Cohorts, deployment, privacy, governance.
The enterprise layer is where measurement meets an organization. Cohorts define the scope. Privacy architecture defines what we do not touch. Governance defines what the results may and may not be used for.
Design a 30-day pilot → Data surface registry25–100 operators. 30-day windows. Company-specific analysis.
A cohort is the unit of enterprise deployment. It is large enough to produce distributions and divergence patterns, small enough to govern and explain.
25–100 operators
Small enough for governance and individual review. Large enough for cohort distributions, divergence analysis, and archetype patterns.
30-day windows
Baseline and post-intervention windows are matched in length. 30 days balances statistical stability against organizational patience.
Company-specific analysis
Every cohort is analyzed against the external reference field and against itself. No two cohorts share a distribution — only a metric definitions.
Field-relative
Each operator and the cohort are compared to the external reference field. Company-specific shape is reported alongside field-relative position.
In-scope declared
Which AI systems and external systems are in scope is declared before ingest. No silent expansion of scope mid-pilot.
Sequential cohorts
Enterprises run sequential cohorts to track movement over time and compare intervention effects across windows.
What we require, and what we do not.
The privacy posture is a table, not a promise. It states exactly which data classes are required, optional, or excluded — and it is enforced structurally, not by policy alone.
| Data class | Requirement |
|---|---|
| Prompt content | NOT REQUIRED |
| Response content | NOT REQUIRED |
| Code content | NOT REQUIRED |
| Email content | NOT REQUIRED |
| Slack content | NOT REQUIRED |
| Token telemetry | CORE |
| Signed numeric snapshots | CORE |
| Approved metadata | OPTIONAL |
| Business outcomes | OPTIONAL JOIN |
Content is never required. Token telemetry and signed numeric snapshots are the core. Everything beyond that is an explicit, governed, optional join — declared in advance and revocable.
Who sees what, by default.
Permissions are set before ingest, not negotiated after results appear. The defaults are conservative; widening requires explicit governance.
Full operator profile
Each employee can see their own full operator profile. The profile is private to the employee and the pilot team. No one else receives individual-level detail by default.
Cohort-level views
Management receives cohort-level views by default: distributions, divergence patterns, archetype concentrations, aggregate intervention results. Not individual profiles.
Explicitly governed
Individual identity exposure to management or wider audiences is explicitly governed. It requires consent and a documented purpose. It is never a default.
No adverse action in pilot
The pilot terms explicitly exclude employment termination and adverse-action use. This is a term of the pilot, not a preference. A pilot that cannot accept this term does not start.
Enterprise identity is separate from public identity.
Enterprise permissions and identity mapping should be separated from public SigRank identity. The two systems serve different populations, different governance, and different trust models.
Public SigRank identity
The open operator evaluation field. Self-asserted, pseudonymous, portable. Governed by the public ecosystem's terms. Not an enterprise directory.
Enterprise identity mapping
The enterprise's internal mapping between operator records and employees. Governed by enterprise policy, consent, and the pilot terms. Never exposed to the public ecosystem.
Conflating the two would leak enterprise personnel data into a public system, or import public self-assertions into enterprise governance. The separation is structural.
What must be locked before production.
Some items must be finalized before any production deployment. Others can remain placeholders — they do not block the pilot and stabilize through repeated runs.
- Canonical metric definitions
- Official formulas
- Reference-field version
- Archetype taxonomy
- Pilot privacy policy
- Individual vs cohort visibility rules
- Intervention claim language
- Enterprise brand name
- Paired brand name
- Full connector catalog
- Executive dashboard design
- Vertical-specific workflow packs
The lock list protects measurement integrity. The placeholder list protects against premature design commitments that repeated pilots would otherwise overturn.
The product wins by being narrower.
Explicit non-goals define the boundary of the product. They are not limitations to be overcome — they are commitments that keep the product honest.
Not building
- Full HRIS replacement
- Employee surveillance suite
- LMS
Not building
- Generic productivity score
- Universal business-outcome predictor
- Content-reading workplace intelligence agent
Not building
- Large dashboard before pilot requirements stabilize
- Automated adverse employment decision system
Enterprise governance starts with a pilot.
The enterprise layer is not a product feature — it is the terms under which measurement is allowed to happen. Designing a pilot is the first step in defining those terms for your organization.
Design a 30-day pilot →