cem_acpt 0.11.0 → 0.11.2

data/specifications/CEM-6715.md

@@ -0,0 +1,133 @@
+# CEM-6715 — Make the Windows GCS bucket configurable
+
+## Summary
+
+Replace the hard-coded `gs://win_cem_acpt` bucket in the Windows test
+path with a configurable value (`platform.gcp.windows_bucket`),
+defaulting to `win_cem_acpt` so internal puppetlabs users see no
+behavior change. Surface a `--windows-bucket` CLI override.
+
+Implements RFC 0006 (`docs/rfcs/0006-configurable-windows-bucket.md`).
+
+## Functional Behavior
+
+### Config schema
+- New nested key `platform.gcp.windows_bucket`.
+- Default: `'win_cem_acpt'` (preserves current behavior).
+- Read via the existing `config.get('platform.gcp.windows_bucket')`
+  dot-notation API.
+
+### CLI flag
+- New `--windows-bucket BUCKET` long flag (no short alias) on the
+  `cem_acpt` command (not `cem_acpt_image` — image-build path does not
+  use the bucket).
+- Sets `options[:platform][:gcp][:windows_bucket] = BUCKET`, which
+  merges into config above the user config / config-file layer.
+
+### Runtime use in `test_runner.rb`
+- New private helper `windows_bucket_uri` that returns
+  `"gs://#{config.get('platform.gcp.windows_bucket')}"` and raises a
+  clear error if the configured value is `nil` or empty.
+- `upload_module_to_bucket` (line 386) uses `windows_bucket_uri` instead
+  of the literal `'gs://win_cem_acpt'`.
+- `cleanup_bucket` continues to operate on `@run_data[:win_remote_module_path]`
+  (unchanged — already derived from upload).
+- `WinNode` (in `lib/cem_acpt/utils/winrm_runner.rb`) is updated to
+  accept the bucket URI and use it on its own hardcoded line 83
+  (`gcloud storage cp gs://win_cem_acpt/#{@mod_name} ...`). Without this
+  fix, the RFC's stated goal is not actually achieved.
+
+## Input/Output Contracts
+
+### `CemAcpt::TestRunner::Runner#windows_bucket_uri` (private)
+- Input: none (reads from `config`)
+- Output: `String` of the form `"gs://<bucket>"`
+- Raises: `RuntimeError` (message: `'platform.gcp.windows_bucket is not configured'`)
+  when the config value is nil or an empty string.
+
+### `CemAcpt::Utils::WinRMRunner::WinNode.new`
+- New signature: `initialize(login_info, mod_name, bucket_uri)`
+- `bucket_uri` is the full URI string, e.g. `'gs://win_cem_acpt'`,
+  passed in from the runner. The class no longer hard-codes the bucket.
+
+### CLI
+- `--windows-bucket BUCKET` writes to nested options as
+  `{ platform: { gcp: { windows_bucket: BUCKET } } }`.
+
+## Edge Cases
+
+- **Empty / nil bucket value (e.g. user sets `windows_bucket: ''` in
+  YAML).** `windows_bucket_uri` raises with a clear message before any
+  `gcloud` call is attempted.
+- **Bucket value with a `gs://` prefix** (e.g. user pastes a URL).
+  Out of scope. We expect a plain bucket name. Document the expectation
+  in the README and rely on `gcloud` to fail loudly if a malformed URI
+  is passed. (Avoiding silent normalization keeps the contract simple.)
+- **Linux-only test runs.** Neither `upload_module_to_bucket` nor
+  `cleanup_bucket` runs (gated by `tests.first.include?('windows')` and
+  `@run_data[:win_remote_module_path]` being set). The new config key
+  is unused — no error is raised.
+- **CLI flag interacting with `-O platform.gcp.windows_bucket=foo`.**
+  `-O` is parsed to nested options before `--windows-bucket`. If both
+  are supplied, the later (in argv order) wins because `OptionParser`
+  processes them sequentially and both write to the same key. Not a
+  new pattern; matches existing flag/`-O` interactions.
+
+## Constraints / Invariants
+
+- Default behavior unchanged for any existing user that does not set
+  `platform.gcp.windows_bucket`.
+- `valid_keys` validation: `platform` is already in
+  `BASE_VALID_KEYS`, so nested `gcp.windows_bucket` is implicitly
+  accepted (validation is top-level only). No schema gate to update.
