- Add LITELLM = "litellm" to LLMProviderType enum (already in HEAD from prev commit)
- Add comment in LLMClient.__init__ explaining standard Bearer auth for LiteLLM
- Add regression test test_provider_type_enum_values — guard against value drift
- Add test_litellm_client_uses_default_bearer_auth — verify no extra headers
- Clean up duplicate @RELATION lines in models.py
- All 20 backend + 4 frontend tests pass
- Add _normalize_timestamp_value to key column values in orchestrator.py and executor.py before SQL generation
- Fix test mocks for .options(joinedload()) chain, explicit None attrs for mock job
- Add global translation run progress indicator (TranslationRunGlobalIndicator + store)
- Fix translate page import missing translationRunStore
- All 208 translate tests pass
- Enable response_format (json_object) for all providers including Kilo/OpenRouter
- Skip reasoning_effort for Kilo/OpenRouter (returns 400 — mandatory reasoning)
- Add structured_outputs fallback: retry once without response_format if upstream
provider (e.g. StepFun) rejects it with 'structured_outputs is not supported'
- Handle model refusal field (refusal) instead of treating as empty content
- Fix None-handling guards for message/finish_reason/content fields
- Apply same fixes to both preview.py and executor.py _call_openai_compatible
- Connect orphaned TermCorrectionPopup, BulkReplaceModal, BulkCorrectionSidebar
DeepSeek v4 Flash returns finish_reason=length when reasoning uses
all output tokens. Partial row extraction via regex recovers complete
rows from truncated JSON instead of failing with 400 error.
Also: extra_body reasoning_effort for Kilo/OpenRouter proxied providers
Added disable_reasoning toggle that works via:
1. API parameter: reasoning_effort: 'none' (DeepSeek, Qwen)
2. System prompt instruction: 'Respond directly with ONLY the JSON.
Do NOT include any reasoning, thinking, or chain-of-thought'
(works for ALL models universally)
Backend: model column, schema, preview+executor payload
Frontend: checkbox in LLM settings, i18n en+ru
New _token_budget.py calculates safe batch_size and max_tokens based on source text length, target languages, dictionary size, and model context window (64K DeepSeek v4 Flash). preview.py: auto-reduces sample_size when texts are long. executor.py: per-batch dynamic max_tokens. 12 new tests.
finish_reason=length from DeepSeek v4 Flash — model runs out of
output tokens due to reasoning_content (CoT) + multi-language (4 langs).
8192 tokens ensures complete JSON response for preview and execution.
Preview + executor now log finish_reason, message_keys, and raw
response structure when LLM returns empty content, enabling
diagnosis of DeepSeek API issues (empty content with finish_reason=length)
Preview + executor now log the raw LLM response (first 1000 chars)
when JSON parsing fails, enabling debugging of malformed LLM output.
Previously ValueErrors from JSON parse failures were silently returned
as 400 without any diagnostic data in logs.
Multi-language save fix:
- MultiSelect.svelte: selected prop changed to $bindable([])
(without $bindable, bind:selected never propagated to parent component,
so target_languages always stayed as default ['en'])
Target table column mapping (INSERT enhancement):
- target_language_column — stores language code (e.g. 'ru', 'en')
- target_source_column — stores original source text
- target_source_language_column — stores detected BCP-47 source language
- SQL generator includes these columns in INSERT when configured
- Frontend: 4 input fields under 'Target Column Mapping' section
with clear labels, placeholders, and i18n hints
Backend: model, schema, service, orchestrator
Frontend: job config page, i18n en+ru
DB: 3 new columns added to production
- Delete migrate_sqlite_to_postgres.py (one-time script, no longer needed)
- Fix @INVARIANT in mappings.py: SQLite → PostgreSQL
- Remove deprecated source_language column from TranslationJob model
+ Alembic migration 8dd0a93af539 to drop column if exists
- Remove source_language from TranslateJobResponse schema
Fixes recurring 'source_language does not exist' scheduler error
CR1: Add needs_review, language_overridden to translation_languages migration
CR2: Add per_language_metrics to translation_metric_snapshots migration
H1: BulkFindReplaceService.preview() now uses replacement_text (not find pattern)
H2: get_run_records now includes per-language data via selectinload
H3: Preview per-language approve/reject uses correct (jobId, rowId, lang) args
H4: _compute_config_hash now includes target_languages
- Auto-detection of source language per row via LLM (US6)
- Multi-target translation — one LLM call for N languages (US1-US3)
