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
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) |
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.
Idea stage. Problem signal present; solution and team signals weak. Focus: problem definition and team formation.
Solution and market signals strengthening. Evidence dimension is the binding constraint. Focus: customer discovery and pilot design.
Balanced signal across all six dimensions. Evidence score above threshold. Trajectory toward investment-readiness.
Previously progressing; signal degraded or stagnant across two or more sessions. Intervention required.
Significant shift in Problem or Solution dimension signals. Normal and healthy — tracked and labeled explicitly.
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
- 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)
- 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)
- 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.