docs(i18n): inventory backend and Python translation boundaries

This commit is contained in:
cesnimda
2026-11-17 09:01:41 +01:00
parent 07fbd5f72c
commit 5d8d5d6ff7
6 changed files with 3685 additions and 3 deletions
+2
View File
@@ -1,5 +1,7 @@
# I18N-00 baseline audit
17 November 2026 addendum: [B01 backend message census](BACKEND_MESSAGE_CENSUS_B01.md), on `codex/i18n-message-census` at immutable source `07fbd5f72c3605d6b2b5b31332ba5786f9446309`, adds 110 contextual source loci from seven owners and eight source findings. Seven raw Git-blob hashes and 110 expression/line hashes passed 117 integrity checks; unique IDs/loci and counts reconciled. Product tests were not run. This is partial research; full I18N-00 acceptance, formal stages01–08 and publication remain blocked by their existing gates. EU24 plus wider European scope is unchanged. Historical 188 dynamic-call reviews are complete at `4172c23d7ce13e8327a5e51bbb09f296c24fb9b8` per central tracking; the older remaining-80 paragraphs below describe the original November2 checkpoint, not current work. No application/catalogue/schema changes, translations or approval decisions occurred.
Date: 2026-11-02. Verdict: **IN_PROGRESS; partial research checkpoint, not full I18N-00 acceptance. I18N-01 through 08 BLOCKED. Production NO-GO.**
Research only, authorised by master prompt and continuous-progress instruction. Branch `feature/i18n-00-audit`, worktree `C:/Users/Connor/.codex/worktrees/i18n-00-audit/Job tracker`, baseline `5cb903b989742b7a96cb22ddf4c0e24993b18980`. Application/API/tests/UI/sidecar diff against `ba04d817` was empty. This is the accepted conditional development handoff, not blanket runtime, language or production acceptance. No controller or approval state was changed. All ten supplied I18N prompts were read. Their implementation requirements are future specifications, not authority to execute them.
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,47 @@
# B01: backend and processing message census
17 November 2026. Research-only checkpoint on `codex/i18n-message-census`, based on immutable A3 source `07fbd5f72c3605d6b2b5b31332ba5786f9446309`. The [inventory](BACKEND_MESSAGE_CENSUS_B01.json) binds seven exact Git blobs by SHA-256 and 110 individually reviewed source loci by line, exact expression and line hash. Its unit is a source locus, not a unique sentence, translated key, test pass or defect. A locus can contain multiple literals; related duplicate/caller lines do not add records. `sourceLiteralSegments` are raw lexical captures preserving source escapes, including protocol identifiers and fragments; they are not a translation list. The complete expression is authoritative where interpolation contains nested quotes.
Each record partitions those raw captures into `translationCandidateSegments`, `nonTranslatableSegments` and `unresolvedSegments`: 155 segment occurrences = 102 prose/template candidates + 47 non-translatable segments + six unresolved segments. These are occurrence counts, not unique messages or accepted translations. Action/error/dictionary/status identifiers (including `login`, parser codes and `Applied`/`Waiting`/`Offer`), separators, the product signature and the date-format pattern are excluded from translator candidates. Date formatting remains a formatter contract rather than a translated sentence. `embeddedProtocolTokens` separately protects `en`/`nb-NO` and `min_length`/`max_length` inside prose; placeholders and proper names retain the documented context and private/runtime exclusions.
Four diagnostic segments remain unresolved pending display/exposure/owner decisions. Two greeting fragments at reminder L84 remain unresolved because the lexical extraction cannot parse nested C# interpolation; its `there` fallback is present in the exact expression but is not separately captured. No inferred replacement segment or translation guidance was invented. A partition of the captured segments does not establish complete language parsing or message extraction.
The scope remains all 24 EU official languages plus broader European national/official, script and regional/minority candidates in the existing country/language matrix. This batch identifies source-language contracts those locales will need; it adds no language pack or country relationship and changes no catalogue/schema or release status.