+- Existing `Config::CemAcpt#defaults` already has `platform: { name: 'gcp' }`
+  — extend it to include `gcp: { windows_bucket: 'win_cem_acpt' }`.
+
+## Error Handling
+
+- Missing/empty bucket → `RuntimeError` from `windows_bucket_uri`,
+  raised before any `gcloud` invocation.
+- `gcloud` upload/remove failures continue to be handled exactly as
+  today (`Shell.run_cmd` raises on upload; `raise_on_error: false` on
+  cleanup).
+
+## Non-Goals
+
+- Refactoring how the Windows path detects a Windows run
+  (`tests.first.include?('windows')`). Out of scope; covered by
+  separate RFCs.
+- Generalizing the bucket idea to other platforms (no AWS S3, etc.).
+- Validating bucket-name format (`gcloud` does that already).
+- Reworking `WinNode` to take a richer config object — minimum
+  threading only.
+
+## Acceptance Criteria
+
+From the RFC:
+
+- [x] Config schema (`defaults`) extended with
+      `platform.gcp.windows_bucket = 'win_cem_acpt'`.
+- [x] CLI flag `--windows-bucket BUCKET` parsed in `Cli`.
+- [x] All references to the literal `'gs://win_cem_acpt'` removed
+      from `test_runner.rb` **and** `utils/winrm_runner.rb`.
+- [x] RSpec coverage for `windows_bucket_uri` (default, override,
+      empty/nil) and that `upload_module_to_bucket` /
+      `cleanup_bucket` use the configured bucket.
+- [x] `docs/ARCHITECTURE.md` §13 updated to describe the new key
+      and the line-220 ASCII diagram updated.
+- [x] README updated under "Testing with sce_windows" to mention
+      the new key/flag.
+
+## Files Touched
+
+- `lib/cem_acpt/config/cem_acpt.rb` — extend `platform` defaults.
+- `lib/cem_acpt/cli.rb` — add `--windows-bucket` flag.
+- `lib/cem_acpt/test_runner.rb` — add `windows_bucket_uri`, use it,
+  pass to `WinNode`.
+- `lib/cem_acpt/utils/winrm_runner.rb` — accept bucket URI, drop
+  hardcoded literal.
+- `spec/cem_acpt/test_runner_spec.rb` — coverage for the new helper
+  and call sites.
+- `docs/ARCHITECTURE.md` — §13 prose + line-220 diagram.
+- `README.md` — document new key/flag in the sce_windows section.
+- `specifications/CEM-6715.md` — this file.

data/specifications/CEM-6716.md

@@ -0,0 +1,160 @@
+# CEM-6716 — Tighten `--quiet` semantics and fix logging typos
+
+Implements RFC 0007 (`docs/rfcs/0007-logging-quiet-and-typos.md`).
+
+## Summary
+
+Two related logging defects:
+
+1. `--quiet` is silently a no-op when used without `--log-file`. The
+   CI-mode override that re-adds `$stdout` is undocumented in
+   `--help`.
+2. `lib/cem_acpt/test_runner.rb#upload_module_to_bucket` (line 401)
+   references `@run_data[:module_pakage_path]` — a typo for
+   `module_package_path` — which interpolates to an empty string in
+   the log line.
+
+Fix both, document the CI-mode override, add coverage.
+
+## Functional Behavior
+
+### `lib/cem_acpt.rb#initialize_logger!`
+
+Rewrite the body so that:
+
+- `--quiet` always drops `$stdout` from `logdevs`.
+- `--log-file FILE` always adds the file to `logdevs`.
+- If `config.ci_mode?` is true and `$stdout` was dropped, re-add it
+  and emit a one-time `warn`-level message that `--quiet` was
+  overridden so `::group::` directives can reach Actions stdout.
+- If `logdevs` is still empty after the above (i.e. `--quiet` outside
+  CI with no `--log-file`), raise a `RuntimeError` with the message
+  `--quiet without --log-file would silence all output; pass --log-file or drop --quiet`.
+- Verbose / level / formatter handling is unchanged.
+
+The warn is emitted via the freshly-built logger, which is safe
+because the warning destination is exactly the destination the user
+will end up with (`$stdout` and/or the log file). No new "warn-once"
+helper is introduced; `initialize_logger!` runs once per process, so
+"once" is implicit.
+
+### `lib/cem_acpt/test_runner.rb#upload_module_to_bucket`
+
+Rename `@run_data[:module_pakage_path]` → `@run_data[:module_package_path]`
+in the `logger.info` call on line 401. The non-typoed key on the next
+line already exists in `run_data` (set in `pre_provision_test_nodes`
+at line 188).
