Morpheus ecosystem model

Morpheus has five emission pillars, a claim rate and a retention rate on each, a yield-funded buyback, a compute pool whose stipend formula sets what providers are paid, and a queue of MOR that is owed and not yet minted. This models all of it over the deployed schedule, to February 2040. Provider pay is not a dial here: the contract pays providers in proportion to the MOR users have staked for compute, so the pool damps pay on its own.

Start on the first panel, which follows a single day's issuance from what the schedule releases to what reaches the loose float. The other four take one question each. The retention and claim dials default to dated on-chain readings, and the marked value on each slider is that reading, so moving one reads as a departure from the evidence. The two staking dials are labelled assumptions: the model's only guesses, and it says so on each.

Is MOR locked up faster than it is issued?from the ecosystem model, 2026-08-25

Of the 12,130 MOR the schedule releases on a day in 2026-08, 2,537 reaches the loose float. Providers are paid 807 MOR for serving roughly 976M tokens of inference that day, at the measured rate of 1.2M tokens per MOR (only 30% of payments report token counts).

Collateral providers must hold, against all MOR staked10%295k MOR needed, of the 2.86M staked anywhere in Morpheusthe contract pays a provider at most what it has staked, per year. Keeping807 MOR/day of pay flowing needs 807 x 365 = 295k staked across providersall MOR staked anywherethe same ratio across the whole schedule100%never passes 100%, peaks 18% in 2027-07, then falls back as pay fallsDoes serving inference force MOR buying?No: providers fund the collateral from the pay itselfthe cap forces HOLDING, not buying: pay cannot be sold until the provider's yearrolls. The one outside need is a new provider's seed stake: 12k MOR in the last 24 days.And the user side: staking 336 MOR buys 1 MOR a day of compute:That cost only rises as the pool tightens. The cut-off-return dial is theonly brake on it; the contract has none.
Funded by deposit yield, not by emission, so it never decaysThe emission contract. It creates new MOR every day on a fixed, declining schedule that reaches zero in early 2040. Nothing can change it.The schedule12,130/day24% of each day's new MOR, reserved for people who deposit stETH and stablecoins with the protocol.Capital2,911/day24% of each day's new MOR, reserved to pay the operators who serve inference. It builds up in a pool until providers earn it.Compute2,911/day24% of each day's new MOR, for code contributors. Weights are assigned by hand by the protocol's multisig.Code2,911/day24% of each day's new MOR, paid to builder subnets in proportion to the MOR staked on each.Builders2,911/day4% of each day's new MOR: the protocol's insurance fund for bugs, audits and user losses.Protection485/dayThe people who deposited capital, claiming the MOR this pillar owes them.Depositors1,654/dayThe operators serving inference. They are paid from the compute pool in proportion to the MOR users have staked; the token figure is roughly how much inference that pay bought.Providers807/day ~ 976M tokCode contributors claiming their rewards. Under half of what this pillar owes has ever been claimed.Contributors1,310/dayBuilder subnets receiving rewards. Most of what they are paid ends up staked back.Subnets2,038/dayThe insurance fund itself. It is claimed in full by the protocol's own multisig into protocol wallets, so it flows to treasuries, never to holders.The fund485/dayAll MOR locked somewhere: staked on builder subnets, held as provider collateral, or staked for compute access. Locked MOR cannot be sold, so this pot is not supply in any practical sense.Staked stock2.9M MORMOR in nobody's known bucket: ordinary holders and exchanges. What is left of supply after everything staked, burned or held by the protocol is taken out. The sellable part of the token. Growth in the MOR staked for compute is bought out of THIS bucket, which is how usage demand drains it.Loose float2.1M MORThe protocol's own balance: distributor reserves, Safes and the 16-year lock, holding minted MOR nobody has claimed yet. When recipients draw faster than the schedule mints, the outflow is paid FROM this balance, which falls but never goes below zero.Treasuries4.0M MOR, +3,524/dayThe dollars that deposited stETH and stablecoins earn as yield. The protocol spends this yield buying MOR on the market: the only recurring buyer in the system.Deposit yieldstETH and AaveMOR bought on the market with the deposit yield. Fixed in dollars, not in MOR, so it does not shrink with the schedule. It can only buy MOR that is loose: once the float cannot supply it, the fill falls short of what the yield wants to spend.Buyback498/dayHalf of every filled buyback is sent to the burn address, permanently.Burned249/dayThe other half of every filled buyback, locked until 2040 to seed the next emission era. Never withdrawn so far.16-year lock249/day
Month 2026-08

Every node is somewhere MOR sits and every ribbon is MOR moving, at the width and the rate the model gives that flow. The five pillars are drawn apart because they behave nothing like each other: capital claims about half its pillar and stakes none of it, compute is paid the stock-model draw and posts back more than it is paid, and the protection fund is claimed in full by the protocol's own Safe, so it flows to treasuries rather than to the float. The buyback runs on its own track at the foot, because it is funded by capital-provider yield rather than by emission. Drag the month or press play and the whole system thins as the schedule decays.

Treasuries are the protocol's own balance: MOR minted to distributors and Safes that no end recipient has claimed yet, including the pre-minted reserve that pays providers. When recipients draw faster than the schedule mints, the difference is paid out of this balance, which falls but never goes below zero. The three buckets on the right are sized against the measured-default run, so moving a dial changes their fill rather than resizing the scale under you. And the model conserves MOR exactly, month by month:

Audit, 2026-08: 296k burned + 2.86M staked + 3.96M treasuries + 303k AMM inventory + 2.12M loose float = 9.54M MOR, against 9.54M modelled supply (the 42M cap at the measured 77.0% mint rate). The float is the residual of that identity, so MOR can never appear or vanish between buckets.

Every retention and claim default is a dated on-chain reading, not a choice. Compute retention sits above 100% because collateral grew faster than earnings, so MOR came into the pillar from outside it. One provider earns 70% of the pillar and sits at exactly 100%, because the contract caps a provider's annual earnings at its own posted stake and it has to top up to keep earning. Take that provider out and the rest of the field retains more, not less.

The two staking dials are the model's only assumptions. The growth in MOR staked for compute has no measured history yet: the series is too short and too concentrated to yield a rate, so 6x is an analogy to token growth, not a reading. The cut-off return has no basis in data by construction: staking is observed to continue at a 108% return and nothing below that has ever been seen. Retention cannot outrun the emission funding it, which is the point the whole model turns on.

Where the numbers come from

The emission series is the exact step sum of rewardPools(n) on the deployed contract, identical across Ethereum, Base and Arbitrum. Claim rates, retention, the claim lag and the lock-release curve are measured from chain over dated windows. The model itself is a workbook checked against 21 contract reads, and this page reproduces it: a check runs on every build asserting that the browser engine returns the workbook's own answers at five points across the schedule, to a millionth of a MOR. If it drifts, the build fails.

Two things to hold on to. No panel carries a price, market cap or valuation, so this says nothing about what MOR is worth. And it forecasts nothing: the growth in MOR staked for compute is the largest judgement call in it, and it is an assumption rather than a measurement, because its own history is too short and too concentrated to yield a rate. Treat the dials as a way to ask what would have to be true.

Every setting is linkable. Move the dials, use Copy link to these settings, and the URL carries the whole scenario.

The reasoning behind the model, and what it changed about our view of MOR, is in How MOR Actually Works. The project review is at Morpheus, and the live inference figures the compute panel starts from are at Morpheus inference activity.

Model built 2026-08-25 from the deployed contracts.