Research systems / 01

Read the footnote.
Test the intuition.

An academic research note on whether company-specific fundamental knowledge retains value when machine-generated breadth enters the experiment.

The operatingREALITYcontext before conclusion
01Companypolicy & practice
02Industrystructure & cycle
03Footnotethe hidden variable

Every business leaves a different trail.

The useful signal is rarely found in a single headline metric. It lives in the choices a company makes, the language it repeats and the small disclosures that explain how an industry actually works. This project treats that specificity as a research variable.

ncompany contexts
mindustry contexts
document relationships

Can experience still
carry explanatory power?

Two artifacts, one controlled comparison: does domain-specific fundamental research retain an advantage when the same question is approached with a general-purpose machine system?

A / Human baseline
A

random pork oinks

Built without AI: supply-chain and holding-structure graph research, informed by more than a decade of experience across tier-1 trading companies.

Material
Pure earnings, SEC filings and footnotes
Orientation
Hand-made company, ownership and industry graphs
Purpose
Apply accumulated fundamental-research experience
B / Control group
B

pork with memories

An ensemble of Qwen models adapted with supervised fine-tuning (SFT) and LoRA on twenty-five years of publicly sourced research material.

Corpus
Earnings-call transcripts, SEC filings, and company and industry reports
Method
Model-generated outputs rather than manually constructed rules
Purpose
Test whether fine-tuned models can outperform accumulated fundamental-research knowledge
Central empirical testDo fundamental research and an understanding of the economy and industry still matter — or can they be replaced entirely by AI built by outsiders?

A · random pork oinks · without AI

From document to
decision context.

Research is assembled as a chain of evidence, not a pile of excerpts.

01

Collect

Gather the company’s own vocabulary across filings, notes and reporting periods.

02

Situate

Place each disclosure inside the operating structure of its industry.

03

Interpret

Turn connected observations into a view that can be inspected and challenged.

Current layer / collectStart with the source’s own terms.

Recurring definitions, accounting policies and changes in presentation become the first map of what deserves attention.

Relational research map / working exampleHuman-written knowledge becomes structure.

Filter the hand-made graph by relationship type, then select a node to inspect the research judgment encoded before any conclusion is drawn.

6 nodes / 9 human-written relationships

Context is a connected object.

Choose a company below, then follow one finding from source language to industry implication.

Choose a company example
Source document
Annual filing · Note 7

“Certain customer tooling is retained on-site and depreciated over a twelve-year useful life.”

FY 2024 / Harbor Components
Cross-reference

Capacity expansion / contract manufacturing

Research interpretation
Company-specific

The asset is a relationship signal.

A long useful life implies durable tooling, repeat programs and switching costs that may not appear in the headline margin.

Industry context

A structural feature, not a one-off.

In precision components, customer qualification cycles can turn fixed assets into evidence of embedded demand.

End of Track AContinue to Track B — the fine-tuned model ensemble

B · pork with memories · fine-tuned model condition

How the model ensemble is
trained, then assembled.

Three stages: the corpus the models read, the supervised fine-tuning that adapts them, and the synthesis that brings their outputs into one comparison. The middle stage is where the work sits.

Design schematic / not a performance result
Stage 01

The corpus the models read

Twenty-five years of publicly sourced material across three document types.

Input
Research horizon / year 25 of 25
Stage 02

Supervised fine-tuning with low-rank adapters

Curated pairs are passed through a frozen base model. Only the low-rank adapters are trained, and the pass is repeated once per ensemble member — this stage, not the merge, is the cost of Track B.

Where the work sits
Training pass / adapter 1 of 5
Base weights held frozen · d × dTrained low-rank adapters · r ≪ d
Stage 03

Ensemble synthesis

The fine-tuned members’ outputs are combined into the single response that enters the comparison.

Output
Fine-tuned runsSynthesisControl

The comparison stays open.

Both tracks read the same filings and answer the same question. Nothing on this page is a result — each method is set out so it can be inspected before either artifact is scored.

ECONOMICS · FINANCE FUNDAMENTAL RESEARCH
Independent researcher / economics & finance

Research formed by study and practice.

Academic training in finance at MIT Sloan, and in economics and computer science at Columbia University, followed by approximately a decade of fundamental research across tier-1 trading companies — the same filings and footnotes this note argues from, read under professional conditions.

Academic training
MIT Sloan School of ManagementFinance
Academic training
Columbia UniversityEconomics and computer science
Applied practice
Senior rolesFundamental research as a core input to systematic investment and trading