Method
How growth is measured
Short version: by storing star counts every day and subtracting. Everything else on this page is the detail of what happens before there is enough history to subtract.
The three bases for a growth figure
| Shown as | Means | Trusted for | Repos |
|---|---|---|---|
| measured | Stars now minus stars at the snapshot nearest to seven days ago, divided by the exact number of days between them. A span under five days or over ten is refused rather than scaled up. | Any repository, any age. This is the real thing. | 7.8k |
| measured 30d | The same subtraction against the snapshot nearest to thirty days ago, accepted between fourteen and forty-five days. Used when nothing sits in the short window. | Any repository. Slower to react than the 7-day figure. | 13 |
| lifetime avg | Total stars divided by the repository's age in days. | Only repositories younger than 420 days. For a project born three months ago, its lifetime average is its current rate; for a six-year-old project it would be meaningless, so it is not used. | 297 |
| stale | Nothing. Shown as —. | Repositories that dropped out of the top 100 of every topic they belong to, so we no longer get a fresh star count for them. Their last known count is frozen; subtracting from it would read as "no growth", which would be false, so no rate is published. Excluded from trending. | 1.2k |
| not yet | Nothing. Shown as —. | Older repositories we have not watched long enough. They appear under "Most starred" but are excluded from every trending board and their momentum is capped at 35. | 9 |
A short-window rate needs two snapshots at least five days
apart, and a long-window rate needs fourteen. That gate exists so a one-day
difference is never multiplied up and presented as a weekly trend. Right now there are
27 days
of history, starting 2026-10-10.
The momentum score
Growth rate alone would put a two-week-old repository gaining 40 stars a day above a mature project gaining 300. Momentum is the blend used for ranking, and every repository page shows its own arithmetic.
| Component | Weight | Definition |
|---|---|---|
| Velocity | 55 | The growth rate above, log-scaled against a ceiling of 750 stars/day. Log-scaled because the gap between 1 and 10 a day matters more than between 300 and 310. The ceiling is set from the collected population — the fastest 0.1% of repositories run at roughly 650 stars/day — because an earlier ceiling of 50 left the top forty repositories in a dead tie. |
| Acceleration | 15 | The 7-day rate divided by the 30-day rate. Only earns points above 1.0 — speeding up, not merely fast. Needs both windows, so it is zero for most repositories today. |
| Freshness | 15 | Days since the last push, decaying linearly to zero at 180 days. |
| Scale | 15 | Star count, log-scaled against 150,000. A credibility floor so a three-star repository with three stars a day cannot top a board. |
| Penalties | × | Archived repositories keep 40% of their score, forks 70%. Both are also filtered out at collection time, so these only catch repositories archived after we first saw them. |
Where the repositories come from
Each category is a list
of GitHub topics. GitHub's search API rejects OR between
qualifiers, so every topic is queried separately and the results are merged
here. Each topic is asked three questions per run: its most-starred
repositories, those created in the last year, and those created in the last 45
days. Forks and archived repositories are excluded at the source.
| Category | Topics queried | Tracked |
|---|---|---|
| AI Agents | ai-agents · llm-agent · autonomous-agents · agentic-ai · multi-agent | 1.1k |
| RAG | rag · retrieval-augmented-generation · vector-database · embeddings · semantic-search | 707 |
| Agent Skills & MCP | mcp · model-context-protocol · agent-skills · llm-tools · function-calling | 1.2k |
| LLM & Inference | llm · large-language-models · llm-inference · llm-serving · fine-tuning | 883 |
| Robotics | robotics · robot · slam · motion-planning · robotics-simulation | 546 |
| ROS2 | ros2 · ros | 224 |
| Mechatronics | embedded · microcontroller · firmware · cnc · stm32 · esp32 · 3d-printing | 815 |
| Computer Vision | computer-vision · object-detection · image-segmentation · opencv | 539 |
| SaaS | saas · saas-boilerplate · multi-tenant · stripe · subscription | 386 |
| Business | crm · erp · e-commerce · invoicing · accounting · business | 396 |
| Finance | finance · fintech · algorithmic-trading · quantitative-finance · trading-bot | 574 |
| Dev Tools | developer-tools · cli · devops · kubernetes · self-hosted | 1.6k |
| Security | security · cybersecurity · pentesting · security-tools | 779 |
| Data | data-engineering · etl · data-pipeline · database · duckdb | 580 |
| Learning & Roadmaps | roadmap · tutorial · awesome-list · interview-questions · learning-resources | 768 |
Known limits
- A repository is only re-read while it still ranks in the top 100 of one of its topics. Drop out of that and its star count goes stale — the page shows when it was last read.
- Stars measure attention, not quality or adoption.
- A repository with no GitHub topics set cannot be found by this method, however good it is.
- Category topics are chosen by hand, so coverage is a judgement call. Topics returning almost nothing are dropped rather than left as empty rails.
Collection log
| Started | Status | Requests | Repos seen | New | Errors | Notes |
|---|---|---|---|---|---|---|
| 2026-11-05 11:40 | ok | 235 | 7.9k | 93 | 0 | scored 9332 repos; auth=anon |
| 2026-11-04 11:40 | ok | 248 | 7.9k | 50 | 0 | scored 9239 repos; auth=anon |
| 2026-11-03 11:40 | ok | 243 | 7.9k | 67 | 0 | scored 9189 repos; auth=anon |
| 2026-11-02 11:40 | ok | 229 | 7.9k | 70 | 0 | scored 9122 repos; auth=anon |
| 2026-11-01 11:40 | ok | 231 | 7.9k | 107 | 0 | scored 9052 repos; auth=anon |
| 2026-10-31 10:40 | ok | 222 | 7.9k | 60 | 0 | scored 8945 repos; auth=anon |
| 2026-10-30 10:40 | ok | 227 | 7.9k | 40 | 0 | scored 8885 repos; auth=anon |
| 2026-10-29 10:40 | ok | 231 | 7.9k | 67 | 0 | scored 8845 repos; auth=anon |