Deep Science Ventures · Assessment Pipeline
Deep Science Ventures · Assessment Pipeline
Four sequential assessment stages — from broad outcome screening to a single-company decision · v1 · 2026-06-17
To make an investment decision before the venture exists, we assess what can be known and build conviction in layers — de-risking the opportunity long before any company is formed. We start from outcomes that matter in the world (these exist independently of any company), then progressively test ever sharper questions: is this a venture-grade gap? can a specific approach close it? — and only then proactively complete the team to execute it. Each of the four assessments is narrower and deeper than the last, killing weak candidates cheaply. The team is built into the process: a Founder-in-Residence is hired once an Opportunity Area is chosen and co-develops the approach, and co-founder(s) are recruited once an approach is selected.
Unit of analysis shifts each stage: Stage 0 scores an outcome leaf; Stage 1 decomposes a surviving outcome into pathways and scores each optimistically then pessimistically; Stage 2 scores one chosen approach in a single pass; Stage 3 scores one company executing that approach. Indicators recur and deepen — the Indicator matrix tab shows exactly which indicators are assessed at which stage, and the Indicator detail by stage tab gives the full 0–3 rubric for every node.
Full 0–3 scoring rubric for every node, organised by category. Switch stages below.
Rows are the seven assessment dimensions; columns are the four stages. Each cell lists the indicators scored for that dimension at that stage, with the category aggregation rule beneath. Empty cells mean the dimension is not yet (or no longer) assessed at that stage. Read across a row to see how a single dimension deepens as the funnel narrows.
Source rubrics: Pre-screen — dsv-oa-review.pages.dev (Scoring guide tab, rubric v2.1);
Opportunity Area — oa-evaluator-v19/shared/scoring-reference.md;
Approach — ap_evaluator_design_v1.md / ap-pipeline-v1.html;
Company — framework provided 2026-06-17 (Series A valuation model).
Every indicator (rows, grouped by dimension) against the four stages. A filled cell means the indicator is scored at that stage, coloured by stage. Read across a row to see an indicator’s coverage; read down a column to see a stage’s full indicator set.
The four venture questions drawn as left-to-right evaluation trees (mirroring the Approach-evaluator schematic). Each baseline branches into categories, then into a Possibility branch (is the opportunity real and big enough?) and, where relevant, a Competency branch (can the team deliver it?). Operator tags show how leaf nodes combine. Click any leaf for its rubric; use the filter to dim indicators a given stage doesn’t assess. Sales (DSV) and DSV fit sit outside these four core questions and appear only in the heatmap above.
The same indicators regrouped around the Series-A outcome structure. Founder baseline and Impact stand alone; then three outcomes each pair a Possibility branch (is it achievable?) with a Competency branch (can the team deliver it?) — Outcome 1 · Commercial (Economic Upside + Commercial tractability), Outcome 2 · Technical (Neglect + Technical tractability), and Outcome 3 · Funding. Click any leaf for its rubric; use the filter to dim indicators a stage doesn’t assess.
How the final Investment Committee rates a company: each Series-A outcome gets a Possibility and a Competency rating on the 1–4 SOL ladder, computed from the 0–3 indicator scores via gate conditions then category aggregation. The derived formulas roll these into per-outcome probabilities and the in-house valuation. Source: Notion Rating Template ‘26 (Portfolio Ratings Entries DB) · transcribed 2026-07-23.
The 1–4 IC ladders are a band-for-band relabel of the 0–3 indicator scale — the translation is a pure
+1 shift (rating = indicator band + 1): 0→1 CON · 1→2 HCON · 2→3 HSOL · 3→4 SOL.
Gate conditions below are expressed on the 0–3 indicator scale; outcome ratings are stated on the 1–4 ladder.
If ANY of the following are true, Commercial = CON (1). Stop here.
| Indicator | Gate condition |
|---|---|
| 2a — Market size (TAM) | Score ≤ 1 |
| 2b — Revenue potential | Score ≤ 1 |
| 2c — Profit potential (NOAK) | Score ≤ 1 |
| 2d — Scalability ceiling | Score ≤ 1 |
| 3a — Buyer motivation | Score = 0 |
| 3b — Buyer ability | Score = 0 |
| 3c — Startup sellability | Score = 0 |
| 3e — Customer validation | Score ≤ 1 |
If no gate fired, calculate:
Final score = average of Cat 2 and Cat 3
If ANY of the following are true, Commercial = CON (1). Stop here.
