[Ljunqiang]

Fixes #1112

Root cause

buildable.rs's kinds_conflict opens with a (Float, Int) => false arm that exists solely for Go's untyped-constant conversion (var f float64 = 1 builds) — its own comment concedes "the kind model cannot tell an untyped constant from an int variable". So every value position that compares kinds let a typed int variable through: var y int = 3; var x float64 = y checks clean (0 diagnostics) and then fails go build with cannot use y (value of type int) as float64 value in .... Issue #1112 enumerates the surfaces; all 10 were reproduced accept-then-go build-fail before this change.

Fix

New Collector::value_conflicts(want, got, value) = kinds_conflict(want, got) || (want == Float && got == Int && !self.is_untyped_constant(value)), guarding all 20 value positions:

site	shape
typed init (1-value + parallel)	var f float64 = y
struct literal field	point{f: y}
fn arg + pointer-inner auto-& (x2)	g(y) / take(y) on *float64
method arg + pointer-inner auto-& (x2)	pt.scale(y)
report_return_conflict (4 callers, value expr threaded through)	return y / return g()
append	append(s, y)
assign single + parallel	x = y / pf, xi = y, y
compound += and -=/*=//=	x += y
map index key	m[y] on map[float64]T
array / slice literal	[2]float64{y} / []float64{y}
map literal key + value	map[float64]string{y: ..} / map[string]float64{..: y}
copy elements	copy(fs, is)
new: Stmt::Send (narrow)	ch <- y on chan float64

is_untyped_constant is structural and conservative: int/rune literals, -x, a op b over constants, and references to an untyped const — file-level only when declared without a type annotation (const k int = 3 stays typed and rejects; fn-local consts fail open as untyped, since Define records no annotation — documented fail-open, same direction as kinds_conflict). Anything that reads a variable, calls, receives, or indexes is typed by construction in Go → false, i.e. value_conflicts is stricter, never looser, than kinds_conflict.

The Stmt::Send check is deliberately narrow (Float elem ← non-untyped Int only): kind-level chan-element conflicts stay owned by HYB-E3011's shape rules, so no double-reporting.

Untouched on purpose: kinds_conflict itself, equality_kinds_conflict (1 == 1.0 still legal), conversion_kinds_conflict (float64(y) still legal), kind-level container recursion ([]float64 vs []int variables — out of issue scope).

Side fix: the array-literal checker's literal-only const_int_value guard false-rejected an untyped file const (const k = 1 + [2]float64{k} → E1002; Go accepts). Subsumed by value_conflicts; now accepts, and typed/variable elements still reject.

Validation

• New hy-check/tests/typed_int_float_coercion_issue_1112.rs: 29 tests — 20 reject faces (every table row above, incl. both send forms and f32) + 7 accept-control groups (literal/neg/binary/rune, untyped file and local const at init/assign/return/arg/append/array/slice/map/send/+=, float→float, float64(x) conversion, 1 == 1.0, classic return 1).
• The issue's 10 repros + make-chan/array probes all flip accept → reject with HYB-E1002; literal controls stay accept.
• cargo test --workspace green except the pre-existing spec_t3_strings failure, reproduced on clean HEAD (it reads SPEC-LANGUAGE.md's T3 fence — unrelated).
• Corpus: new pin typed-int-into-float64-reject.hy (EXPECTED-VERDICT: REJECT BUILDABLE + EXPECTED-DIAGNOSTIC: HYB-E1002); census 126 → 127 on all 8 surfaces; harness 127/127, tripwire not fired; negative flip verified (exit 2 flipped / exit 0 restored).
• Differential loop: 40/40 accept (build+run) + 87/87 reject.
• Baseline check: without this change the new pin accepts (bug reproduced) while accept-side is byte-identical 40/40 — no accept regression.

Credit

• wminno — surface checklist in [Soundness] hy-check: assigning a typed int variable to a float64 variable passes check, fails go build (HYB-E1002) #1112.
• Stale companions: zidirahim617-del's PR checker: reject variable-to-variable assignments of different scalar types (closes #1112) #1130 (partial fix: 3 call sites + tests/assign_types.rs) and mhatipog's PR Fix/typed int float assignment 1112 #1117. This PR supersedes both with full-site coverage — happy to fold their test files in if the maintainer prefers.

[Ljunqiang]

Post-merge validation on upstream/main @ a785e49 (session-35: #1419/#1432/#1439/#1441 — disjoint files from this PR, corpus still 126): cargo test --workspace green (the local-only spec_t3_strings failure is a working-tree CRLF artifact in SPEC-LANGUAGE.md — reproduced on clean HEAD, passes on LF checkouts), census 127, harness 127/127 (tripwire not fired), differential 40/40 accept + 87/87 reject.

Census coordination: #1428 and #1430 also bump 126 → 127 with their own pins — whichever lands last should rebase the count to 128/129 on all 8 surfaces (same pattern as the #906/#905 sibling take).

[limitlessprojects]

Queue-reset sweep: this repo now runs a rolling conveyor, and open PRs
outside the active review window are being closed as not planned to keep the
queue honest — a stale queue wastes every contributor's time, including yours.

This is not a verdict on the work. If this change still matters:

1. rebase against current main (main has landed a lot since this PR opened),
2. confirm the diff still applies and the affected verdicts still hold,
3. reopen — the triage board picks reopened PRs up as fresh arrivals.

Wildcard holds (verdict window 2026-11-01) and Pareto land candidates are
unaffected. Thanks for the effort — high-volume repos only stay reviewable
with a queue that actually drains.