# CURRENT_TASK.md

**Task:** Add a fourth, non-terminal outcome to the suggested-fix banner — **Awaiting verification** (`applied_pending`). Today the verifying banner forces a synchronous verdict (worked / didn't / partial), but a lot of real fixes are async — engineer ran the script but is waiting on the client to power-cycle, AD replication, an O365 license sync. Without a fourth state the banner sits stale or the engineer guesses wrong.

**Status:** ✅ **Engineering complete; PR #156 open.** Backend tests, prompt guardrail, frontend tsc/build clean; Alembic migration applies. Pending browser QA.

**Branch:** `feat/fix-pending-verification` (off `main` after the Escalation Mode merge).

**PR:** https://gitea.resolutionflow.com/chihlasm/resolutionflow/pulls/156

## What ships

| Layer | Change |
|---|---|
| Schema | New `FixStatus="applied_pending"` + new `pending_reason` Text column on `session_suggested_fixes`. |
| Migration | `c0f3a4b7e91d` — adds `pending_reason`, extends status CHECK constraint. |
| API | `PATCH /suggested-fixes/{id}/outcome` accepts `applied_pending`, requires `notes`, stamps `applied_at` only (NOT `verified_at`). Pending in/out transitions allowed (parked, like partial). |
| Generators | `resolution_note_generator` and `escalation_package_generator` system prompts handle the new status without real-looking examples; resolution note frames the fix as provisional; escalation package surfaces pending verification as the leading hypothesis with reference to what's being waited on. |
| Frontend | New `BannerMode='pending'` + `PendingBanner` component (info-tone, mirrors `PartialBanner`) with worked / didn't / update reason / dismiss actions; "Waiting to verify…" overflow option in `VerifyingBanner`; nudge "Still checking" now records pending with a reason instead of just silencing; `AssistantChatPage` banner-mode derivation maps `applied_pending → 'pending'`; page-level Resolve/Escalate now handle pending honestly. |
| Tests | 4 new integration tests in `test_fix_outcome_endpoint.py`: pending-requires-notes, pending stores reason + applied_at-not-verified_at, pending→success transition, pending_reason update on re-PATCH. 21/21 pass. Prompt anti-parrot guardrail passes. Frontend `tsc -b` and Vite build pass via Docker. |

## Out of scope (intentionally)

- Cross-session "Follow-ups" dashboard rollup. The chat-anchored `PendingBanner` is the per-session reminder. Add the dashboard surface only if engineers report losing track across multiple pending sessions.
- Optional follow-up timer ("remind me in 30m"). Nice but not the wedge.

## Resume point — DO THIS NEXT

**Browser QA:** verify the new flow end-to-end in dev:
1. Trigger a suggested fix, click Apply, then open the verifying banner overflow → "Waiting to verify…" → enter a reason → confirm `PendingBanner` renders with the reason.
2. From `PendingBanner`, click "It worked" → confirm transition to terminal success and banner dismissal.
3. From `PendingBanner`, click "Update reason" → confirm reason updates server-side.
4. From `PendingBanner`, click "Dismiss" → confirm banner dismissal and no terminal timestamps.
5. Trigger the nudge state (3+ post-apply messages) → click "Still checking" → enter a reason → confirm pending state takes over.
6. From a pending fix, use page-level Resolve → confirm the fix is patched to `applied_success` before the resolution note.
7. From a pending fix, use page-level Escalate → confirm the outcome intercept appears before the conclude modal.

After QA passes, commit/push the local review fixes and merge PR #156.

## Just-shipped (2026-04-30)

**PR #155 — Escalation Mode wedge** merged into main as `ac42f97`. The wedge for ResolutionFlow's GTM (first paying-customer push). Senior tech sees structured handoff context in seconds via the magic-moment screen. Plan: [`docs/plans/2026-04-27-escalation-mode-wedge-design.md`](../docs/plans/2026-04-27-escalation-mode-wedge-design.md).

## Two-metric framing (Escalation Mode — read before quoting numbers)

The in-product `GET /analytics/flowpilot/escalations` endpoint measures *post-claim time-to-first-action*. The "minutes recovered" sales claim is `manual_baseline − in_product_metric`. Manual baseline comes from the founder's stopwatch on the next 5 escalations. Don't roll the in-product number alone into "minutes recovered" — that's the apples-to-oranges miscount Codex caught.

## Kill-switch (Escalation Mode)

Week 8: if 0 of 3 pilots produce a verifiable hours-saved-per-week number above 1.0, revisit the wedge.