# Issue — Implement `should_trigger_reasoning()` Phase 4 gate

[Devnil434]

Description

Before triggering the VLM/LLM reasoning layer, the pipeline must determine
whether a track is important enough to reason about.

Without this gate, the VLM would run on every frame and significantly slow
down the pipeline.

Implement:

should_trigger_reasoning(seq: TrackSequence) -> bool

inside:

services/memory/trigger.py

---

Requirements

Return True only when ALL conditions are satisfied:

1. Track is inside a restricted zone

2. Dwell time exceeds configured threshold

3. At least one suspicious action exists:

   • LINGERING
   • NEAR_KEYPAD
   • REPEATED_APPROACH

4. Track is not inside cooldown window

---

Additional Function

Also implement:

reset_cooldown(track_id: int)

to clear cooldown state after reasoning completes.

---

Constraints

• Pure in-memory logic only
• No Redis calls
• Cooldown duration: 5 seconds
• Thresholds should come from centralized settings

---

Acceptance Criteria

• Returns False below dwell threshold
• Returns False during cooldown
• Returns True for suspicious sequences
• Add 6 unit tests covering all branches
• Existing tests remain green

---

Helpful Context

Relevant modules:

services/memory/trigger.py
services/memory/pipeline.py
libs/schemas/memory.py

[saurabhhhcodes]

Hi maintainer, I would like to work on this under GSSoC. My plan is to implement the Phase 4 should_trigger_reasoning gate as pure in-memory logic, read dwell/cooldown thresholds from central settings, add reset_cooldown(track_id), and cover below-threshold, cooldown, zone, action, and positive-trigger branches with unit tests. Please assign this to me.

[dilshadalikhan2004]

Hi @Devnil434 , I would like to work on this issue under GSSoC’26.

I have gone through the requirements and relevant modules (trigger.py, pipeline.py, and memory.py). My plan is to implement the Phase 4 should_trigger_reasoning() gate using lightweight pure in-memory logic to avoid unnecessary VLM/LLM inference overhead.

Implementation plan:

-validate restricted-zone presence

• check dwell-time threshold from centralized settings
• verify suspicious actions (LINGERING, NEAR_KEYPAD, REPEATED_APPROACH)
• enforce the 5-second cooldown window using in-memory state only
• implement reset_cooldown(track_id: int) for cooldown cleanup after reasoning completes

I will also add comprehensive unit tests covering:

• below-threshold dwell rejection
• cooldown rejection
• restricted-zone validation
• suspicious action filtering
• successful trigger conditions
• cooldown reset behavior

I will ensure the implementation remains clean, deterministic, lightweight, and consistent with the existing codebase while keeping all current tests green.

Please assign this issue to me.

[SARTHAKJINDAL1]

hi @Devnil434 ,I would like to work on this under GSSoC.