| Indicator | Gate condition |
|---|---|
| 3a — Buyer motivation | Score = 0 |
| 3b — Buyer ability | Score = 0 |
| 3c — Startup sellability | Score = 0 |
| 3e — Customer validation | Score = 0 Suggested: Score ≤ 1 (1 LOI or MOUs/non-binding interest; <20% of forecast; vague terms) |
If no gate fired, calculate:
Apply weighted average for final score = (Cat1 × 0.25) + (Cat2 × 0.25) + (Cat3 × 0.50)
If ANY of the following are true, Technical = CON (1). Stop here.
| Indicator | Gate condition |
|---|---|
| 4a — Scientific & engineering feasibility | Score = 0 |
| 4b — Development time | Score = 0 |
| 4c — Manufacturability | Score = 0 |
| 4d — Expert validation | Score = 0 |
| 5c — Defensibility | Score = 0 |
If no gate fired:
Final score = average of Cat 4 and Cat 5
If ANY of the following are true, Fundraising = CON (1). Stop here.
| Indicator | Gate condition |
|---|---|
| 6d — Investor feedback | Score ≤ 1 |
| 6e — Capital efficiency | Score ≤ 1 |
If no gate fired:
Final score = Cat 6
Final score = Cat 6 (no gates in v.1)
Transcribed from the Portfolio Ratings Entries database schema (data source 456580d0…79dcc).
| Section | Fields & allowed values |
|---|---|
| DSV fit | DSV Capability — High / Low · DSV Influence — High / Low |
| 🌳 Impact (gate) | Impact score — Pass / Fail / NA. The highest single impact score across 1a–1f wins (+ justification); score only the dimensions relevant to the approach, mark the rest N/A |
| 👫 Founder baseline | Founders 1–3, each: indicator Pass / Fail / NA + evidence summary on Magnetism (concise storytelling with empathy, track record of influence) · Determination (bias for action under uncertainty, pace of execution) · Resilience (cooperation with ownership & empathy, self-awareness & active growth) — all assessed in-role |
| Outcomes 1–3 | Per outcome (📈 1 Commercial · 🧪 2 Technical · 🤑 3 Fundraising): Possibility Score (1–4 SOL ladder) · Competency Score (1–4) · Evidence Confidence — High / Low · Commentary (specific outcome to Series A + evidence summaries for Possibility and Competency) · Probability (formula). Outcome 4 deprecated 2026-07-23 — properties remain in the schema but are no longer filled in |
| Risk & Speed grid | Six dimensions — Market · Value Capture · Tech · Neglect · Team · Funding — each rated Risk (High / Med / Low) and Speed (High / Med / Low), with commentary per dimension |
| ⚠️ Validation | Outcomes & Valuation confidence — Low: Guesstimate 🤷 · Med: Comps OR funder · High: Funder AND Comps |
| 💰 Valuation | Target Series A valuation (£) · Valuation calculation narrative (benchmark to sector → comparators’ Series A valuations → Seed/Series-A investor perspectives) |
| Evidence links | Slide deck · TEA · TEAs/LOIs · CRM · FTO/IP strategy · Technology Development Roadmap · Hiring plan · Fundraising plan · Unit economics & scaling plan · Outcomes-to-Series-A table |
| Period tracking | Investment raised in this period (£) · Grants secured in this period (£) · Number of employees (FTE) |
Formula properties on the rating record, in dependency order. Expressions were transcribed from Notion’s formula editor on 2026-07-23 (the API does not export them); rows still marked ⚠ await transcription. The full In-house-valuation chain is drawn in the calculation-flow panel below.