## Reviewed owner batch
| Owner | Loci | Reviewed scope |
| --- | ---: | --- |
| AuthController.cs | 24 | Selected login/registration validation, language preference, security mail, delivery errors and avatar failures |
| SmtpEmailSender.cs | 4 | Plaintext transport boundary, suppressed-send diagnostic and configuration exceptions |
| FollowUpReminderHostedService.cs | 15 | Subject/body fragments, date/missing-value formatting, reason/status boundaries |
| UserNotificationStore.cs | 1 | Owner-scope diagnostic; the store does not construct notification prose |
| AiOperationQueue.cs | 7 | Admission, timeout and upstream message propagation |
| CvProcessingQueue.cs | 5 | Disabled worker, invalid/missing run, cancellation/timeout and failure fallback |
| tools/summarizer/app.py | 54 | Selected routing errors, all 24 entries in `_PARSER_ERRORS`, parser/upload errors and generated-summary headings/frames |
| Total | 110 | Partial contextual census, not exhaustive even within these files |
Categories reconcile exactly: 86 controlled-message loci; eight controlled-output headings; four controlled-output templates; six upstream-message boundaries; four diagnostic templates; one private-content boundary; one machine-code locus. Machine codes also appear beside prose in mixed loci; the category counts do not count every code token. Every record has proposed namespace, owner, context, source binding, typed parameter/sensitivity notes where present and separate unperformed native/runtime review states. No actual private values were read or exported.
## Source findings
| ID | Evidence | Required future handling |
| --- | --- | --- |
| B01-F01 | AuthController L124–178, L608/L611 | Literal English validation coexists with invariant codes and upstream Identity descriptions. Preserve authentication semantics; map controlled display separately from codes. `en`/`nb-NO` is the present endpoint limitation, not the target language scope. |
| B01-F02 | AuthController L343–344, L655/L667/L725, L1065–1066, L1176–1177; SMTP L22/L47–49 | Security email templates have no locale argument in the inspected send contract. Preserve token-bearing URLs as opaque runtime parameters, proper names and security meaning; inspect all template callers and recipient-language ownership before implementation. |
| B01-F03 | Reminder L79–109 | English frames, long dates without explicit recipient culture, raw stored status and rules-engine reason are composed together. Separate controlled labels/formatting from original job/company/name content and stable status identities. Rules-engine producer remains unreviewed. |
| B01-F04 | Auth L178/L611/L1184; AI queue L80/L203–210; CV L98/L105/L114 | Identity, usage, handler and CV outcome messages cross owner boundaries. This batch does not establish their complete value domains, parameter safety or client display consumers. Trace producers and consumers before cataloguing arbitrary exception text. |
| B01-F05 | SMTP L37/L41/L42; NotificationStore L54 | Diagnostic text is distinct from user notifications and mail. SMTP suppression log parameters include email/subject metadata; exclude their runtime values from translation exports. Preserve owner-scope validation. This is classification, not a security acceptance finding. |
| B01-F06 | AI queue L51–70/L218; CV queue L37/L67/L84/L88/L98 | Queue messages require stable category/status/retry metadata independent of translated display. Cancellation and timeout are distinct outcomes; persisted messages need an explicit locale policy. |
| B01-F07 | Python L759–767/L1210–1235/L1309/L1348/L1400–1405 | Provider/parser codes select English HTTP detail. Keep codes/statuses invariant; translate controlled text at an approved owner boundary. Duplicate direct parser validations still need reconciliation and full caller review. |
| B01-F08 | Python L1103–1153 | Generated summaries combine English headings/suggestion frames and source-derived advertisement/skill content. Output language is an independent contract; changing interface locale must not rewrite source facts. Prompts and generated-content semantic quality are unqualified. |
## Validation and limits
Root review independently repeated all117 immutable file/record checks and110 disjoint segment-partition checks; the [validation receipt](BACKEND_MESSAGE_CENSUS_B01_VALIDATION.json) records155 raw literal occurrences:102 translation candidates,47 non-translatable and6 unresolved. These are occurrence classifications, not approved messages, translations or product tests. Application, locale catalogue and schema diffs are empty. The six unresolved captures remain excluded from translation-ready guidance.
