FAQ: common issues during training

Self-debug reference for the training track. If the scenario is correct but platform behavior is inconsistent — contact ConnectiveOne support.

Fast Line Pro

AI does not reply or {{ai_answer}} is empty

AI replies incorrectly / hallucinations

• Check the FAQ document content (short txt for training)
• Add a fallback edge and a message like “Please rephrase the question”

Branching (Router / switch and edges)

Always takes one branch

• Runs → value of the variable in the condition (e.g. {{user_choice}} is empty)
• Make sure the MessageKeyboard button writes a value into the variable
• Comparison type: string 'priority' vs number 1 vs boolean true
• In Router, enable default branch for remaining cases
• In MessageKeyboard, connect a default edge for free text

Found legacy if_else in an old scenario

• Logic is the same as Router: conditions + else
• For new training flows use Router or switch

Router / switch “does not work” after registered_users__get

• In training, Use Case 2 does not use registered_users — start with a simple menu via edges or Router on user_choice
• If you still use __get: edge error = customer not found — that is an expected branch, not a bug

→ Conditional branches (Router) · Built-in actions — switch

Handoff to an operator from the scenario (Use Case 3)

Edge limit does not fire / customer is in queue but silence

• In connect parameters: status: "queue" (if needed)
• After edge limit there must be a MessageKeyboard with text for the customer
• Check Runs: on which edge execution stopped

Connect returns error

• subject_alias does not match the subject alias in Settings → Operator Panel → Subjects
• Subject not created or alias has a typo (case, space)

Connect success, but the dialog is not where you expected

1. First compare subject_alias in the scenario with the subject in Settings
2. Then — Operator Panel documentation (routing, operators)

CSAT / NPS (Use Case 7b)

CSAT does not start after dialog close

• Check chat_closed_alias in all connect blocks — it must point to the CSAT section alias
• No duplicated alias entrypoint in two sections
• CSAT section exists, alias in Latin characters

Error writing to Custom Data / INTEGER

• Field rate must be type INTEGER in the CD model
• Free-text on the keyboard went into __set — protect the source-handle (disconnect or Router / validation)
• In Runs: send text instead of a button — expect edge error, not success on write

Score is not collected in email / a channel without keyboard

• Check supportedMessengers on the keyboard node
• Add a separate branch or skip CSAT for channels without buttons

NPS / score not after close in widget

• Check alias chat_closed / call_node path in the connect block
• Audit the alias in Runs after simulating close

Operator Panel / queue (reference)

Operator does not see the dialog

• First: subject_alias in the scenario vs subject in Settings
• Then: operator linked to the subject / skill group — OP documentation
• If the scenario and Settings are correct — ConnectiveOne support

Action Jail / phone (Use Case 5)

Three UA phone formats

If one format fails — expand the regex / Action Jail; check validation on WaitForInput (phone vs none).

Validation on the node vs Action Jail

• Validation on input — rejects invalid input before the action
• Action Jail — normalizes allowed variants to one format
• For production you often need both

CRM / mock API (Use Case 4)

Which CRM to use for training?

Mock. See Mock vs production API. A real CRM — after the basic level (Use Case 1…9).

401 / empty response

• Token in headers via {{instance_settings.…}}
• Instance Settings key exists and is filled
• save_responce (or response_mapping) is set on the node — check the variable name in Runs

In legacy if_else always goes to the “not found” branch

• Compare status: string "200" vs number 200 — check the actual type in Runs
• Check that save_responce wrote the expected fields into {{crm_data.*}}

Chatbot vs Agent in Fast Line Pro

Tickets (Use Case 10)

Where to create the categorizer — main bot or Copilot?

• AI catalog (optional): CD lookup in the same bot — see AI catalog
• Use Case 12: separate Copilot — overview

Ticket is not created / error edge

• subject_alias does not match the subject in Settings
• Check action_tickets_create vs action_tickets_create_v2 in Inspector
• Runs → constants after create: ticket_id or equivalent

“Sticky” chat_room_id / repeated create

• Check create-action parameters: whether an old room is reused without logic
• New Run vs continuing the same dialog — different behavior
• In Runs: which chat_room_id is in constants before and after create

Ticket custom field not visible in tabular OP

• Subject fields in Settings → Tickets
• Does not block the Use Case 10 happy path in Runs (ticket_id is enough)

Broadcast (Use Case 11)

Link button — Action Jail or MessageKeyboard?

