No public source in this audit demonstrates that Man AHL has trained a finance-specific foundation model.
- No job listing is treated as proof of a completed system, production promotion, research improvement, alpha, or autonomous investment authority.
- No predictive trading model is relabeled as a language model without explicit architecture and training evidence.
- Company-disclosed performance and adoption are not independent attribution.
- “No public evidence found” is a search result, not a claim that a private firm lacks the capability.
What This Means for the System Under Consideration
The public precedent does not support assuming that a single proprietary finance GPT is the necessary starting point. The reviewed sources separate four jobs: frontier reasoning for difficult research, proprietary context and tool access, hard evaluation gates tied to investment methodology, and smaller adapted models or embeddings for narrow high-volume tasks.
That sequencing preserves optionality. Build the research harness and its evaluation contracts first. Capture every hypothesis, code artifact, data lineage record, backtest configuration, rejection reason, and human override. Fine-tune only after the traces show a repeated task where a smaller model can beat the frontier baseline on quality-adjusted cost.
The operating design questions are visible across the public record: research-loop traces, governed runtimes, training and deployment controls, narrow-model economics, and retrieval quality. The relevant test is whether a proposed stack makes model output testable in the organization’s own data, permissions, and evaluation environment.
If this raises questions specific to your organization, I’d welcome the conversation — brandon@brandonsneider.com
Verification Ledger
| Claim family | Evidence family | Confidence |
|---|---|---|
| Man AHL agentic workflow | First-party articles and partnership announcement | High for workflow; low for independent outcomes |
| CFM narrow fine-tuning | First-party case study | High for implementation; medium for reported benchmark without replication |
| AQR-affiliated LM training | Technical paper plus official leadership page | High for research; low for deployment |
| Hiring signals | First-party job boards where available; marked mirrors otherwise | High for first-party capability intent; medium-low for mirrors |
| Investor deployment | First-party firm disclosure or named Reuters reporting | Medium-high for adoption; low for investment impact |
| Quiet-firm negative findings | Search audit and official careers pages | Low as a statement about actual private capability |
Next Research Lanes
- Archive firm-owned job artifacts for the mirror-only Acadian, WorldQuant, PDT, and Voleon findings; keep Numerai’s firm-published model disclosure separate from the older role mirror.
- Resolve BlackRock’s model architecture and Citadel’s ambiguous “trained on” wording.
- Add a model-artifact check to every landscape refresh: named weights, embeddings, retrieval systems, agent tools, and general finance benchmarks must remain separate.
- Capture full QRT AI-platform role descriptions and identify whether the platform supports internal fine-tuning or only model serving; preserve fork provenance for
quant-mind. - Re-run the quiet-firm job audit quarterly, including archived and expired roles for Systematica, Marshall Wace, Brevan Howard, Caxton, Aspect, Squarepoint, Renaissance, Winton, and D. E. Shaw’s investment business.
Sources
Full URLs, retrieval dates, promotion notes, and claim boundaries are recorded in the cross-audit ledger, job-listing ledger, podcast practitioner ledger, title-blind topical ledger, leadership personnel ledger, and worker-layer ledger. The broader historical practitioner record remains in the existing buy-side source ledger.
August 2026