Add retention policy knowledge to the privacy domain (#177)
Some checks are pending
Validate knowledge index / validate-index (push) Waiting to run
Validate AL review fixtures / validate-review-fixtures (push) Waiting to run
Validate skill index and report schemas / validate-contract (push) Waiting to run
Validate frontmatter and structure / validate (push) Waiting to run

* Add retention policy knowledge to the privacy domain

Two articles covering retention policies for extension-owned tables,
the gap that lets high-volume log tables grow unbounded:

- register-owned-log-tables-for-retention-policies: an extension's own
  log tables must be added to the allowed-tables list from install AND
  upgrade code, guarded by IsAllowedTable plus an upgrade tag, with a
  mandatory minimum retention where audit needs one.
- ship-a-default-retention-policy-setup: registration only makes a table
  selectable; nothing is deleted until a Retention Policy Setup record
  exists, so ship one (disabled by default) as the platform's own
  Retention Policy Installer does.

Each ships good/bad AL samples. Claims verified against the BC admin
docs and the Retention Policy module in microsoft/BCApps.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011gvTjm746MtJEVWeRTbG46

* Address review feedback on retention policy knowledge

- Scope both articles to bc-version [17..] (retention policies shipped in v17).
- Allowed-tables sample: add OnRefreshAllowedTables subscriber with a
  ForceUpdate path; the upgrade tag now gates one-time setup only.
- Default-policy sample: use Retention Policy Setup.FindOrCreateRetentionPeriod
  instead of a hand-rolled lookup-then-insert that can collide on code.
- Anti-pattern now keys on append-only tables rather than table names.
- al-privacy-review: add retention-policy tokens and deterministic routing
  for both articles, with a bounded per-table text search for delete and
  registration paths.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Address second review round on retention policy knowledge

- Scope both articles to bc-version [22..]: FindOrCreateRetentionPeriod
  first appears in the 21.1 System Application and OnRefreshAllowedTables
  in 22.
- Reframe the default-setup article as optional guidance; registration
  without a setup is valid. The anti-pattern and review routing now cover
  only false claims that registration alone cleans up data.
- Make all four samples self-contained: declare Contoso Activity Log in
  each, add the Retention Policy Setup permission, and guard the default
  setup on IsAllowedTable.
- Let evaluation overrides list additionalArticles and register both
  retention pairs as extra privacy cases (38 cases, existing IDs unchanged).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Revert evaluation harness change for multiple articles per domain

The harness intentionally evaluates one paired article per domain. Keep it
as designed; how the retention pairs join privacy evaluation is left to the
maintainers.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Register retention policy pairs in the privacy evaluation override

Use main's articles override so both retention article pairs get positive
and clean cases alongside no-pii-in-telemetry-message-string.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Jeremy Vyska <jeremy@sparebrained.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Jeremy Vyska 2026-10-05 15:02:04 +02:00 • committed by GitHub
parent 7726d5d8d8
commit 02e7ab15b0
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
8 changed files with 333 additions and 2 deletions

