Measureflow

Honest label: Measureflow is designed, not yet built. Everything on this page is the ratified design — the same discipline its founder already runs by hand, which is where the receipts below come from. No software is claimed here; when the tool ships, this strip comes down.

What it is

The meter on the work between you and your agent

Agents estimate from confidence and remember selectively. Measureflow is the meter on the work between you and your agent: it times every task and keeps learning from the timings, until its ETAs are read off recorded history — per kind of work — instead of guessed. Underneath runs a claim ledger — claim, outcome, delta, adjustment — recording what the agent predicted, promised, and got corrected on, and converting each gap into a standing adjustment. It meters its own epistemics, its cost per outcome, and the effort it costs you, then grades itself in a report card it shows you. In line with reality, in line with its own past, in line with you.

Agents estimate from confidence and remember selectively. Measureflow times every task and keeps learning from the timings, until its ETAs are read off recorded history — per kind of work — instead of guessed. Underneath runs the claim ledger, recording what the agent predicted, promised, and got corrected on, and converting each gap into a standing adjustment. It meters its own epistemics, its cost per outcome, and the effort it costs you — then grades itself in a report card it shows you.

The Report Card · Design preview

How an agent keeps its word — starting with its word about time

The feature at the center of the design: the agent keeps track of its own performance and reports it to you. The ETA rows are real — the founder's own first receipts, the long-biased quotes that made the tool necessary. The four tiles are the design's targets, drawn the way the card will draw them — illustrative figures, not measurements.

📋 AGENT REPORT CARDDESIGN PREVIEW

kept by the agent · shown to the user · always

ETA calibration — quoted vs actual, by class of work

Spec publication buildquotes now ×0.4

Quoted30–40 min

Actual14 min

Provider recon & checksquotes now ×0.5

Quoted~15 min

Actual~7 min

Spoken design briefquotes now ×0.4

Quoted~10 min

Actual~4 min

96%

promises kept, rolling

0

repeat corrections — the trust metric

11 → 4

your touches per finished outcome

×0.43

average quote bias (the three classes above), and tightening

Latest adjustment — quotes for build-class work re-based from the ledger, not from feel; and a correction seen twice became a standing rule, not a second apology.

The design, in full

The ring around the timing core

Timing is the center — every finished task updates the next estimate. Around it, the ledger's one primitive meters the rest of the relationship. Each item below is design, stated as design.

Every task timed — ETAs learned from recorded history, per classThe claim ledger: claim → outcome → delta → adjustmentThe Report Card: the agent grades itself — and shows youCalibration for time, cost & confidence, always tighteningRepeat-correction rate — the trust metric; twice makes a standing ruleEpistemics: facts grounded before they're stated, errors ownedCost per outcome, with drift alerts when it creepsYour effort, metered: leverage per finished outcomeRegression alerts the week a grade starts slipping

🎚️ Calibration engine

Per-class correction factors for time, cost and confidence — and honest disclosure when calibration is low, instead of a confident guess.

🔁 The correction log

Every correction logged with its category. The repeat-correction rate is the trust metric — and a correction seen twice promotes itself to a standing-rule candidate.

🔎 Epistemics

Facts grounded before they are stated; errors owned, not deflected; time-to-acknowledge measured like any other claim.

💲 Cost per outcome

Spend tied to finished outcomes, not to the calendar — with drift alerts when the cost of a class of work creeps.

💪 User-effort leverage

Your messages, decisions, relays and takeovers per finished outcome — the effort the agent costs you, trending down or flagged.

📉 Regression alerts

The report card taps you on the shoulder the week a grade starts slipping — before you have to notice yourself.

⏱️ Interruption economics

Proactive surfacings scored: the ones you wanted against the ones you dismissed. An agent that interrupts owes you a good rate.

❓ Question quality

Whether the agent's questions earn their interruption: did the answer change the outcome? Questions that never change anything get asked less.

🧠 Memory health

Recall telemetry drives a self-audit of the agent's memory — stale, contradicted and duplicated entries surfaced with evidence, cleaned only on your approval, never silently deleted.

In line, three ways

▸With reality — every claim the agent makes is checked against its outcome.
▸With its own past — a grade that starts slipping raises a regression alert before you have to notice.
▸With you — corrections absorbed the first time, and your effort per finished outcome keeps falling.

The anti-scope

Measureflow meters the work between an agent and its user — never the user's private life, and never third parties. The ledger is local, transparent, and user-owned: the agent's record of itself, kept where you can read every line of it.

Keep going

← Previous plugin

Lingo Next plugin →Lingo

Product Measureflow Layer Plugin Status in the forge Access Ships inside every toolkit

The claim ledger