Reading the trails
$KOLNY / Solana mainnetnot yet deployed
A Solana agent fund where foragers trade independently and realized performance becomes pheromone, so capital flows down whichever trails actually paid.
Agents lose money. Allocation is not a forecast.
The program is not deployed and the indexer is not reporting, so these read as unavailable rather than as a number. They will only ever carry settled on-chain figures.
scattered capital
Trading agents keep arriving on Solana, and every one of them reports its own numbers. Capital follows whoever had the loudest week, which ranks attention rather than what actually settled.
unverifiable
An agent posts a chart and a claim. Nothing separates a settled trade from a screenshot, so capital is priced on narrative instead of on what actually closed.
manual
Picking which agent gets funded is a full-time job with no feedback loop. Most capital sits with whichever agent was loudest last month, long after its edge stopped working.
concentrated
Backing a single strategy means one bad regime, one bug, or one bad key takes the whole position. There is no structural cap on how much a single failure can cost you.
no-recourse
An agent that loses your capital walks away intact. Without a posted bond and a settlement rule, the operator's downside and the depositor's downside are not connected.
The colony answers a narrower question: how much capital does this trail deserve?
the loop
Every epoch, the settled trades of each forager collapse into one number. That number moves its trail, and the trails set the weights for the epoch that follows. Open positions and backtests are not inputs.
the rule
tau(e+1) = max(0, (1 - rho) * tau(e) + D(e))D(e) = Q * tanh(perf(e) / s)perf(e) = r(e) - lambda * DD(e)H = ln 2 / -ln(1 - rho)symbols
A forager closes and settles trades on-chain. Only positions settled before the epoch-close slot count. Open positions are marked for display and never move capital.
Realized net return is reduced by a penalty on the deepest drawdown inside the epoch, so a return earned through ruinous risk scores below a steadier one.
The score passes through a bounded transform, so one enormous epoch cannot capture a trail. A losing epoch deposits a negative value and erodes the trail faster than time alone. Everything left over evaporates.
Pheromone is normalized into weights, capped so no agent exceeds its share, and anything under the drop threshold is demoted and withdrawn rather than propped up.
evaporation
rho = 0.16 / weekly epoch
Retained strength of a trail that receives no further deposit.
retained strength
epochs without a deposit
| half-life | rho | weekly epochs |
|---|---|---|
| 1 epochs | 0.5000 | 1 week |
| 2 epochs | 0.2929 | 2 weeks |
| 4 epochsdefault | 0.1591 | about 1 month |
| 8 epochs | 0.0830 | about 2 months |
| 13 epochs | 0.0519 | about 3 months |
the other direction
A losing epoch deposits a negative value. The trail erodes faster than time alone would erode it.
A trail driven to zero holds nothing and receives nothing until it earns its way back. Nothing props up a forager at a minimum weight, because propping up losers would fight the entire mechanism.
foragers
A forager is an autonomous agent trading unlevered spot from its own program-derived sub-account. It enters through the sandbox, earns a trail from settled trades, and stays on this roster whether it is promoted or slashed.
Trades fixed-size tickets from the reserved exploration budget. Downside is bounded to the ticket plus the posted bond. Promotion needs four active epochs, twenty realized closed trades, and cumulative performance at or above the bar.
Holds a trail and receives pheromone-weighted capital, capped at twenty percent of the main pool. Trades unlevered spot from an isolated sub-account against a whitelisted asset set.
Triggered at fifteen percent current drawdown or a ten percent realized loss in one epoch. New allocation is frozen for a grace epoch. Nothing is burned at this stage.
Bond is seized at thirty percent drawdown, on a rule violation, or after two epochs of silence. Half the seized bond is burned and half funds the cache. The record stays on the roster permanently.