+
+### CLI help (`lib/cem_acpt/cli.rb`)
+
+Update the `-q / --quiet` description from `'Do not log to stdout'`
+to mention the CI override and the requirement to pair with
+`--log-file`. Target text:
+
+```
+'Drop stdout (requires --log-file outside CI; CI mode keeps stdout for ::group:: support)'
+```
+
+### Documentation (`docs/ARCHITECTURE.md` §15)
+
+Replace the "conditional / silently a no-op" prose with the new
+behavior:
+
+- `--quiet` *with* `--log-file FILE`: stdout dropped, logs go only to
+  `FILE`.
+- `--quiet` *without* `--log-file`: `RuntimeError` at startup with a
+  clear message.
+- CI mode (`GITHUB_ACTIONS`, `CI`, or `-I`): `--quiet` is overridden
+  so `$stdout` is re-added; a single warn line is emitted at startup
+  noting the override.
+
+Also remove or update the §19 item 6 ("worth a behavioral sanity
+check") — it has been resolved by this fix.
+
+## Input/Output Contracts
+
+### `CemAcpt#initialize_logger!` (private, instance method on the singleton)
+
+- **Input:** none (reads `@config`).
+- **Output:** `CemAcpt::Logging::MultiLogger` (unchanged return shape).
+- **Raises:**
+  - `RuntimeError("Config must be loaded before logger can be initialized")` — unchanged.
+  - `RuntimeError("--quiet without --log-file would silence all output; pass --log-file or drop --quiet")` — new.
+
+### `lib/cem_acpt/test_runner.rb#upload_module_to_bucket`
+
+No signature change; only the interpolation key in the log line changes.
+
+## Edge Cases
+
+| Case | `--quiet` | `--log-file` | CI mode | Result |
+| --- | --- | --- | --- | --- |
+| 1 | unset | unset | no | logs to `$stdout` |
+| 2 | unset | set | no | logs to `$stdout` and file |
+| 3 | set | unset | no | **raises** at startup |
+| 4 | set | set | no | logs only to file |
+| 5 | unset | unset | yes | logs to `$stdout` (CI does not add a duplicate) |
+| 6 | unset | set | yes | logs to `$stdout` and file |
+| 7 | set | unset | yes | logs to `$stdout` (override warn emitted) |
+| 8 | set | set | yes | logs to `$stdout` and file (override warn emitted) |
+
+CI mode is detected by `Config::Base#ci_mode?` (sees env vars
+`GITHUB_ACTIONS` / `CI` or the `-I` flag); the same source of truth
+as today.
+
+## Constraints / Invariants
+
+- The logger is always the `MultiLogger` returned by `new_logger` —
+  no logger-shape changes leak to callers.
+- The warn-once is emitted *after* the logger is built and *before*
+  `initialize_logger!` returns, so it lands in the same destinations
+  as the rest of the run's logs.
+- Behavior outside `initialize_logger!` (level, verbose, signal
+  handlers, trap-context handling) is untouched.
+
+## Error Handling
+
+- The new `RuntimeError` is allowed to bubble out of `run_cem_acpt` /
+  `run_cem_acpt_image`. `OptionParser` already surfaces option errors
+  by writing to stderr and exiting non-zero; we want analogous
+  behavior here. We do **not** catch and re-raise.
+
+## Non-Goals
+
+- Promoting `--quiet` to fully silence logs in CI (would break
+  `::group::`).
+- Adding a `--silent` / "really quiet" mode.
+- Changing `--verbose`, `--debug`, or `-I` behavior.
+- Refactoring `MultiLogger` or `Logger`.
+- Making `logger_warn_once` a real helper — not needed because
+  `initialize_logger!` runs once per process.
+
+## Acceptance Criteria
+
+From the RFC:
+
+- [ ] `lib/cem_acpt.rb#initialize_logger!` rewritten; comment cites
+      RFC 0007.
+- [ ] `:module_pakage_path` typo fixed in `test_runner.rb`.
+- [ ] RSpec covers the four logging combinations listed in RFC
+      (`-q`, `-q --log-file`, `--log-file` alone, `-q` in CI mode)
+      plus the new failure case (`-q` outside CI without
+      `--log-file`).
+- [ ] RSpec smoke test for `upload_module_to_bucket` asserts the log
+      line contains the actual `module_package_path`.
