Deterministic. Reproducible. Not a model wrapper.
The CommunityOS scoring engine ranks every account on four archetypes using rules that you can read, audit, and reproduce. The same inputs always produce the same outputs. No prompts, no model drift, no black box.
60 percent language. 40 percent normalized metrics.
The split was calibrated on real scans, not invented. Linguistic substance carries the heavier weight because reach without substance is exactly what KOL spend already buys.
What they actually say.
- Topical depth in posts about your category
- Reply substance, not reply count
- Original-content ratio versus reposts
- Thread structure and argument depth
- Voice consistency across time
Reach, normalized.
- Followers, normalized against account age and posting cadence
- Engagement rate, normalized against follower count
- Reach-per-post variance (signal vs noise)
- Quote-tweet pickup as a distribution proxy
Farms and shells never reach the queue.
Pre-campaign filtering removes bots, engagement farms, and inactive shells before scoring happens. The Act Now queue never shows them, so reports built on the queue can't include them either.
Accounts below a posting cadence floor are filtered out before scoring. Inactive shells inflate vanity engagement; we remove them at the door.
Engagement farms post in tight bursts on identical topics with copied phrasing. The engine catches the pattern and excludes the cluster.
Extreme imbalances are a known farming pattern. Caught in pre-filter, never seen in the queue.
Bot networks reuse phrasing across accounts. Linguistic similarity across the cluster is a kill signal.
Every account scored on all four.
Reach and substance are different currencies. The Archetype tells you which one each person holds.
Speaks about you with depth and conviction. The highest-value advocates.
Carries your message into rooms you can't enter. Wide audience, fast distribution.
Ships things: write-ups, tools, threads with original work in them.
Shows up before it's obvious. Tests, validates, gives the honest first signal.
CommunityOS is not a language-model wrapper.
The category is full of products that pipe data into a prompt and ship the response as a feature. That isn't a methodology, it's a roundtrip. Here's why we built the engine differently.
The same audience produces the same scores. Always. Methodology you can defend to a CMO or a CFO without saying "the model decided."
Rules-based scoring doesn't degrade when a vendor updates a model. Your last quarter's report is comparable to this quarter's.
The features and weights are documented. Every score can be traced to the inputs that produced it.
No per-token cost on every classification. The engine runs on commodity infrastructure, not per-request inference.
Methodology paper publishing under CC-BY-SA before public launch. We will release the feature list, weights, and validation set so anyone can reproduce the scoring on a held-out sample.
Want to see the engine on your audience?
Limited pilot slots ahead of public launch. Leave a work email and we will reach out.