[cemililik]

Summary

The purge reference and the user manual said that the implicit --output-dir fallback "emits a WARNING naming the resolved dir". It did not: nothing was logged. The fallback is the corpus's directory, and the memorisation check (data.erasure_warning_memorisation) looks for <output_dir>/final_model/ there, a place a training run rarely writes its model. A default forgelm purge --row-id X --corpus data/train.jsonl therefore ended with a completed-erasure record and no memorisation event, so a reader of the audit chain alone would conclude that no model had consumed the row.

--row-id without --output-dir now logs one warning naming the resolved directory and the final_model/ path it will look for, which makes the documented behaviour true. The documentation also described the condition as "a final_model/ exists for any run that consumed the corpus"; the code checks one literal directory and cannot tell which runs consumed the corpus. It now says that.

Related issues

Refs #102 — the warning and the documentation. Reading training.output_dir from --config and scanning final_model* / checkpoint-* is the larger fix; it changes the content of an erasure audit event (additively) and wants its own review, so it stays open.

Changes

• forgelm/cli/subcommands/_purge.py: when --output-dir is omitted, a WARNING names the corpus directory used for the audit log, the salt and the memorisation check, the final_model/ path that check looks for, and tells the operator to pass --output-dir <training output dir> (also needed to correlate with forgelm reverse-pii).
• tools/check_module_size.py: the purge module's budget goes from 1216 to 1224 lines (+8) with a history note. This follows the project's documented option ("land the split or raise the budget with a budget_history note") and the previous raise's reasoning (C-18): the message has to sit where the default is resolved.
• Documentation, English and Turkish: the audit-event catalogue, the purge reference, the GDPR erasure guide and user-manual page, and the risk-treatment plan.

Type

• Bug fix

Testing

• tests/test_gdpr_erasure.py::TestImplicitOutputDirWarning: the warning fires without --output-dir, names both paths, and is absent when --output-dir is given.
• The first gauntlet run of this branch failed tests/test_check_module_size.py (the budget); that is fixed by the second commit. The full CLAUDE.md gauntlet from a clean worktree of the final state: exit 0.

Checklist

• English and Turkish mirrors updated together
• CHANGELOG entry under [Unreleased]
• Tests added

🤖 Generated with Claude Code

Summary by Sourcery

Make implicit purge output-directory behavior visible and align memorisation-check documentation with the implementation.

Bug Fixes:

• Log a warning when purge --row-id implicitly uses the corpus directory as --output-dir, including the exact final_model/ path checked for memorisation.

Enhancements:

• Clarify that memorisation detection checks only for a literal final_model/ directly under --output-dir and cannot identify which training runs consumed the corpus.

Build:

• Increase the purge module size budget to accommodate the warning logic and document the budget change.

Documentation:

• Update the English and Turkish purge references, audit-event catalogues, GDPR erasure guides, user manuals, risk-treatment plans, and changelog to describe the actual output-directory and memorisation-check behavior.

Tests:

• Add coverage for implicit-directory warning content, silence with an explicit output directory, and absence of a memorisation event when no model exists in the resolved directory.

[sourcery-ai]

Sorry @cemililik, you've used your own review budget of 250,000 diff characters for the last 7 days.

You can request another review in 12 hours and 58 minutes by commenting @sourcery-ai review. Upgrade to get a review now.

[coderabbitai]

Warning

Review limit reached

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

Next included review available in 1 minute.

Check out review usage here.

View limit details

Limit details: You’ve used all 3 included reviews currently available. Your 48 included PR review attempts over the past 7 days set your current allowance at 3 reviews per hour.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

• Configuration used: defaults
• Review profile: CHILL
• Plan: Essentials
• Run ID: 40657c29-bbf1-468e-a98b-f94452503c65

📥 Commits

Reviewing files that changed from the base of the PR and between 56b4f0d and 9d1045c.

📒 Files selected for processing: 14

• CHANGELOG.md
• docs/guides/gdpr_erasure-tr.md
• docs/guides/gdpr_erasure.md
• docs/qms/risk_treatment_plan-tr.md
• docs/qms/risk_treatment_plan.md
• docs/reference/audit_event_catalog-tr.md
• docs/reference/audit_event_catalog.md
• docs/reference/purge_subcommand-tr.md
• docs/reference/purge_subcommand.md
• docs/usermanuals/en/compliance/gdpr-erasure.md
• docs/usermanuals/tr/compliance/gdpr-erasure.md
• forgelm/cli/subcommands/_purge.py
• tests/test_gdpr_erasure.py
• tools/check_module_size.py

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

[sourcery-ai]

Reviewer's Guide

This PR makes the documented implicit --output-dir behavior real by logging the corpus-directory fallback and the exact final_model path inspected, adds focused regression tests, and updates English/Turkish operational documentation to clarify that memorisation detection is limited to a literal final_model directory directly under the selected output directory.

