Your thesis as rules, with the reason behind every result
Publish your thesis as rules and preferences. Each deal shows met, mismatch or unknown per criterion, with the source, and unknowns become questions.
- Rubrics, fit reports, overrides Pilot
- Thesis sandbox Later
- Thesis fit
- 2 of 3 rules met
- 1 unknown, 0 mismatches
- Evidence quality
- Mostly founder reported
- Revenue not yet backed by a document
- Diligence readiness
- 2 items missing
- Round size, revenue schedule
Hard requirements
- StageRule: Pre-seed or seedSlide 1Met
Found: Seed
The cover says "Seed round".
- GeographyRule: India or United StatesSlide 9Met
Found: Pune and Austin
Team slide lists both offices.
- Round sizeRule: USD 1M to 3MUnknown
Found: Not in the deck
No amount on the use of funds slide. Added to the interview as a question.
Four kinds of rule, written the way your partners talk
A thesis in a memo is hard to apply the same way to every deck. As a rubric, each part becomes something Delsivo can check and explain.
| Kind | Example | How it is applied |
|---|---|---|
| Hard requirements | Stage is pre-seed or seed. Sector is B2B software. Headquarters in India or the United States. Round size USD 1M to 3M. | Checked in code against extracted facts: Met, Mismatch or Unknown |
| Weighted preferences | Evidence of paying customers (weight 3). Founder has sold to this buyer before (weight 2). Capital-efficient growth (weight 1). | Scored per criterion by the model with a rationale and citations; weights and the total are computed in code |
| Exceptions | Pre-revenue companies are fine if the founders have built in this market before. | Shown beside the rule it relaxes, so the analyst sees why a rule was not applied |
| Questions | If round size is unknown, ask it in the interview. Ask how customers are counted. | Turn Unknowns into questions for the interview or a request |
The examples are a made-up seed fund's rubric. You publish your rubric as a version; drafts never touch live deals.
Three results, and Unknown is never a fail
Each criterion shows the rule, the value found, the source and the reasoning. The result is a suggestion until an analyst accepts it.
- Met
The deal satisfies the rule, with the source shown.
- Mismatch
The deal explicitly falls outside the rule. Surfaced for an analyst to review, never an automatic rejection.
- Unknown
The value isn't known yet. It turns into a question for the founder, not a failure.
Unknown has a dashed border and is never red, in every screen. A missing value is a question for the founder, not evidence against them.
Fit, evidence and readiness, never blended into one number
A single score hides the reason. Delsivo keeps the three questions a partner actually asks apart.
Thesis fit
How the deal matches your published rubric: hard rules met, mismatched or unknown, plus weighted preferences.
Evidence quality
How much of what matters is backed by a cited source rather than a bare statement.
Diligence readiness
What is still missing before the deal can move to the next stage.
Analysts can change any result, with a reason on record
Extraction can be wrong and exceptions are real. An override keeps the original result, the new one, who changed it and why.
- Thesis fit
- 3 of 4 rules met
- 1 unknown, 1 override with a reason
- Evidence quality
- Improving
- Paid contracts corroborated
- Diligence readiness
- 1 item missing
- Round size
Hard requirements
- StageRule: Pre-seed or seedSlide 1Met
Found: Seed
The cover says "Seed round".
- SectorRule: B2B softwareSlide 2Met
Found: Freight logistics
The problem slide describes freight brokers, read as a logistics company.
SOSam Okafor changed Mismatch to Met: It sells software subscriptions to brokers, not freight services. See slide 6.
- GeographyRule: India or United StatesSlide 9Met
Found: Pune and Austin
Team slide lists both offices.
- Round sizeRule: USD 1M to 3MUnknown
Found: Not in the deck
No amount on the ask slide. Added to the interview.
Weighted preferences
- Paying customersRule: Weight 307:42Met
Found: 8 paid, 12 unpaid pilots
Paid contracts confirmed in the room; pilots not yet converted.
- Prior sales to brokersRule: Weight 2Unknown
Found: Not stated
Neither the deck nor the form says. Suggested question for the first meeting.
Sample data. The fund, founder, company and numbers are made up.
Overrides need a reason
No silent edits. The reason is stored with the user and time, and shows on the criterion.
Re-runs keep the old run
Re-screening creates a new assessment. The earlier one stays, with its rubric version and evidence snapshot.
Rubric versions
Every run records which rubric version, model and prompt version produced it.
Totals computed in code
Weights are applied by the product, not the model, and unscored criteria are left out of the total.
What AI suggests, and what your team decides
Screening is where it is most tempting to let a model decide. Delsivo doesn't.
AI suggests, with a citation
- Reads facts from the deck, each with its slide
- Applies your hard rules and reports Met, Mismatch or Unknown
- Scores weighted preferences with a rationale and citations
- Suggests questions for every Unknown
- Explains each result in a sentence an analyst can check
People decide, with a reason
- Which rubric version is published
- Accepting or overriding each result, with a reason
- Every pass, always with a recorded reason
- Every stage change and every message to a founder
- Whether a Mismatch is a real no or an exception
One fund, more than one way to invest
Different stages and programmes need different rules.
- Pilot
Rubrics per strategy
Keep a pre-seed rubric and a seed rubric, or one per programme, each with its own versions and questions.
- Pilot
Strategic criteria
Corporate venture teams can add strategic fit criteria beside financial ones, shown as their own group.
- Later
Thesis sandbox
Try a rubric change against past deals before publishing it, and see which results would change.
Questions about screening
Test your rubric on deals you already know
In a pilot we write your thesis as a rubric with you, run it on past deals and compare the results with your team's own calls.