Read exact bytes with `git show 07fbd5f72c3605d6b2b5b31332ba5786f9446309:<path>`; source SHA-256 covers raw Git bytes. An inline Python standard-library assertion check re-read all seven blobs, compared every selected expression and UTF-8 line hash, checked all related line ranges, unique record IDs/source loci and reconciled counts: **117 file/record binding checks passed (seven files plus 110 records)**. After segment refinement, all 117 bindings passed again; all 110 record partitions were checked as disjoint multiset partitions/subsets of their raw captures, and the 155/102/47/6 segment totals reconciled. Source expressions, record IDs and hashes were unchanged from the original B01 artifact. Embedded protocol tokens were verified against exact source expressions. These are inventory-integrity checks, not product tests. JSON parsed successfully. No new runner/dependency was added; temporary preparation files live outside the repository on E:.
Caller coverage is deliberately bounded to the producing owner and explicitly recorded same-file related lines. External response consumers, notification producers, Identity/usage/rules-engine producers, full auth/provider/upload validation, model prompts, lexical dictionaries, document renderer/font/PDF output, email HTML/preheaders and broader frontend visible/ARIA states remain open. No runtime route, queue, email delivery, provider, parser process, model, database, build or browser was invoked. Graph navigation located owners before source inspection; no graph rebuild ran.
The historical 188 dynamic frontend call domains are already accounted for at `4172c23d7ce13e8327a5e51bbb09f296c24fb9b8` (central tracker); they were not re-reviewed or counted in these 110 loci. Neither count establishes full visible-text coverage. I18N-00 remains partial; formal I18N-01–08 entry, Phase10/Product Review, human/native/legal/device and locale publication/release gates remain unchanged. Production NO-GO.
@@ -0,0 +1,15 @@
{
"status": "inventory-integrity-only",
"sourceCommit": "07fbd5f72c3605d6b2b5b31332ba5786f9446309",
"fileAndRecordBindingsPassed": 117,
"segmentPartitionsPassed": 110,
"segmentOccurrences": {
"sourceLiteralSegments": 155,
"translationCandidateSegments": 102,
"nonTranslatableSegments": 47,
"unresolvedSegments": 6
},
"nativeRuntimeTests": "NOT RUN",
"localeAcceptance": "NOT GRANTED",
"scope": "Partial seven-owner I18N-00 census; no app/catalogue/schema changes"
}
+8 -2
View File
@@ -6,7 +6,7 @@ Completed: source-owner audit draft; ordered implementation contracts; source-bo
Next authorised work, without implementation approval:
1. Resolve 188 dynamic/unknown calls, direct translation object access and wrappers. Review the 853 entries without static callers without treating them as unused.
1. Preserve the completed 188 historical dynamic-call domain reviews (`4172c23d7ce13e8327a5e51bbb09f296c24fb9b8`, central tracker; 108 original plus 80 supplemental, zero pending). Review current direct access/wrappers and the historical 853 entries without static callers without treating them as unused; do not repeat the accepted census.
2. Inventory hardcoded frontend/ARIA/empty/error/export text and backend/Python/email/queue strings. Assign real namespace, translator context, parameter sensitivity and owning component rather than inferring from keys.
3. Complete national/regional official-record checks beyond the initial Norway/Switzerland/Serbia checks. Expand territory/minority coverage where the current research envelope omits required contexts.
4. Evaluate actual installed/locked library compatibility and Montenegrin canonicalisation before choosing a runtime migration. Do not install packages now.
@@ -14,6 +14,12 @@ Next authorised work, without implementation approval:
The delegated researcher was interrupted by a service usage limit; the parent continued, corrected references to nonexistent deliverables and produced the actual inventory/catalogue/matrix. No agent is working in the background. Full stage completion must not be inferred from the presence of the expected filenames.
