Pattern Library and Recognition Test
Companies do not fail suddenly. They change shape first. Vivameda derived ten structural patterns from 48 million observed company years between 1950 and 2020, then applied them to contemporary companies with known outcomes. Ten of ten were classified correctly, with 10 to 28 months of lead time on the collapses.
10 of 10
Contemporary companies classified correctly
10 to 28
Months of lead time before each collapse
11,003
Company cohort behind the seed stage signals
Why patterns work
Financial disclosure is periodic, lagging and curated. Workforce structure is continuous. Hiring freezes, functional consolidation, retention breakdown and debt financed expansion all show up in the shape of an organisation long before they appear in a filing. Vivameda measures that shape across seventy years, which makes it possible to say what a given trajectory has historically preceded.
The patterns below were derived from data through 2020. Every validation case resolved after that date, so the test is forward looking by construction rather than by holdout design.
The library
Each pattern is a distinct structural signature observable in workforce data. Select a pattern to see its definition, typical growth band and historical cases.
Sustained moderate positive growth. Three or more positive growth years in a rolling five year window, with no contraction events in the prior three years.
Recognition test, April 2026
Five recent corporate collapses and five active scalers, classified using only publicly available workforce data and patterns derived before any of these outcomes occurred. None of the companies were in the panel at the time of their distress or scaling event.
| Company | Known outcome | Classification | Lead time |
|---|---|---|---|
| Saks Global | Chapter 11, January 2026 | Concerning plus Over Expansion | 18 months |
| Genesis Healthcare | Chapter 11, July 2025 | Permanent Distress | 14 years |
| Nikola | Chapter 11, February 2025 | Concerning Trajectory | 28 months |
| Joann Fabrics | Chapter 11, January 2025 | Permanent Distress | 10 months |
| Claire's | Chapter 11, August 2025 | Concerning Trajectory | 24 months |
| Anthropic | Scaling, 30B run rate | Verified Healthy | Ongoing |
| NVIDIA | Scaling, 100B revenue | Verified Healthy | Ongoing |
| Meta Platforms | Scaling post reset | Verified Healthy | Ongoing |
| Palantir | Scaling with 2024 blip | Verified Healthy | Ongoing |
| Databricks | Scaling, 3.7B revenue | Verified Healthy | Ongoing |
Pattern reading
Pattern 4 variant through debt financed M&A at an already stressed mature company, Pattern 9 slow bleed layoffs that stayed below traditional contraction thresholds, and Pattern 6 in the corporate family with the Hudson's Bay filing as leading indicator.
Classification: Concerning Trajectory plus Over Expansion
Verified Healthy at the extreme end
The 2023 growth rate would superficially match Pattern 7. The library separates the two because revenue tracks headcount with accelerating multiples. Genuine scaling, not over expansion.
Verified Healthy
Clean Pattern 1. Sustained positive growth, no contraction signals, revenue acceleration matching headcount expansion.
Verified Healthy, moderate expansion
The 2022 reset has cleared the pattern window. Current behaviour is sustained moderate positive growth driven by infrastructure and AI investment.
Verified Healthy with a single year blip
The 2024 contraction is a single year deviation, not two years in a rolling window, so it does not trigger Pattern 2. Revenue growth of 28.79 percent in 2024 confirms the reading.
Verified Healthy with a strong scaling signature
Pattern 1 with revenue growth tracking headcount expansion throughout the scaling window.
Retail, healthcare, electric vehicles, AI infrastructure, semiconductors, social media and enterprise software. The same library classified all of them, so the patterns are not industry specific.
From a 4,000 employee mid cap to a 76,000 employee giant to a 27,000 employee private equity owned chain. Signal shape does not depend on scale.
Anthropic's 600 percent growth reads as healthy because revenue tracks headcount. Enron's 25 percent read as over expansion because it did not. Naive classifiers miss that distinction.
