toga-ai 1.0.384 → 1.0.385

package/knowledge/sessions/2026-07-20-bdr-vapi-golive-tcox.md

@@ -0,0 +1,76 @@
+---
+type: session
+slug: bdr-vapi-golive
+title: BDR VAPI call loop — self-heal fix, dialer diagnosis, go-live prep
+author: tcox
+repos: [bdr, ai-bdr, worker2, api2, dbchanges2]
+framework: "2.0"
+client: shared
+status: active
+created: 2026-07-20
+updated: 2026-07-20
+---
+
+# Session: bdr-vapi-golive
+**Date:** 2026-07-20
+**Project/Repo:** bdr + worker2 + api2 + dbchanges2 (2.0)
+**Task:** Get VAPI phone calls working end-to-end on the new BDR Next.js site (fix the primary-phone stranding bug, unblock the dialer, prove the call loop) and prepare same-day go-live for a real campaign.
+
+---
+
+## What WORKED
+- **BDR primary-phone self-heal shipped** — `src/server/toga.ts` + `test/togaPrimaryPhone.test.ts` (10 regression tests written FIRST, all failed on old code; 114/114 suite green after; tsc + eslint clean). Detects a write response echoing `primaryContactPhoneNumber` null OR digit-mismatched, finds the orphaned phone-row uuid (echo → GET /contacts/{uuid} → POST /contact-phone-numbers), re-links via identifier-only `PUT {primaryContactPhoneNumber:{uuid}}` (single-key identifier objects hit api2's forced-MATCH path). `requestCall` throws on failure (route → 500); upsert absorbs + warns. Commit `4ac579e` on `BDR-Phase-2`, merged into deploy branch `_dev-sandbox`.
+- **api2 root cause located precisely** — `Component/Api/V2/V2.php` UPDATE path lines **4985–4996** inject a reverse `contactId` back-reference into nested child objects (commit `91d803f`), defeating the single-uuid forced-MATCH fast-path (L6952) and diverting `getForeignKeyValue()`'s return before the FK assignment (L5047). CREATE path (L4601-4616/4647) has no injection. Verified by an Explore subagent read; documented in knowledge (`api2 nested-relationship-writes.md`).
+- **Dialer blockage root-caused from worker2 source + prod reads** — campaign 2 ('26.05 Ryan Nitti') blocked SOLELY by `dateEnd=2026-06-09` (thrown by `validateCampaignReady` L784-786); its assistant (CampaignAssistants id 3 'Alex') is valid; the dialer cron (Core.CronJobs id 12 'AI BDR', `* * * * *`) is ACTIVE and has errored every minute since June 9. Contact 248 (tcox) is fully eligible: primary phone linked (row 132), `dtNextContactRequested` set, 0 prior attempts.
+- **Test-campaign migration written, reviewed, pushed** — `Client_True/2026-07-20a - AiBdrTcoxTestCampaign.sql` on `feature/ai-bdr-test-campaign` (`e99a367`, pushed). Fences campaign 2, creates '26.07 - AI BDR - TCOX TEST' (copy of campaign 2 config, NO CampaignCallWindows rows → only requested calls possible), links ONLY contact 248. sql-reviewer verdict SHIP (idempotent guards match real UNIQUE constraints).
+- **Accidental `_main` push recovered cleanly** — revert commit `56055a0` pushed (never force-push); verified via read-only DB queries the executor had NOT applied the migration first; feature branch rebuilt by cherry-pick on reverted `_main`.
+- **Read-only diagnosis tooling** — BDR's api2 client creds (from `BDR/.env.local`) work for REST probes; the "TOGa Database Integration" MCP connector (`toga_query`, prod + sandbox envs, read-only) works for SQL. ClickUp MCP works for ticket lookups.
+- **/capture published** 4 knowledge docs incl. new workflow `2.0/apps/ai-bdr/workflows/safe-call-loop-testing.md`.
+
+## What did NOT work — DO NOT RETRY THESE
+- **Treating `Campaigns_Contacts.status = PENDING` as the blocker** — PENDING is a NORMAL dialable state; the eligibility SQL checks status only when `dtNextContactRequested IS NULL` (Vapi.php L237), and the webhook resets links to PENDING for retries.
+- **Assuming the campaign date window is checked in `getActiveCampaignsForDialing`** — it is NOT (that query filters only `isActive=1` + assistant INNER JOIN); the date gate lives in `validateCampaignReady`. Half-wrong first theory cost a probe cycle.
+- **Using the dialer's `dryRun` as a "safe" production preflight** — `createContactAttempt` (sets `dtStarted`, leaves `dtEnded` NULL) runs BEFORE the dryRun early-return (Vapi.php L438-466): a prod dry-run writes real attempt rows (counts toward maxAttempts) and WEDGES contacts as "in progress" until `dtEnded` is set. Unwedge: `UPDATE ContactAttempts SET dtEnded=NOW() WHERE contactId=? AND dtEnded IS NULL`.
+- **Beta end-to-end call test** — impossible as wired: Vapi is one production account whose assistants' serverUrl targets production `webhook.togahub.com`, so a beta-placed call's end-of-call report can never find its `contactAttemptUuid` (it lives in the beta DB). Also `worker2/Config/beta.ini` still points `aws_worker_queue_url` at WorkerProductionQueue. Beta is for SQL/eligibility rehearsal via a LOCAL worker2 run only.
+- **Reviving campaign 2 by extending its dateEnd** — rejected: 125 PENDING links, up to 86 contact-level-dialable real prospects, and the every-minute cron starts dialing within 60s of eligibility.
