Growth is a lathe,
not a funnel.

The work that never comes back

Here is how growth work dies at a small company. Someone notices a number sagging in a dashboard. Someone else remembers a complaint from support, or a Reddit thread, or a churned customer's email. A hunch forms in a meeting, a change ships in a hurry, an experiment maybe gets set up — in a different tool, with a different definition of the metric — and six weeks later nobody can say what happened, because the answer is spread across analytics, a payment dashboard, an ad manager, a spreadsheet, and someone's memory of a Slack thread.

Then the quarter turns over and the whole thing starts again, from zero. The evidence is gone. The reasoning is gone. Whether it worked is genuinely unknowable. The work didn't compound; it evaporated.

The standard answer is more tools — a sharper dashboard, a better experimentation suite, another integration. But the failure isn't in any single tool. It's in the shape of the work: a line. Lines have ends. Everything poured down a funnel leaves through the bottom.

A word from the workshop

Torno is Italian for “I return.” It comes from tornare — to turn — and further back, from the lathe: the oldest machine tool, the one that spins the work so a maker can shape it. The word holds both ideas at once: the returning, and the machine that makes returning useful.

A lathe doesn't move the work forward. It brings the same material back under the tool, revolution after revolution, and the shape emerges from accumulated passes. No single pass is heroic. The precision comes from the returning.

That is the wager behind Torno: growth work should be mounted on something that turns.

The Loop

In Torno, the unit of growth work is a Loop — a durable, addressable state machine wrapped around one objective, with one target Metric at its center:

Signals → Analyzers → Insights → Hypotheses → Changes (+ Exposure)
   ↑                                                    |
   └──────────── new Signals ← Results ← Decisions ←────┘
                     one Metric at the center

Everything the world tells you — product events, support email, public threads, revenue movements, a Metric watch firing — lands as an immutable Signal with its provenance attached. Analyzers turn Signals into Insights that carry confidence and citations. A Hypothesis must fill a forced grammar: because this evidence, we believe this change, for this audience, will move this Metric, by this much, within this window. A Change ships with a declared exposure — or is honestly labeled unmeasurable. A readout wears exactly its evidence state. And a Decision — ship, kill, iterate, extend — closes the turn with rationale and evidence snapshots, appended to a log that cannot be rewritten.

Then the Loop returns to listening. Not to the same place: the decision log is longer, the definitions are versioned, the next proposal is required to read what happened before. A circle that merely repeats would be failure — this is a helix. Same axis, higher thread.

Honesty as a mechanical property

Growth software has a lying problem. Not fraud — just a thousand small overstatements: the dashboard that renders “no data” as zero, the experiment page that crowns a winner at p = 0.049, the attribution model that quietly claims credit for everything. Confidence is cheap to render.

Torno treats honesty as a property of the instrument, not a virtue of the operator. A Metric is one versioned definition, read through one envelope that carries its value, its bounds, its basis, its freshness — every surface reads the same envelope, so a number can't drift between tools. Readouts speak a closed vocabulary: insufficient evidence, trending, significant win, significant loss, guardrail breach — and nothing else. “No data” and “below the privacy threshold” are rendered answers, never zeros. A Change that declared no exposure simply cannot appear in a causal claim.

The instrument never reads more than the evidence. That is what makes acting on it feel safe — and it's the one promise a growth tool should refuse to bend.

Agents at the lathe, humans at the controls

The second wager: the operator of this machine is increasingly not a person. Founders already delegate research, drafts, and campaign mechanics to agents. So Torno is built agent-native rather than agent-tolerant: every registered capability generates its web, API, CLI, and MCP bindings from one capability map. An agent working through torno sees the same product a human sees — not a reduced subset bolted on afterward.

But operating is not governing. Anything irreversible, sensitive, or money-spending passes through policy: allow, deny, or hold for a human. An agent must attach its reasoning, which is stored as a claim — never treated as instruction. Ad publishing waits for human approval by default. Spend has ceilings. The lathe spins fast; the controls stay under your hands.

One turn, concretely

A storefront product notices trials stalling. The funnel shows the plan-picker step bleeding; two support emails ask which plan includes custom domains; a public thread says the pricing page is confusing — that one arrives quarantined, labeled, fenced. Analysis clusters the voice, decomposes the funnel, and produces an Insight with its confidence and citations. A Hypothesis proposes collapsing three plans to two, for new trials, targeting the trial-to-paid Metric, expecting half a point in three weeks. The Change declares its exposure tag and goal event, policy allows it, delivery is a signed local bundle, and measurement starts at first confirmed exposure. Two weeks in, the readout says trending — not a win, and it says so. The Decision, when the evidence is sufficient, will be recorded by name, with the readout attached, in a log the next turn must consult.

Nothing in that story is heroic. That's the point. It's one clean revolution of a machine built to revolve.

Why the batteries are included

Loops break in the gaps between tools — where the metric definition changes hands, where the experiment lives apart from the revenue, where the ad manager doesn't know what the funnel knows. So Torno bundles the instruments the Loop needs: collection, listening, analytics, experiments, Stripe-derived revenue, ad drafting, briefs, and governance, behind one semantic layer.

The boundary is drawn honestly: payment rails and ad auctions stay with their providers, specialist tools will out-depth Torno inside their specialty, and your data remains exportable — reads, raw events, webhooks — because a system of record that locks you in is just a nicer trap.

I return

The name is the promise, in first person, made by the product: every Signal comes back as a Decision; every Decision comes back as context; every turn comes back with more.

Torno. I return — with more.