Glossary

Routing Reputation

Routing reputation scores Lightning nodes based on reliability, uptime, and payment success rate to optimize path selection.

Key Takeaways

  • Routing reputation is a local scoring system that routing nodes and wallets use to track the reliability of other nodes, improving payment path selection over time based on past outcomes.
  • Reputation data is entirely local: each node maintains its own view of the network, and no global reputation score is shared between peers. This protects privacy but means every node starts from scratch.
  • LND's Mission Control is the most widely deployed reputation system, using probability estimators to predict which channels can successfully forward a given payment amount.

What Is Routing Reputation?

Routing reputation refers to the data a Lightning Network node collects about the reliability of other nodes and channels it has interacted with during payment forwarding. When a node sends or routes a payment, it records whether each hop along the path succeeded or failed, the amounts involved, and how quickly the peer responded. Over time, these records build a local picture of which nodes reliably forward payments and which ones frequently fail.

This concept exists because senders face a fundamental information asymmetry: the gossip protocol reveals total channel capacity but not how the channel balance is distributed between the two sides. A channel with 1 BTC of capacity might have 0.9 BTC on one side and 0.1 BTC on the other, but senders cannot know this in advance. Routing reputation helps fill that gap by learning from past attempts which channels can handle specific payment sizes.

Unlike centralized credit scores or public ratings, Lightning routing reputation is maintained independently by every node. Node A's assessment of Node B may differ entirely from Node C's assessment, because each has routed different payments through different channels at different times.

How It Works

Routing reputation operates at the pathfinding layer of a Lightning node. When a node needs to find a route for a payment, it combines gossip data (channel capacities, fees, timelocks) with its own historical records of past payment outcomes. The process works as follows:

  1. A payment attempt is initiated, and the node selects a route using a modified Dijkstra's algorithm that weighs both cost (fees and timelocks) and estimated success probability
  2. If the payment succeeds, the node records a success for each node pair along the route at that payment amount
  3. If the payment fails at a specific hop, the node records a failure for that node pair and amount, then recalculates and tries an alternative route
  4. Over time, success and failure records narrow the estimated liquidity range for each channel, making future path selection more accurate

The key insight is that reputation is tracked per directed node pair, not per channel. If Node A and Node B share multiple channels, their combined forwarding capacity is what matters. A failure on one channel does not necessarily mean the entire relationship is unreliable: the node pair may still have liquidity in another channel between them.

Mission Control in LND

LND's Mission Control is the most prominent implementation of routing reputation. It records the outcome of every HTLC forwarding attempt and uses these records to estimate the probability of future payments succeeding through the same channels.

For each directed node pair, Mission Control tracks two critical data points:

  • The highest amount that has successfully been forwarded (the success bound)
  • The lowest amount that has failed (the failure bound)

These two bounds define a range: amounts below the success bound are likely to succeed, amounts above the failure bound are likely to fail, and amounts in between are uncertain. Mission Control uses a probability estimator to assign success probabilities within this range.

Probability Estimators

LND offers two probability models for interpreting Mission Control data:

  • Apriori estimator: the original default model, which assigns a base success probability (commonly 0.6) to each hop with no prior information. After observing outcomes, it adjusts this probability based on the success and failure bounds. Failed node pairs are penalized with a time-decaying penalty: a configurable half-life (default 3,600 seconds) controls how quickly the node "forgives" past failures, since channel balances shift constantly as other payments flow through
  • Bimodal estimator: introduced as experimental in LND v0.16, this model assumes that channel liquidity tends to cluster at the extremes rather than being uniformly distributed. In practice, channels often have most of their balance on one side, and the bimodal model accounts for this skew. It produces more accurate probability estimates for large payments and was promoted to recommended default in later releases

Both estimators feed into the pathfinding cost function, which calculates an effective cost for each potential route. The cost combines actual fees, timelock penalties, and the virtual cost of a failed attempt:

effective_cost = routing_fees
  + timelock_penalty
  + attempt_cost / success_probability

The attempt_cost represents the opportunity cost of a failed payment: the time and locked liquidity wasted on a route that does not complete. Dividing by the success probability means unreliable routes incur a higher effective cost even if their nominal fees are low. This creates a natural tradeoff between cheap routes and reliable routes.

Memory Decay and Freshness

Reputation data goes stale quickly. A channel that failed a 500,000 sat payment five minutes ago may have received an inbound payment since then, restoring its forwarding capacity. Mission Control addresses this through time-based decay:

  • Failed node pairs gradually recover their reputation over time, controlled by a half-life parameter
  • A minimum failure relaxation interval prevents the node from immediately retrying a recently failed channel
  • Mission Control caps the number of historical records it stores, discarding the oldest entries to prevent unbounded memory growth

This decay is essential because Lightning channel balances are constantly shifting. A reputation system that remembered failures indefinitely would blacklist nodes that are only temporarily low on liquidity.

Other Implementations

Core Lightning (CLN) takes a different approach with its renepay plugin, which uses a minimum-cost-flow algorithm that blends success probability with fee optimization. Rather than maintaining explicit reputation scores, renepay learns from failed attempts within a payment session and adjusts its liquidity estimates accordingly. CLN's default pay plugin also applies capacity checks and cost functions that balance fees against expected reliability.

