Leverage measurement for AI-native engineering
Fulcrum logs every agent-assisted task at the time of the work: the senior-engineer hours it would have taken, the agent minutes it took, and the operator minutes you spent. It turns “we use Claude Code” into an execution factor you can defend, publish, and improve.
$1,199 per year · installed in your environment · API, dashboard, CLI, and MCP server · included for a year with every Vantalect leverage engagement
Fulcrum is the instrument behind the public leverage record at charlessieg.com/leverage: 3,552 tasks since February 22, 2026, every row recomputable. Measured on our own build, not promised for yours. Yours gets measured.
What a record holds
Every field belongs to one of four evidence classes, and the dashboard never lets an estimate pose as a measurement.
Captured at the time of the work, not reconstructed afterward.
How long the work would have taken without the agentic path, with its basis and author on the record.
The circumstances the result was produced under, recorded as boundaries rather than multipliers.
Work that needed substantial new learning, or was not feasible unaided, is classified rather than assigned an invented hour count.
Inside the product
The dashboard gives the summary, the leverage trend, and the project breakdown. The cost page prices the same work by model and by activity. Both run in the dark and light themes, and the records list reads on a phone.
Demo data. These screenshots come from a demonstration installation. Every figure in them belongs to StrataForge Robotics, a fictional company, and none of it is a customer's record.
How it works
The record is written while the work happens, by the tool that did the work, and confirmed by the person who supervised it.
At the end of a task the agent calls record_leverage over MCP, or you run leverage add. A client-generated record id makes every retry idempotent: the same task never lands twice.
Drafts become confirmed records in the dashboard. Corrections keep the original row and supersede it, so the ledger is auditable end to end.
Confirmed, reviewed records feed the summary, the time series, the team leaderboard, and, if you choose, an immutable public snapshot with a manifest and hashes.
Every surface your team already uses
One PostgreSQL-backed service behind four front doors, so the record gets written by whatever did the work.
record_leverage to get_fatigue_assessmentleverage add from any shell, idempotent by record idFrom an agent session
# the agent, at the end of a task, over MCP record_leverage( task="Split services and certifications into pages", human_estimate_hours=12, estimate_basis="qualified_senior", agent_session_minutes=41, wall_clock_minutes=41, operator_minutes_total=5, tokens=404000, token_source="exact", capability_class="time_compression", client_record_id="8DDB198D-…") # or from any shell leverage add --task "…" --human-hours 12 --agent-session-minutes 41
The part nobody else measures
By the thirteenth task of the day the operator has made hundreds of supervisory decisions. Fulcrum counts them, scores the day, and hands the agents guidance to match: fewer questions, more conventional defaults, completed work instead of options.
| Records today | Level | What the agents do |
|---|---|---|
| 0–3 | Low | Normal operation. Present options, ask clarifying questions. |
| 4–7 | Moderate | Fewer questions. Reasonable defaults when intent is clear. |
| 8–12 | High | Minimize decision points. Front-load the reasoning. |
| 13+ | Very high | Maximum autonomy within the constraints. Present finished work. |
get_fatigue_assessmentPublished metrics
The ratio of sums is not the mean of ratios, and Fulcrum refuses to blur them. Every published figure carries its formula identifier and its coverage.
60 × sum(H) / sum(A) The primary execution metric: senior-hours saved per agent minute, summed over every record.
median(60 × H / A) The middle individual task, immune to the outliers that dominate a mean.
60 × sum(H) / sum(O) Senior-hours per operator minute, over records with complete operator-time coverage, with the coverage count published beside it.
60 × H / wall_clock How much sooner the work landed. At period level the denominator is the union of intervals, never a naive sum.
mean(60 × H / A) Kept for continuity and labeled as the task mean, never presented as the headline.
counts and shares How much of the work was skill extension or otherwise infeasible unaided, reported as counts, never as an invented factor.
Licensing
Fulcrum is licensed for installation where your engineers already work. The record it builds is yours, whether or not you renew.
The API, the dashboard, the CLI, and the MCP server, installed in your environment. Updates for the term of the license.
Request a licenseBoth the Assessment and the Embed install Fulcrum on day one, so week six is measured with the same instrument as week one. Then $1,199 per year, only if you choose to continue.
See the engagementsTerms, what you need to run it, and the questions people ask first
Licensing
One annual license installs Fulcrum in your environment: the API, the dashboard, the CLI, and the MCP server your agents already know how to call. $1,199 per year. Included for a year with every Vantalect leverage engagement.
Installed in your environment, on your hosts, under your identity provider.