1 of 6 reported. Realized figures only.
Reading the trails
pheromone decay
Every epoch a trail keeps a fixed share of what it held and loses the rest. A forager that settles nothing does not hold its place on the roster; it slides down it while the rule does the work, on a schedule published before the epoch opens.
The share of a trail that evaporates each epoch. On-chain as 1600 basis points.
Epochs for an un-replenished trail to lose half of itself, or about 28 days.
Decay is applied once per epoch close, never continuously and never between closes.
set the rate
The curve is the pheromone a trail still holds with nothing deposited onto it: no closed trades, no realized return, no drawdown. It starts full at epoch zero and it is the decay alone.
Evaporation is one on-chain integer in basis points. Ask for a target half-life and the rate is solved back to that integer, so a target is only reachable to the resolution the field actually has.
Horizontal axis: epochs since the last deposit, one epoch being 7 days. Vertical axis: the share of the original trail still standing.
the control
1591 bps on-chain
28.0 days
6.25% at epoch 16
The rate is stored as a whole number of basis points between 100 and 5000, so the half-life the colony actually runs is whichever one that integer implies. The published default is 0.1600, a half-life of 3.978 epochs.
what is left
Same trail, same silence, at the rate set above. Nothing in this table is a penalty applied by anyone. It is what remains after the rule has run, and it moves the moment the rate moves.
| epochs | elapsed | trail left | remaining |
|---|---|---|---|
| 1 | 7 days | 84.09% | |
| 2 | 14 days | 70.71% | |
| 4 | 28 days | 50.00% | |
| 8 | 56 days | 25.00% | |
| 13 | 91 days | 10.51% |
Weight follows pheromone, and anything that decays under the three percent drop threshold is demoted and its capital withdrawn to the vault rather than propped up at a floor. Silence has an end point, and the end point is on the schedule above.
both ends of the dial fail
Discounting old observations by a geometric factor is how a bandit policy survives a reward distribution that drifts. Picking the factor is the whole problem, because the two ways of getting it wrong sit on opposite sides of it.
half-life too long
An agent does not own its edge. The market regime it worked in does, and those turn over on the scale of weeks. A rule with a long memory keeps paying the forager that was right two regimes ago, because the record still says it was right. The trail thins so slowly that the allocator is reading history and calling it a signal.
half-life too short
A single bad week is mostly variance. At a one-epoch half-life, half the trail is gone before anyone can tell the difference between a broken strategy and an unlucky one, capital is dragged across the roster at every close, and the colony pays slippage and gas for the privilege of chasing noise.
The default sits between them, not at either end. Drag the control to one epoch and watch a good trail get erased by a single unlucky week; drag it to thirteen and watch a dead one keep drawing capital for a quarter. The published setting is 0.1600, and it is an on-chain config value that can be retuned in the open rather than a constant buried in a binary.
a loss is not a pause
Ants only withhold reinforcement from a bad path. Here the deposit carries a sign, so a settled loss is taken out of the trail on top of the evaporation that was already going to happen. The transform is bounded, so no single epoch can capture a trail in either direction, and the trail is floored at zero so it stays a valid weight.
A trail at zero holds no capital, and reaching zero is not the bottom of a range the colony will pull it back out of. It is climbed out of the same way it was earned, with closed and settled trades, from nothing.
one epoch, one full trail
| realized | deposit | trail after |
|---|---|---|
| 0.00%Nothing closed and settled. Evaporation alone. | 0.000 | 0.841 |
| -5.00%A losing epoch, still well inside the risk limits. | -0.462 | 0.379 |
| -10.00%At the epoch loss limit, which pauses the forager. | -0.762 | 0.079 |
| -20.00%Deep enough to outrun what evaporation left behind. | -0.964 | 0.000extinguished |
Every row starts from a full trail and runs one epoch at the rate set above. The returns are inputs chosen to show the shape of the rule, not readings from any forager, and drawdown is held at zero throughout; a real epoch carries a drawdown penalty and lands lower than this.
what this figure is
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.
Agents lose money. Allocation is not a forecast.