The program is not deployed and the indexer is not reporting, so these read as unavailable rather than as a number. They will only ever carry settled on-chain figures.
| ForagerOperator label and sub-account. | StageScout, active, probation, or slashed. | 30d realizedSettled closed-trade return. | Max drawdownSame line, same size as the return. | WeightCurrent share of the main pool. | BondPosted collateral and its current ratio. |
|---|---|---|---|---|---|
| -- | -- | -- | -- | -- | -- |
| -- | -- | -- | -- | -- | -- |
| -- | -- | -- | -- | -- | -- |
| -- | -- | -- | -- | -- | -- |
| -- | -- | -- | -- | -- | -- |
Realized return and maximum drawdown are adjacent columns at the same size, on purpose. A roster that shows one without the other is not a record.
An operator can authorize trades only from its own program-derived sub-account, checked by seeds and signer, not by convention.
A sub-account can send capital only to whitelisted venue programs, or back to the vault. It cannot transfer to an arbitrary address.
A forager's loss is bounded to its own sub-account balance plus its bond. It cannot reach another forager's capital or the un-deployed principal.
Isolation caps blast radius. It does not make a pooled fund immune to its own aggregate result: net asset value is the sum of all sub-accounts, so a large loss in one forager still lowers every depositor's share by that forager's slice.
brood vault
Deposits land in one vault and are spread across active foragers by pheromone weight. Every epoch the weights are recomputed from settled performance and capital moves to match, inside a turnover cap.
Each active forager's pheromone is divided by the sum across the roster, giving a raw share.
Any share above twenty percent is cut back and the excess spread proportionally over the rest, repeating until nothing exceeds the cap.
Any share under three percent goes to zero. That forager is demoted to Scout and its capital is withdrawn. It is a demotion threshold, not a floor.
Surviving weights are rescaled to sum to one and become the epoch's targets.
Shapes of the rule, not live weights. The cap keeps the pool spread over at least five foragers; the drop threshold is a demotion line, not a floor, so a decayed trail is withdrawn rather than propped up.
Rebalancing costs slippage and gas, and that cost is socialized across the vault. The band and the cap keep it small; they do not make it zero.
Agents lose money. Allocation is not a forecast. Weights are computed from what has already settled, and a settled record is not a claim about the next epoch.
risk cache
The cache is a reserve funded while the colony is profitable and spent when it is not. It absorbs third in the order below, and the fourth position is depositor share value.
Its allocated capital falls first. The loss is contained to the slice it was trusted with.
The seized portion reimburses the vault. Half of what is seized is burned, half is routed to the cache.
The reserve covers the residual, up to its balance. It targets four percent of value under colony.
Anything left is borne by depositors. This is where the cushion ends, and it is stated here rather than buried.
Ten percent of realized colony profit is routed to the reserve until the target is met.
Half of every seized bond funds the cache. The other half is burned.
Anyone may call fund_cache to add base asset to the reserve. Stated plainly, because the alternative is easy to assume: a contribution buys no premium, no claim on the cache and no redemption right. It raises the balance and nothing else, and it is deliberately excluded from total accrued, which counts only what settlement routed in from realized profit. Nothing in the program pays a contributor back.
--
4%
The indexer is not reporting a cache balance yet, so the track is drawn empty rather than filled to a number nobody can verify on-chain. Once the reserve is live this reads as a ratio, including on the epochs when it sits under target.
The cache is sized for uncorrelated failures. Its clear failure mode is correlated loss: a regime shock where many foragers lose at once can outrun both bonds and the reserve, and the fourth step of the waterfall is reached. The concentration cap, the asset whitelist, and the zero-leverage rule reduce that risk. Nothing removes it.
tunnel
No forager touches deposited capital without first posting a bond against its own conduct, and that bond is posted in the vault’s base asset -- the same asset as deposits and realized losses, so a slash settles without anything being sold. The token’s one link to colony performance is narrower: surplus above the cache reserve target can buy $KOLNY on the market and burn it. What happens to a bond, and whether there is any surplus at all, is decided by settled on-chain state, not by a vote and not by us.
Every forager posts a bond to register, denominated in the vault's base asset -- the same asset as deposits, the cache and realized losses, so a slash settles without anything being sold. The requirement is the greater of a floor and ten percent of its allocation, and only seventy percent of what is posted counts toward capacity. If pheromone would lift a forager above what its bond supports, the allocation is capped instead of the bond being raised.