- Language-aware storage: TranslationLanguage, per-language stats
- Multilingual dictionaries with language-pair-aware filtering (US7)
- Inline correction on any run result + submit-to-dictionary (US8)
- Context-aware dictionary: auto-capture row context, usage notes,
Jaccard similarity, priority flagging in LLM prompts (US8b)
- Configurable preview sample size 1-100, cost warning at >30
- Per-language history & metrics with MetricSnapshot preservation
- 36 files, +5022/-373, all specs GRACE-Poly v2.6 compliant
Translations:
- FR-045: new-key-only execution mode — translate only rows with unseen keys
- Compare source row keys against TranslationRecord.source_data from last succeeded run
- Baseline expired (>90 days) fallback to full mode with baseline_expired event
- run_noop early return when zero new keys (skip LLM + SQL)
Scheduler reliability:
- Stale PENDING protection: runs older than 1h auto-marked FAILED, no longer block schedule
- load_schedules() reloads active translation schedules from DB on restart
- add_translation_job/remove_translation_job register/unregister with APScheduler
- execution_mode column on translation_schedules with additive DB migration
Automation view:
- GET /settings/automation/translation-schedules endpoint
- Translation schedules displayed on Automation page below validation policies
Trace propagation:
- seed_trace_id() in all background entry points:
TaskManager._run_task/_flusher_loop, SchedulerService._trigger_backup,
websocket_endpoint, IdMappingService, all standalone scripts
Migrations:
- dictionary_entries: origin_run_id, origin_row_key, origin_user_id
- translation_schedules: execution_mode
Tests:
- 3 new test modules (22 tests): core_scheduler, executor_filter, scheduler_execution+guard
- Fix pre-existing translate test isolation (conftest.py)
- Fix test_list_runs_filter_status parameter
- Fix ModuleNotFoundError: 4 lazy imports in _schedule_routes.py changed
from 3-dot to 4-dot relative imports (src.api.dependencies -> src.dependencies)
- Fix TypeError: update_schedule() param name mismatch (timezone -> timezone_str)
- Add add_translation_job/remove_translation_job to SchedulerService
to register translation schedules with APScheduler via execute_scheduled_translation
- Add GET /settings/automation/translation-schedules endpoint
returning all translation schedules with joined job names
- Update Automation settings page to display translation schedules
(cron, timezone, active badge, last run) below validation policies
Replace [Entry]/[Exit]/[Action]/[COHERENCE:OK] with [REASON]/[REFLECT]/[EXPLORE]
in belief_scope context manager, BeliefFormatter, and 34 source files.
Per molecular-cot-logging skill invariants.
- Add environment_id to TranslationJob model/schema/API + DB migration
- Pass environmentId from page to TranslationPreview and fetchPreview
- Fix _fetch_sample_rows: use result_type='samples' to include virtual cols
- Add 'kilo' and 'openrouter' to supported LLM provider types
- Make response_format conditional (skip for non-OpenAI upstream providers)
- Remove source_table field from form (redundant with datasource)
- Restore datasourceSearch display from saved job on page load
- Add request/response logging for LLM API calls
repo.is_dirty() and repo.index.diff("HEAD") internally call git diff --cached,
which fails with exit 129 (unknown option) in some environments.
Introduce _parse_status_porcelain() using git status --porcelain,
a self-contained command that avoids --cached entirely.
All 25 git route tests continue to pass.
Move dataset review clarification into the assistant workspace and
rework the review page into a chat-centric layout with execution rails.
Add session-scoped assistant actions for mappings, semantic fields,
and SQL preview generation. Introduce optimistic locking for dataset
review mutations, propagate session versions through API responses,
and mask imported filter values before assistant exposure.
Refresh tests, i18n, and spec artifacts to match the new workflow.
BREAKING CHANGE: dataset review mutation endpoints now require the
X-Session-Version header, and clarification is no longer handled
through ClarificationDialog-based flows
Add readiness hints, blocker summaries, and progress cues across
the dataset review workspace to clarify partial recovery and
launch gating states.
Highlight read-only preview snapshots, pending semantic review,
clarification queue status, and unresolved validation blockers to
guide the next recommended action.
Document known limitations in backend test doubles so permission,
task dispatch, and uncovered persistence branches are explicit.