Sequence diagram for purge output directory fallback and memorisation check

sequenceDiagram
    participant Operator
    participant Purge as forgelm purge
    participant Filesystem
    participant Audit as Audit log

    Operator->>Purge: _run_purge_row_id(args, output_format)
    Purge->>Purge: Resolve output_dir from corpus directory
    Purge-->>Operator: logger.warning(resolved output_dir and final_model path)
    Purge->>Filesystem: os.makedirs(output_dir, exist_ok=True)
    Purge->>Filesystem: Check output_dir/final_model
    alt final_model exists
        Purge->>Audit: data.erasure_warning_memorisation
    end
    Purge->>Audit: data.erasure_completed

Loading

Flow diagram for purge memorisation detection scope

flowchart TD
    A[forgelm purge --row-id] --> B{--output-dir given?}
    B -- No --> C[Use corpus parent directory]
    C --> D[Log WARNING with resolved directory and final_model path]
    B -- Yes --> E[Use selected output directory]
    D --> F[Check exact output_dir/final_model directory]
    E --> F
    F --> G{final_model exists directly under output_dir?}
    G -- Yes --> H[Emit data.erasure_warning_memorisation]
    G -- No --> I[No memorisation warning]
    H --> J[Complete erasure audit]
    I --> J

Loading

File-Level Changes

Change	Details	Files
Warn when row purges implicitly resolve the output directory from the corpus path.	Resolve the fallback directory before creating audit artifacts. Emit a WARNING containing the resolved directory and the exact final_model path checked for memorisation. Recommend supplying the training output directory and preserve silence for explicit --output-dir usage. Add coverage for warning contents, explicit-directory silence, and the absence of a memorisation event when the model is elsewhere.	forgelm/cli/subcommands/_purge.py tests/test_gdpr_erasure.py
Correct documentation to describe the literal memorisation-check behavior and its operational limitation.	Replace claims about detecting any run that consumed the corpus with the actual direct-child final_model/ check. Document that the fallback is the corpus directory and that operators should provide the training output directory. Synchronize English and Turkish audit catalogs, purge references, GDPR guides, user manuals, and risk-treatment plans.	docs/reference/audit_event_catalog.md docs/reference/audit_event_catalog-tr.md docs/reference/purge_subcommand.md docs/reference/purge_subcommand-tr.md docs/guides/gdpr_erasure.md docs/guides/gdpr_erasure-tr.md docs/usermanuals/en/compliance/gdpr-erasure.md docs/usermanuals/tr/compliance/gdpr-erasure.md docs/qms/risk_treatment_plan.md docs/qms/risk_treatment_plan-tr.md
Record the bug fix and accommodate the purge module's enforced size budget.	Add an unreleased changelog entry covering the warning and corrected memorisation semantics. Increase the deferred module budget by eight lines and document the rationale in budget_history.	CHANGELOG.md tools/check_module_size.py

Possibly linked issues

• Memorisation warning looks only at <output_dir>/final_model, and output_dir defaults to the corpus folder, so a default invocation never warns (CR-A10-022) #102: The PR addresses this issue’s default-output-dir warning and documentation; broader model discovery and audit-event fixes remain open.
• #unknown: The PR directly resolves DOC-D1B05-016 by adding the warning the documentation described; it does not address the other listed findings.
• #DOC-TR08-002: The PR updates both GDPR guide mirrors to state the check only looks for final_model/ directly under --output-dir and cannot identify consuming runs.

---

Tips and commands

Interacting with Sourcery

• Trigger a new review: Comment @sourcery-ai review on the pull request.
• Continue discussions: Reply directly to Sourcery's review comments.
• Generate a GitHub issue from a review comment: Ask Sourcery to create an
  issue from a review comment by replying to it. You can also reply to a
  review comment with @sourcery-ai issue to create an issue from it.
• Generate a pull request title: Write @sourcery-ai anywhere in the pull
  request title to generate a title at any time. You can also comment
  @sourcery-ai title on the pull request to (re-)generate the title at any time.
• Generate a pull request summary: Write @sourcery-ai summary anywhere in
  the pull request body to generate a PR summary at any time exactly where you
  want it. You can also comment @sourcery-ai summary on the pull request to
  (re-)generate the summary at any time.
• Generate reviewer's guide: Comment @sourcery-ai guide on the pull
  request to (re-)generate the reviewer's guide at any time.
• Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
  pull request to resolve all Sourcery comments. Useful if you've already
  addressed all the comments and don't want to see them anymore.
• Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
  request to dismiss all existing Sourcery reviews. Especially useful if you
  want to start fresh with a new review - don't forget to comment
  @sourcery-ai review to trigger a new review!

Customizing Your Experience

Access your dashboard to:

• Enable or disable review features such as the Sourcery-generated pull request
  summary, the reviewer's guide, and others.
• Change the review language.
• Add, remove or edit custom review instructions.
• Adjust other review settings.

Getting Help

• Contact our support team for questions or feedback.
• Visit our documentation for detailed guides and information.
• Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

[cemililik]

Superseded by #999: the wave-0 fixes are combined by theme into five pull requests to cut the review load. This branch's commits are included there unchanged apart from the CHANGELOG.md merge; the branch stays on the remote.