Discovery breaks at scale. This is the repair we are working towards.

Human progress is constrained less by intelligence than by coordination.

At every scale, advancement depends on the right people discovering one another at the right time, under conditions of sufficient trust. Scientific collaboration, institutional formation, capital allocation and talent development are all coordination outcomes.

When coordination fails, potential goes unrealised despite capacity.

In small societies, discovery operated inside bounded networks.

• Reputation was directly observable.
• Intent was socially contextualised.
• Intermediation was accountable — the person who introduced you was known.

Trust scaled because scale was limited. Discovery happened within visible relational boundaries.

High trust. Low reach.

As societies expanded, relational density could not scale with them.

• Reputation became abstract.
• Context fragmented.
• Visibility emerged as a substitute signal.

Discovery shifted from tribal routing to broadcast exposure. The mechanism changed from who vouches for you to who can see you.

High reach. Low signal integrity.

Modern coordination systems optimise for reach. Trust does not scale proportionally with it.

When reach expands faster than accountability, signal integrity deteriorates and legitimacy is obscured by visibility. This is the divergence every failure below descends from.

From the divergence between reach and trust, three structural distortions emerge.

Visibility inflation

Popularity becomes a proxy for legitimacy. Quantitative metrics displace qualitative judgement — ten thousand followers outrank three people who can actually speak to your work.

Cold routing

Requests travel without contextual bridges. A cold message carries no endorsement, so trust is consumed rather than transferred. The recipient pays the cost of evaluating it, and learns to stop paying.

Informal intermediation

Warm introductions do work — but they are memory-dependent. The risk sits entirely with the person making them, and the judgement involved never compounds into anything. Good connectors accumulate no credit for being right.

Effective large-scale discovery requires three things at once:

• Contextual transfer — the request arrives with its context intact.
• Accountable forwarding — passing something on costs something.
• Compounding judgement — being right repeatedly has to accrue.

No system in wide use optimises for all three simultaneously.

What is absent is a structured trust-routing layer: one where intent travels through accountable paths, forwarding carries measurable consequence, reputation reflects routing decisions, and discovery requires opt-in consent at each step.

Ladders is an attempt to build towards such a layer.

The intended system defines how intent moves between people through accountable paths. Three properties would hold everywhere in it:

• Every transmission requires explicit consent.
• Every forward carries reputational exposure.
• Every decision compounds.

The vocabulary below is how we currently think about the problem. It has been neither implemented nor tested against real routing behaviour, and we expect parts of it to be wrong.

Each node would hold a reputation R in the range 0 to 1, starting neutral at 0.5.

When a forward resolves:

…where W is the intent's risk weight, somewhere between 0.1 and 1.0. Vouching for a high-stakes ask earns more when it lands and costs more when it does not.

The only claim we are willing to make about these numbers is their shape: loss is roughly twice gain. Trust should be earned slowly and lost quickly, because that is how it behaves between people who actually know each other.

Every real question here is open. Whether reputation should be a single scalar at all, whether it should decay, whether it should be domain-specific rather than global, how it survives a cold start — none of these are settled, and a system that got them wrong would be worse than no system.

Nodes do not see the graph. A participant sees their direct connections, intent that has been routed to them, and their own local slice of the path.

• The full path is not visible by default.
• A seeker cannot see nodes ahead of the intent.
• A finder cannot see the path behind it.
• A bridge sees only the previous node and the next.

This is a privacy constraint before it is a design one. A routing layer that exposed everyone's graph to everyone would be a worse product than the address book it replaced.

Any routing system with reputation attached is a system somebody will try to farm. The intended defences are structural rather than editorial — limits in the protocol rather than moderators reading messages.

We would rather say this plainly than overclaim: constraints like these raise the cost of gaming a system. They do not make it ungameable, and any design that says otherwise has not met enough adversaries yet.

A coordination layer that only works when people are generous is not a system, it is a mood. Each role has to come out ahead by behaving in the way the network needs.

The intended pipeline is a progressive filter. At each hop, a human decides whether to spend standing on passing something along.

The proportions above are drawn to illustrate attenuation, not measured. We have no throughput data, because there is nothing running to measure.

The substantive claim is the one underneath them: layered human judgement catches what filtering cannot — context mismatch, wrong timing, the ask that is reasonable but not from this person to that one.

Ladders is not a social network. It does not optimise for content, followers or engagement.

It is not a marketplace. It does not match supply and demand through exposure.

It is meant to be a coordination layer: something that structures how intent moves across trust boundaries.

Capability is widely distributed. Access is not. Broadcast systems amplify that asymmetry; structured routing is an attempt to reduce it. It would not equalise outcomes — only the pathways by which someone can legitimately become visible.

Traditional networks amplify noise as they grow. The intent here is the opposite.

Growth should increase:

• Path diversity
• Signal quality
• Accumulated context

Growth should not increase:

• Broadcast exposure
• Unbounded visibility
• Unfiltered inbound volume

Whether that inversion actually holds is the central empirical question of the whole design, and it is not one that can be answered from a document.

Stated plainly, because a direction that only lists its upside is marketing.

• It does not eliminate inequality of access. It reorganises how it is mediated.
• It does not guarantee good decisions. Bridges will be wrong, confidently.
• It does not prevent collusion. It makes it more expensive.
• It depends entirely on participant integrity. Cultural decay degrades routing quality faster than any parameter can compensate for.
• It does not replace judgement. It gives judgement somewhere to accumulate.

Saying every question here is open is cheap unless the questions have answers that could arrive. These are the observations that would settle them against us.

They are about the routing and reputation design specifically. The premise underneath Rungs — that dormant ties survive and that people cannot rank their own — is tested separately, on the evidence page.

None of these carries a threshold yet. There is nothing running to calibrate one against, and a number chosen now would be the same guesswork the standing note at the top of this page warns about. The thresholds get fixed, and published, before anything is measured.

None of the above. What is built is Rungs: a system that reads the network you already have, remembers what you know about the people in it, and tells you who can actually move the thing you are working on.

That is the first honest step towards this — before intent can be routed through strangers' judgement, it has to be worth routing through your own. We would rather ship that and be held to it than publish a protocol nobody can run.