# Galen pre-seed public proof ledger

Pre-seed investor workspace: https://preseed.usegalen.com (live mirror: https://www.usegalen.com/preseed)
Pre-seed public Markdown packet: https://preseed.usegalen.com/llms.txt (live mirror: https://www.usegalen.com/preseed/llms.txt)
Pre-seed structured JSON brief: https://preseed.usegalen.com/investor-brief.json (live mirror: https://www.usegalen.com/preseed/investor-brief.json)
Pre-seed AI diligence prompt: https://preseed.usegalen.com/ai-diligence-prompt.txt (live mirror: https://www.usegalen.com/preseed/ai-diligence-prompt.txt)
Pre-seed concise partner memo: https://preseed.usegalen.com/partner-memo.md (live mirror: https://www.usegalen.com/preseed/partner-memo.md)
Pre-seed public proof ledger: https://preseed.usegalen.com/proof-ledger.md (live mirror: https://www.usegalen.com/preseed/proof-ledger.md)
Pre-seed first-call diligence questions: https://preseed.usegalen.com/diligence-questions.md (live mirror: https://www.usegalen.com/preseed/diligence-questions.md)
Pre-seed forwardable intro email: https://preseed.usegalen.com/intro-email.txt (live mirror: https://www.usegalen.com/preseed/intro-email.txt)
Private packet request: https://preseed.usegalen.com#data-room

This public-safe ledger summarizes what Galen is claiming, what stage each claim is in, how investors should interpret it, and what private diligence should be requested before making a technical judgment.

## Public proof scorecard

### ~90%: Hidden mechanism support recovered in the strongest public summary tests.
Label: architecture proof
Galen's current flagship evidence shows the architecture can recover and use cause-and-effect structure when a realistic single-cell-style experiment contains identifiable signal.
Private packet focus: Exact setup, baselines, confidence intervals, ablations, and leakage checks are in the approved proof packet.

### 15 / 15: Control settings behaved as expected.
Label: control discipline
The evidence practice matters as much as the benchmark: Galen records when evidence supports a claim, when confidence is earned, and when a control outcome remains a control outcome.
Private packet focus: Approved investors can review the control settings, control outcomes, and claim-boundary notes.

### built: The experiment-selection software is already part of the system.
Label: experiment choice
In controlled dry-lab tests, Galen selected experiments expected to teach the model more than simple same-budget alternatives. The pre-seed loop tests that claim in living cells.
Private packet focus: The packet includes the active-design verdict, comparison policies, and the real-lab scoring plan.

### prospective: The round funds predictions committed before results.
Label: next proof
The next investor-grade evidence is forward-looking: define the benchmark, lock the baselines, run the experiment, and score the model after biology answers.
Private packet focus: Approved investors receive the validation contract, use-of-proceeds detail, and protocol-level exclusions.

## Proof summary

### validated: Controlled causal recovery
Galen recovered masked cause-and-effect structure in realistic single-cell tests and used it to predict previously unseen responses.

### building: Experiment selection
Galen is measuring whether model-selected experiments teach more than same-budget random or conventional plans.

### next: Forward-looking human-cell proof
The round funds a real human-cell learning loop with predictions committed before results and a second experiment queue chosen from what the model learns.

## Claim ledger

### Causal recovery in realistic tests
Status: validated
Evidence: In realistic test settings based on measurements from individual cells, Galen recovered the causal relationships behind cell responses, beat matched comparison models in two positive settings, and correctly handled 15 control settings.
Investor interpretation: The model can recover and use cause-and-effect structure when the data contains enough signal to identify it.
Current application: Current evidence supports controlled causal recovery under realistic measurement constraints.
Next proof: Run the same kind of test on actual intervention data, with predictions and scoring rules fixed before results are known.

### The cause-and-effect path matters
Status: validated
Evidence: When Galen used its full path for inferring causes, combining effects, and predicting the resulting cell state, it outperformed comparison models built to test whether the signal was real and correctly labeled.
Investor interpretation: The result supports Galen's thesis that causal structure improves intervention prediction.
Current application: Current evidence is strongest in controlled settings where the evaluator can score the true causal structure.
Next proof: Carry the same discipline into the first real human-cell study.

### Experiment-selection software
Status: building
Evidence: In a software benchmark, Galen chose a same-budget experiment that taught the model more than random selection.
Investor interpretation: The experiment-selection rule is worth testing in the lab.
Current application: This result supports the lab study design for testing model-selected experiments under real experiment cost and noise.
Next proof: Compare model-selected experiments against random, diverse, uncertainty-driven, and conventional plans under real experiment cost and noise.

### Integrated virtual-cell machinery
Status: building
Evidence: The codebase connects cell-response modeling, uncertainty estimates, experiment selection, and reporting into one workflow.
Investor interpretation: Galen is building a coherent system for prediction, experiment choice, measurement, and learning.
Current application: The current system is ready to be refined against real human-cell data.
Next proof: Use real human-cell data to decide which parts of the system deserve scale-up.

### Real human-cell learning loop
Status: next
Evidence: The pre-seed roadmap defines a real human-cell study with predictions recorded before results, comparison models fixed in advance, model-selected experiments, and measured outcomes.
Investor interpretation: This is the next value-creating proof.
Current application: This study turns Galen's modeling evidence into a living-cell learning loop.
Next proof: Run the first forward-looking study in human CD4+ T cells, a central immune-cell type, and score previously unseen interventions before revising the model.

### General programmable virtual-cell platform
Status: long-term
Evidence: The platform map defines predict, design, and verify as the long-term product layer as focused models become reliable before results are known.
Investor interpretation: The company can become much larger if the loop compounds across contexts.
Current application: Galen earns platform breadth by expanding reliable prediction across contexts, interventions, and customer workflows.
Next proof: Earn platform language through real performance before results are known, reliable confidence estimates, learning efficiency, and customer workflow integration.

## Ground rules
- This ledger is public and intentionally omits protocol-level details, exact baselines, intervals, ablations, leakage checks, budget detail, round terms, and proprietary implementation detail.
- Treat validated, building, next, and long-term claims as distinct stages.
- Request the private proof packet for exact technical diligence before making a final investment decision.