jobtrackingapp/docs/work-programmes/decisions.md

JobTracker programme decisions and assumptions

DEC-001 — Programme source paths

• Date: 2026-08-02
• Decision: The source files are exactly docs/todo/work.md and docs/todo/ollama.md; no fallback path search was needed.
• Reason/evidence: Both exact files exist and were read completely (922 and 825 lines respectively).
• Alternatives considered: docs/work-programmes/ content-based discovery, required only if names differed.
• Consequences: Original references in the master plan use these paths and line ranges.
• User approval required: No; factual discovery.
• Reversible: Yes, update paths if the sources move.

DEC-002 — One authoritative plan, compatibility pointer only

• Date: 2026-08-02
• Decision: docs/work-programmes/master-work-plan.md is authoritative. The older requested docs/plans/post-audit-ux-reliability-program.md will point to it instead of duplicating the checklist.
• Reason/evidence: The current request explicitly requires one authoritative record; work.md:46-59 requires the older plan path, and docs/plans/ did not exist.
• Alternatives considered: duplicate both full plans; rejected because they would drift.
• Consequences: Legacy references remain valid without a second status source.
• User approval required: No; directly reconciles both instructions.
• Reversible: Yes.

DEC-003 — Canonical origin precedes identity recovery

• Date: 2026-08-02
• Decision: SEC-001/SEC-002 precede Microsoft legacy relinking and email change/recovery.
• Reason/evidence: JT-002 security URLs currently fall back to request Host; recovery built first could send attacker-controlled links.
• Alternatives considered: follow the suggested identity-first order; rejected as unsafe.
• Consequences: SEC-001 is the first implementation package.
• User approval required: No; safer dependency ordering was requested.
• Reversible: No practical reason to reverse.

DEC-004 — Scoped production authority

• Date: 2026-08-02
• Decision: Production mutation authority applies only to the production-local-AI inventory/benchmark/install/configuration/rollout described in ollama.md, and only after its backup/rollback gates. Other production deployment still requires explicit instruction.
• Reason/evidence: work.md:15 prohibits production deploy; ollama.md:3 and the current request authorize scoped AI work.
• Alternatives considered: treat either instruction as globally overriding the other; rejected as over-broad.
• Consequences: repository work continues; PROD items remain blocked without documented access.
• User approval required: No; this is the narrow intersection of explicit instructions.
• Reversible: Yes with new authority.

DEC-005 — ADR-004 is superseded for the scoped local-first programme

• Date: 2026-08-02
• Decision: Implement one central local-first routing policy with controlled external fallback, while retaining existing models/configuration for rollback.
• Reason/evidence: ADR-004 and current architecture choose one deployment provider; ollama.md:398-451,574-608 explicitly requires ordered routing/fallback and the user asked to execute it.
• Alternatives considered: keep one provider and ignore the new programme; scatter fallback in callers; both rejected.
• Consequences: ADR-004 must later be superseded/updated in its own cohesive package. No provider change occurs before privacy/entitlement/queue controls.
• User approval required: Already supplied by the programme request.
• Reversible: Yes; old provider path/config remains for rollback.

DEC-006 — Free means no AI in the target policy

• Date: 2026-08-02
• Decision: The target external plan model is Free (no AI, core non-AI tracking) and Pro (defined AI capabilities). Preserve existing user data and internal Premium compatibility while migrating behavior.
• Reason/evidence: work.md:636-719 is explicit; current roadmap/code instead permits limited Free AI and uses Premium.
• Alternatives considered: retain existing limited Free AI; rejected because it contradicts the new programme.
• Consequences: POL-001 requires a deliberate server-side policy/compatibility package; copy changes cannot precede enforcement.
• User approval required: Already supplied.
• Reversible: Product policy is reversible later; data migration should remain additive.

DEC-007 — One durable operation foundation

• Date: 2026-08-02
• Decision: Strategy Snapshot, CV processing and other long AI work share OPS-001/AI-001; feature packages supply handlers/results/UI only.
• Reason/evidence: Both programmes specify the same state, idempotency, retry, notification and restart requirements and explicitly warn against duplicate implementations.
• Alternatives considered: separate CV and Strategy queues; rejected as duplicated infrastructure and inconsistent behavior.
• Consequences: AI-003/AI-004 depend on the common foundation.
• User approval required: No; explicitly requested.
• Reversible: The model can be extended; duplicate queues should not be introduced.

DEC-008 — New-table ownership is unresolved until provider-safe design

• Date: 2026-08-02
• Decision: Do not create operation/deletion tables until their package selects one schema owner and proves fresh/upgrade SQLite and MariaDB behavior.
• Reason/evidence: docs/infrastructure/database-ownership.md mandates reconciler DDL for current cross-provider tables, while audit JT-019 identifies dual schema ownership as a risk and the Phase 0 identity design prefers a real EF migration.
• Alternatives considered: silently follow either document; rejected because the conflict is material.
• Consequences: no migration is created in SEC-001; OPS-001 records the eventual choice and rollback evidence.
• User approval required: No now; ask only if the evidence leaves materially different safe choices.
• Reversible: Yes before schema deployment.

DEC-009 — Production access is currently blocked, not guessed

• Date: 2026-08-02
• Decision: Mark production inventory/benchmark/rollout blocked while continuing safe repository work.
• Reason/evidence: existing backup/verification docs say this environment has no route or production credentials; only CI secret names and /opt/job-tracker/app are documented.
• Alternatives considered: guess SSH host/user or scan; explicitly prohibited.
• Consequences: PROD-001/003/004 and REL-001 cannot be DONE; sanitized templates/harnesses can still be built.
• User approval required: No.
• Reversible: Immediately when documented access is provided.

DEC-010 — One canonical origin replaces provider callback overrides

• Date: 2026-08-02
• Decision: App:PublicBaseUrl now builds Gmail and Microsoft Graph callback URLs; the old provider-specific redirect variables are removed from Compose and the example environment.
• Reason/evidence: Independent callback origins contradicted SEC-001's single-origin trust boundary and allowed configuration drift. Provider registrations must use the documented paths under the canonical origin.
• Alternatives considered: Accept overrides only after equality validation; rejected because it preserves duplicate configuration with no supported separate-origin use case.
• Consequences: Operators with legacy redirect variables must register/use <APP_PUBLIC_BASE_URL>/api/gmail/oauth/callback and /api/microsoft-graph/oauth/callback. Existing variables no longer affect the application.
• User approval required: No; this implements the approved canonical-origin prerequisite.
• Reversible: Yes by restoring validated overrides, but only if a real separate-origin requirement is established.

DEC-011 — Dedicated one-hop nginx trust network

• Date: 2026-08-02
• Decision: Nginx and backend share a dedicated internal WEB_PROXY_SUBNET; forwarded headers are accepted for one hop only from that configured CIDR. Production Compose has no host-bound app or Ollama ports; development port bindings require docker-compose.dev.yml explicitly.
• Reason/evidence: Docker's dynamic/default/shared networks cannot safely identify nginx as the trusted hop, and the old auto-loaded override disabled proxy trust while exposing ports.
• Alternatives considered: clear all known proxy collections; trust the external shared network; assign a single static container IP. The first two trust too much, while a dedicated CIDR tolerates container replacement without pinning one container address.
• Consequences: production must inventory Docker networks and set a non-overlapping WEB_PROXY_SUBNET before deploy. External Traefik still requires separate operator verification.
• User approval required: No; this is the scoped security prerequisite.
• Reversible: Yes via version rollback and previous environment, but rollback must not reopen host ports unintentionally.

DEC-012 — Nginx Host is derived, not separately configured

• Date: 2026-08-02
• Decision: The frontend container derives nginx's exact server name at startup from APP_PUBLIC_BASE_URL; no second host setting exists. Unknown Hosts receive nginx 444 except /health.
• Reason/evidence: Hard-coding the current domain or introducing an independent nginx-host variable would violate the single canonical-origin contract and create drift.
• Alternatives considered: application-only Host rejection; separate APP_EXTERNAL_HOST; operator-only Traefik filtering. Each leaves one repository boundary weak or duplicates authority.
• Consequences: the frontend container fails fast on malformed/unsupported origins; IPv6 host literals are currently outside the deploy preflight contract.
• User approval required: No.
• Reversible: Yes; broaden only with matching parser, nginx and deployment tests.

DEC-013 — Microsoft sign-in has one tenant-qualified trust path

• Date: 2026-08-02
• Decision: Microsoft ID tokens are accepted only by the exchange/link validator, which requires the configured account mode plus exact GUID tid/oid and issuer agreement. The raw Microsoft bearer scheme is removed. Email-like claims are metadata, not verified ownership.
• Reason/evidence: The UI already exchanges Microsoft tokens for local sessions; maintaining a second issuer-disabled API bearer path duplicated and weakened the trust decision. oid alone is tenant-scoped.
• Alternatives considered: harden both bearer and exchange paths; retain common implicitly in Production; use sub or email as identity. Each adds duplicated policy or preserves the audited ambiguity.
• Consequences: undocumented raw-token API clients stop working; Production must choose a tenant mode. Legacy identity ownership remains unresolved until SEC-004 and Microsoft must not be enabled there first.
• User approval required: No; this is the validated P0-1A contract.
• Reversible: The account mode is configurable; raw bearer support should return only with a documented requirement and the same policy/tests.

DEC-014 — Split email ownership from session revocation

• Date: 2026-08-02
• Decision: Split the original SEC-005 into SEC-005A (session/recovery revocation) and SEC-005B (registration/pending email plus migration/UI).
• Reason/evidence: The revocation work uses the existing schema and is independently testable/rollbackable; pending email requires a coordinated database and frontend contract. Keeping both under one active item violated the requested small-package cycle.
• Alternatives considered: keep one broad item; rejected because its status could not accurately distinguish verified security behavior from an unstarted migration/UI flow.
• Consequences: SEC-004 depends on both children. Original source/audit references remain on each, so no requirement was lost.
• User approval required: No; this is tracking granularity within approved scope.
• Reversible: Yes by presenting them as one release, but their verification remains separate.