Eclair, the Lightning implementation maintained by ACINQ, optimizes primarily for low fees but also considers channel age and capacity in its routing decisions. Research comparing Lightning pathfinding strategies has found that LND's approach tends to favor payment reliability, while Eclair tends to find paths with lower fees.

Why It Matters

Payment success rates directly determine the usability of the Lightning Network. A payment that fails three times before succeeding creates a poor user experience and wastes time and locked liquidity at each intermediate hop. Routing reputation transforms pathfinding from a blind search into an informed decision, improving success rates with every payment attempt.

For routing node operators, reputation has economic implications. Nodes that maintain high uptime, adequate channel balances, and fast response times earn better reputation scores on other nodes, which means more payments are routed through them and more routing fees are collected. Unreliable nodes are gradually bypassed as the network learns from failures.

Layer 2 protocols like Spark take a different approach by eliminating multi-hop routing entirely. Because Spark uses a shared UTXO model with cooperative signing rather than payment channels, there is no need for pathfinding or reputation systems. Payments settle directly without intermediaries, avoiding the success rate challenges that routing reputation exists to solve.

Use Cases

Wallet Path Optimization

Mobile wallets and Lightning gateways benefit significantly from routing reputation. A wallet that connects through a Lightning Service Provider can track which outbound routes from its LSP consistently succeed for certain payment sizes. Over time, the wallet learns the network topology from its own perspective, reducing failed attempts and improving the end-user payment experience.

Large Payment Routing

Routing reputation is especially valuable for larger payments. A 1 million sat payment requires sufficient directional liquidity at every hop along the route. Without reputation data, the sender must rely on channel capacity alone, which reveals nothing about balance distribution. Mission Control's success and failure bounds directly indicate whether a channel has handled similar amounts recently, making multi-hop payments of significant size much more reliable.

Multipath Payment Splitting

When a payment is split across multiple paths using multipath payments (MPP), reputation data helps the sender decide how to divide the amount. If reputation records show that one route reliably handles 200,000 sats but fails at 500,000, the sender can split accordingly rather than discovering these limits through trial and error.

Channel Jamming Mitigation

Routing reputation plays a role in proposed defenses against channel jamming attacks. The HTLC endorsement proposal (BOLT PR #1071) combines local reputation tracking with an endorsement signal: nodes track how their direct peers behave and then signal downstream whether an HTLC comes from a well-reputed source. This binary endorsement flag is added to the update_add_htlc message, allowing intermediate nodes to prioritize HTLCs from peers with strong track records while throttling unendorsed traffic during congestion.

Risks and Considerations

Cold Start Problem

New nodes have no reputation data for any peer. Every first payment attempt is essentially a guess, and building reliable reputation requires many payment attempts across diverse routes. This creates a bootstrapping challenge: new nodes must accept lower success rates while their reputation database populates. Meanwhile, nodes with extensive payment history enjoy a compounding advantage in routing accuracy.

This problem is partially addressed by tools like Rapid Gossip Sync, which helps new nodes quickly download the network graph. However, the graph provides channel capacity, not balance distribution, so reputation must still be built through actual payment attempts.

Privacy Implications

As a node routes more payments, its reputation database accumulates information about liquidity distribution across the network. A node with extensive routing history effectively learns approximate balances for many channels, which can be a privacy concern. Techniques like payment probing can accelerate this process deliberately, mapping channel balances across the network without actually completing payments.

Reputation data also reveals payment patterns. If a node observes that certain channels consistently deplete in one direction at specific times, it can infer payment flow patterns for the peers involved. This is one reason why blinded paths and onion routing are important: they limit how much routing nodes can learn about payments they forward.

Gaming and Manipulation

Because reputation is built through observed behavior, it can be gamed. A malicious node could maintain reliable forwarding during low-fee periods to build strong reputation, then exploit that reputation by failing payments or extracting higher fees when it matters. Since reputation systems rely on past behavior predicting future behavior, sudden changes in node behavior can cause unexpected payment failures until the reputation adjusts.

Whitewashing attacks are another concern: a poorly reputed node can shut down and restart with a new identity, resetting its reputation to the default baseline. This is a known limitation of local-only reputation systems where node identity is cheap to change.

Local-Only Limitations

The strictly local nature of Lightning routing reputation means there is no way to share reputation data between nodes. While this preserves privacy, it also means that every node must independently discover which peers are reliable. A new node cannot benefit from the years of routing history that established nodes have accumulated. Research has noted that local-only schemes are weaker for edge nodes (which route fewer payments) compared to large routing hubs that see high payment volumes and build reputation faster.

Stale Data

Reputation data has a short shelf life. Channel balances can shift dramatically within seconds as large payments flow through the network. A node pair that successfully forwarded 1 BTC a minute ago may have lost all directional liquidity since then. The half-life decay in systems like Mission Control addresses this, but it creates a constant tension: decay too quickly and you lose valuable history, decay too slowly and you route through channels that have changed.

For a deeper look at how Lightning pathfinding algorithms use reputation data, see the research article on Lightning Network routing. For more on the peer scoring mechanisms that complement routing reputation, see the glossary entry on peer scoring.

This glossary entry is for informational purposes only and does not constitute financial or investment advice. Always do your own research before using any protocol or technology.