Nascens

The Atlas Venture Spine

Seven stages. One signal.

The AVS converts a free-text pitch, a mentor session transcript, or a structured intake form into a graded signal vector, an opportunity state, a trajectory estimate, and actionable guidance — consistently, at every interaction, across your entire venture population.

The spine

01
Intake
Raw venture input
02
Normalize
LLM → signal vector
03
Opportunity State
7-state taxonomy
04
Stage Transition
Advance or flag
05
Trajectory Estimate
Population match
06
Guidance
Ranked interventions
07
Track
Outcome + compound

Stage 01

Intake — Any format, one entry point.

Nascens accepts venture input in three forms: a structured intake questionnaire, a free-text pitch submission (deck summary, elevator pitch, or written application), or a session transcript from a mentor conversation. All three enter the same normalization pipeline.

Form intake

Structured questionnaire

Founder completes a standardized intake form. Fields map directly to the 6-dimension signal vector. Fastest normalization path.

Text intake

Free-text submission

Pitch deck summary, written application, or elevator pitch. LLM extraction required. Confidence labeled accordingly.

Session intake

Mentor session capture

Transcript or structured notes from a mentor meeting. Nascens extracts signal updates and flags state transitions since the last session.

Stage 02

Normalize — Free-text becomes signal.

The normalization stage is the gate through which all downstream processing flows. The LLM layer (Anthropic claude-sonnet-5) extracts a score on each of the six dimensions from whatever input arrives. No normalization means no downstream operation.

Dimension What is scored Confidence basis
Problem Clarity, specificity, and evidence of problem existence LLM-derived (novel) until population comparable found
Solution Technical feasibility, differentiation, and solution-problem fit LLM-derived (novel)
Market Size evidence, timing, and accessibility LLM-derived (novel)
Team Domain expertise, execution evidence, and team completeness LLM-derived (novel)
Evidence Customer discovery, pilots, LOIs, and validation artifacts LLM-derived (novel) → comparable as population grows
Assumptions Explicit vs. implicit assumption load; riskiest assumption rank LLM-derived (novel)
Confidence labeling is non-negotiable. Every normalized score carries a confidence tier: novel (LLM-derived, no comparable), comparable (matched to population cohort), or established (empirically grounded, 20+ comparable transitions). Guidance based on LLM first-principles is never presented as empirically validated.

Stages 03–04

State and Transition — Where is this venture, and where is it going?

The normalized signal vector maps to one of seven opportunity states. State determines what transition probability applies and which interventions are queued.

Forming

Idea stage. Problem signal present; solution and team signals weak. Focus: problem definition and team formation.

Validating

Solution and market signals strengthening. Evidence dimension is the binding constraint. Focus: customer discovery and pilot design.

Progressing

Balanced signal across all six dimensions. Evidence score above threshold. Trajectory toward investment-readiness.

Stalling

Previously progressing; signal degraded or stagnant across two or more sessions. Intervention required.

Pivoting

Significant shift in Problem or Solution dimension signals. Normal and healthy — tracked and labeled explicitly.

Terminal

Venture concluded — graduated, pivoted out of program, or dissolved. Enters population record. Outcome tracked.

Stages 05–06

Trajectory and Guidance — What happens next, and what should the mentor do?

Trajectory estimation compares the current venture's signal vector against the anonymized population record. For novel ventures (no comparable yet), trajectory is LLM-derived. For comparable ventures, it references empirical transition records.

Guidance is the terminal output of the spine: a ranked list of interventions, labeled by the dimension most likely to unlock the next state transition. The guidance is addressed to the program director, the assigned mentor, and the founder — three different framings of the same signal.

Illustrative guidance output · EduTech startup · Cohort 4 · Session 2

Binding constraint: Evidence (0.29)

The venture has strong Problem signal (0.82) and acceptable Market signal (0.71). The binding constraint on a Validating → Progressing transition is Evidence. Team is a secondary constraint (0.44).


Ranked interventions

  1. Design a 2-week customer discovery sprint. Target 15 structured conversations with the identified buyer persona. Produce a written synthesis. This is the fastest path to an Evidence score above 0.50. (novel)
  2. Address team completeness — technical co-founder or advisor. Team signal will become a co-binding constraint at the next stage. Identify and engage a technical advisor within this cohort period. (novel)
  3. Do not over-invest in solution development at this stage. Solution signal (0.58) is sufficient for the current state. Further development ahead of customer validation adds assumption load without Evidence benefit. (novel)

Stage 07

Track — Every outcome feeds the population.

The final stage closes the loop. When a venture concludes — graduated, dissolved, pivoted out, or invested — its full signal history and outcome are committed to the anonymized population record. Every record makes the next trajectory estimate more grounded.

For the program director

Cohort-level signal summaries. Which ventures are stalling and why. Which mentor assignments are producing the best transitions. Year-over-year population trends for funder reporting.

For the mentor

Per-venture signal history. Which of their interventions correlated with positive state transitions. Matching signal — which ventures in the current cohort map to their domain expertise.

See the full architecture.

The AVS runs on a spine proven across four domains. The technical details are documented for design partners.