## TL;DR
Map every voicemail recording to the exact sequence and step it belongs to, and make the drop step look up the recording by the lead's actual current step - not by a default. Test the mapping in staging before any campaign, because the wrong script to the wrong lead is worse than no voicemail at all.

```
agent's voicemail drop played the wrong recording  -  the first-touch script went to a lead on step 5
```

## Steps

1. Build a recording map: each sequence step points to exactly one approved audio file, with the file name, duration, and the script text it reads.
   Expected: the map covers every step that uses voicemail drops, and every entry names a real file.

2. Change the drop step to resolve the lead's current sequence step first, then pick the recording from the map. Remove any fallback that plays a default recording when the lookup fails - a missing mapping should block the drop, not play the wrong one.
   Expected: a test drop for a step-5 lead plays the step-5 recording, and a lead with no mapping gets no drop and a logged warning.

3. Dry-run the full sequence in staging with test leads at different steps, and listen to what actually plays at each step.
   Expected: every step's audio matches the map, verified by listening, not just by file name.

4. Name recordings unambiguously - sequence name, step number, and version in the file name - so a human picking a file cannot grab the wrong one.
   Expected: file names sort by sequence and step, and no two files share a confusingly similar name.

5. For leads who already got the wrong recording, note it on the record so the next human touch can acknowledge the confusion naturally.
   Expected: affected records carry a flag, and the next step's context includes what the lead actually heard.

## Use this when
- voicemail drops play the wrong script for the lead's sequence step
- recordings are picked by default instead of by step lookup
- a campaign's audio was never tested end to end

## Not for this skill when
- the dialer fails to connect or drops calls - that is a telephony problem
- live-call transcription is wrong - see the transcription skill
- the recording audio quality is bad - that is a production issue, not a mapping bug

## Variant phrasings
- voicemail drop played the first-touch script to a late-stage lead
- wrong recording fired on the drop step
- drop step used a default recording instead of the step's one
- audio mismatch between sequence step and voicemail

## Why it happens
Voicemail drops get wired up once - "play this file when voicemail answers" - and the file is whatever was handy during setup. As sequences grow steps, nobody updates the wiring, and the drop step has no idea which step the lead is on. The default recording becomes the recording for everyone.

## Edge cases
- A lead re-enters a sequence at a later step - the drop must use the re-entry step, not the original step 1.
- Recording files get replaced with new versions - the map should point at versioned files so an overwrite does not silently change what plays.
- Some steps should have no voicemail at all - represent that explicitly in the map rather than leaving the step unmapped.
- Compliance: some regions restrict automated voicemail drops - check the rules for the lead's location before enabling drops there.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_oRi_2-SDO7jOrIOIiuR5xw