+- **`isOkayToCall=0` mass-flip plan** — verified it WOULD work (hard-required at Vapi.php L231) but rejected: two mass writes to real prospect rows + restore risk (5 contacts already have `isOkayToCall=0` — genuine DNC — and must never be blanket-restored).
+- **Assuming the BDR site was deployed** — no Amplify app exists; ClickUp `868kdf4tj` "BDR - create amplify and sandbox route" is still 'to do' (ClickUp says Alex Peterson; team says Jeff took it — ClickUp stale).
+
+## Not tried yet (candidates for next session)
+- **The actual test call** — blocked on the dbchanges2 PR merge (branch pushed, PR not confirmed opened/merged).
+- **UI test A:** Call Now with a DIFFERENT number → proves the self-heal live (pointer re-links to new number; dialer calls the new number).
+- **UI test B:** one scheduled call ~10 min out → the call-later path and `formatCstDateTime` CST conversion have NEVER been runtime-tested.
+- **Purest bug repro:** phone-less HubSpot test contact through the funnel (needs someone with HubSpot access to create it).
+- **api2 engine fix PR** — gate the V2.php 4985-4996 injection to true reverse/has-many relations; confirm `RecordFields.childPolicy` for `Contacts.primaryContactPhoneNumberId` during verification.
+- **worker2 dry-run fix** — move `createContactAttempt` below the dryRun return.
+- **Fix CampaignAssistants id 1 'Mark'** — corrupt 37-char `assistantIdentifier` (Alex's uuid + extra trailing '4'); would 400 at Vapi if assigned.
+- **Real-campaign migration** — awaiting business params (name, dates, timezone, attempt caps; recommended: web-funnel-only = NO CampaignCallWindows rows). Entry links need `hsCampaignId=<new campaign uuid>` for attribution.
+- **Cleanup migration** deactivating the test campaign — must NOT merge in the same window as the create file (executor runs folder files alphabetically in one pass).
+- **Stranded-contact backfill** — 40 of campaign 2's 132 linked contacts lack a primary phone pointer (only matters if that list is revived).
+
+## Current file state
+| File | Status | Notes |
+|------|--------|-------|
+| BDR/src/server/toga.ts | committed `4ac579e`, on `BDR-Phase-2` + `_dev-sandbox` | self-heal (TogaPhoneRow, ensurePrimaryPhone, strict requestCall) |
+| BDR/test/togaPrimaryPhone.test.ts | committed `4ac579e`, same branches | 10 regression tests, fake api2 at fetch boundary |
+| dbchanges2/Client_True/2026-07-20a - AiBdrTcoxTestCampaign.sql | on `feature/ai-bdr-test-campaign` `e99a367`, pushed; PR pending | fences campaign 2, creates test campaign, links contact 248 only |
+| dbchanges2 `_main` | revert `56055a0` pushed | accidental direct push recovered; DB was untouched |
+| knowledge (team repo) | 4 docs published via /capture | web-funnel-app, vapi-webhook-handler, nested-relationship-writes, safe-call-loop-testing |
+
+## Decisions made
+- **BDR-side self-heal now, api2 engine fix later** — funnel unblocked without a platform-wide risky change; engine fix deferred to its own reviewed PR. Rejected: waiting on api2.
+- **requestCall strict-throws / upsert absorbs** — a "successful" submit whose call can't happen must not report success; but lead priming must never break page render.
+- **One-contact test campaign over `isOkayToCall` flip** — zero writes to prospect data; test campaign has no call windows so only requested calls can ever fire. Rejected: mass flip (restore risk), reviving campaign 2 (mass dialing).
+- **Migration via dbchanges2 PR, not direct SQL** — review + normal per-env process; test call fires ~60s after the executor applies it (timing note is in the file header + PR body).
+- **Skipped the elevated dbchanges2 architecture-doc edit** about missing `_main` branch protection (developer: repo owner already aware).
+
+## Blockers
+- **dbchanges2 PR** from `feature/ai-bdr-test-campaign` not yet opened/merged (tcox) → blocks the call test.
+- **Amplify app doesn't exist** (ClickUp `868kdf4tj`, urgent, Sprint 82; Jeff has it per team word) → blocks URL-live. `_dev-sandbox` already contains the fix; needs SSR hosting + the 5 env vars (values' location: `info/.env.local`).
+- **Real-campaign business params** not yet provided (tcox/team).
+
+## Exact next step
+> Open + merge the dbchanges2 PR from `feature/ai-bdr-test-campaign` (`e99a367`). When the executor applies it, answer the call from +1 (216) 616-0067 (contact 248's primary phone rings within ~60s), then verify write-back with read-only queries: `ContactAttempts` for `contactId=248` has `dtEnded` + `c_callOutcome` + `c_vapiCallIdentifier` set, and `Contacts.dtNextContactRequested` is cleared. If the call happens but nothing writes back, suspect the webhook (`VAPI_SERVER_SECRET` signature or `end-of-call-report` type guard) in `worker2/Worker/Vapi.php`.
+
+---
+_Saved by /session-save on 2026-07-20_

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "toga-ai",
-  "version": "1.0.384",
+  "version": "1.0.385",
   "description": "TOGA Technology Team Claude Knowledge System — shared AI coding harness with skills, knowledge base CLI, and project installer for Claude Code.",
   "keywords": [
     "claude",