DEC-015 — Pending email uses Identity tokens and one provider-aware EF migration

• Date: 2026-08-02
• Decision: Store only the proposed address and request time, rotate the Identity security stamp for each replacement request, and use the built-in change-email token. Add both fields through one EF migration whose SQLite and MariaDB column types are explicit; do not duplicate these Identity columns in the startup reconciler.
• Reason/evidence: The active email must remain authoritative until proof. Identity already binds tokens to user, new email, purpose and security stamp; stamp rotation makes the latest request win without another token table. Dry-run SQL showed SQLite-scaffolded types were unsafe for MariaDB until the migration branched by provider. Audit/schema decision P0-4B assigns these Identity fields to EF.
• Alternatives considered: immediately replace Email; store plaintext confirmation tokens; add a request table/nonce abstraction; add the same columns to startup reconciliation. The first two are unsafe, and the latter two add duplicate state/ownership without a demonstrated need.
• Consequences: deployment must run 20260802205800_AddPendingEmailChange before the new API version. SQLite uses TEXT; MariaDB uses varchar(320) and datetime(6). Rolling-version and MariaDB execution still require verification.
• User approval required: No; this is the smallest implementation of the approved ownership contract.
• Reversible: Yes via the migration Down before data relies on pending requests; active email data is unchanged.

DEC-016 — SEC-005B remains short of local browser verification

• Date: 2026-08-02
• Decision: Mark SEC-005B IMPLEMENTED — NOT VERIFIED, not VERIFIED LOCALLY or DONE.
• Reason/evidence: all automated suites and an isolated API runtime pass, but the in-app browser denied localhost because its administrator policy check could not be verified. SMTP and MariaDB execution are also unavailable.
• Alternatives considered: infer browser behavior from component tests or claim the existing Docker UI; rejected because the running containers are old images and the programme forbids inferred test claims.
• Consequences: SEC-004 repository work can proceed, but SEC-005B retains explicit browser/provider/production acceptance checks.
• User approval required: No; truthful status accounting is required.
• Reversible: Yes immediately after the blocked checks pass.

DEC-017 — Canonical Microsoft ownership never backfills legacy evidence

• Date: 2026-08-02
• Decision: Add nullable bounded MicrosoftTenantId/MicrosoftObjectId with one unique composite index, use only that pair for ownership, and leave every legacy subject/email row null-canonical until explicit dual-proof relinking.
• Reason/evidence: oid is tenant-scoped, legacy subjects may be oid or sub, and provider email is mutable metadata. A disposable migration rehearsal preserved duplicate legacy values and enforced unique proven pairs. The recovery ceremony requires both a purpose-bound token delivered to the confirmed app email and a fresh Microsoft token for the same pair.
• Alternatives considered: backfill from common, legacy subject, email or next-seen token; retain email fallback; add a general identity-provider framework. Each would merge unproven identities or add unrelated abstraction.
• Consequences: ambiguous/unconfirmed legacy users require operator assistance; production needs a counts-only inventory and relink window. Link/unlink revoke sessions, and passwordless unlink is refused until a safe provider reauthentication path exists.
• User approval required: No; this implements the approved JT-001 safety contract.
• Reversible: Additive schema is reversible before canonical data is relied upon. Application rollback must never restore email auto-linking.

DEC-018 — Keep MariaDB date work in SQL and bound SQLite fallback by owner/job

• Date: 2026-08-02
• Decision: Branch only at the affected DateTimeOffset query roots: SQLite materializes owner/job-scoped rows before ordering or range comparison; MariaDB keeps server-side ordering, filtering, aggregation and pagination.
• Reason/evidence: the real SQLite provider throws before execution, while Pomelo generates the required SQL. Changing all timestamp storage would require a risky migration and editing response endpoints individually would leave CV cleanup/reprocess siblings broken. Real-provider tests and fresh HTTP rehearsal pass.
• Alternatives considered: global timestamp conversion/schema rewrite; always materialize on every provider; modify historical migrations; catch-and-retry translation exceptions. These add migration risk, production performance cost or hide separate JT-019 ownership drift.
• Consequences: SQLite work is bounded by tenant/job and current entitlement limits but still happens in memory; production MariaDB behavior is unchanged. If per-user AI history grows materially, a later UTC scalar column/index migration may be measured and designed.
• User approval required: No; this is the smallest root-cause correction within CORE-001.
• Reversible: Yes; revert the provider branches and relational test, with no data rollback.

DEC-019 — Keep the editable interview board canonical and move the generated brief

• Date: 2026-08-02
• Decision: Keep GET /interview-prep for the durable editable InterviewPrepItem board, move the distinct cached/generated InterviewPrepNote response to GET /interview-prep/brief, and delete the unused flat timeline action so the grouped/filterable timeline remains canonical.
• Reason/evidence: both interview representations have active repository UI callers and incompatible DTOs, while only the newer timeline has a caller. Overloading by query parameter or deleting one live feature would preserve ambiguity or break behavior. Reflection and isolated HTTP tests now show one action per method/path with owner 200, other user 404 and anonymous 401.
• Alternatives considered: delete either interview feature; retain the old path with a discriminator; rename the editable board; merge DTOs. Each creates unnecessary compatibility state or breaks the current workspace contract.
• Consequences: undocumented direct consumers of the generated brief must adopt /brief; repository clients/docs are updated. No data or schema changes.
• User approval required: No; this is the smallest cohesive repair of confirmed JT-004.
• Reversible: Yes by restoring the route/client and flat action, but rollback also restores the exploitable availability defect and is not recommended.

DEC-020 — Attachment final paths are the durable operation identity

• Date: 2026-08-02
• Decision: Represent recoverable attachment mutations as <final>.uploading and <final>.deleting; do not add a separate JSON journal, schema or periodic worker. Reconcile once after database initialization and preserve unknown plain files.
• Reason/evidence: every final path is already generated, unique and root-bounded. The suffix plus row existence encodes every required recovery decision, and real-SQLite failure injection proves upload promotion, delete restore/purge and idempotent restart behavior. A second journal would introduce dual durable state and another crash-ordering problem.
• Alternatives considered: JSON operation journal; database operation table; object-store abstraction; periodic/multi-replica reconciler. None is required for the current single-filesystem deployment and each adds coordination or migration cost.
• Consequences: recovery retries on safe restart rather than a timer; persistent failures remain observable until restart/operator action. Multi-replica or high-volume deployments must first add a lease and measured periodic reconciliation. SEC-009 must reuse these root/quarantine rules.
• User approval required: No; this is the smallest implementation of the approved JT-010 recovery contract.
• Reversible: Yes after draining/reviewing all suffix markers. No schema rollback exists; unknown plain orphans must never be guessed away.

DEC-021 — Worker enumeration bypasses filters once, then re-enters owner scope

• Date: 2026-08-02
• Decision: BackgroundTenantRunner may ignore tenant filters only to enumerate non-empty job owners. Each owner is processed sequentially in a new scope whose live CurrentUserService restores all normal query filters. Rules, reminders, daily export and enrichment receive separate default-false switches.
• Reason/evidence: hosted scopes have no HTTP identity and therefore returned zero rows. Leaving services on while fixing that root cause would unexpectedly start status mutations, file exports, email and AI calls. Real-SQLite tests prove owner filtering, per-owner settings, failure isolation and fake-only side effects.
• Alternatives considered: unfiltered queries throughout each worker; a privileged DbContext; automatic activation under legacy settings; a distributed scheduler. These enlarge the trust boundary, create rollout risk or solve unmeasured scale.
• Consequences: workers remain inert until explicitly enabled after OPS/POL prerequisites. Execution is sequential/single-instance; leasing and durable operations are the next package. Existing backup, probe and CV-run handling remain unchanged.
• User approval required: No; this is the requested safe foundation and does not activate production work.
• Reversible: Yes by keeping switches false and reverting the runner/service changes. Already-generated exports or user-visible mutations require separate reviewed rollback.

DEC-022 — Durable operations are EF-owned and store references, not payloads

• Date: 2026-08-02
• Decision: Split OPS-001 into independently reviewable OPS-001A state/schema, OPS-001B notifications and OPS-001C APIs/UI. UserOperations is owned only by a provider-conditional EF migration; the startup reconciler does not create or alter it. The row stores bounded policy/subject references and no generic raw payload.
• Reason/evidence: a manually branched migration produces correct SQLite and bounded MariaDB DDL, avoiding the repository's old SQLite-type failure while reducing JT-019 dual ownership. CV/Strategy inputs already have durable domain IDs; copying private text into a queue row is unnecessary. Concurrent SQLite tests prove one idempotent row and one lease winner.
• Alternatives considered: extend CvExtractionRun; reconciler plus no-op migration; generic JSON payload; external queue/Redis; one large schema/UI package. These duplicate feature state, preserve dual ownership, increase private-data copies/infrastructure, or prevent small rollbackable review.
• Consequences: feature producers reference domain rows and must re-check entitlement/privacy before work. Notifications/API/UI follow without changing the operation identity. MariaDB execution remains a deployment gate.
• User approval required: No; this resolves the recorded DEC-008 conflict using repository evidence and the requested smallest reliable design.
• Reversible: Yes before consumers rely on rows; stop producers/drain rows before Down. Additive table may remain during application rollback.