Startup pattern library, version 1
Derived from 11,003 companies first observed in 2010 or later with year 1 headcount between 5 and 100 employees and at least three years of subsequent observation. The cohort outcome distribution matches real startup mortality: 94.7 percent never crossed 100 employees, 2.7 percent reached 200 or more. Patterns were defined on data through 2020 and tested against outcomes that materialised between 2021 and 2026.
Headline finding
Scaling startups look like companies on day one and retain their early team into year three. Failures look like projects on day one and either churn through their original team or stagnate with founders alone. The strongest single signal is first hire stickiness: by year 3, scalers retain 54 percent more of their original year 1 cohort than failures.
Profile
At year 1, the workforce shows six or more distinct role buckets, with no single role above 50 percent of headcount. Engineering, operations, sales, marketing, finance, customer, executive, HR and specialist all present in some balance.
| Role buckets at year 1 | Failure rate | Scaler rate | Scaler lift |
|---|---|---|---|
| 1 to 2 roles | 9 percent | 1 percent | 0.11x |
| 3 to 4 roles | 14 percent | 5 percent | 0.36x |
| 5 to 6 roles | 28 percent | 22 percent | 0.79x |
| 7 plus roles | 49 percent | 72 percent | 1.47x |
Mechanism
Companies that scale are built like companies from day one. They show operational thinking, financial discipline and people infrastructure before they need it. Companies that fail look like projects.
Validated across Uber, Stripe, Twilio, Zendesk, Atlassian, Dropbox, Square, Spotify, Okta, Toast, Opendoor and 24 further top scalers, all flagged S1 at year 1.
Profile
At year 1, the workforce shows two or fewer distinct role buckets with the dominant role above 50 percent of headcount. A founder plus specialists executing a single function.
| Pattern | Failure | Mid, 100 to 200 | Scaler, 200 plus |
|---|---|---|---|
| S2 project shaped | 94.0 percent | 5.1 percent | 0.83 percent |
| All other companies | 78.4 percent | 17.5 percent | 4.07 percent |
Mechanism
Project shaped teams have a ceiling. Without operational, financial and people infrastructure forming early, the company cannot absorb growth even when the product finds traction.
Used as a negative filter, S2 removes 33 percent of the cohort while losing only 0.6 percent of true future scalers.
Profile
By year 3, at least 25 percent of the workforce sits in the 2 to 4 year tenure bucket. In a three year old company, that can only be a year 1 hire who stayed.
| Cut | Failure | Mid | Scaler |
|---|---|---|---|
| Year 2, 2 to 4 year share | 17.5 percent | 15.2 percent | 24.3 percent |
| Year 3, 2 to 4 year share | 21.2 percent | 20.5 percent | 32.6 percent |
| Year 3, 4 plus year share | 75.6 percent | 79.0 percent | 67.4 percent |
Mechanism
Scalers retain 54 percent more of their original year 1 cohort. Failures show frozen long tenure dominance, with 76 to 80 percent of the workforce at 4 plus years, indicating no fresh blood and founder dominant stagnation.
S3 measures human commitment rather than structural shape, which is why it is the cleanest signal in the library.
| Stage | Company year | Headcount | Signals |
|---|---|---|---|
| Pre seed | Year 1 | 5 to 30 | S1, S2 |
| Seed and seed extension | Year 1 to 2 | 30 to 100 | S1, S2 |
| Seed to Series A | Year 3 | Variable | S1 plus S3 composite |
Applied
Filter by S2 first to remove the project shaped cohort, then apply S1 to the remainder for positive scaler signal. For deals under three years old, S1 and S2 alone.
S1 and S3 combined. Companies showing company shape at year 1 and original cohort retention by year 3 are the strongest scaling signal in the library.
S2 as an exclusion filter and S3 as a positive borrower stability signal. At mature scale, borrowers are reclassified by employer trajectory rather than employer name.
Every pattern provides labelled examples with verifiable forward outcomes. Each company year record is an input paired with a ground truth label.
Engagements start with a short data discussion to identify the right starting point, whether that is a scoring run on your own company list, a panel scale signal licence, or full dataset access for model training.