04 — MemoDecision system

What an investor actually receives

Not a score and not a summary. A memo whose every claim carries its source, whose unresolved questions are stated rather than smoothed over, and which keeps being revised after it ships.

04.1 — Output

Specimen — illustrative company, not a real filing

Something an investor can actually use before the meeting.

Investment memo

Company under analysis

Investment signal82^/100

Meridian Runtime, Inc.

SeedDelaware C-corpSF Bay AreaConfidence: high

Primary thesis — Proprietary inference architecture emerging underneath a wrapper narrative, carried by an unusually strong founder network.

Core scores

Dimension	Score
Founders	91
Timing	93
Market	88
Technical moat	86
Defensibility	84
Capital efficiency	80
Product	74
Traction	72
Distribution	59

Predicted trajectory

12 MO Enterprise platform with usage-based compute pricing.

36 MO Vertically integrated inference stack sold beneath the application layer.

Alpha

01Hiring pattern suggests a transition into custom inference infrastructure, not model fine-tuning.

02unpublished but directly relevant prior IP held by the CTO, filed under a previous employer's assignment.

03independently hiring around the product — three early customers posting integration roles that name it explicitly.

0412–18 months behind is where the nearest competitors sit on the public benchmark artifacts.

Hidden risks

01Patent ownership may overlap with a founder's former employer ; assignment is not recorded with USPTO.

0261% of revenue appears attributable to two customers, both in one vertical.

03Pricing may not support inference costs at scale under the observed usage mix.

US context — activated

Delaware C-corp, two Form D filings. No sector regulator gates the product today; export-control exposure becomes live if the runtime ships as on-prem hardware. Talent is the binding constraint — compiler engineers in the Bay Area are the scarcest input in the plan.

Then it writes the questions.

Not generic diligence prompts. Questions that target the specific uncertainty the evidence produced — which is also how a founder can tell whether you did the work.

What's your go-to-market strategy?Your customer announcements imply considerably faster enterprise adoption than your sales hiring would suggest. Are deployments primarily founder-led, channel-led, or expanding organically inside existing accounts?

Do you have IP?The memory-scheduling application lists an inventor who was employed elsewhere at the filing date, and we can't find a recorded assignment. How was that handled?

How do you think about pricing?Pricing moved from per-call to per-compute-unit in March. Was that a margin correction, or the first step toward selling the runtime itself?

04.2 — Δ thesis

Research never closes.

New employee, new patent, new customer, new lawsuit, new round, new competitor, new release, new paper, founder departure, pricing change. Each pass recomputes the thesis and diffs it against the last.

Investors are not notified that research ran. They are notified when the number moved and why — which is the only alert anyone keeps subscribed to. The engine stores every delta with its detection timestamp, so a prediction that held is as legible as one that failed.

Material change detected · 14 days

72→84

The enterprise expansion predicted in April appears to be occurring. Four Fortune 500 customer signals detected, three of them from job postings naming the product. Technical-moat score unchanged; distribution revised upward.

Prediction heldIP flag still open