Continuation: CONTEXT_TRIAGE.md records all 188 dynamic calls, 108 finite-branch resolutions (240 existing-key references), four direct catalogue reads and 26 source-context-reviewed JSX/accessibility literal candidates. Seven concrete planning findings I00-13 through I00-19 added. Remaining dynamic call queue is 80; prior 188 count is the original inventory total, not the remaining queue. No application changes.
Historical continuation: CONTEXT_TRIAGE.md recorded 108 finite-branch resolutions (240 existing-key references), four direct catalogue reads and 26 JSX/accessibility literal candidates, with seven findings I00-13 through I00-19. Its remaining-80 accounting is superseded by the completed 188 historical reviews above. No application changes in that research checkpoint.
17 November 2026 continuation: `codex/i18n-message-census` starts at exact A3 `07fbd5f72c3605d6b2b5b31332ba5786f9446309`. [B01](BACKEND_MESSAGE_CENSUS_B01.md) adds 110 reviewed backend/Python source loci across seven owners, eight source findings and seven exact blob bindings. Inline standard-library assertions passed 117 file/record integrity checks plus unique IDs/loci and reconciled counts; no product test ran. This is a partial controlled-message/context census, with upstream and external consumers still open. Locale catalogue/schema, European scope, application code and all gates remain unchanged. No commit/push was performed by the census worker; root review/publication remains pending.
- [x] B01 bounded backend/Python/email/queue message classification and immutable source/line/hash evidence.
- [ ] Extend full visible-text census; trace upstream Identity/usage/rules-engine and notification producers plus external message consumers.
- [ ] Complete full locale/surface/native/market acceptance and applicable stage entry decisions. The missing local TODO.md was not created; central ROADMAP_TODO.md I18N-00 stays open.
Product tests NOT RUN; source equality and planning checks PASSED within their documented limits. Preserve existing PDF evidence. No application models, production services, orchestrator configuration or signing/approval material accessed.
+3 -1
View File
@@ -1,6 +1,8 @@
# Surface inventory and message accounting
All rows need a source-bound contextual census before I18N-00 acceptance. Existing-catalogue inventory is in `MESSAGE_INVENTORY.json`; partial JSX/accessibility literals are now recorded in `FRONTEND_LITERAL_CANDIDATES.json`, with direct reads and dynamic calls in separate inventories. This is not a 100% coverage claim. Namespace assignments remain provisional until caller review; translator context cannot be inferred safely from key names alone. The AST inventory identifies syntactically named `t` calls, not symbol-resolved calls. See CONTEXT_TRIAGE.md for reviewed literal context and the 108 resolved conditional calls; 80 table/template/wrapper calls remain.
All rows need a source-bound contextual census before I18N-00 acceptance. Existing-catalogue inventory is in `MESSAGE_INVENTORY.json`; partial JSX/accessibility literals are recorded in `FRONTEND_LITERAL_CANDIDATES.json`, with direct reads and dynamic calls in separate inventories. This is not a 100% coverage claim. Namespace assignments remain provisional until caller review; translator context cannot be inferred safely from key names alone. The AST inventory identifies syntactically named `t` calls, not symbol-resolved calls. The historical 188 dynamic-call domains are accounted for at `4172c23d7ce13e8327a5e51bbb09f296c24fb9b8` per the central tracker (108 original plus 80 supplemental, zero pending in that historical census); older CONTEXT_TRIAGE.md remaining-80 text is superseded accounting, not a current queue.
17 November 2026: [backend message batch B01](BACKEND_MESSAGE_CENSUS_B01.md) adds 110 source-context-reviewed loci from seven backend/Python owners at A3 `07fbd5f72c3605d6b2b5b31332ba5786f9446309`. Its [machine-readable records](BACKEND_MESSAGE_CENSUS_B01.json) preserve source hashes/expressions, proposed namespace, owner, typed parameters/sensitivity, categories and separate review states. Producing-owner/same-file caller coverage is bounded; upstream producers and external consumers remain open. No surface or additional locale is accepted by this increment.
| Surface / proposed namespace | Source owner / inspection target | Content boundary and required evidence |
|---|---|---|