+- [ ] `docs/ARCHITECTURE.md` §15 updated; §19 item 6 removed (or
+      noted as resolved).
+- [ ] `--help` output for `--quiet` mentions the CI override and the
+      `--log-file` requirement.
+- [ ] `bundle exec rake spec` and `rubocop` are clean.
+
+## Files Touched
+
+- `lib/cem_acpt.rb` — rewrite `initialize_logger!`.
+- `lib/cem_acpt/test_runner.rb` — fix typo in
+  `upload_module_to_bucket`.
+- `lib/cem_acpt/cli.rb` — update `--quiet` description.
+- `docs/ARCHITECTURE.md` — §15 prose, §19 item 6.
+- `spec/cem_acpt_spec.rb` — `initialize_logger!` coverage.
+- `spec/cem_acpt/test_runner_spec.rb` — log-line smoke test.
+- `specifications/CEM-6716.md` — this file.

data/specifications/CEM-6717.md

@@ -0,0 +1,239 @@
+# CEM-6717 — Namespace dynamically-created platform classes
+
+## Summary
+
+Move the dynamically-generated platform class (currently `::Gcp`) out
+of the top-level constant table and into the `CemAcpt::Platform`
+namespace. Also move the per-platform mixin out of top-level
+(currently `::Platform`, defined in `lib/cem_acpt/platform/gcp.rb`)
+into a sibling namespace `CemAcpt::Platform::Mixin::<Name>`. Use
+`const_defined?(name, false)` for the cache check to avoid ancestor
+lookup, and camel-case multi-word platform names correctly. Update
+`docs/ARCHITECTURE.md` §6 and §19 to reflect the new behavior.
+
+Implements RFC 0008 (`docs/rfcs/0008-namespace-platform-classes.md`),
+plus the additional mixin relocation discussed during spec review.
+
+## Functional Behavior
+
+### `CemAcpt::Platform::Mixin` (new namespace)
+
+Add a new namespace module `CemAcpt::Platform::Mixin` that holds the
+per-platform mixin modules. The dynamic class will be loaded from the
+sibling `CemAcpt::Platform::<Name>` constant; the mixin lives at
+`CemAcpt::Platform::Mixin::<Name>`. Splitting the two avoids a
+self-shadowing constant (we cannot have both
+`CemAcpt::Platform::Gcp` the class and `CemAcpt::Platform::Gcp` the
+mixin) and keeps both names purely descriptive.
+
+The namespace is declared up front in `lib/cem_acpt/platform.rb`
+alongside the existing `module CemAcpt::Platform`:
+
+```ruby
+module CemAcpt
+  module Platform
+    module Mixin; end
+    # ... existing content
+  end
+end
+```
+
+### `lib/cem_acpt/platform/gcp.rb`
+
+Rename the module from top-level `module Platform` to
+`module CemAcpt::Platform::Mixin::Gcp`. Method bodies, requires, and
+the file path are unchanged.
+
+```ruby
+# Before
+module Platform
+  def platform_data
+    ...
+  end
+  ...
+end
+
+# After
+module CemAcpt
+  module Platform
+    module Mixin
+      module Gcp
+        def platform_data
+          ...
+        end
+        ...
+      end
+    end
+  end
+end
+```
+
+The full body of the existing top-level `Platform` module — every
+public and private method — moves verbatim into
+`CemAcpt::Platform::Mixin::Gcp`.
+
+### `CemAcpt::Platform.platform_class(base_type, platform)`
+
+Replace the body of `platform_class` so that:
+
+1. `const_name` is built from the platform name with multi-word
+   support: `platform.split(/[_-]/).map(&:capitalize).join`
+   (e.g. `gcp` → `Gcp`, `aws_govcloud` → `AwsGovcloud`,
+   `foo-bar` → `FooBar`).
+2. The cache check looks for the constant **directly on
+   `CemAcpt::Platform`**, with the second argument `false` so the
+   ancestor chain is not consulted:
+   `CemAcpt::Platform.const_defined?(const_name, false)`.
+3. On cache hit, return `CemAcpt::Platform.const_get(const_name, false)`.
+4. On miss:
+   a. `require_relative "platform/#{platform}"` to load the file.
+   b. Resolve the mixin via
+      `mixin = CemAcpt::Platform::Mixin.const_get(const_name, false)`.
