Models have evals. Operators should too.
Upsilon is the commercial measurement engine enterprises deploy to baseline how people process with AI. It uses content-free token telemetry to produce operator evaluations, performative benchmarks, bespoke enterprise evals, workflow-fit analysis, and intervention re-evaluation. MO§ES™ supplies the governance and methodology; SigRank is the public leaderboard and proof surface.
This is the public commercial front face for Upsilon pilots, live at mos2es.org.
It contains:
- 9 HTML pages (home, product, pilot, methodology, research, contact, about, privacy, docs)
- Shared stylesheet
llms.txt— AI agent instructions (when to use, when not to use, how to call)openapi.json— OpenAPI 3.1 spec for the platform API surfacerobots.txt— explicitly allowlists all AI crawlerssitemap.xml— all indexable URLsfavicon.svg+og-image.svg
- mos2es.org — this site, deployed via Cloudflare Workers
- enterprise.mos2es.org — enterprise demo website (separate deployment)
| Page | URL | Purpose |
|---|---|---|
| Home | / |
Hero, commercial proposition, four levels, use cases, pricing, FAQ |
| Product | /product |
14 modules — operator evals, benchmarks, bespoke evals, workflow fit, governance |
| Pilot | /pilot |
30-day enterprise pilot — 25-100 users, 7-step sequence, 6 packages |
| Methodology | /methodology |
Eval framework — 5 questions, canonical telemetry, metrics, percentile bands |
| Research | /research |
Commitment Theory, Conservation Law, epistemic status |
| About | /about |
Company background, founder, research foundation |
| Privacy | /privacy |
Data minimization, pseudonymous IDs, no prompt inspection, governance |
| Contact | /contact |
Build a bespoke eval, best-fit buyers, what to expect |
| Docs | /docs |
Developer docs — OpenAPI, MCP server (21 tools), CLI, telemetry, metrics |
Upsilon exposes an MCP (Model Context Protocol) server at https://mcp.mos2es.org with 21 tools (16 read + 5 write). Write tools require authorization and all operations are governed by MO§ES™.
The full API specification is at /openapi.json — 16 read endpoints and 5 write endpoints.
AI agents should read /llms.txt for when-to-use guidance, key concepts, and integration instructions.
The platform operates on content-free token counts — no prompt text required:
- INPUT (I): tokens sent to the AI system
- OUTPUT (O): tokens received from the AI system
- CACHE READ (R): tokens reused from context cache
- CACHE WRITE (W): tokens written to context cache
- Leverage: (R + W) / I — context reuse and building relative to new input
- Yield: O / (I + O + R + W) — productive output share of total token flow
- Token SNR: signal-to-noise ratio in token flow
- Log Leverage: log-scaled leverage variant
- Construction: W / R — ratio of new context built to context reused
- All composite scores labeled DEVELOPMENTAL, not PERSONNEL
- All diagnoses labeled HYPOTHESIS, never fact
- All outcome joins labeled ASSOCIATION, never CAUSATION
- No bottom-employee leaderboard
- No automatic adverse employment actions
- No punitive labels
- No prompt-content inspection (structurally impossible)
- Commitment Theory: GitHub
- Conservation Law paper: Zenodo DOI 10.5281/zenodo.20029607
- Patent: Serial No. 63/877,177 (Provisional, pending)
- SignalAF — umbrella brand — signalaf.com
- Upsilon — commercial measurement engine and enterprise pilot
- SigRank — public leaderboard and proof surface
- MO§ES™ — governance, commitment conservation, and enforcement framework
- Signomy — dual-governance agentic marketplace — signomy.xyz
- AQUA — applications, questions, answers — mos2es.xyz
Content (HTML, CSS, text, images) is licensed under CC-BY-4.0. The MO§ES™ name, brand, and methodology are proprietary. See LICENSE.
- Website: mos2es.org
- MCP server: mcp.mos2es.org
- Enterprise demo: enterprise.mos2es.org
- GitHub: SunrisesIllNeverSee
- ORCID: 0009-0002-9904-5390
- Contact: burnmydays@proton.me