A gatherer walks past the site it is standing on two times in three #124

New issue

Closed

opened 2026-08-05 23:52:14 +00:00 by sickday · 0 comments

sickday commented

2026-08-05 23:52:14 +00:00

Owner

Bot.Goals.Fish's choosing_ground sorts the grounds it can work by distance,
takes the nearest three and picks one at random:

grounds
|> Enum.sort_by(&Sites.chebyshev(&1.position, snapshot.position))
|> Enum.take(3)
|> Enum.random()

The intent is fine — bots should not all pile onto one spot — but the ground the
bot is standing on gets no weight at all. A fisher spawned at a perfectly good
ground walks away from it two times in three, and fishing grounds are sparse
enough that the two runners-up are routinely hundreds of tiles off rather than
tens.

Observed

A harpooner and a cager spawned exactly on the Fishing Guild ground
(2598,3420) both set off east on the same tick, to a ground ~240 tiles away,
and wedged in the mountains without ever casting. The same pair spawned at
Catherby went to work immediately — the difference was only which of the three
the roll returned.

From the earlier net-fisher sitting: a fisher hiked past a spot 13 tiles away.

Why this is worth fixing rather than tuning

On its own it costs a bot some walking. Combined with the straight-line hop in
Bot.Travel it is the difference between a working bot and a permanently
wedged one, because the long trip it chose for no reason is exactly the trip
that wedges. The two together made a live sitting read as "the new kinds do not
work" when both kinds were fine — the fix for that sitting was to spawn a fleet
and keep whichever bots happened to roll the ground under their feet, which is
not a thing an operator should have to know.

Shape of a fix

Keep the spread, lose the pointless journey. Either:

Prefer the ground underfoot: if one of the candidates is within a short
radius of the bot, take it rather than rolling; or
Weight the roll by distance so a ground ten times further is picked far
less often than an equally-close alternative.

Bot.Goals.GatherAndBank picks sites the same way and deserves the same look —
tree and rock sites are dense enough that it has not bitten there yet, which is
luck rather than design.

`Bot.Goals.Fish`'s `choosing_ground` sorts the grounds it can work by distance, takes the nearest three and picks one at **random**: grounds |> Enum.sort_by(&Sites.chebyshev(&1.position, snapshot.position)) |> Enum.take(3) |> Enum.random() The intent is fine — bots should not all pile onto one spot — but the ground the bot is *standing on* gets no weight at all. A fisher spawned at a perfectly good ground walks away from it two times in three, and fishing grounds are sparse enough that the two runners-up are routinely hundreds of tiles off rather than tens. ### Observed A harpooner and a cager spawned exactly on the Fishing Guild ground (`2598,3420`) both set off east on the same tick, to a ground ~240 tiles away, and wedged in the mountains without ever casting. The same pair spawned at Catherby went to work immediately — the difference was only which of the three the roll returned. From the earlier net-fisher sitting: a fisher hiked past a spot 13 tiles away. ### Why this is worth fixing rather than tuning On its own it costs a bot some walking. Combined with the straight-line hop in `Bot.Travel` it is the difference between a working bot and a permanently wedged one, because the long trip it chose for no reason is exactly the trip that wedges. The two together made a live sitting read as "the new kinds do not work" when both kinds were fine — the fix for that sitting was to spawn a fleet and keep whichever bots happened to roll the ground under their feet, which is not a thing an operator should have to know. ### Shape of a fix Keep the spread, lose the pointless journey. Either: * **Prefer the ground underfoot**: if one of the candidates is within a short radius of the bot, take it rather than rolling; or * **Weight the roll by distance** so a ground ten times further is picked far less often than an equally-close alternative. `Bot.Goals.GatherAndBank` picks sites the same way and deserves the same look — tree and rock sites are dense enough that it has not bitten there yet, which is luck rather than design.

sickday referenced this issue from a commit

2026-08-05 23:58:42 +00:00

fix(bot): stop a gatherer walking past the site it is standing on

sickday closed this issue

2026-08-05 23:58:42 +00:00

sickday referenced this issue from a pull request that will close it,

2026-08-05 23:59:04 +00:00

fix(bot): stop a gatherer walking past the site it is standing on #125