+      A `NameError` here means the loaded file did not define
+      `CemAcpt::Platform::Mixin::<Name>` — let it propagate; the
+      message is already actionable.
+   c. Build the class with the resolved mixin:
+      `Class.new(baseklass) { include mixin }`.
+   d. Register via `CemAcpt::Platform.const_set(const_name, klass)`.
+5. No top-level `Object.const_set` call. The `::<Platform>` constant
+   must not be created as a side effect.
+
+The previous `include Platform` lexical-lookup hack is removed — the
+mixin is now resolved by an explicit `const_get` against the new
+namespace, which is what RFC 0008 was reaching for.
+
+### Logging
+
+`platform_class` already logs the cache hit / miss / final-class lines.
+Keep them; they remain useful for `-d` runs. The wording can stay the
+same — `class_name` becomes `const_name` in the log.
+
+## Input/Output Contracts
+
+### `CemAcpt::Platform.platform_class(base_type, platform)`
+- Inputs unchanged: `base_type` is `:base` or `:test`; `platform` is a
+  string from the file basenames in `lib/cem_acpt/platform/`.
+- Output unchanged: a `Class` object that subclasses
+  `CemAcpt::Platform::Base` (`:base`) or
+  `CemAcpt::Platform::TestBase` (`:test`).
+- New invariant: the returned class is reachable via
+  `CemAcpt::Platform.const_get(const_name, false)` and is **not**
+  registered on `Object`.
+- Existing error semantics preserved: invalid `base_type` raises
+  `CemAcpt::Platform::Error`.
+
+### Public API (`use`, `get`)
+- No signature changes. Callers continue to receive the same class
+  objects. Anything that holds a reference to the returned class keeps
+  working; there is no top-level `::Gcp` reference to break.
+
+### Platform file contract (new)
+- Each `lib/cem_acpt/platform/<name>.rb` MUST define
+  `CemAcpt::Platform::Mixin::<CamelCaseName>` and define
+  `#platform_data` and `#node_data` on it.
+- Files are loaded by `platform_class` via `require_relative`; loading
+  happens at most once per platform per process.
+
+## Edge Cases
+
+- **Cache hit across calls.** Memoization still works because
+  `CemAcpt::Platform.const_defined?(const_name, false)` returns true
+  on the second call. `const_get(..., false)` returns the same object.
+- **Multi-word platform names.** `aws_govcloud` → `AwsGovcloud`;
+  `azure-arm` → `AzureArm`. Both the dynamic class
+  (`CemAcpt::Platform::AwsGovcloud`) and its mixin
+  (`CemAcpt::Platform::Mixin::AwsGovcloud`) follow the same form.
+- **Same name registered elsewhere.** A future `class CemAcpt::Foo` or
+  top-level `class ::Gcp` will not collide because:
+  - We register on `CemAcpt::Platform`, not `Object`.
+  - We use `const_defined?(name, false)`, which ignores ancestors.
+- **Mixin missing or misnamed.** If a platform file does not define
+  `CemAcpt::Platform::Mixin::<Name>`, `const_get` raises `NameError`.
+  The error is allowed to propagate — its message names the missing
+  constant, which is enough for a contributor to fix the file.
+- **Both `:base` and `:test` requested for the same platform.** Today
+  the second call returns the cached class regardless of `base_type`
+  — the first call's `base_type` wins. This is preserved (it is
+  out of scope for this ticket; the RFC does not call it out).
+- **Re-loading.** Tests that exercise the dynamic path multiple times
+  may need to clear `CemAcpt::Platform.<Name>` between examples; the
+  new spec covers this with a `before`/`after` that removes the
+  constant if defined. They should not need to clear `Object`
+  constants any more.
+
+## Constraints / Invariants
+
+- `CemAcpt::Platform::Base` and `CemAcpt::Platform::TestBase` are
+  unchanged.
+- `platforms` enumeration logic is unchanged.
+- `const_defined?(name, false)` does not look up ancestors, so a
+  reopened `Object::Platform` (unlikely but possible) cannot leak in.
+- The mixin contract is now explicit: each platform file defines a
+  module under `CemAcpt::Platform::Mixin`. There is exactly one
+  platform file in the repo today (`gcp.rb`), so the migration is
+  self-contained.
+
+## Error Handling
+
+- Unknown `base_type` still raises `CemAcpt::Platform::Error` with the
+  existing message `"Base type #{base_type} is not supported"`.
