Commit 27d23e8

- [x] PR [#402](https://github.com/NobleFactor/devlore-cli/pull/402) opened; every leg green

except `test (windows-latest)`, as expected.

Measured against head `dd72af41`'s check-runs, uncapped: **45** `--- FAIL` lines, zero

`[build failed]`. The framework change delivered **−3, not −18**, and the estimate this plan

carried was wrong.

**Why.** 15 of the 45 still carry the leak signature (`TempDir RemoveAll cleanup: … being used by

another process`), and every one of them is a **test-side runtime environment that is never

closed** — not a production path. The framework contract is that whoever constructs an

environment closes it; these tests construct one and never do. That was true before this change

(the environment held `spec.Root`, unclosed) and remained true after (the environment mints its

own Root, unclosed). Relocating the mint could not fix it.

The proof is a natural experiment inside one package: `lifecycle_e2e_test.go`,

`lifecycle_starlark_test.go`, and `result_flow_starlark_test.go` each close their environment and

**none of their tests appear in the failure list**; `provider_test.go` (2 mints),

`lifecycle_api_test.go` (10), `gather_api_test.go` (1), `git_resume_test.go` (1), and

`receipt_integration_test.go` (1) close nothing and **all of their tests do**. 15 unclosed mint

sites, 15 leak failures.

**The estimate's error, named:** the #393 diagnosis attributed these 18 failures to production

pre-Run error paths and the resume family. Those production leaks were real and are fixed (lore's

closed-Root loop, verify/readback never closing, devlore-test's aliasing) — but they were not what

these tests were failing on. The verification that preceded the option-4 ruling asked whether

anything dereferences `spec.Root` between spec construction and Run (it does not, and that holds);

it never asked who closes the environment in the failing tests. Inherited estimates get their own

enumeration — see [[verify-ledger-facts-by-enumeration]].

**No regressions:** the other 30 failures are the standing write/path/chown grind plus the known

singles, unchanged, and no failure anywhere traces to the new mint or preflight code paths.

- [x] `t.Cleanup(func() { _ = environment.Close() })` at all 15 mint sites across the five files

above, matching the idiom the three already-passing helpers use.

- [x] Verified by enumeration rather than assumption: repo-wide, every test file that sets a root

now closes it. The files that mint without a root (`pkg/op` internals, `instance_test.go`,

`builder_test.go`) hold no handle; `TestNewRuntimeEnvironment_BadAnchorFails` mints nothing

because the mint is what fails.

- [x] `make test` zero failures; `make lint-all` 0 issues per GOOS; gofmt clean.

- [ ] Second windows measurement. Expected 45 → ~30 — and unlike the first estimate, this one

names its mechanism: 15 unclosed handles, 15 cleanups.

t.Cleanup(func() { _ = environment.Close() })

t.Cleanup(func() { _ = environment.Close() })

t.Cleanup(func() { _ = environment.Close() })

t.Cleanup(func() { _ = environment.Close() })

t.Cleanup(func() { _ = environment.Close() })

t.Cleanup(func() { _ = environment.Close() })

t.Cleanup(func() { _ = environment.Close() })

t.Cleanup(func() { _ = environment.Close() })

t.Cleanup(func() { _ = environment.Close() })

t.Cleanup(func() { _ = resumeEnvironment.Close() })

t.Cleanup(func() { _ = host.Close() })

t.Cleanup(func() { _ = environment.Close() })

t.Cleanup(func() { _ = environment.Close() })

t.Cleanup(func() { _ = environment.Close() })