[LucHeart]

Summary

• Some collars sold as the Wellturn T330 are actually a different variant: they read the shock level from the low nibble (0-15) of the intensity byte, and the upper bits carry a toggle bit and a 3-bit rolling counter ([toggle:1][counter:3][level:4]). The stock T330 takes a raw 0-100 value.
• Adds this protocol as a new WellturnT330Alt1 model ("wellturnt330alt1" / "t330alt1"). The stock WellturnT330 encoder is unchanged from develop.
• The toggle bit flips every ~1s and the counter steps every ~2s within a transmission. A gap of more than 200ms between frames starts a new transmission, and the base counter goes up by one. Rolling state is kept per shocker ID.
• Sequence stores the last command and has a new refill(nowMs). RFTransmitter calls it before each payload frame. It does nothing for models without rolling state.
• The encoder takes the current time as a parameter instead of calling millis(), so the protocols component stays free of a clock dependency and the rolling logic can be host-tested without an esp_timer mock.
• Shock maps 0-100 to a 0-15 level; like D80, a non-zero intensity never rounds down to level 0.
• If refill() fails, the sequence is terminated and logged instead of resending the previous frame.
• Vibrate and sound are encoded the same way as on the stock T330.

Limitations

• The new model is only reachable through serial and local JSON commands. The flatbuffers ShockerModelType schema has no value for it (D80 has the same gap), so commands from the hub or backend are dropped until schemas, the backend and the frontend are updated.

Known issues to follow up

• With about 5 or more active sequences, one pass of the transmit loop takes longer than 200ms, so every frame counts as a new transmission.
• A new transmission uses baseCounter + 1, which can repeat a counter value already sent during a long previous press.
• The rolling state is an unsynchronised global std::unordered_map that never removes entries.

Test plan

• Host tests pass (components/protocols, components/common, components/device_control)
• Shock at various intensities on a T330 Alt1 collar and check it responds at the right levels
• Hold a shock for more than 2s and check the collar keeps responding
• Vibrate and sound still work on the Alt1 collar
• The stock wellturnt330 model behaves exactly as it does on develop

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[fresh-bites]

should this have a floor at 1? 1-6 produce level 0 in the final 4-bit field

we might want similar logic as:

Firmware/components/protocols/src/D80Encoder.cpp

Line 26 in 59f45da

if (intensity > 0) intensity = std::max((intensity * 15) / 100, 1);

[LucHeart]

Good catch, fixed in ea68d05. This encoding now lives in its own WellturnT330Alt1 model (the stock T330 is back to develop), and shock uses the same floor as D80: any non-zero intensity is at least level 1. Added test cases for 1, 6 and 14%.

[fresh-bites]

we're ignoring the return value, not sure it matters since it looks a little edge-casey but if it returns false we might want to log an error or skip the transmit for that round, vs now i think the prior contents of the payload buffer would be transmitted

[LucHeart]

Fixed in ea68d05. If refill() fails, the transmit loop now logs an error and sets the sequence's end to now, so it goes straight to the terminator instead of resending the previous payload. That matches what modifySequence already does when fill() fails.