+- Unknown `platform` is still rejected by `use` / `get` before reaching
+  `platform_class` (existing checks).
+- Missing mixin in a loaded file raises `NameError` from `const_get`.
+  No new error classes; the stock `NameError` message identifies the
+  missing constant.
+
+## Non-Goals
+
+- Switching to a registration DSL (`Platform.register :gcp do ... end`).
+  The RFC defers this; we keep the dynamic-class approach.
+- Caching `:base` and `:test` variants under separate constant names.
+  The cache currently keys on platform alone; preserved.
+- Any change to GCP behavior. `gcp.rb`'s methods move byte-for-byte
+  into the new namespace.
+
+## Acceptance Criteria
+
+From RFC 0008 (`docs/rfcs/0008-namespace-platform-classes.md`), plus
+the mixin relocation:
+
+- [ ] `Platform.platform_class` registers the dynamic class under
+      `CemAcpt::Platform`, not on `Object`.
+- [ ] No top-level `::Gcp` constant is created at runtime, verified
+      by spec.
+- [ ] No top-level `::Platform` mixin is created at runtime — the
+      mixin lives at `CemAcpt::Platform::Mixin::Gcp`. Verified by
+      spec.
+- [ ] Multi-word platform names (`aws_govcloud`, `azure-arm`) produce
+      valid CamelCase constant names.
+- [ ] `docs/ARCHITECTURE.md` §6 and §19 (item 5) updated.
+- [ ] `spec/cem_acpt/platform/gcp_spec.rb` updated to reference the
+      new mixin name (`CemAcpt::Platform::Mixin::Gcp`) and continues
+      to pass.
+
+## Files Touched
+
+- `lib/cem_acpt/platform.rb` — declare `module Mixin` namespace;
+  rewrite `platform_class` per the proposal.
+- `lib/cem_acpt/platform/gcp.rb` — relocate the mixin from top-level
+  `module Platform` to `CemAcpt::Platform::Mixin::Gcp`.
+- `spec/cem_acpt/platform_spec.rb` — **new** spec covering the
+  namespace, const-defined behavior, cache, multi-word naming, and
+  the absence of top-level `::Gcp` / `::Platform` constants.
+- `spec/cem_acpt/platform/gcp_spec.rb` — update `RSpec.describe`,
+  `host.extend(...)`, and any other references from `Platform` to
+  `CemAcpt::Platform::Mixin::Gcp`.
+- `docs/ARCHITECTURE.md` — §6 prose update (drop the "top-level
+  constant `Gcp`" note, describe the new
+  `CemAcpt::Platform::Gcp` / `CemAcpt::Platform::Mixin::Gcp` split)
+  and §19 item 5 update / removal.
+- `specifications/CEM-6717.md` — this file.
+
+## Test Plan
+
+1. `bundle exec rake spec SPEC=spec/cem_acpt/platform_spec.rb` for the
+   new namespacing tests.
+2. `bundle exec rake spec SPEC=spec/cem_acpt/platform/gcp_spec.rb` to
+   confirm the GCP mixin spec passes after the rename.
+3. `bundle exec rake spec` for the full suite.
+4. `rubocop lib/cem_acpt/platform.rb lib/cem_acpt/platform/gcp.rb spec/cem_acpt/platform_spec.rb spec/cem_acpt/platform/gcp_spec.rb`.

data/specifications/CEM-6718.md

@@ -0,0 +1,120 @@
+# CEM-6718 — Wire up or remove orphaned Bolt log formatters
+
+Implements RFC 0009 (`docs/rfcs/0009-bolt-log-formatter-cleanup.md`),
+**Option B** (delete the orphaned formatters).
+
+## Summary
+
+Two formatter classes — `BoltOutputFormatter` and `BoltErrorFormatter`
+— exist under `lib/cem_acpt/test_runner/log_formatter/` but are never
+matched by `LogFormatter.new_formatter`'s `case`. `BoltOutputFormatter`
+is `require_relative`'d in `log_formatter.rb` but unused;
+`BoltErrorFormatter` is neither required nor referenced. Both have
+shipped in this dead state for several releases.
+
+Per the RFC's recommendation, delete both files and the dangling
+`require_relative`. Add a one-line comment on
+`BoltSummaryResultsFormatter` documenting that it is the canonical
+Bolt formatter. Update `docs/ARCHITECTURE.md` §11 and §19 to reflect
+the cleanup.