| Formula property | Role & definition |
|---|---|
O1–O4: P# / O1–O4: C# (O4 deprecated) | ✓ toNumber(substring(prop("… Score"), 0, 1)) — leading digit of the selected 1–4 option label |
Outcome 1–4: Probability (O4 deprecated) | ✓ if(empty(Possibility) or empty(Competency), 1, round(1/(1+exp(−0.9×P#))² × C#/4, 2dp)) — squared sigmoid on Possibility × linear Competency factor; unrated outcome → 1 (neutral); Evidence Confidence not used |
Average possibility / Average competency | ✓ round((ΣP#) / (4 − count of empty P#s) ×10)/10 and round((ΣC#) / (4 − count of empty C#s) ×10)/10 — mean of the non-empty outcomes’ scores, 1 dp |
Probability of solving all Outcomes | ✓ round(P(O1)×P(O2)×P(O3)×P(O4) ×100)/100; forced to 0 when Outcome 1’s probability is empty or Portfolio Category contains “Please” (data-completeness gate) |
Overall Confidence (previously “risk mitigation”) / Overall Speed | ⚠ Rollups of the six-dimension Risk / Speed grid |
In-house valuation | ✓ Most recent valuation + (💰 Target Series A valuation − Most recent valuation) × Probability of solving all Outcomes |
Confidence weighted valuation | ✓ {Low → 0.33 · Med → 0.66 · High → 1 · unset → 0} × In-house valuation — the ⚠️ Validation tier applied as a discount multiplier |
Inception or Project Valuation | ✓ Stage switch on the entry title: contains “IC” → In-house valuation; contains “Project” → In-house valuation ÷ 2 (project-stage haircut); otherwise falls back to Number of employees (FTE) — likely a placeholder branch |
Portfolio Category | ✓ Portfolio triage (only for entries titled “23”/“24”): ⭐ BEACON/Follow-on (AvgComp ≥ 3.33 & AvgPoss ≥ 3) · 🙈 LEAVE (Influence Low, or AvgComp < 2.66 & AvgPoss < 3) · 🦾 SUPPORT (AvgComp ≥ 2.66 or AvgPoss ≥ 3, Influence & Capability High) · 🛠 Improve DSV Capability · 🤓 “Please complete…” (the P(all) gate). Non-vintage entries → “” — no completeness gate for current entries (audit №1) |
Total Score Bar | ⚠ Presentation layer over the metrics above |
Every step and input from the raw IC ratings down to the valuation figure. All formulas in the chain are
transcribed verbatim from the Notion formula properties (2026-07-23) — verified for Outcomes 1–3; Outcome 4’s
Probability and P#/C# are assumed identical (pattern-confirmed). Only the Portfolio Category
expression remains untranscribed — and it matters: see the equation audit below.
The model is an expected-value interpolation: the company is worth its last marked valuation,
plus the step-up to the Series A target weighted by the probability that all outcomes are solved.
toNumber(substring(prop("… Score"), 0, 1))Average possibility =
round(( O1:P# + O2:P# + O3:P# + O4:P# )
/ (4 − count of empty P#s) ×10)/10
Average competency =
round(( O1:C# + O2:C# + O3:C# + O4:C# )
/ (4 − count of empty C#s) ×10)/10if(empty(Possibility Score) or empty(Competency Score), 1,
round( 1/(1+exp(−0.9 × P#))² × C#/4 ×100)/100)| P\C | C1 | C2 | C3 | C4 |
| P1 CON | 0.13 | 0.25 | 0.38 | 0.51 |
| P2 HCon | 0.18 | 0.37 | 0.55 | 0.74 |
| P3 HSol | 0.22 | 0.44 | 0.66 | 0.88 |
| P4 Sol | 0.24 | 0.47 | 0.71 | 0.95 |
only if title contains "23"/"24":
AvgC≥3.33 & AvgP≥3 → ⭐ BEACON/Follow-on
Infl=Low or (AvgC<2.66 & AvgP<3) → 🙈 LEAVE
(AvgC≥2.66 or AvgP≥3) & Infl+Cap=High → 🦾 SUPPORT
Cap=Low → 🛠 Improve DSV Capability
else → 🤓 "Please complete…"
otherwise → "" (no gate)toNumber(if(empty(prop("📈 Outcome 1: Probability")), 0,
if(contains(prop("Portfolio Category"), "Please"), 0,
round(((prop("📈 Outcome 1: Probability")
× prop("🧪 Outcome 2: Probability"))
× prop("🤑 Outcome 3: Probability"))
× prop("Outcome 4: Probability") ×100)/100)))prop("Most recent valuation")
+ (prop("💰 Target Series A valuation")
− prop("Most recent valuation"))
× prop("Probability of solving all Outcomes"){ Low → 0.33 · Med → 0.66 · High → 1 · unset → 0 }
× prop("In-house valuation")title contains "IC" → In-house valuation
title contains "Project" → In-house valuation ÷ 2
otherwise → Number of employees (FTE)Worked example: Most recent = £2M, Target = £20M. Ratings: O1 Commercial P3 HSol/C3 → 0.66; O2 Technical P4 Sol/C4 → 0.95; O3 Fundraising P2 HCon/C3 → 0.55; O4 deprecated/unrated → 1. P(all) = round(0.66 × 0.95 × 0.55 × 1, 2dp) = 0.34 → In-house valuation = 2M + (20M − 2M) × 0.34 = £8.12M. Structural properties worth knowing: (1) individual outcome probabilities never reach 0 — the floor is 0.13 (P1×C1) and the ceiling 0.95 (never certainty) — though the 2dp rounding of the product can hit 0.00 for uniformly weak companies; hence the PF+SA note proposing a 0-based possibility scale to allow hard-zeroing. (2) P(all) is dominated by the weakest outcome because probabilities multiply. (3) A company at P(all) = 0 is still worth its current valuation, not £0 — the interpolation floors at Most recent valuation. (4) The sigmoid compresses the top: moving Possibility 3→4 adds less probability (+0.07 at C4) than moving Competency 3→4 (+0.24 at P4) — team upgrades move the valuation more than possibility upgrades in the upper bands.