DEC-023 — One current generic notification is atomic with terminal operation state

• Date: 2026-08-02
• Decision: Store one owner-scoped UserNotification per current terminal operation outcome, committed in the same relational transaction. Use generic bounded text, no email delivery and no private operation/failure content. A manual retry removes the prior notification so the next terminal outcome can replace it.
• Reason/evidence: a unique nullable operation foreign key supplies database idempotency; a forced notification-write failure proves the terminal update rolls back. This reuses OPS-001A rather than introducing an outbox framework or a second queue.
• Alternatives considered: transient frontend notifications; email outbox; multiple immutable notifications per retry attempt; generic event bus. Transient state fails restart recovery, email is not authorized, and the latter two add delivery/history machinery not required by either programme.
• Consequences: OPS-001C can expose stable read/dismiss state without creating another notification model. Historical retry-attempt notifications are not retained; operation attempt/failure fields remain the technical state. MariaDB execution remains a deployment gate.
• User approval required: No; this implements the approved persistent-notification prerequisite without external side effects.
• Reversible: Yes before API/UI consumers rely on it. Older application versions tolerate the additive table; schema Down deletes notification state.

DEC-024 — Operation APIs expose state, not worker internals

• Date: 2026-08-02
• Decision: Expose authenticated owner list/detail/cancel/retry and notification list/count/read/dismiss APIs, but no generic operation-create endpoint. Return only bounded user-facing state; omit idempotency keys, lease tokens, provider/model fields, raw failure text and result references. Use bounded polling rather than realtime infrastructure.
• Reason/evidence: feature producers must enforce entitlement/privacy and durable subject references before admission, so a generic create endpoint would bypass later policy. Existing Axios, MUI and browser events cover the UI without a dependency. Two-user HTTP checks prove copied identifiers return 404.
• Alternatives considered: WebSockets/SignalR; a generic JSON task API; exposing the full entity; merging reminders and operation notifications. These add infrastructure, unsafe authority or misleading counts without a current need.
• Consequences: AI-003/004 own feature admission and result navigation. The operation page polls every 15 seconds while mounted; the shell polls unread count every 60 seconds. Realtime delivery can be reconsidered only if measured UX/load requires it.
• User approval required: No; this is the smallest implementation of both programmes' stable status and notification contract.
• Reversible: Yes. UI/API removal leaves durable operation/notification data intact; application rollback can retain additive tables.

DEC-025 — Public Pro policy uses live roles while retaining legacy billing identifiers

• Date: 2026-08-02
• Decision: Expose only free and pro; make Free AI entitlement and limits zero; authorize explicit AI actions with one live database-role policy; recheck queued/background work at execution; preserve the internal Premium Identity role and Stripe:PricePremium configuration key for billing/data compatibility.
• Reason/evidence: the new programme supersedes the old limited-Free-AI model. Claim-only authorization would let an already-issued session retain AI after downgrade, while renaming the persisted role/config now adds migration and rollback risk without changing user-visible behavior. Core job create/detail and deterministic enrichment were traced separately because they must still work when their optional model call is skipped.
• Alternatives considered: static RequireRole; guards copied into every controller; renaming the Identity role/config; wrapping every summarizer call in a new provider abstraction; blocking whole controllers. These leave stale-claim/worker bypasses, add scattered checks, create needless migration risk, pre-empt POL-002/AI-002, or hide existing non-AI data.
• Consequences: explicit locked APIs return stable pro_required; Admin maps to Pro; UI receives ai/proThemes; existing AI history and non-AI editing remain accessible. Usage accounting outside AI Workspace remains incomplete and blocks provider rollout/full POL-001 verification. PRODUCT-001 still owns removal of invented landing-page price/tier claims.
• User approval required: No; the programme explicitly requires Free=no-AI and centralized enforcement.
• Reversible: Yes as one repository-only policy/UI change with no schema update. Workers must remain off while rolling back to avoid restoring an entitlement bypass.

DEC-026 — Classify synthetic AI workloads before finalizing privacy routing

• Date: 2026-08-02
• Decision: Advance PROD-002 immediately after POL-001 and before POL-002, even though the recommended list placed the general privacy policy first.
• Reason/evidence: POL-002 explicitly depends on task type, privacy class, payload shape, latency and fallback suitability. POL-001 produced the reachable-call inventory, and PROD-002 can safely classify it and create synthetic fixtures without production access or provider calls. Writing policy first would either duplicate this inventory or invent categories without fixtures.
• Alternatives considered: keep POL-002 next and revise it later; perform PROD-001 hardware inventory first; start queue implementation. The first creates churn, while the latter two are blocked by production access or need the privacy contract.
• Consequences: PROD-002 is the sole in-progress item; POL-002 follows with evidence-backed classes. No production/provider action is introduced.
• User approval required: No; the user directed dependency-aware reordering and conflict recording.
• Reversible: Yes; documentation/fixtures can be revised before routing code depends on them.

DEC-027 — External AI needs two administrator/user gates and defaults to local

• Date: 2026-08-03
• Decision: Persist AiEnabled and ExternalAiProcessingAllowed per user; preserve AI-enabled behaviour for existing accounts, default external consent to false, and permit an external /cv/* request only when backend and sidecar administrator gates, a supported configured provider, live Pro entitlement, AI-enabled preference and explicit user consent all agree. Background calls without an authenticated request fail safe to local.
• Reason/evidence: the existing global sidecar AI_PROVIDER could route full CV data externally without a user decision. One backend policy plus a sidecar permission header closes that execution path without deleting rollback provider configuration or inventing the final AI-002 router.
• Alternatives considered: remove Gemini/Groq; rely on UI/local storage consent; trust one environment flag; refactor every AI interface now; silently use external when configured. These either break rollback compatibility, are bypassable, or prematurely duplicate AI-001/002.
• Consequences: disabling AI is enforced from live database state; external processing is off by default and needs deliberate two-sided configuration. True local-first fallback triggers, operation policy snapshots, provider/reason persistence, payload minimization and cost controls remain explicit AI-001/002 gates, so POL-002 is not overstated as fully verified.
• User approval required: No; this is the smallest safe implementation of the programme's explicit privacy controls and preserves provider configurations.
• Reversible: Yes. Application rollback should leave the additive preference columns in place; turning both administrator gates off immediately restores local-only processing without data loss.

DEC-028 — Durable AI work extends UserOperations with a default-off typed worker

• Date: 2026-08-03
• Decision: Reuse OPS-001A/B/C for every long AI task. Add one task-handler worker and one admission service; expose no generic create endpoint. Admission stores only a domain subject reference and policy snapshot, enforces Pro/privacy/capacity/idempotency/deadline, and returns the existing stable status URL. Start at one worker and keep it disabled until real handlers and rollout checks pass.
• Reason/evidence: the existing operation store already supplies persistent states, atomic claims, leases, retries, cancellation, restart recovery, notifications and owner APIs/UI. A second CV/Strategy queue or Redis would duplicate proven state. A generic create API would let callers bypass task-specific ownership and payload validation.
• Alternatives considered: separate in-memory channel; Redis/Hangfire; one worker per feature; synchronous provider calls; generic public task creation; unbounded hosted-service parallelism. These lose restart state, duplicate infrastructure, expand authority or fail the congestion requirement.
• Consequences: AI-003/004 only add typed handlers/producers. Current capacity serialization is process-local for the documented single-backend deployment; database reservation is required before multiple backend replicas. Provider/model semaphores and circuit/provenance remain AI-002 responsibilities.
• User approval required: No; this follows both programmes' explicit instruction to reuse the smallest reliable existing infrastructure.
• Reversible: Yes. Keep the worker switch false, remove admission/worker registrations, and retain operation rows/API history. No new schema was added in this slice.

DEC-029 — One sequential sidecar router owns local-first fallback

• Date: 2026-08-09
• Decision: Keep provider execution behind the existing sidecar boundary, make Ollama the default primary, permit at most one sequential external fallback, and carry the backend's rechecked privacy/task decision through the AI-001 execution scope. Reuse existing operation provider/model/progress fields for provenance; add no queue/provider schema or dependency.
• Reason/evidence: every generative /cv/* path already converges on one _provider_generate family, while deterministic tasks and /summarize must remain local. Sidecar fake-transport tests prove local success, consent/config/task/cost denials, schema fallback, circuit behavior, external failure and no parallel duplicate call. Backend tests prove policy propagation and success/failure provenance.
• Alternatives considered: provider selection in each controller; browser-selected providers; a second provider abstraction in .NET; simultaneous local/cloud racing; a new circuit/attempt table; increasing synchronous timeouts. These scatter policy, expose authority, duplicate the established boundary, risk double charge/output, add unneeded schema, or mask the queued-operation root cause.
• Consequences: AI_ROUTING_MODE supports local_only, local_first and explicitly gated external_only; invalid values fail closed. New durable task IDs stay local until allowlisted. The current circuit is process-local and one AI worker is the effective single-model concurrency limit. Per-request prompt size limits external spend/exposure, but complete monthly cross-feature accounting and model selection remain rollout gates.
• User approval required: No; this directly implements the approved local-first programme without invoking a provider or production service.
• Reversible: Yes. Set EXTERNAL_AI_ENABLED=false or AI_ROUTING_MODE=local_only; the older AI_PROVIDER/model configuration is retained. Existing nullable operation fields and AI history remain readable.

DEC-030 — Strategy generation is one typed operation and one structured inference

• Date: 2026-08-09
• Decision: Make Focus Plan GET cache-only and move generation to typed strategy.snapshot work on the shared queue. Encode only job and at most four attachment IDs, reuse active operations, and replace four sequential model calls with one bounded JSON response whose complete shape is validated before the unique cache row is updated. Keep Strategy absent from the external fallback allowlist.
• Reason/evidence: the traced button/GET path owned four serial model calls and had no restart/cancel/retry identity. AI-001/002 already provide every needed state/policy/routing primitive. One structured inference reduces timeout exposure and makes publication atomic without a second queue or schema.
• Alternatives considered: increase HTTP/proxy timeouts; keep GET as a command; add a Strategy queue/table; store prompts/private text in operation payload; four model calls inside the worker; race external/local providers. These retain the root failure, duplicate infrastructure/private data/output, or violate established routing safety.
• Consequences: existing cached FocusPlanDto remains readable, while generation now returns 202 and the UI resumes by operation ID/context. Retry can overwrite only the same unique result row. Worker activation and real-model tuning remain rollout gates; operation records provide deduplication/provenance, not full monthly billing accounting.
• User approval required: No; both programmes explicitly require durable Strategy work and consolidation with AI-001/002.
• Reversible: Yes. Keep the worker off and revert a621226; no schema/dependency changed. Cancel or drain queued strategy.snapshot rows before removing the handler.

