Key takeaways
- Hard rules are yes-or-no checks against facts. Weighted preferences need judgement, so a reviewer scores them from 0 to 10.
- A fact missing from the deck is Unknown, not a fail. It becomes a question for the founder.
- A mismatch is flagged for a person to review against your exceptions. It is never an automatic rejection.
- Weights are scaled to add up to 100, and preferences nobody could score yet are left out of the weighted score instead of counting as zero.
How to use the investment thesis builder
- 1Name the fund or strategy and pick the currency for round sizes. A sample thesis for a fictional fund is filled in to start from.
- 2Switch on the hard rules you hold to, and tick the stages, sectors and geographies in your mandate. Set a round size range, or leave one end open.
- 3Weight each preference from 0 to 10, or add up to three of your own. A weight of 0 leaves a preference out.
- 4Write your exceptions and any extra questions, one per line. A question for each hard rule is added for you.
- 5Change the sample company's facts to see each rule read Met, Mismatch or Unknown, score a few preferences, then copy the rubric as Markdown or JSON.
The four parts of a thesis rubric
| Part | What it is | Example | How it is applied |
|---|---|---|---|
| Hard rule | A requirement you can check against a fact | Stage is Pre-seed or Seed | Met, Mismatch or Unknown, per company |
| Weighted preference | Something you favour that needs judgement | Founders have worked on this problem before (weight 8) | Scored 0 to 10 by a reviewer, with the evidence for the score |
| Exception | When you would look past a hard rule | A partner may sponsor a company outside our sectors | Read by a person when a rule shows Mismatch |
| Question | What to ask when a fact is missing | How much are you raising, and on what instrument? | Sent when a rule shows Unknown |
Why Unknown is not a fail
Early decks leave things out. A pre-seed deck may not state the round size or where the team will be based, and that says little about the company. If a missing fact counted as a fail, the rubric would reject companies for what they didn't mention rather than for what they are.
So a rule with no fact to check reads Unknown, and the builder turns it into a question. A rule that clearly doesn't match reads Mismatch, and that is flagged for a person to check against the exceptions, not rejected automatically. People decide what happens next.
How the preference weights work
Weights are scaled so they add up to 100: each weight is divided by the total of all weights. The sample's six preferences have weights 8, 9, 6, 5, 6 and 4, a total of 38, so customer demand (weight 9) counts for 9 ÷ 38, or 23.68%.
The weighted score is the sum of each score times its weight, divided by the total weight of the preferences that were scored. If a reviewer scores only founder experience (7) and customer demand (4), the score is (7 × 8 + 4 × 9) ÷ (8 + 9) = 5.41 out of 10, from 2 of 6 preferences. The rest are left out instead of counting as zero, and the tool shows how much of the weight was scored.
| Preference | Weight | Share of 100 |
|---|---|---|
| Evidence of customer demand, paid or committed | 9 | 23.68 |
| Founders have worked on this problem before | 8 | 21.05 |
| Bottom-up market large enough for a venture outcome | 6 | 15.79 |
| Hard-to-copy advantage (data, distribution or technology) | 6 | 15.79 |
| Capital-efficient path to the next milestone | 5 | 13.16 |
| A clear first sales channel | 4 | 10.53 |
Writing a thesis that screens consistently
- Keep hard rules to what you would really never invest outside of. Everything else is a preference.
- If one preference carries most of the weight, it is probably a hard rule. The builder warns you when one reaches 60 of the 100.
- Describe what earns a high and a low score for each preference, so two reviewers reading the same deck land close together.
- Write exceptions down. An exception that lives only in a partner's head is applied to some companies and not others.
- Give the rubric a version and a date, and keep the old version when you change it, so earlier decisions can be read against the rules of the time.
Thesis rubrics in Delsivo
Delsivo's thesis screening is planned for the pilot and uses the same four parts. By design, a fund publishes its rubric as a version, hard rules run in code against facts taken from the deck with slide citations, and each criterion shows the rule, the observed value, the rationale and the source. AI suggests preference scores with citations; analysts accept, change or override them, and an override needs a reason. Thesis fit, evidence quality and diligence readiness are shown separately.
There is no import for the builder's JSON yet. Use it as a record of your thesis, or as the brief when you set up any screening tool. See how thesis screening works.