Work Transformers Labs · Research

Project Ghostmap®

Research into how investment teams can move faster without losing rigour, and the hard questions that sit underneath it.

Applied researchInvestment and capitalOpen questions

Ghostmap is an internal research and development programme exploring how investment and real estate teams assess decisions more consistently. It is research, not a product.

Why this matters
A system is only as good as the reasoning beneath it.

It is easy to make a tool that fills a template. It is far harder to make one an investment committee can trust. The difference is the reasoning underneath, and some of the hardest parts of that reasoning are still unsolved research problems. Project Ghostmap is where we study them, so the investment systems we deploy are built on solid ground rather than clever prompts.

Applied research
The hard problems beneath investment reasoning
Reasoning & logic
How rigorous deal judgement actually works
Systems
Built on that reasoning, not generic scripts
Investment decisions
Faster, consistent, defensible at committee
01
Systems need reasoning

A script repeats steps. It has no judgement of its own. The judgement has to be researched, then designed in.

02
The reasoning is hard

Consistent data, trustworthy numbers and lasting context are genuine technical problems, not things a prompt solves.

03
Investment needs it most

Committee decisions have to be both fast and defensible. There is little room for output that only looks right.

The research question

Can a team move faster without losing rigour?

For most investment teams the constraint is not pipeline, it is capacity. Underwriting eats senior time, packs vary in quality between people, and the committee has to be able to defend every figure. Speed and rigour are usually treated as a trade-off. The research question is whether they can reinforce each other instead.

The open problems we are researching

Four hard questions, openly stated.

These are the foundations. None of them are fully solved, and most of the easy answers do not hold up under real deal conditions.

01
Working with messy, inconsistent data
Open question
The problem

Every deal arrives differently: data rooms, PDFs, spreadsheets, emails, and no two are structured the same. Before anything can be reasoned about, that mess has to become consistent. The translation is more fragile than it looks.

What we are exploring

How to turn varied, inconsistent source material into a clean, dependable structure a system can reason over, without losing the detail that changes a decision.

Why it matters

Consistency is the foundation of trust. If the inputs are not dependable, nothing built on top of them can be.

02
Reasoning about financial models
Open question
The problem

Language models are good with words, not arithmetic. Getting consistent, defensible figures out of an analytical model is a real research problem, and the obvious approaches drift.

What we are exploring

How reasoning and computation can work together so a figure stays the same everywhere it appears and holds up to committee scrutiny.

Why it matters

An investment committee has to trust every number. Output that merely looks right is not good enough to sign off.

03
Decision memory
Open question
The problem

Most tools forget context the moment a session ends. A real investment decision evolves over weeks, across people, documents and changing assumptions.

What we are exploring

How a decision can carry its assumptions, reasoning and history over time, so context never has to be rebuilt from scratch.

Why it matters

Less time re-explaining the deal, more time on the judgement that actually moves it. Memory is where rigour compounds.

04
Source traceability
Open question
The problem

When a number cannot be traced back to where it came from, it cannot really be trusted, and rigour quietly erodes under time pressure.

What we are exploring

How to keep a clear, automatic line from any figure or claim back to the document and assumption behind it.

Why it matters

Traceable work is defensible work. It is what lets a team stand behind a recommendation with confidence.

Project Ghostmap is early-stage applied research. We share the questions we are working on, openly. The methods we develop to answer them stay in the lab, because that work is what will power what we do next.

Why now

The pressure is rising on both sides.

01
AI is changing the analyst

It is reshaping which parts of underwriting are manual and which are not, faster than most teams can adapt.

02
Speed expectations are up

Deal flow rewards teams that can assess more of what matters, sooner, without adding headcount.

03
Scrutiny is rising too

Committees and investors expect more rigour and clearer evidence, not less. Both pressures bite at once.

From research to deployment
This research powers the reasoning inside our investment systems.

The point of the research is not the research. It is to make the systems we deploy genuinely dependable: built on consistent data, figures a committee can trust, and context that lasts. When the hard problems beneath a decision are taken seriously, the system stops being a clever assistant and starts being something an investment team can actually rely on.

See how investment teams use us →
The system the research feeds
Research is one layer. The system around it is what makes it dependable.

The questions on this page do not stay on the page. They feed the reasoning layer of the operational systems we deploy, beneath the work a team actually uses.

Fig. 02 · The operational system we buildSecurity and governancepermissions, human review, traceabilityData abstractionmessy inputs made consistent and retrievableReasoning and domain logicjudgement, powered by applied researchOperational systemsreviews, memos, reportsOperational UIwhere your team works with the systemApplied researchWork Transformers LabsFrontier modelsflexible, no lock-inDocumentsCompany dataConnectorsGraph retrievalINPUTS & CONNECTORSTHE SYSTEM WE BUILDYOUR WORK
How it fits together
Research is one layer. The system makes it dependable.

Every finding is sourced, scored, explained and ends in a recommendation a committee can defend.

Data roomBill of quantitiesComparablesFinancial modelTender packSourced findingsConfidence scoresRisk flagsIC-ready paperAudit trailDefensible callExtractSourceScoreExplainCross-checkFlag riskRecommendLogTraceINPUTSTHE REASONING SYSTEMWHAT REVIEWERS GET
How we publish

What we share, and what we keep.

Project Ghostmap is open research into the reasoning beneath faster, defensible investment decisions. We publish the questions we are wrestling with, the patterns that recur across real deployments, and the principles that hold up under pressure. We do not publish client data, or the parts of the method still being worked out. It is research, not a product: the questions are the interesting part, and the answers are earned in the work.

Project Ghostmap

Research that earns the right to be trusted.

We work on the hard problems first, so the systems we deploy hold up where it counts.