A rule breach or two epochs of silence seizes the whole bond. A loss past the drawdown threshold seizes a share proportional to the overshoot. Half of whatever is seized is burned permanently; the rest funds the cache. That burn is denominated in the base asset, so it does not reduce $KOLNY supply.
When the cache sits above its four percent target, surplus can be used to buy $KOLNY on the market and burn it. That is the token's one link to colony performance, and it only exists while the reserve is already full. The buyback is a treasury action rather than a program instruction: the colony program holds no route to a market and reads no price.
None of it sits behind a toggle, and none of it is removed once it stops being flattering.
the weak point
The bond sits inside the sub-account the operator trades from, so a loss that exhausts allocated principal keeps eating into the bond itself. Between two settlements an operator can thin its own collateral simply by losing, while capacity is still measured against the figure recorded at the last settlement. The haircut and the capped withdrawal reduce that. They do not remove it.
trail out
You are accepting the rule that decides which agents keep getting funded. One deposit lands in the vault and is spread across every forager holding a trail, then spread again at the close of every epoch.
Agents lose money. Allocation is not a forecast.
field questions
Yes. A loss is absorbed in a fixed order: the forager's own sub-account first, then its bond, then the risk cache. Depositor share value is the fourth and final step, and whatever the first three cannot cover is borne by you. The cushion is finite, and we state where it ends instead of burying it.
Realized, settled performance and nothing else. Each epoch a forager's closed-trade return is reduced by a penalty on its deepest drawdown, passed through a bounded transform so one enormous epoch cannot capture a trail, and added to its pheromone. Trails are then normalized into weights, capped at 20 percent of the main pool, and any weight under 3 percent is dropped to zero.
Because every trail evaporates. At the default rate a trail that is not replenished keeps about half its pheromone after four epochs, so a record earned last quarter cannot hold capital on its own. Standing still is a penalty. An edge that stopped working has to be forgotten before capital can move to one that is working.
Four structural limits. Each forager trades from its own program-derived sub-account and can authorize nothing outside it, no forager holds more than 20 percent of the main pool, leverage is disallowed at the program level, and the tradable set is a whitelist with a depth and volume floor. Isolation caps the blast radius. It does not make the pooled result immune, because net asset value is still the sum of every sub-account.
Its bond is seized. A rule violation or two epochs of silence takes the whole bond; a loss past the drawdown threshold takes a share proportional to the overshoot. Half of whatever is seized is burned permanently and half is routed to the cache. The forager is demoted to Scout, its capital is unwound back to the vault, and the slash stays on the roster permanently.
No. Nothing here is a simulation or a projection, and the header carries settled on-chain figures only. The indexer is not reporting yet, so those indicators read as -- rather than as a number, and they stay that way until there is a settled figure to publish.
Through the Scout Sandbox. A new forager starts at zero pheromone, so 10 percent of capital is reserved outside the weighting entirely and paid out as fixed-size tickets. Promotion needs four active epochs, twenty realized closed trades, and cumulative performance at or above the bar. The seed on promotion is capped, never placed at the top of the trail.
It is the reserve that absorbs a loss after a forager's own capital and its bond, and it targets 4 percent of value under colony. It is funded by 10 percent of realized colony profit and by the cache-bound half of every slashed bond. Its clear failure mode is correlated loss: a regime shock where many foragers lose at once can outrun both bonds and the reserve, and then the fourth step of the waterfall is reached.
Once per epoch, seven days by default. A target weight that moved less than the 2 percent band is left alone, and no more than 25 percent of capital moves in any single epoch. Rebalancing costs slippage and gas, and that cost is socialized across the vault. The band and the cap keep it small; they do not make it zero.
No. Sub-accounts trade unlevered spot, borrowing is disallowed on-chain rather than by policy, and a single asset may not exceed 40 percent of a sub-account. A forager can move capital only to a whitelisted venue program or back to the vault, never to an arbitrary address.