DEC-031 — CV extraction runs own review data; UserOperations own execution

• Date: 2026-08-09
• Decision: Keep CvExtractionRun as the artifact/result/review record and make one typed cv.process operation reference its numeric ID. Replace the unbounded channel and separate hosted service with AI-001 admission/leases/retries/cancellation/notifications. Upload returns 202 after persistence; no raw CV payload is duplicated into operation state.
• Reason/evidence: the complete trace showed useful persistent review state but two competing execution mechanisms: synchronous upload and a process-local channel. Reusing both existing models gives restart-safe orchestration without a new queue/schema and preserves the mandatory accept/discard gate.
• Alternatives considered: raise proxy timeouts; keep synchronous upload; add a second CV queue/table; store raw CVs/prompts in operation payloads; replace extraction runs with generic operations. These retain the 504/lost-wakeup path, duplicate infrastructure/private data, or discard domain review/version history.
• Consequences: all four long CV actions share one default-off worker and operation UI. Parser-version/process isolation remains SEC-006/007; browser/model/MariaDB/production gates remain before rollout. Existing clients must accept the upload endpoint's 202 operation response.
• User approval required: No; both programmes explicitly require one durable operation foundation and preservation of human review.
• Reversible: Yes. Keep the worker off, revert c3c5af8, and retain operation/extraction rows. Cancel or drain cv.process rows before removing the handler.

DEC-032 — Provider account management and signed-out authentication share logic, not presentation

• Date: 2026-08-09
• Decision: Add an explicit sign-in presentation to the existing Google and Microsoft components. It reuses provider token exchange and two-factor handling but skips signed-in account discovery, link/unlink controls and linking copy. Keep the full account presentation unchanged for authenticated profile/settings surfaces.
• Reason/evidence: the programme requires a conventional single sign-in card without implying account linking. Duplicating provider callback code would risk divergence from the hardened tenant/linking path, while rendering account-state panels on login creates the prohibited clutter and misleading relationship.
• Alternatives considered: retain tabs; create duplicate login-only provider clients; hide copy with CSS; combine provider exchange and account linking. These preserve the UX defect, duplicate sensitive logic, hide rather than remove inaccessible state, or weaken the identity boundary.
• Consequences: login presentation becomes simpler without changing backend identity ownership. Provider account management remains available only in its existing authenticated surfaces. Real-provider and production verification are still required.
• User approval required: No; this is the smallest implementation of the explicit UX-001 requirement and preserves the prior security contracts.
• Reversible: Yes. Reverting UX-001 restores the tabbed presentation; no provider link, session, schema or configuration data changes.

DEC-033 — Theme preference is application-owned; MUI only renders resolved mode

• Date: 2026-08-09
• Decision: Keep the existing themeMode:<user> storage keys as the only preference store, resolve user → anonymous → System explicitly, and feed the resulting Light/Dark mode into MUI without MUI storage or provider/router remounts. Use a dedicated auth-user event and read-only cross-tab storage subscription; apply the same resolution in a pre-paint Next script.
• Reason/evidence: the trace proved the general auth event fired before the new user key was stored and no later theme event occurred. Refresh therefore changed namespace and appeared random. Provider keys/router dependencies also discarded state, while MUI's default storage could become a second source of truth.
• Alternatives considered: keep provider keys; emit another general auth event; store one global preference; let MUI own mui-mode; add a server profile migration. These retain remounts/request loops, lose user isolation, create competing precedence or add unnecessary backend scope.
• Consequences: users without a saved scoped choice inherit the explicit anonymous choice, otherwise the documented default is System. Explicit Light/Dark ignore OS changes. The pre-paint script must remain behaviorally aligned with themePrefs; tests cover both.
• User approval required: No; this directly implements the approved UX-002 contract without schema, dependency or production changes.
• Reversible: Yes. Revert 11734ee; existing preference values remain unchanged and readable.

DEC-034 — Important-term quality stays deterministic and on demand

• Date: 2026-08-09
• Decision: Harden the existing shared JobCvMatchService rather than add a model or stored analysis version. Clean HTML/chrome at the matcher boundary, use bilingual/general filler categories, rank bounded phrase runs before singleton terms, and extend the canonical skill vocabulary for punctuation-sensitive technologies. Rename UI output to “important terms.”
• Reason/evidence: the complete path proved the reported Norwegian words came from English-only deterministic token ranking. Imported descriptions may be clean, but manual descriptions/notes reach the same matcher. Results are recomputed and the generated learning sync already preserves user decisions, so a cache/schema migration would solve a nonexistent storage problem.
• Alternatives considered: hardcode the five examples; call Ollama for keywords; add a result table/version column; filter only in the UI; silently delete learning items. These are incomplete, less reliable, duplicate state, leave API consumers dirty or discard user history.
• Consequences: term changes appear on the next request; obsolete pending generated learning items auto-complete under existing behavior while done/dismissed decisions remain. The curated vocabulary remains intentionally bounded and test-driven.
• User approval required: No; this is the deterministic-first implementation explicitly required by both programmes.
• Reversible: Yes. Revert da1aa8b; no schema, cache or provider state changes.

DEC-035 — Career Workspace derives actions from existing domain state

• Date: 2026-08-09
• Decision: Render one action-oriented overview from the completeness and durable import state already owned by CareerProfilePage; load only the separate recent-CV list. Link actions to the existing profile/import anchors, builder and saved-job workflow, and retain extraction Apply/Discard as the sole merge gate.
• Reason/evidence: duplicating /career/profile and /profile-cv/runs requests in the route wrapper would create competing loading/polling state. The existing page already has correct durable run, review and version data, while CV variants are a separate bounded list. The old wrapper duplicated headings and explanation without exposing next actions.
• Alternatives considered: a second workspace data loader; moving all profile state into a new global store; direct one-click job-CV creation without a job choice; rewriting the profile editor. These add inconsistent state, premature architecture or bypass the established job/application boundary.
• Consequences: the page presents concise state-aware navigation without changing API, profile persistence or imported-data approval semantics. Browser and production verification remain before DONE; deeper builder/job interaction remains in CAREER-002/JOBS-001.
• User approval required: No; this is the smallest cohesive implementation of the approved programme requirement.
• Reversible: Yes. Revert the CAREER-001 implementation commit; no data/configuration migration exists.

DEC-036 — CV autosave uses revision ordering and explicit discard warnings

• Date: 2026-08-09
• Decision: Keep the existing debounced variant autosave API, but track the latest name/settings and a monotonic client revision. Only the newest request may update the visible save state. Offer an explicit retry and block route/unload navigation until the user chooses to discard pending or failed edits.
• Reason/evidence: the editor already has versioned server saves, but a route change during the debounce and overlapping responses could lose work or falsely display Saved. Blank names were ignored by the backend while remaining blank in the UI. The existing API is sufficient; the defect is client coordination and feedback.
• Alternatives considered: replace autosave with a manual-only form; add a queue/schema; extend timeouts; silently flush during unload; refactor variant versioning. These regress the working interaction, add unrelated infrastructure or cannot reliably complete during page teardown.
• Consequences: users receive a native discard decision for in-app navigation and browser-standard warning on unload. A blank name stays unsaved until corrected. The deeper builder redesign and browser checks remain CAREER-002 work.
• User approval required: No; this is a reversible data-loss safeguard within the approved programme.
• Reversible: Yes. Revert b58cc19; server variants and version history require no migration.

DEC-037 — Variant entries override profile facts; they do not delete them

• Date: 2026-08-09
• Decision: Keep profile-backed entries read-only at the ownership boundary: a CV variant may reorder, hide or override them, but cannot delete the Career Profile item. Provide full add/edit/reorder/confirmed-delete behavior only for variant-owned custom entries using the existing string[] model.
• Reason/evidence: the architecture and resolver define variants as lenses over stable Career Profile item keys. Treating Delete in the Builder as profile deletion would violate separation and risk changing every CV. The custom-section model already owns document-specific content and needs no schema change.
• Alternatives considered: delete profile entries from the Builder; clone profile entries into variants; add a second custom-entry table; keep the opaque one-item-per-line textarea. These violate ownership, duplicate data/infrastructure or fail the approved interaction requirement.
• Consequences: original entries use Hide rather than Delete; custom entries receive explicit destructive confirmation. Existing render/public/PDF/version data remains compatible.
• User approval required: No; this preserves the programme's explicit Career Profile separation and existing data.
• Reversible: Yes. Revert a5b74e0 and 2043349; settings JSON remains backward-compatible.

DEC-038 — Job email has one canonical route with filtered compatibility redirects

• Date: 2026-08-09
• Decision: Make /correspondence the canonical Job email hub and represent recruitment review as ?view=review. Redirect the legacy /correspondence/review route into that filtered view while reusing the existing review component and APIs.
• Reason/evidence: the two pages were separate navigation surfaces over related linked/review workflows, while the job workspace already reuses the same correspondence domain. Route composition removes duplicate information architecture without prematurely rewriting mature Gmail decision logic.
• Alternatives considered: delete the old route immediately; copy review cards into the inbox; merge backend endpoints before a provider capability trace; keep both pages indefinitely. These break bookmarks, duplicate behavior, expand risk or preserve the approved UX defect.
• Consequences: existing links remain compatible and there is one user-facing hub. Provider-neutral thread state and explicit draft/send remain separate MAIL-001 increments and are not implied by this routing change.
• User approval required: No; the product decision to consolidate and preserve route compatibility is explicit.
• Reversible: Yes. Revert 6008b4a; no persisted correspondence/provider data changes.

