Merge origin/master into v1: per-model free slots rule, task 08 step 1b

docs/plans/v1/05-proxy.md

@@ -53,7 +53,11 @@ func (h *Hosts) SetDraining(name string, on bool)
 
 // Chooser adapts config, health and limiter to lease.Chooser using choose.Best:
 // Info{Healthy, Draining: false (the lease table already filtered), Loaded: model in Loaded,
-// CanServe: cfg.Serves, Free: lim.FreeSlots(host), Queued: lim.Queued(host, model), Weight}.
+// CanServe: cfg.Serves, Free: free slots FOR THIS MODEL on this host =
+// cfg.Hosts[host].Models[model].Parallel - lim.InFlight(host, model) (never below 0; 0 when the
+// host does not list the model), Queued: lim.Queued(host, model), Weight}.
+// (Corrected 2026-09-25: an earlier version said lim.FreeSlots(host), which sums every model's
+// slots and let a host win on slots the requested model cannot use.)
 func Chooser(cfg *config.Config, h *health.Table, l *limiter.Limiter) lease.Chooser
 
 func New(cfg *config.Config, h Health, leases *lease.Table, lim *limiter.Limiter, rec Recorder, log *slog.Logger) *Handler

docs/plans/v1/08-smoke-readme.md

@@ -30,6 +30,13 @@ check failed and crossbar's log. A failure is a defect in tasks 01–07 **or in
 code only when a rule from an earlier task was broken; if the script's expectation contradicts a
 task rule, stop and report which.
 
+- [ ] **1b. One known owner finding to fix before the smoke can pass.** `proxy.Chooser`
+  (task 05, `internal/proxy/hosts.go`) computes `Free` with `lim.FreeSlots(host)`, which sums the
+  slots of *every* model on the host; the rule now reads: `Free` = free slots **for the requested
+  model** = `cfg.Hosts[host].Models[model].Parallel - lim.InFlight(host, model)`, floored at 0,
+  and 0 when the host does not list the model. Make that change (only that), run
+  `go test -race -count=1 ./internal/proxy/` (must stay `ok`), then re-run `make smoke`. This is
+  the one code change this task makes; record it in Deviations as an owner-directed fix.
 - [ ] **2. Update `README.md`.** Keep the seven v0 sections; change: the intro (leases, not "first
   healthy host"); `## Configure` gets `db`, `lease_idle`, `retention`, `queue_max` in the table and
   the new `example.toml`; `## Point clients at it` adds the `X-Crossbar-Route` header
@@ -40,7 +47,7 @@ task rule, stop and report which.
 - [ ] **3. Run the gate.** `make gate`. **4. Log and commit.** Row `v1/08-smoke-readme`, with the smoke line in Notes.
 
 ```sh
-git add tools/smoke.sh cmd/fakeupstream/main.go README.md docs/implementer-log.md
+git add tools/smoke.sh cmd/fakeupstream/main.go internal/proxy README.md docs/implementer-log.md
 git commit
 ```
 

docs/plans/v1/README.md

@@ -119,3 +119,9 @@ tools/run-plan.sh docs/plans/v1          # from a clean checkout on master
   "does not fail the gate", but `go vet ./...` compiles `main.go`, whose two-argument
   `admin.Handler` call no longer matches — the gate does fail. Ornith noticed while reading.
   Task fault (mine): step 5 now allows the one-call edit to `main.go`.
+- 2026-09-25, task 08, first session: smoke check 1 sent conversation A to alpha, not beta. Cause:
+  my task 05 rule `Free: lim.FreeSlots(host)` sums a host's slots across all its models, so
+  alpha's six `small-9b` slots outscored beta's two `ornith` slots for an `ornith` request. The
+  smoke's expectation (per-model slots) is the right semantics. Task fault (mine): task 05's rule
+  is corrected and task 08 gained step 1b, the one code change allowed in it. Ornith had
+  diagnosed the summing correctly before it was stopped.