+
+## Functional Behavior
+
+### File deletions
+
+- Delete `lib/cem_acpt/test_runner/log_formatter/bolt_output_formatter.rb`.
+- Delete `lib/cem_acpt/test_runner/log_formatter/bolt_error_formatter.rb`.
+
+### `lib/cem_acpt/test_runner/log_formatter.rb`
+
+Remove the dangling `require_relative`:
+
+```ruby
+require_relative 'log_formatter/bolt_output_formatter'  # remove
+```
+
+The `case` dispatch is unchanged — it already routes
+`CemAcpt::Bolt::SummaryResults` to `BoltSummaryResultsFormatter` and
+falls through to `StandardErrorFormatter` for `StandardError`.
+
+### `lib/cem_acpt/test_runner/log_formatter/bolt_summary_results_formatter.rb`
+
+Add a one-line YARD comment above the class declaration noting that
+it is the canonical Bolt formatter. Replace the existing one-line
+comment so the file does not gain extra noise. Target text:
+
+```ruby
+# Canonical formatter for Bolt subsystem results. Wired into
+# LogFormatter.new_formatter for CemAcpt::Bolt::SummaryResults.
+class BoltSummaryResultsFormatter < Base
+```
+
+### `docs/ARCHITECTURE.md`
+
+**§11 (Result aggregation & log formatting):** drop the paragraph at
+lines 691–694 that calls out `BoltOutputFormatter` /
+`BoltErrorFormatter` as orphaned files. The formatter table above it
+is unchanged.
+
+**§19 (item 6):** remove item 6 entirely (the "BoltOutputFormatter
+and BoltErrorFormatter are dead" entry). Renumber the subsequent item
+(currently item 7, "`lib/terraform/image/gcp/windows/` is an empty
+directory") to item 6.
+
+## Input/Output Contracts
+
+No public-API changes. The dispatcher signature
+`LogFormatter.new_formatter(result, *args, **_kwargs)` and its return
+contract are unchanged. The deleted classes were never wired in; no
+caller constructs them.
+
+## Edge Cases
+
+- A caller that pushes a hypothetical `Bolt::Cmd::Output` or
+  `Bolt::Cmd::OutputError` onto the results queue would have hit the
+  `StandardError` branch (or the `else` raise) before this change and
+  continues to do so after. Behavior is unchanged.
+- No spec file references either deleted class (verified by grep);
+  removing them does not break the test suite.
+
+## Constraints
+
+- RuboCop must remain clean.
+- `bundle exec rake spec` must pass with no new failures.
+
+## Invariants
+
+- `LogFormatter.new_formatter` continues to dispatch the same four
+  result types it dispatches today (`Goss::Api::ActionResponse`,
+  `Bolt::SummaryResults`, `StandardError`, else-raise).
+- The remaining formatter files under
+  `lib/cem_acpt/test_runner/log_formatter/` (`base.rb`,
+  `bolt_summary_results_formatter.rb`, `goss_action_response.rb`,
+  `goss_error_formatter.rb`, `standard_error_formatter.rb`) are
+  unchanged in behavior.
+
+## Error Handling
+
+No new error paths. The pre-existing `else raise ArgumentError`
+branch in `new_formatter` is unchanged.
+
+## Non-Goals
+
+- Re-implementing `BoltOutputFormatter` / `BoltErrorFormatter`. If the
+  need re-emerges, RFC 0009 documents Option A as the alternative
+  path; reintroducing them would be a small, separate patch.
+- Refactoring or restructuring `BoltSummaryResultsFormatter` itself.
+- Touching the Goss formatters.
+- Updating the RFC's status (the RFC remains the historical record of
+  the decision; the implementation closes the ticket).
+
+## Acceptance Criteria
+
+- [ ] `lib/cem_acpt/test_runner/log_formatter/bolt_output_formatter.rb` deleted.
+- [ ] `lib/cem_acpt/test_runner/log_formatter/bolt_error_formatter.rb` deleted.
+- [ ] `require_relative 'log_formatter/bolt_output_formatter'` removed from `lib/cem_acpt/test_runner/log_formatter.rb`.
+- [ ] One-line comment added to `BoltSummaryResultsFormatter`.
+- [ ] `docs/ARCHITECTURE.md` §11 paragraph about orphaned formatters removed.
+- [ ] `docs/ARCHITECTURE.md` §19 item 6 removed; subsequent item renumbered.
+- [ ] `bundle exec rake spec` passes.
+- [ ] `rubocop` is clean.