7. Troubleshooting and Known Limits

Where people get stuck, and an honest account of what the sequencer does not do yet.

Setup problems

"Step 7 says it can't run." The step 6 find-contacts automation is still a draft. Publish it in the automation review UI. Most common blocker in setup, and the error points at step 7, not the cause.

"Step 5 won't score anything." Same problem one step earlier. Publish the step 4 scoring automation.

"It says my ICP files are missing." Steps 3 onward read the knowledge base. Run Build ICP and Build scoring rubric first. See the Setup: Your Knowledge Base article.

"I can't create the automations." Creating automations needs workspace admin. Someone with admin access runs steps 4 and 6.

"I lost my place in the setup." You did not, progress is tracked in a file. Type prepare the workspace for APG in a new chat and it resumes from the first incomplete step.

Scoring problems

"Scores are all clustered in the same range." Your rubric weights are not discriminating. Fix the rubric and re-score rather than adjusting individual accounts. Usually the fix is more weight on fewer dimensions.

"The scores don't match my judgment." Read the score rationale on a few accounts. If it is leaning on a field you do not trust, that field is the problem. If it is reasoning correctly from a premise you disagree with, your ICP is the problem.

"Half my accounts came back unscored." Step 7 surfaces these separately and lets you score them, keep them, or drop them. Nothing is silently cut.

List problems

"My list has poor email coverage." The most common bottleneck in the program. Run a verification pass, drop what you cannot reach, and consider a LinkedIn-only approach for the rest. Never enroll a list you have not verified.

"Verification passed but I'm still bouncing." Catch-all domains accept everything at the server level, so verification cannot confirm a specific address. Flag them and treat them as unverified.

"The agent screened out accounts I would have kept." Your anti-icp.md is over-broad. It also shapes the low end of your scoring bands, so an over-aggressive anti-ICP suppresses accounts twice.

Sequence problems

"My sequence won't publish and it mentions LinkedIn." LinkedIn is not connected. Go to Settings > Sequences > your name and connect it at the bottom. Any sequence with a LinkedIn step is blocked until you do.

"A test run failed saying no sender is available." Usually not your fault, and it happens even with a healthy sender selected. Tell the agent your only sender is a connected account and ask it to set the sequence to connected-account sending. Do not publish a draft just because something suggested it, that is a common wrong guess and does not fix this.

"Sending stopped and nothing looks broken." Check for a paused enrollment first. Emails are quality-checked before they send, and a draft that fails the check pauses the enrollment, which does not resume by itself. Ask the agent to resume it. The usual trigger is an em dash in the copy, which is blocked outright.

"My sequence isn't showing in the Sequences panel." It is a draft. Publish it, and republish after every edit. The first publish leaves it paused, so publishing is not the same as sending.

"I enrolled contacts and nothing sent." You have not hit Activate. It is the only action that starts sending.

"I edited the copy but the old version is still going out." Republish. The agent usually does this itself, but confirm.

"My LinkedIn step failed on contacts I know." You cannot send a connection request to an existing connection. Rebuild that cohort as DM-only and skip the invite step.

"I asked for bold text or a hyperlink and it didn't work." Sequence emails are plain text by design. Paste the raw URL. Plain text also outperforms formatted email on reply rate.

"I want to rename something across all my sequences at once." Bulk edits on published sequences can reset the internal step references and strand contacts mid-flow. Edit them one at a time, or pause first.

"The agent asked me four questions and I don't know the answers." You do not need them. Say which list you are targeting, say you are unsure on the rest, and ask for options. Reacting to three proposals is easier than answering from scratch, and this is the most common place a first sequence gets abandoned.

"The copy sounds like AI." Two fixes, in order: hand the agent the ban list from the Building Sequences article, and drop in two of your own emails as a voice reference. Concrete bans and real examples do far more than "make it sound human."

"Every email in the preview has the same opener with different names." The personalization is not finding real signals. Either your contacts have thin data, or you have not told the agent to research a signal per contact with a fallback. Read five previews across cohorts to see how widespread it is.

"I want to change the gap between two steps." Timing edits go through chat, not a field.

Performance problems

"Replies dropped off and I don't know why." Check whether your sequences finished. Flat pipeline about three weeks after a sequence completed is the most common cause, and the easiest to miss because a finished sequence looks identical to a healthy one.

"Open rate is low." Under 30% is a deliverability problem, not a subject-line one. Check authentication, then verification. Above 30%, ignore open rate.

"Good reply rate, nothing books." Either the ask is too big or you are reaching people who cannot buy. Read the replies: "sounds interesting, but I'd need to loop in X" is a targeting problem; polite non-committal replies usually mean the ask was too heavy.

"Most of my replies are people saying no." Normal, not a problem to fix. About half of all replies are out-of-office, wrong person, or not interested. What matters is the share genuinely interested.

"Bounce rate spiked mid-campaign." Pause the affected inbox, not the whole campaign. Isolate the failing segment, re-verify it, and resume from a healthy inbox. Full procedure in the Sending Safely article.

"LinkedIn stopped letting me send requests." You hit the weekly cap, roughly 100 per rolling seven days whatever tier you pay for. It resets seven days after the request that opened the window. If it happened early, your acceptance rate probably dropped below 30% and LinkedIn tightened your ceiling.

What this is not built for

Not bugs, and there is no setting to find:

• Sequences run in a straight line. No conditional branching.

• Email and LinkedIn only. No SMS, no dialer.

• Step timing and content are edited by talking to the agent, not a field.

• Lists do not refresh themselves as signals fire. Re-run them.

• Not built for a rep sending a thousand emails a week with rotating inboxes.

What it is built for

Reaching a few hundred specific people, each with a message that reflects something real about them. Lightfield screens companies against your own win history without getting sloppy on company 300, then finds the people inside them who match the buyers you have actually closed. Sending less to better-chosen people is the higher-performing strategy, not a compromise.

Still stuck

Ask the agent in chat. It can read your workspace state, which is faster than working through a list. Describe what you expected and what happened instead.