• First: native URL in MessageKeyboard (open-url) — see Use Case 11
• Action Jail — only if native does not cover the case

Telegram OK, Viber error / not delivered

• Check open-url vs open-web-app
• Text-only vs image+button — simplify for diagnostics
• Document the result in the result template (Viber attempt)

send_after / one_time

• send_after: create success in Runs ≠ instant delivery in the channel — wait or check the delay
• one_time: a repeated trigger must not duplicate the message
• Do not confuse with action_broadcast__subscribe (topic for a series of broadcasts)

Why action_broadcast__subscribe?

• User subscription to a topic for future broadcasts, not a one-off send from the “Clients” tab
• One-off broadcast in Use Case 11 — via Broadcast template + alias

→ Broadcast subscribe

Priority segment (optional)

Different branches do not fire

• Node order: first check the segment, then subject_alias
• Variable type: '1' (string) vs 1 (number) vs true (boolean) — see Runs
• Mock API / Registered Users / CD — align the segment flag source

→ Priority segment

Copilot / process bot (optional)

Where does the catalog categorizer live (Use Case 12)?

• In the Copilot bot (a separate bot or process sections)
• The main bot only routes (“Catalog” button)

Copilot vs macro Use Case 8

Deeplink

Deeplink + registered_users — do you need get_command + jail?

• No on the first pass: deeplink params in constants on Start → registered_users__get → branches
• Details — in Use Case 6

Operator Panel / filters

“In queue” vs “Unassigned”

• In queue: dialogs waiting for an operator in the queue
• Unassigned: without an assignee; they may overlap — check Settings → Operator Panel and the Runs connect block

Advanced Fast Line Pro (optional)

• fastline_completion, streaming — optional Advanced Fast Line Pro

Registered Users

Data is not stored “forever”

• Only state was used — you need registered_users__set
• Fields were not added to the Registered Users model in Settings

How to test across sessions?

1. Finish the Run after registration
2. New Run → registered_users__get at start
3. Optionally — the same user via Telegram

→ Use Case 6 — registration and return visit

Macros

Operator does not see “Run macro”

1. Section is marked “Used as a macro”
2. Settings → Operator Panel → Process macro → bot with macros selected
3. Dialog is open in Operator Line (not only scenario preview)
4. Macro variables are declared in the section parameters

→ Integrator overview — macros

Channels

Telegram does not receive messages

• Settings → Bots → Channels → token is valid
• You tested preview but expected Telegram — those are different paths
• For Use Case 1 the channel is optional; preview is enough for logic

Keyboard is not shown

• supportedMessengers on the keyboard node
• The channel supports inline buttons

Different behavior widget vs Telegram

• Separate branches or validation by {{channel}}
• Test critical flows in both channels if the bot is multichannel

URL in the message as text, not a link

• Message mode (HTML / plain) in MessageKeyboard
• For widget — check link display parameters in Inspector

→ Connect Telegram

AI Assistant (Instance Agent) {#ai-assistant-instance-agent}

This is not Fast Line Pro in the bot — see concepts — Instance Agent.

Training rule

1. UC first by hand → self-check in Runs.
2. Optionally after — rebuild the same UC via Agent and compare edges / alias / variables.
3. Trust Runs, not the agent’s answer.

Agent built a scenario, but it does not work

• Refresh Scenario Builder (F5 / Cmd+R) after agent edits
• Check Runs — edges, variables, which edge actually fired
• Fix by hand; Agent is for explanations and optional rebuilds

Agent answer contradicts Runs

• Trust Runs and the test in the channel / OP
• Ask the agent to explain a specific node by title, then verify in Inspector

No “spark” button in Scenario Builder

• Prerequisites — Instance Agent Integration + AI Assistant module in the role
• → Configure access

Read-only or no write permission

• The agent only explains and suggests a UI tour — you make the changes yourself
• Ask mode is enough for most training UCs

General debug {#general-debug}

Details: Use Case 9 — tooling tour. Outside the scenario: when it is not the scenario.

If after self-debug the scenario is correct but platform behavior is inconsistent — ConnectiveOne support.

Related materials

• Actions glossary
• Testing matrix
• When it is not the scenario
• Key concepts