Pilot Readout — Demo Results

Pilot readout from a 50-operator synthetic AI operator evaluation. Real output from the MO§ES™ platform: cohort stats, composite score distribution, metric distributions, divergence findings, and intervention outcomes. Every number below was computed by the actual scoring, benchmark, and diagnostic engines — not hand-written for marketing.

Synthetic data. All data on this page is from a synthetic 50-operator demo dataset, labeled SYNTHETIC throughout. No real operator data is shown. Governance labels apply: scores are DEVELOPMENTAL, diagnoses are HYPOTHESIS, outcome joins are ASSOCIATION.

Pilot Overview

Operators

Pseudonymous IDs (op_001 – op_050)

Observations

1,668

Token telemetry events over 30 days

AI Providers

ChatGPT, Claude, Codex, Copilot, Cursor

Interventions

Targeted interventions with outcomes

Window

30d

2026-07-01 → 2026-07-30

Data Quality

Blocking issues · 50 OK · 1882 warnings

Composite Score Distribution DEVELOPMENTAL

The AI Operator Development Index is a 0–100 composite score weighted across four canonical metrics: Leverage (30%), Yield (30%), Token SNR (20%), Construction (20%). The score is DEVELOPMENTAL — for development use, not a personnel performance rating.

Median

48.6

Mean

49.8

Range

2.4 – 94.0

Min → Max

IQR

36.8 – 65.9

Q1 → Q3 (middle 50%)

Cohort Metric Distributions

Five canonical metrics computed from content-free token telemetry. No prompt text required — only INPUT, OUTPUT, CACHE READ, and CACHE WRITE token counts.

Metric	P10	P25	Median	P75	P90	Mean	Std	Outliers
Leverage (R+W)/I — context reuse	3.54	9.85	12.18	23.51	29.49	14.66	8.63	—
Yield O/(I+O+R+W) — output share	0.71	5.48	6.01	9.54	15.96	7.68	4.42	1
Token SNR signal-to-noise ratio	0.17	0.28	0.34	0.35	0.53	0.34	0.10	4
Log Leverage log-scaled leverage	0.55	0.99	1.09	1.37	1.47	1.07	0.30	—
Construction W/R — new context built	0.82	1.16	1.73	2.75	4.02	1.96	1.05	4

Sample Operator Profile — op_001 SYNTHETIC

Team

Product Engineering

Role

Software Engineering

Level

Associate

Primary AI

Claude

Composite Score: 48.6 DEVELOPMENTAL

Component	Value	Percentile	Weight	Status
Leverage	11.82	46.0	30%	CANONICAL
Yield	6.01	49.8	30%	CANONICAL
Token SNR	0.34	46.4	20%	CANONICAL_WITH_INTERPRETATION_LIMIT
Construction	1.84	52.7	20%	CANONICAL_WITH_INTERPRETATION_LIMIT

Benchmark: Peer benchmark selected (49 similar operators available). Selection reasoning: ≥5 similar operators available — peer benchmark (step 5). This is the benchmark engine's selection algorithm picking the best available comparison class given data constraints.

Usage–Operation Divergence HYPOTHESIS

The divergence engine compares each operator's usage volume against their operation performance. Large gaps flag potential workflow friction, tool mismatch, or hidden expertise.

MIXED (balanced)

LOW–LOW (underused)

LOW USAGE / HIGH OPERATION

HIGH USAGE / LOW OPERATION

Top divergent operators (all diagnoses are HYPOTHESIS, never fact):

Operator	Class	Usage %ile	Yield %ile	Divergence
op_003	LOW_USAGE_HIGH_OPERATION	10.2	79.5	−69.3pp
op_034	LOW_USAGE_HIGH_OPERATION	12.2	80.0	−67.8pp
op_016	LOW_USAGE_HIGH_OPERATION	14.3	80.6	−66.3pp
op_007	LOW_USAGE_HIGH_OPERATION	16.3	79.7	−63.4pp
op_019	LOW_USAGE_HIGH_OPERATION	18.4	81.1	−62.7pp

These operators show high operation performance relative to their usage volume — a pattern that may indicate hidden expertise, efficient workflows, or underutilized talent. All diagnoses are HYPOTHESIS pending further investigation.

Intervention Outcomes ASSOCIATION

12 targeted interventions were assigned across the cohort. Outcomes are labeled ASSOCIATION — the platform measures what changed, not what caused the change.

SUCCESS

NEGATIVE

PARTIAL

NO EFFECT

ID	Operator	Catalog	Target Metric	Outcome
int_001	op_047	COA-001	yield	PARTIAL
int_002	op_004	COA-001	yield	NO_EFFECT
int_003	op_030	COA-001	yield	SUCCESS
int_004	op_045	COA-001	yield	SUCCESS
int_005	op_031	CTX-001	leverage	NEGATIVE

Showing 5 of 12 interventions. The full intervention catalog includes targeted training (COA-001), context architecture redesign (CTX-001), and more. Negative outcomes are reported honestly — the platform does not cherry-pick.

Data Quality

Blocking Issues

No data quality blockers

OK

Operators passing all checks

Warnings

1,882

Non-blocking quality warnings

Explore This Data Yourself

All the data on this page is available via the live MCP server. AI agents can call these tools right now:

# Get pilot status
curl -X POST https://mcp.mos2es.org/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/call","params":{"name":"get_pilot_status","arguments":{}},"id":1}'

# Get composite score for any operator
curl -X POST https://mcp.mos2es.org/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/call","params":{"name":"get_composite_score","arguments":{"operator_id":"op_003"}},"id":2}'

# Find divergent operators
curl -X POST https://mcp.mos2es.org/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/call","params":{"name":"find_usage_operation_divergence","arguments":{}},"id":3}'

See developer docs for the full 21-tool list, or read llms.txt for AI agent instructions.

Governance

All composite scores labeled DEVELOPMENTAL — not a personnel performance rating
All diagnoses labeled HYPOTHESIS — never presented as fact
All outcome joins labeled ASSOCIATION — never causation
No bottom-employee leaderboard — cohort distribution only
No automatic adverse employment actions
No prompt-content inspection — structurally impossible (only token counts)
Negative intervention outcomes reported honestly