DEC-039 — Connected email capabilities are explicit and provider-neutral

• Date: 2026-08-09
• Decision: Consume the existing IEmailProviderRegistry through one owner-scoped read controller and expose provider status/search/thread/plain-text detail without changing scopes or adding a second provider abstraction. Report send as unavailable until a provider has an implemented, tested send contract.
• Reason/evidence: Gmail, Microsoft Graph and IMAP already implement the same read seam, but only Gmail consumed it and the hub could not state actual capabilities. Their current OAuth/service contracts are read-only. The separate follow-up action uses application SMTP and cannot honestly represent connected-provider send.
• Alternatives considered: copy Gmail controller behavior for each provider; claim SMTP as provider send; add speculative provider-action interfaces; broaden OAuth scopes before designing re-consent/audit/uncertain delivery; return untrusted HTML through the shared detail API. These duplicate logic, mislead users, widen privileges prematurely or create an unsafe rendering path.
• Consequences: the hub can truthfully identify connected Gmail/Outlook/IMAP accounts and a shared API exists for later UI composition. Provider-native state changes and send remain unavailable and must be introduced with capability flags, re-consent and delivery-state tests. Message detail returns plain text only.
• User approval required: No; this is a local, fake-tested repository increment within MAIL-001 and invokes no provider or email service.
• Reversible: Yes. Revert 536d403; no schema, dependency, OAuth scope, provider token or persisted message changes.

DEC-040 — Message detail is plain text with an explicit saved-copy fallback

• Date: 2026-08-09
• Decision: Fetch live detail through the neutral provider endpoint when a row has a provider/message ID, but render only plain text. If provider access fails, warn and fetch the tenant-scoped persisted correspondence copy. Manual entries use the persisted path directly.
• Reason/evidence: imported correspondence remains useful after token expiry or provider outage, while returning/rendering provider HTML would widen the untrusted-content surface. The existing correspondence row is already the local job record and needs no duplicate store.
• Alternatives considered: render provider HTML; fail the whole view when reauthorization is required; silently fall back; include full message bodies in every inbox list response; add a message-cache table. These increase XSS/payload/state risk, hide stale provenance or duplicate existing data.
• Consequences: users can distinguish live and saved content availability, malformed legacy metadata cannot break detail, and late responses cannot populate another selected row. The saved copy may be stale and is labeled as such when live access fails.
• User approval required: No; this is local read-only behavior with synthetic/mocked tests and no provider invocation.
• Reversible: Yes. Revert a20775c; no schema, dependency, provider scope or persisted data changes.

DEC-041 — Delivery idempotency is a content-free tenant ledger

• Date: 2026-08-09
• Decision: Persist one EmailSendAttempt per owner/client UUID with a SHA-256 payload hash and strict pending → sending → sent/failed/uncertain transitions. Terminal and uncertain attempts cannot restart; a new explicit review must use a new request UUID. Store no recipient, subject or body.
• Reason/evidence: SMTP/provider calls cannot be made atomic with a database commit. A crash or transport interruption after acceptance is inherently uncertain, so retrying the same attempt can duplicate email. The ledger must exist before any provider scope/button is enabled and must preserve only the metadata needed for deduplication/audit.
• Alternatives considered: rely on disabled buttons; store request IDs on correspondence; reuse AI operations; retry on every timeout; save full draft content in the audit row; add a message outbox that assumes provider idempotency. These do not prevent concurrent/direct-API duplicates, conflate domains, risk duplicate sends/private-data retention or promise atomicity the providers do not offer.
• Consequences: API integration can reserve a unique attempt before external I/O and fail closed on pending/uncertain records. A provider success followed by database failure remains reconcilable rather than blindly retried. User export must include the non-sensitive metadata; job/account deletion cascades it. The migration is additive and EF-owned with explicit SQLite/MariaDB types.
• User approval required: No; the approved MAIL-001 programme requires idempotent/uncertain send safety. This increment is inert and used no provider.
• Reversible: Yes. Before reverting 653f011, stop send admission, reconcile/drain attempts and downgrade 20260809195014_AddEmailSendAttempts. No production migration has been applied.

DEC-042 — Send permission is explicit; transport uncertainty fails closed

• Date: 2026-08-09
• Decision: Request Gmail send and Graph Mail.Send scopes on new/reconnected accounts, derive CanSend from the stored granted scope, keep IMAP read-only, and expose one neutral delivery contract. Treat an HTTP rejection as known failed-before-delivery, but any network interruption/cancellation as uncertain.
• Reason/evidence: read access cannot authorize send, and existing tokens must not be assumed upgraded. Provider HTTP acceptance is outside the database transaction; after a broken transport the application cannot safely prove that no message was accepted. Mocked HTTP tests prove scope, payload and classification without contacting providers.
• Alternatives considered: reuse application SMTP; silently expand existing token authority; mark all errors failed/retryable; expose provider error bodies; add SMTP credentials to the IMAP connection; enable a route before adapter tests. These misrepresent identity, risk duplicate delivery/data exposure, widen secret storage or invert the required dependency order.
• Consequences: existing connections show read-only until explicit reconnect consent. Gmail supports its provider thread ID; Graph currently sends a new message and does not claim reply-thread semantics. The later API must reserve the ledger before calling either adapter and surface uncertain state for manual reconciliation.
• User approval required: No; this repository-side programme requirement used fake transports only. Real account consent/send still requires an explicitly authorized synthetic provider account.
• Reversible: Yes. Revert e9937ac to stop requesting/using send permission. Already granted provider permission is managed by the provider/user connection and is not automatically revoked by a code rollback.

DEC-043 — Send admission reserves before delivery and never retries ambiguity

• Date: 2026-08-10
• Decision: Admit provider delivery only through an authenticated, user-rate-limited route that requires an owned job, explicit confirmation and a canonical client UUID. Reserve and begin the content-free ledger before provider I/O; persist sent correspondence, a content-free job event and the sent state in one local transaction. Return existing sent results but reject every other duplicate, especially uncertain attempts.
• Reason/evidence: provider acceptance cannot share the database transaction. Owner-scoped SQLite tests prove malformed, unconfirmed and cross-tenant requests do not reserve or send; duplicate, rejected and interrupted attempts call the fake provider at most once. Canonical UUID formatting closes a simple deduplication bypass.
• Alternatives considered: call the provider before reserving; retry timeouts; rely on a disabled button; store message content in the ledger/event; use legacy application SMTP; mark local persistence failure as sent. These can duplicate delivery, lose audit state, expose content or bypass connected-provider consent.
• Consequences: a successful provider call with failed local persistence is intentionally uncertain and requires manual reconciliation. A process stop after admission can leave a sending row; a later repository increment must age it into an explicit uncertain/manual-review state without redelivery. Existing read-only connections cannot send until re-consented.
• User approval required: No; MAIL-001 explicitly authorizes local implementation and fake verification. Real provider consent/send remains gated.
• Reversible: Yes. Disable admission/UI, then revert 123fc55. No schema rollback is needed for this route-only increment.

DEC-044 — Replies remain bound to their provider thread

• Date: 2026-08-10
• Decision: Offer an editable reply only for a message whose exact provider connection currently has send consent. Keep provider/from identity and thread read-only while allowing recipient, subject and body edits; require the shared app-owned confirmation before POST. Treat client/network ambiguity as uncertain and expose no retry action.
• Reason/evidence: a Gmail thread ID has no valid Graph meaning, and silently switching providers would misrepresent reply semantics. Mocked UI tests prove cancel invokes no API, the visible reviewed fields match the request, the UUID is stable for the attempt, and uncertainty disables resend.
• Alternatives considered: allow cross-provider thread switching; use browser confirm; hide recipient/thread; auto-retry network failures; enable manual correspondence through application SMTP; add a new draft framework before proving the flow. These weaken provenance, accessibility, consent or duplicate safety.
• Consequences: read-only connections require explicit reconnect consent. This increment supports replies, not durable refresh recovery or a new-message composer; those remain tracked rather than being implied. Basic email remains ungated by Pro and no AI path can send.
• User approval required: No; this is synthetic local MAIL-001 implementation. Real provider/email verification remains gated.
• Reversible: Yes. Revert 449faeb; API, ledger, scopes and saved correspondence remain unchanged.

DEC-045 — Interrupted sends age to terminal safety states without retry

• Date: 2026-08-10
• Decision: Run a post-readiness safety scan every five minutes. After 15 minutes, pending attempts become failed-before-provider and sending attempts become uncertain. Use conditional cross-owner updates in one local transaction and create one content-free owner notification; never invoke or enqueue provider delivery.
• Reason/evidence: a process can stop between ledger reservation, provider acceptance and local completion. The application can prove an old pending attempt never reached the provider, but cannot prove the same for sending. Real-SQLite two-owner tests prove classification, fresh-row preservation, notification isolation and repeat-run idempotency.
• Alternatives considered: retry on restart; leave rows indefinitely; mark every row failed; mark every row sent; scan provider mailboxes; add provider-specific reconciliation. These risk duplicate email, misleading state, excessive provider authority or unresolved user state.
• Consequences: interrupted sending requires manual Sent-folder review. Multiple replicas may select the same candidate, but the status predicate allows only one update/notification. The five-minute query intentionally avoids a new migration in this increment; large-ledger performance remains a measured rollout check.
• User approval required: No; this is local safety recovery with no provider/external call.
• Reversible: Yes. Revert ee5ef7e; existing terminal states and notifications remain truthful and should not be rewritten.

