docs: replace removed Ames %heed/%jilt with the backpressure protocol

[sigilante]

Fourth PR from the audit against urbit/urbit@08026c84b2. This is the batch that makes an agent or vane author write code which crashes, so it is kept separate from the larger card-coverage work.

Companion PRs: #266 (mechanical), #267 (Hoon signatures), #268 (user manual).

%heed and %jilt are gone

Neither appears in +$ task:ames (lull.hoon:844-903). The only trace left in lull is a comment recording that they were "introduced in state %4, and removed in %21", and ames.hoon carries the migration that drops them ($: %21 :: remove %heed and %jilt at :2825, with adapters at :1317-1403 and :3314).

A vane passing either today crashes Ames — but ames/tasks.md still documented both, with a "Returns" section, as though they were live. This is the single worst thing the audit found: the docs actively instruct you to send a task that no longer exists.

Replaced with the cards that supersede them:

Card	Shape	Meaning
%halt	$>(%halt deep)	halt a flow after a remote %flub
%goad	[%goad =ship]	restart halted flows once the peer is live

plus their Gall counterparts [%flub $@(~ [blocked=? dap=(unit term)])] and [%spur ~] (lull.hoon:3623-3624). The flow-state comment at lull.hoon:1732 gives both directions — forward, Gall passes a %flub to Ames; backward, a $plea gets %flubbed over the wire — and Gall's +mo-halt (gall.hoon:310-319) is what emits the remote %flub as a %boon.

Note there are two distinct %halts in source. The message-pump → packet-pump one (lull.hoon:1280) is unrelated internal congestion control and is deliberately not documented here.

%clog is no longer an Ames gift

+$ gift:ames (lull.hoon:927-946) has no %clog. It is now [%clog id=*] in +$ task:gall (lull.hoon:3629), which Ames may pass (ames.hoon:165). The README's narrative described the old %heed → %clog gift loop.

For what it's worth, lull's own gift comment at :911 still lists %clog — stale upstream, and out of scope here.

$peer-state and $alien-agenda lost .heeds

Current +$ peer-state is lull.hoon:1160 (peer-state-31): no .heeds, plus a new halt=(set bone), =chain and tip=.... Current +$ alien-agenda is lull.hoon:1132: no .heeds, keens regrew as (jug [path ints] duct), and a new chums.

Worth flagging for reviewers: ames.hoon still contains 14 heeds=(set duct) hits, but every one is a versioned migration state (peer-state-5 and friends), not the live type. Checking only ames.hoon would give the wrong answer.

Verified against a running ship rather than source alone:

> .^(ship-state:ames %ax /=//=/peers/~zod)
... snd={} rcv={} nax={} closing={} corked={} keens={} chain={} tip={} halt={}

No .heeds. The stale heeds={...} line in the scry.md sample output was corrected to the real trailing fields.

Added a $deep section

%halt is written $>(%halt deep) and $deep was documented nowhere, so the task reference had nothing to point at.

Dill %meld payload

Changed from [%meld ~] to [%meld $@(~ [memo=? ford=?])] in Dill's $gift, $task and $flog (lull.hoon:2864, :2882, :2955). A bare [%meld ~] still type-checks, so the new options were invisible rather than broken.

Traced the runtime side to document what they actually select: u3_meld_all (vere melt.c:205-217) frees the persistent memo cache for memo (u3R->cax.per_p) and the Ford build cache for ford (u3R->cax.for_p), both before the dedup runs.

Also fixed a copy-paste error where %pack said "does not return a gift in response to a %meld task".

Not changed

core-academy/ca08.md quotes %heed/%jilt source at length, but Core Academy states it is a snapshot at Zuse 412k. Rewriting quoted source there would misrepresent the snapshot, so it is left alone.

All anchor links in the touched files were checked to resolve.

🤖 Generated with Claude Code