View file

@ -37,11 +37,17 @@ Discard files that are not applicable. Retain conditionally applicable files (an
Narrow the relevant files to the subset that applies to the changes under review. Exclude test codeunits, test libraries, test helper code, files under test/Test/Tests paths, and objects with `Subtype = Test`; test data is synthetic and does not ship to customers. For each relevant file, compute overlap against:
- The changed AL object names and types — especially tables and tableextensions (for `DataClassification` on fields), codeunits that call `Error`, `Session.LogMessage`, or `FeatureTelemetry`, codeunits performing outgoing HTTP requests with customer data, migration codeunits, and objects reading or writing `IsolatedStorage`.
- The changed AL object names and types — especially tables and tableextensions (for `DataClassification` on fields), codeunits that call `Error`, `Session.LogMessage`, or `FeatureTelemetry`, codeunits performing outgoing HTTP requests with customer data, migration codeunits, objects reading or writing `IsolatedStorage`, and install/upgrade codeunits or new tables relevant to retention policies.
- The changed procedures and triggers, weighted toward those that call `Error`, construct `ErrorInfo`, call `Session.LogMessage`, `StrSubstNo`, `GetLastErrorText`/`GetLastErrorCallStack`, `FeatureTelemetry.LogUsage`/`LogUptake`/`LogError`, `HttpClient.Post`/`Get`, `IsolatedStorage.Set`/`SetEncrypted`/`Get`, or `PrivacyNotice.GetPrivacyNoticeApprovalState`.
- Tokens extracted from the diff that relate to privacy (`DataClassification`, `CustomerContent`, `EndUserIdentifiableInformation`, `EndUserPseudonymousIdentifiers`, `SystemMetadata`, `ToBeClassified`, `PrivacyNotice`, `ErrorInfo`, `GetLastErrorText`, `GetLastErrorCallStack`, `TelemetryScope`, `FeatureTelemetry`, `CustomDimensions`, `LogUsage`, `LogUptake`, `LogError`, `ErrorText`, `ErrorCallStack`, `alErrorText`, `alErrorCallStack`, `HybridSL`, `HybridGP`, `HybridBC`).
- Treat `ErrorInfo.Message`, `ErrorInfo.DataClassification`, `ErrorInfo.ErrorType`, and `ErrorInfo.DetailedMessage` as qualified member signals: accept a call or assignment only when symbol resolution proves that its receiver expression or variable has type `ErrorInfo`. Normalize those accesses to `errorinfo-message`, `errorinfo-dataclassification`, `errorinfo-errortype`, and `errorinfo-detailedmessage` retrieval tokens. Bare `Message` or `DataClassification` tokens MUST NOT trigger this article; do not emit the qualified tokens for `Message(...)` dialog calls, table or table-field `DataClassification` properties, or similarly named members on other types. Resolve the receiver's declaration from the containing object when it is outside the changed hunk.
- Worklist ErrorInfo privacy guidance only from those typed `ErrorInfo` member tokens or from construction of an `ErrorInfo` value. For every `FeatureTelemetry.LogError`, inspect the dedicated error text and call-stack arguments in addition to explicit custom dimensions.
- Retention-policy signals. Emit `addallowedtable` for any call to `Codeunit "Reten. Pol. Allowed Tables".AddAllowedTable`, `retention-policy-setup` for any use of `Record "Retention Policy Setup"`, and `retention-policy` for either. Emit `append-only-table` for an extension-owned table (never a tableextension) that the diff adds, or into which the diff adds a new `Insert` path, when the table also passes this bounded check: search the app's `.al` files for the table's object name once each for `AddAllowedTable`, `Delete`, and `DeleteAll`, and emit the token only when none of the three matches. This is a text search per candidate table — do not read or resolve the rest of the app. Emit `deleteall` when the diff adds a `DeleteAll` on an extension-owned table filtered by a date or datetime field inside a job queue codeunit or other scheduled cleanup and the table has no `AddAllowedTable` call.
Route retention-policy guidance deterministically:
- `register-owned-log-tables-for-retention-policies.md` is worklisted by `append-only-table` or `deleteall`, or when an `AddAllowedTable` routine is reached only from an install codeunit, or is guarded by an upgrade tag with no `OnRefreshAllowedTables` subscriber that bypasses the tag. Do not use the table's name as the trigger: an append-only table is a candidate whatever it is called ("Incoming Data" as much as "Activity Log"). A name ending in `Log`, `Entry`, `Archive`, `History`, or `Buffer`, an `AutoIncrement` integer primary key, or inserts from event subscribers, job queue codeunits, or API/web-service handlers raise confidence to `medium`; without any of these the finding stays `low`. Findings from `append-only-table` are advisory (`minor`) because table volume cannot be observed from source.
- `ship-a-default-retention-policy-setup.md` is worklisted by `addallowedtable` or `retention-policy-setup`, but registration without a `Retention Policy Setup` is valid and MUST NOT produce a finding. Report only a false claim: a `Message`, notification, label, page `AboutText`/`InstructionalText`, or setup text in the diff stating that the registered table's data is now cleaned up or deleted automatically when no enabled `Retention Policy Setup` is created for it.
A file enters the candidate worklist when its `keywords` intersect the extracted tokens or its topic (derived from the index entry's `path`, `title`, and `description`) matches a changed object type. Apply the topic-specific gates above after this overlap check; in particular, bare `Message` and `DataClassification` tokens cannot admit ErrorInfo guidance. Read an article's full file — its `## Best Practice` / `## Anti Pattern` bodies — only after it makes the worklist; candidate selection uses the index alone.