DEC-046 — Resolve the audit gate with supported patches, then the smallest router major

• Date: 2026-08-10
• Decision: Update transitive js-yaml/nanoid within their existing major lines and move React Router to 7.18.2 because the later two React Router advisories have no patched 6.x release. Adapt only the obsolete RouterProvider flag and Jest encoding globals; keep the existing route model.
• Reason/evidence: 6.30.4 fixed the originally reported protocol-relative redirect but remained affected by two newer advisories, so npm audit still failed. React 19 and Node 22 satisfy v7 requirements. Audit, focused data-router tests, all frontend tests and the production build pass on 7.18.2.
• Alternatives considered: npm audit fix --force without review; suppress moderate findings; stay on 6.30.4; redesign routing. These either hide the resolution, leave the deployment gate red or expand scope unnecessarily.
• Consequences: the frontend now requires Node 20 or later through React Router v7. CI/live deployment and navigation smoke remain required before DONE.
• User approval required: Yes; the user explicitly requested repair of the reported live deployment audit failure and continuation.
• Reversible: Technically yes by reverting b55a592, but that restores known advisories and the failed gate. Prefer a forward compatibility fix if a deployment-only issue appears.

DEC-047 — Follow-up drafting stays; direct SMTP delivery is retired

• Date: 2026-08-10
• Decision: Preserve grounded follow-up draft generation/edit/copy, replace its direct send control with a link to canonical Job email, and make the legacy send endpoint return 410 without an application SMTP dependency. Keep scheduled reminder notification email unchanged.
• Reason/evidence: the legacy action bypassed provider identity/re-consent, explicit final confirmation, the idempotency ledger and uncertain-delivery handling already implemented in MAIL-001. Drafting itself is safe and useful. Focused worker tests prove the separate scheduled reminder path remains registered and functional with fakes.
• Alternatives considered: route the legacy body directly into /api/email/send; retain both send paths; remove drafting/reminders; silently redirect the POST. These either bypass the reviewed confirmation context, preserve conflicting authority, remove unrelated functionality or misrepresent a state-changing API response.
• Consequences: users copy the draft or open Job email and review provider/from/recipient/content before sending. Old API callers receive an explicit terminal response and must migrate; no correspondence/follow-up date is falsely recorded as sent.
• User approval required: No; this is the smallest safe completion of the already active MAIL-001 programme and sends no email.
• Reversible: Yes by reverting 8fe3903, but that reintroduces the unsafe SMTP bypass. Prefer forward migration of any remaining caller to /api/email/send with explicit confirmation.

DEC-048 — Export delivery state, not content fingerprints

• Date: 2026-08-10
• Decision: Add one explicit send-attempt export record to both existing owner export surfaces. Include provider/request/status/provider-message/failure/timestamps, but omit the internal payload hash. Verify the existing job foreign-key cascade with real SQLite; leave complete identity/account deletion to SEC-009.
• Reason/evidence: users need readable delivery history, while the ledger intentionally stores no recipient, subject or body. The payload hash exists only for deduplication/conflict checks and is not meaningful portable data. Two-owner daily files and decrypted backup tests prove coverage/isolation; the cascade test proves one hard job deletion does not affect the other tenant.
• Alternatives considered: export the entity directly; include the payload hash; omit attempts; claim UserManager identity deletion is complete; add a second cleanup routine. These expose internal correlation data, lose audit history, overstate the current account lifecycle or duplicate the database cascade.
• Consequences: export schemas gain an additive EmailSendAttempts collection. Complete account/database/file/token/backup deletion remains an explicit SEC-009 release item rather than hidden in MAIL-001.
• User approval required: No; this is the approved MAIL-001 export/cascade requirement using synthetic local data only.
• Reversible: Yes. Revert aff34cc; no schema or stored data changes. Existing export files remain valid historical artifacts under their configured retention.

DEC-049 — Keep Career Workspace smoke behavior-based

• Date: 2026-08-10
• Decision: Replace the removed marketing-copy assertion in the Career Workspace Playwright smoke with an assertion that the visible Open CV Builder action targets /career/builder.
• Reason/evidence: Gitea run 608 proves the dependency audit now passes and the page heading renders, but CI fails on copy that the current unit test explicitly expects to be absent. The route action is always rendered and tests the supported user journey. Full local Playwright passes 4/4.
• Alternatives considered: restore obsolete copy solely for the test; weaken the smoke to heading-only; increase timeout. These would contradict current product behavior, reduce journey coverage or hide no timing defect.
• Consequences: wording can evolve without breaking CI while the page-to-builder navigation contract remains protected. Replacement CI and live deployment verification are still required.
• User approval required: No; this is a focused correction to a stale test exposed by the user-authorized deployment repair.
• Reversible: Yes. Revert the focused smoke change, though run 608 would fail again until the expectation or page behavior is reconciled.

DEC-050 — Share correspondence context without broadening mailbox authority

• Date: 2026-08-10
• Decision: Give the shared correspondence component a minimal company/recruiter/role context contract and supply it from both the job dialog and Application Workspace. Do not add provider mutation buttons or scopes in this increment.
• Reason/evidence: the workspace reused the correct correspondence domain but passed null as any, so its suggestion surface lost job context. Gmail currently grants readonly plus send, Graph grants read plus send, and IMAP is read-only; none has a safe read/write mutation seam. Focused tests and the full frontend/build pass.
• Alternatives considered: duplicate the correspondence implementation; expand the workspace aggregate with a full job object; silently request Gmail modify/Graph read-write scopes; display unsupported actions. These add coupling, authority or misleading behavior.
• Consequences: both application surfaces now generate the same optional contextual searches without automatic linking. Provider category mutations remain explicit remaining work that requires a separately reviewed scope/re-consent design.
• User approval required: No; this is a local shared-view correction inside active MAIL-001 and invokes no provider.
• Reversible: Yes. Revert the context-prop changes and test; stored data, scopes and provider grants are unchanged.

DEC-051 — Unlink the app relationship, never the provider copy

• Date: 2026-08-10
• Decision: Expose confirmed Gmail unlink in the canonical hub through the existing owner-scoped unlink endpoint. Label the data effect explicitly and show no equivalent action for providers without a supported link domain.
• Reason/evidence: unlink already existed in the per-job component but not the hub, creating inconsistent copies of the same workflow. The endpoint removes imported JobTracker correspondence for the owned job and returns the thread to review; it does not call Gmail deletion. UI confirmation and real-SQLite two-user tests pass.
• Alternatives considered: delete provider mail; implement a duplicate endpoint; show disabled Outlook/IMAP actions; remove unlink from the per-job view. These broaden authority, duplicate behavior, mislead users or regress a supported workflow.
• Consequences: a thread can be reconsidered in recruitment review after unlink. Imported JobTracker copies are removed only after confirmation, while the mailbox remains untouched. Relink/move stays on the existing per-job management surface for now.
• User approval required: No; this is local implementation of the approved MAIL-001 unlink workflow and uses only mocked/synthetic data.
• Reversible: Yes. Revert 1dabbeb; the endpoint and per-job unlink remain available, with no schema/provider grant change.

DEC-052 — Report only capabilities the connection can use

• Date: 2026-08-10
• Decision: Distinguish disconnected, connected-without-read, read-only/re-consent, read-plus-send and provider-status failure in the canonical hub. Keep saved correspondence visible during provider-status failure.
• Reason/evidence: the previous chip appended Read only even to disconnected providers and silently removed all provider context when the status request failed. Focused failure/capability tests, full frontend and build pass.
• Alternatives considered: keep the ambiguous chip; hide the entire inbox on provider failure; infer archive/read-write support from provider name. These misstate capability, reduce failure isolation or invent authority not present in installed scopes.
• Consequences: users can distinguish connection state from send consent and know saved data remains available. Mailbox organization actions remain absent until a reviewed provider contract and re-consent path exist.
• User approval required: No; this corrects local state communication and invokes no provider.
• Reversible: Yes. Revert f9e641c; no API, schema, scope or stored state changes.

DEC-053 — Basic email stays Free; private drafts need a tenant-owned store

• Date: 2026-08-10
• Decision: Pin basic provider email to local authenticated access without a Pro policy. Do not persist private email content in browser storage or reuse recruiter-message/job draft fields for provider email recovery.
• Reason/evidence: the current send route and hub have no entitlement gate, while all send safety checks remain. Repository inventory found no email-draft entity that owns provider/thread/recipient/subject/body/idempotency together; existing recruiter drafts are a different product domain.
• Alternatives considered: add Pro to basic email; use localStorage/sessionStorage; overload RecruiterMessageDraft; claim in-memory replies are durable. These violate the product decision, expose content on shared browsers, conflate domains or overstate recovery.
• Consequences: Free users retain non-AI email. Durable/new-message drafts require an owner-filtered database model, bounded API, export/deletion coverage and rollback migration as a separate cohesive increment. Future AI assistance remains Pro/privacy gated.
• User approval required: No; this is repository-local verification and a safety boundary inside approved MAIL-001.
• Reversible: The regression test can be reverted with 7f41cb2; no runtime/schema state changed. Adding a Pro gate later would be an explicit product change.

DEC-054 — Make draft persistence inert before exposing private content

• Date: 2026-08-10
• Decision: Add an owner-filtered, revisioned EmailDraft tied by cascade to an owned job, with explicit SQLite/MariaDB migration paths, but expose no draft API or UI until readable export and complete deletion implications are covered.
• Reason/evidence: refresh recovery needs server-side ownership, while browser storage and recruiter-message fields violate privacy/domain boundaries. A real-SQLite two-owner test proves isolation and job cascade; backend 625/625, model-current and dual-provider up/down SQL checks pass.
• Alternatives considered: ship schema/API/UI together; use local storage; reuse correspondence or recruiter drafts; persist provider tokens or idempotency attempts in the draft. These make a larger private-data boundary harder to review, expose shared-browser content or conflate delivery state with editable content.
• Consequences: migration 20260810075206_AddEmailDrafts is additive and reversible, but production must not expose drafts until export/deletion coverage and a bounded owner/job-validating API are verified. A full blank SQLite migration rehearsal remains blocked by the pre-existing JT-019 historical-chain defect, not this migration.
• User approval required: No; MAIL-001 authorizes repository-local durable drafts and no production migration/provider action occurred.
• Reversible: Downgrade the migration before reverting 14b396a. No reachable application behavior exists in this increment.