Found while transcribing the formulas (2026-07-23). Ordered by consequence — the first three interact.
| Gap | Detail & consequence |
|---|---|
| 1 · CONFIRMED — current entries have no completeness gate | Portfolio Category is the only effective completeness gate in P(all) (the empty(Outcome 1: Probability) branch can never fire — a formula property always returns a value; unrated outcomes return 1, not empty). The transcribed expression shows the “🤓 Please complete…” branch exists only inside the title contains "23"/"24" vintage branch; all other entries return “”, which never matches the gate. Consequence: a 2025/26 entry with missing ratings is valued as if the unrated outcomes were certain — an entirely unrated entry values at the full Series A target. Fix: move the completeness check out of the vintage branch (or add an explicit unrated-check to P(all)). |
| 2 · Unrated = certain | An unrated outcome contributes probability 1 to the product — missing ratings raise P(all) and the valuation. Deliberate so the optional Outcome 4 doesn’t drag, but it means an entry rated on only one outcome values as if the other outcomes were solved — unless gate 1 catches it. |
| 3 · Outcome 4 — RESOLVED: deprecated | Confirmed no longer required (2026-07-23). Its properties remain in the schema and P(all) still multiplies Outcome 4: Probability — harmless while unrated (returns 1, neutral), but a rated O4 would still move the valuation. Hygiene fix when convenient: drop the O4 term from P(all) and archive the O4 properties. |
| 4 · Most recent valuation basis undefined | Sum-rollup over the Company relation: an empty relation yields £0 (valuation collapses to Target × P(all)); linking two company pages would sum their valuations. The underlying definition — pre- vs post-money (Target Series A is pre-money per the Company framework), which round, which currency — is stated nowhere. Mixed bases skew the interpolation. |
| 5 · Validation unset → £0 | Confidence weighted valuation falls to its else-branch (× 0) when the ⚠️ Validation tier is not yet set — a new entry reports a zero confidence-weighted valuation rather than “n/a”. |
| 6 · FTE fallback | Inception or Project Valuation reports headcount as a £ valuation when the entry title contains neither “IC” nor “Project” — a title-hygiene dependency with a nonsensical fallback unit. |
| 7 · The 0.9 constant is undocumented | The sigmoid steepness (−0.9) is what fixes the possibility factor at 0.51 / 0.74 / 0.88 / 0.95 — no definition of its provenance or calibration exists in the template. Changing it re-prices every company. |
| 8 · Qualitative layer mostly has no formula role | Impact Pass/Fail, Founder 1–3 Pass/Fail, Evidence Confidence, the Risk & Speed grid and Overall Recommendation never enter any formula — an Impact or Founder Fail leaves the valuation untouched. Presumably human IC gates; worth confirming that is the intent (cf. gate tables above, which do hard-zero on indicators). Exception: DSV Capability/Influence do feed Portfolio Category (the ⭐/🙈/🦾 triage) — but only for “23”/“24”-vintage entries, per gap №1. |
| 9 · Double rounding | Per-outcome probabilities are rounded to 2 dp before the product is rounded again to 2 dp — a small compounding artefact, noted for completeness. |
The Commercial Cat 3 rule keeps “unless 3a or 3c = 1 (either dominates)” — but OA-stage v19 removed the 3c=1 collapse (weak Startup Sellability now averages in; only 3a=1 dominates). Either the IC logic inherits the v19 rationale, or final IC deliberately stays stricter — currently the two documents disagree. Note 3d (Revenue Ramp) and 3e (Customer Validation) are Approach-stage nodes, consistent with Company inheriting the Approach node set.
A visual read of one company’s IC assessment — valuations and status at the top, each outcome’s Possibility × Competency rating with its highlights and open risks, the milestones to Series A, and the indicator scores with the aggregation logic that produces each rating. Open an entry from the library below, load the New assessment form, or paste a saved JSON payload.
Stored in this browser (independent of Notion). New entries arrive via the New assessment form's Save to library; use Export for backup or to share with a colleague, Import to load their file.
Fill in the IC rating inputs; the verified formula chain computes all derived scores live (identical maths to the
Notion template, including per-step rounding). Save to Notion creates the entry in the Portfolio
Ratings Entries database — available when this dashboard is served with its Cloudflare Pages Function
(/api/assessment); from a local file, use Copy JSON and hand the payload to Claude Code.