DEC-055 — Export drafts through existing owner-scoped backup boundaries

• Date: 2026-08-10
• Decision: Include complete readable email drafts in the authenticated encrypted backup and per-owner daily export, using an explicit DTO and the owning job ID set. Do not add a separate export route.
• Reason/evidence: private draft data must be portable before it is user-reachable. Existing exports already own tenant isolation and readable user data; focused two-owner tests prove the on-demand payload and hashed daily files contain only their owner's draft.
• Alternatives considered: omit bodies; export only metadata; add a draft-only download; delay export until account deletion. These produce an incomplete user export, duplicate authorization or expose a reachable data category without portability.
• Consequences: daily JSON exports now contain draft bodies under the same filesystem-at-rest protections and retention policy as correspondence. SEC-009 must explicitly delete live drafts and define backup/export retention; no export schema version was broken because the collection is additive.
• User approval required: No; this is a repository-local privacy prerequisite using synthetic data and existing export authority.
• Reversible: Revert 2fa4e38; stored drafts remain unchanged, but draft UI/API must not ship without another readable export path.

DEC-056 — Keep draft saves incomplete, bounded and provider-inert

• Date: 2026-08-10
• Decision: Expose local-authenticated draft CRUD under /api/email/drafts. Require an owned job and registered provider at creation, allow incomplete recipient/subject/body for autosave, keep provider/thread/job immutable, and require a matching revision for updates/deletes.
• Reason/evidence: refresh recovery must preserve work before send fields are complete, while provider/thread provenance must not silently change. Atomic revision predicates prevent last-write-wins loss; explicit owner clauses plus global filters deny foreign job and direct draft IDs in real SQLite tests.
• Alternatives considered: require send-valid content on every save; contact the provider during save; allow provider/thread changes; use unconditional updates; hide foreign rows only in UI. These break autosave/offline recovery, add side effects, weaken provenance, lose concurrent edits or fail authorization.
• Consequences: saving never sends or checks connection state. Send still uses the separate explicit-confirmed/idempotent boundary and revalidates all fields/provider consent. UI must surface revision conflicts and treat bodies as untrusted plain text.
• User approval required: No; this is approved local MAIL-001 work with synthetic data and no provider/production action.
• Reversible: Revert a9bb22e; the inert/exported schema remains for a later UI. Existing stored drafts are unaffected.

DEC-057 — Persist one delivery identity for the life of a draft

• Date: 2026-08-10
• Decision: Assign every server draft one canonical client-request UUID, preserve it across edits/refresh, include it in readable exports, and allow listing all current-owner drafts for recovery.
• Reason/evidence: the send ledger keys idempotency by client request ID. Generating a new UUID after refresh would let identical restored content reserve a second delivery attempt. Focused tests prove creation, edit preservation, export and tenant-filtered all-draft listing.
• Alternatives considered: generate UUID only at send; regenerate after every edit/refresh; derive it from content; store it only in browser memory. These permit duplicate delivery after refresh, conflate content changes with attempt identity or lose the safety state on navigation.
• Consequences: one draft maps to one send attempt identity until the user explicitly starts a new attempt after a confirmed failure. A sent/restored draft can only replay the existing ledger result. The additive migration defaults only during the same pre-exposure rollout; no draft UI existed before it.
• User approval required: No; this closes a local safety dependency before UI exposure and sends no email.
• Reversible: Downgrade 20260810080858_AddEmailDraftClientRequestId before reverting 80b5532. Do not deploy draft UI without an equivalent persisted idempotency identity.

DEC-058 — Make draft persistence explicit and conflict-visible in the hub

• Date: 2026-08-10
• Decision: Keep reply editing local until the user chooses Save draft, then adopt the server draft/revision/client-request identity. Offer saved drafts for refresh recovery, require revisioned delete, and surface 409 conflicts without replacing local text.
• Reason/evidence: silent autosave introduces navigation/race semantics that are not yet proven. Explicit save is predictable, supports incomplete drafts and preserves the existing final send confirmation. Focused conflict/recovery tests, full frontend and build pass.
• Alternatives considered: browser storage; silent debounced autosave; last-write-wins; discard on refresh; auto-send after save. These expose private data, risk lost edits or weaken explicit send consent.
• Consequences: unsaved edits are intentionally not refresh-durable; saved replies are. A successful send attempts revisioned draft cleanup, while the persisted client ID keeps any surviving copy duplicate-safe. Compose-new-message and durable failed-attempt identity rotation remain separate increments.
• User approval required: No; this is approved local MAIL-001 UI work with mocked APIs and no email/provider action.
• Reversible: Revert d3d2b67; stored drafts remain available through the API/export but no UI consumes them.

DEC-059 — Rotate draft delivery identity only from a terminal failure

• Date: 2026-08-10
• Decision: Add an explicit revisioned new-attempt action that issues a new draft client-request UUID only when the authenticated owner's matching ledger row is failed. Wire the existing Prepare new attempt UI to this action for saved drafts.
• Reason/evidence: ordinary edits/refresh must preserve idempotency, while a provider-confirmed failure needs a deliberate recovery path. Pending/sending/uncertain/sent states cannot disprove delivery and must not rotate. Real-SQLite owner/stale/status tests and mocked UI pass.
• Alternatives considered: let the browser invent UUIDs; rotate on any error; rotate during save; automatically retry failed sends. These lose durable state, can duplicate uncertain/sent mail or weaken explicit approval.
• Consequences: definitively failed drafts can be reviewed and retried under a new ledger identity; all ambiguous or successful attempts remain non-retryable. Unsaved local drafts retain the existing explicit local new-attempt behavior because no durable ledger relationship exists yet.
• User approval required: No; this is approved safety work with fake/local evidence and no email/provider call.
• Reversible: Revert 29de263; saved failed drafts then have no durable retry rotation and the UI should not offer that action.

DEC-060 — Reuse the reviewed draft boundary for new messages

• Date: 2026-08-10
• Decision: Compose new email starts only after selecting an owned recent job and a connected send-capable provider. It creates a local blank, threadless draft and reuses explicit Save draft plus Review and send; read-only/disconnected providers are not choices.
• Reason/evidence: new messages need the same ownership, persistence, confirmation, idempotency and failure semantics as replies. The existing owner-filtered job list and provider capability endpoint provide those selection inputs without a second send implementation.
• Alternatives considered: free-text job IDs; allow disconnected/read-only providers; bypass persistence; a second composer/send route; infer a provider. These expose authorization errors, promise unavailable delivery or duplicate safety logic.
• Consequences: empty-job and missing-send-consent states explain why compose is unavailable. The hub currently lists the 100 most recent owned jobs; older jobs remain accessible from their job workspace and broader searchable selection is follow-up UX, not a hidden authorization bypass.
• User approval required: No; local UI work uses mocked APIs and sends no message.
• Reversible: Revert b735963; reply drafting and all server draft/send boundaries remain.

DEC-061 — Stop MAIL scope at installed provider authority

• Date: 2026-08-10
• Decision: Mark MAIL-001 IMPLEMENTED — NOT VERIFIED after completing the repository draft/send/link flows. Do not add read/unread, pin/read-later, archive, spam or trash UI because the installed Gmail/Graph/IMAP contracts and granted scopes do not authorize those mutations.
• Reason/evidence: Phase 9 says these states apply where supported. Current Gmail adds read/send, Graph adds read/send, and IMAP is read-only; the neutral interface exposes read/detail/send only. Displaying or simulating category actions would be false capability and production-risking scope creep.
• Alternatives considered: request modify/read-write scopes silently; mutate only JobTracker copies while labelling them provider actions; add disabled controls; keep MAIL indefinitely in progress. These broaden consent, misrepresent mailbox state, clutter UX or block independent work.
• Consequences: real-provider re-consent, provider category design, browser gates, SEC-009 deletion and production verification remain explicit blockers before DONE. JOBS-001 becomes the sole in-progress item.
• User approval required: No for status correction; any future provider-scope/re-consent rollout needs a separately reviewed package and safe account verification.
• Reversible: Yes. MAIL-001 can return to IN PROGRESS when provider mutation authority is approved and testable.

DEC-062 — Expose only verified discovery provenance

• Date: 2026-08-10
• Decision: Treat NAV as verified because the controller itself fetches the official NAV feed; expose a stable source key/display name, label acquisition as searched, stamp the request retrieval time and map a deadline only from NAV's explicit applicationDue field. Do not infer work mode or add a one-option source filter.
• Reason/evidence: the previous card hardcoded NAV and an update date while the response's source field was unused. Phase 4 requires honest source/type/retrieval/deadline data and explicitly prohibits presenting inference as verification. Focused provenance/import tests and full regressions pass.
• Alternatives considered: derive source from listing hostname; infer remote/hybrid from title/location; add a disabled or single-option source filter; label date_modified as retrieval time. These would confuse derived and verified data or add controls without a real choice.
• Consequences: NAV cards now explain where and how the listing was obtained and preserve attribution into tracking. Missing work mode remains visibly absent rather than guessed; source filtering becomes useful only when a second real provider is implemented.
• User approval required: No; this is approved repository work using synthetic fixtures with no external request.
• Reversible: Revert 511a9f6; no schema or stored data changes are involved.

DEC-063 — Close synthetic JOBS scope before resuming theme order

• Date: 2026-08-10
• Decision: Move JOBS-001 to IMPLEMENTED — NOT VERIFIED after its complete synthetic browser and regression matrix, then make UX-003 the sole in-progress item before JOBS-002.
• Reason/evidence: JOBS now has honest provenance/import/result/duplicate behavior plus mocked 375/768/1440, theme, keyboard and long-content evidence. Remaining NAV/production checks are external. Theme corrections precede the job-workspace redesign in the validated programme order and the browser environment is currently available.
• Alternatives considered: keep JOBS in progress while repeatedly retrying unavailable live NAV/production; begin the larger JOBS-002 package; skip the outstanding Kanban theme requirement. These would misstate the active work or ignore the established dependency order.
• Consequences: UX-003 becomes the only in-progress package. JOBS remains explicitly incomplete rather than DONE; its live gates stay visible.
• User approval required: No; this is tracking and safe local sequencing within the approved programme.
• Reversible: Yes; JOBS can resume when live NAV or production verification becomes available.

DEC-064 — Use one accessible Kanban move state model

• Date: 2026-08-10
• Decision: Drive pointer and keyboard moves from the same picked-up job and active-column state. Use live theme variables for every surface/tone, Space to pick up/cancel, Enter or Space on a focused valid column to drop, and update local state only after the API succeeds.
• Reason/evidence: the prior static light palette caused confirmed white columns in dark mode, while native drag alone exposed no target or keyboard semantics and rejected moves became unhandled promises. Shared state makes valid/active/invalid/selected/failure states consistent without replacing the precise status menu.
• Alternatives considered: isolated dark hex colours; CSS-only drop feedback; optimistic moves with rollback; a second keyboard-only status implementation; replacing drag with a new library. These duplicate policy, risk stale UI or add unnecessary dependency/complexity.
• Consequences: pointer and keyboard users receive the same server-confirmed behavior and announcements; custom Other remains visibly invalid. The existing exact-stage menu stays available and now has an accessible name.
• User approval required: No; this is approved local UX/accessibility work with mocked browser APIs.
• Reversible: Revert fb6f17e and 4e5ce0c; no schema/data change is involved.

DEC-065 — Reuse the workspace as a route-backed list overlay

• Date: 2026-08-10
• Decision: Keep /applications/:id as the full-page fallback, but make a table row's primary Open action render that same workspace in a MUI dialog at /jobs?workspace={id}&section={section}. Use a pushed history entry for opening, replace only section changes, Back for UI-originated close and query removal for direct-link close.
• Reason/evidence: V-158 confirmed that the previous quick-dialog/full-page chain discarded list context and could not deep-link the embedded presentation. The workspace already composes the authoritative domain components; reusing it avoids a second workspace. MUI supplies the requested focus trap/restoration and full-screen mobile presentation.
• Alternatives considered: build a second drawer workspace; keep the quick dialog primary; clone workspace sections into JobTable; make the full-page route the only URL. These duplicate ownership or leave the Phase 11 navigation defect.
• Consequences: list state remains mounted while the overlay is open, direct workspace/section URLs and Back/Forward work, and a full-page link remains. Complete URL-backed filters and unsaved-edit guards are separate required increments before JOBS-002 can leave progress.
• User approval required: No; this is approved repository implementation with mocked local data and no production/provider action.
• Reversible: Revert the JOBS-002 overlay commit; no schema, dependency or stored-data change is involved.

DEC-066 — Make theme preference independent of authentication

• Date: 2026-08-15
• Decision: Supersede DEC-033's account-scoped storage portion with one canonical browser key, jobtracker.themeMode. Migrate the current legacy value once, apply the same resolution in the pre-paint script, and ignore auth-user changes for theme state.
• Reason/evidence: the user-visible preference is application chrome, while the auth-derived key is resolved asynchronously and can differ across startup paths. Tying these together retained competing sources of truth and allowed refresh to change scheme. Focused persistence/migration/provider tests and production TypeScript build pass.
• Alternatives considered: add route-specific theme effects; keep user/anonymous fallback ordering; let MUI own a second storage key. These preserve the race or reintroduce multiple owners.
• Consequences: theme remains stable through login/logout/navigation/refresh on a browser. Former account keys remain readable for one-time migration but are no longer written.
• User approval required: No; this corrects the requested persistence defect without schema, dependency or production change.
• Reversible: Remove the canonical read/write and restore auth-key subscriptions; legacy values were not deleted.

DEC-067 — Protect the final administrator at the API boundary

• Date: 2026-08-15
• Decision: Refuse demotion or deletion of the final Admin in UsersController; expose current-user and removal-safety state to the admin UI; require an app-owned destructive confirmation for any demotion and stronger copy for self-demotion/self-deletion. Preserve unrelated roles during an Admin toggle.
• Reason/evidence: confirmation alone cannot protect direct API calls or concurrent UI versions. Existing MUI confirm/prompt primitives already match the application and avoid a second dialog dependency. Four controller and three UI tests cover final-admin protection, other-admin demotion, self cancel and self confirm.
• Alternatives considered: SweetAlert2; silently forbid every self-demotion; UI-only warning. These duplicate the design system, prevent legitimate handover, or fail to enforce the invariant.
• Consequences: the final administrator cannot be removed by supported API paths. A self-demotion remains possible only when another administrator exists and the user explicitly confirms.
• User approval required: No; this is requested safety hardening with no production mutation.
• Reversible: Revert the controller/UI change; no stored data or schema changed.

DEC-068 — Make the job workspace a dedicated canonical page

• Date: 2026-08-15
• Decision: Supersede DEC-065. Route every job/application open action to /jobs/:id, keep /applications/:id only as a query-preserving compatibility redirect, and remove the legacy dialog/expandable-detail path from the applications table. Preserve the URL-owned list location in route state for the workspace return control.
• Reason/evidence: the user explicitly rejected the popup interaction and asked for a scalable application workspace. The existing workspace already composes the authoritative checklist, CV, cover-letter, documents, intelligence, timeline and correspondence domains, so the safe change is canonical routing and richer aggregate data rather than another implementation.
• Alternatives considered: retain the route-backed overlay; make a drawer; copy legacy dialog tools into a new page. These conflict with the requested dedicated-page model or duplicate domain ownership.
• Consequences: rows/cards and contextual shortcuts open one responsive workspace; internal row controls remain independent; list state survives return; old application links still resolve. The former quick dialog remains in source for rollback until broader regression proves it can be safely deleted.
• User approval required: No; explicitly requested.
• Reversible: Restore DEC-065 routing/list presentation; no schema or stored data changed.

DEC-069 — Keep CV ingestion hybrid and isolate status polling

• Date: 2026-08-15
• Decision: Keep deterministic Python/library text extraction and OCR, pass extracted text through the existing Ollama-first normalization/classification routes, then validate/diff/review in C#. Poll extraction-run status independently from profile content and never replace unsaved editor state during background refresh.
• Reason/evidence: the repository already implements the user-proposed Python → Ollama → structured-data flow. Python libraries are the correct boundary for PDF/DOCX/image decoding; Ollama adds value in semantic section recognition, but cannot safely or deterministically replace binary parsing. The observed reset was caused by full-profile polling, not controlled-input behavior.
• Alternatives considered: send binary files directly to Ollama; remove deterministic repair/fallback logic; continue full-profile polling; auto-save on every poll. These reduce format coverage, factuality, review safety or user control.
• Consequences: all career fields retain unsaved edits while processing status changes. AI reconstruction remains explicit, local-first and review-gated; accuracy work can be benchmarked per model without changing the ingestion boundary.
• User approval required: No; this implements the requested behavior within the existing approved local-AI architecture.
• Reversible: Rejoin run/profile loads, though that would restore the confirmed data-loss UX defect; no schema/config/data migration changed.

DEC-070 — Let long CV content paginate instead of shrinking or clipping

• Date: 2026-08-15
• Decision: Keep normal entries together, classify over-height entries/list items as flowable, wrap every user-controlled text boundary, and use physical A4/Letter metrics in the editor. Put custom and profile-backed sections in one persisted order. Serialize autosaves and save before export/public rendering.
• Reason/evidence: the renderer used overflow:hidden, fixed 1fr columns and break-inside:avoid on every entry; the editor hardcoded A4 and rounded page counts. Together these could hide partial pages, clip unbroken values or make a block taller than the printable page impossible to paginate. Pathological Chromium/PDF proof produced nine readable pages with zero horizontal offenders without reducing font sizes.
• Alternatives considered: globally shrink text; truncate content; make every entry freely splittable; create a free-form canvas editor; maintain a separate custom-section order. These reduce readability, damage content, or duplicate state.
• Consequences: preview, public HTML and PDF retain one render path; normal entries avoid awkward splits while large content can cross pages safely. Existing variants remain compatible and acquire shared custom ordering on edit.
• User approval required: No; this implements the requested CV rework without schema, dependency or production changes.
• Reversible: Revert the renderer/editor/resolver checkpoint; stored settings remain compatible because the existing Sections and custom:<key> contract is used.

DEC-071 — Encapsulate application-answer compatibility storage

• Date: 2026-08-15
• Decision: Keep the existing marked answer block in JobApplication.Notes for storage compatibility, but make the backend the owner of extracting, removing and updating it. Expose human notes and application answers as separate workspace fields, and preserve the answer through the general application editor.
• Reason/evidence: the retired modal was the only editor that understood the marker. The dedicated page displayed markers as ordinary notes, and saving the general editor could erase the answer. A new schema/table would add migration risk for one text value while the existing representation remains adequate behind a clean boundary.
• Alternatives considered: expose the marker format in every editor; create a second application-package table immediately; copy the retired modal wholesale. These leak implementation details, add avoidable migration/state duplication, or restore the popup architecture the user rejected.
• Consequences: the dedicated page edits and clears answer/recruiter drafts directly, ordinary notes remain readable, existing rows require no migration, and legacy calls continue to work. A future normalized column/table can migrate behind the same API without another UI change.
• User approval required: No; this is compatibility-safe implementation of the requested dedicated workspace.
• Reversible: Revert the workspace/API boundary; no schema or stored-data rewrite occurred.