Org Web Adapter

Editing hungryroot/brand/customer-dislikes-survey.org

:PROPERTIES:
:ID:       d604a732-3d17-43ab-b564-b9fcb7d65f8e
:END:
#+title: Customer Dislikes — Cuisine, Dish Type, Ingredients, Flavors

* TL;DR

Customers can express dislikes across all four dimensions, but the
mechanism differs by dimension. *Only ingredient dislikes are a hard
filter.* Cuisine, dish type, and flavor dislikes are soft signals that
nudge recommendations rather than excluding products.

| Dimension  | Hard or Soft   | Storage                                  |
|------------+----------------+------------------------------------------|
| Cuisine    | Soft           | =Tag= + =CustomerTag= (TagGroup CUISINE) |
| Dish type  | Soft           | =DishPreference= (signed =rating=)        |
| Ingredient | *Hard*         | =Periodicity= (=NEVER= excludes)            |
| Flavor     | Soft           | =Tag= + =CustomerTag= (multiple groups)  |

* Cuisine

- *Models:* =Tag= + =CustomerTag=
- *Files:* =app/models/tag.py:117-120=
- *Tag group:* =TagGroup.ID.CUISINE= (id =15=)
- *Dislike field:* =CustomerTag.preference= (=SmallIntegerField=, negative
  = dislike); =Tag.dislike_score= (=IntegerField=) used in recommendation
  scoring.
- *Hard or soft:* *Soft.* Feeds pairing recommendation and product
  filtering scoring; does not exclude.
- *API:*
  - =GET/POST /customers/{customer_id}/tags/= → =CustomerTagList=
  - =GET/PUT /customers/{customer_id}/tags/{tag_group_id}/= for
    cuisine-specific operations
  - Serializer: =CustomerTagListSerializer= in
    =app/rest/customer_tag.py=

* Dish Type

- *Model:* =DishPreference=
- *File:* =app/models/dish_preference.py:7-10=
- *Relationship:* =ForeignKey= to =DishType= (line 9)
- *Dislike field:* =rating= (=SmallIntegerField=, default =0=). Negative
  value = dislike, positive = like.
- *Hard or soft:* *Soft.* Used for dish-type personalization in
  survey/preference flows.
- *API:* No dedicated REST endpoint. Surfaced indirectly via the
  preference/survey endpoints.

* Ingredient — Hard Exclusion

- *Model:* =Periodicity=
- *File:* =app/models/periodicity.py:7-9=
- *Relationship:* =ForeignKey= to =Product= (which carries ingredients)
- *Field:* =periodicity= (=PositiveSmallIntegerField=) with choices from
  =PeriodicityConsts=: =NEVER=, =SOMETIMES=, =OFTEN=.
- *Hard or soft:* *HARD EXCLUSION.* When =periodicity = NEVER=, products
  are filtered out entirely via =customer.get_never_products()=, which
  is consumed by pairing/product query paths.
- *API:*
  - =GET/POST /customers/{customer_id}/periodicities/= →
    =PeriodicityListCreateAPIView= in
    =app/rest/customer_periodicity.py=

* Flavor

- *Models:* =Tag= + =CustomerTag= (same machinery as Cuisine)
- *File:* =app/models/tag.py:117-130=
- *Tag groups:*
  - =SPICE_LEVEL= (id =6=)
  - =FLAVOR_PROFILE= (id =11=)
  - =FLAVOUR= (id =22=) — e.g. tags =HOT= (22), =MILD= (20), =MEDIUM= (21)
- *Dislike fields:* =CustomerTag.preference= (negative);
  =Tag.dislike_score=, =Tag.pairing_dislike_score=.
- *Hard or soft:* *Soft.* Scoring-based for recommendations.
- *API:* Same surface as Cuisine — =/customers/{customer_id}/tags/= and
  =/customers/{customer_id}/tags/{tag_group_id}/=.

* Related: Post-Purchase Feedback (not a preference)

- *Model:* =CustomerFoodFeedback=
- *File:* =app/models/customer_food_feedback.py:15-26=
- Captures feedback tags (taste, ingredients, nutrition) *after* the
  customer consumes the item. The =preference= field can be negative
  (dislike), but this is a feedback mechanism, not a pre-purchase
  preference declaration. Worth distinguishing from the four
  preference dimensions above when reasoning about whether a dislike
  filters out a product before it's recommended.

* Open Questions / Follow-ups

- For the soft dimensions (cuisine, dish type, flavor), what are the
  current *scoring weights* and where do they live? Need to trace into
  pairing recommendation scoring to know how strong a "dislike" signal
  actually is in practice.
- Does =CustomerFoodFeedback= feed back into the soft dislike signals
  on subsequent recommendations, or is it purely operational/CX data?
- For UX: only ingredient dislikes hard-block products. If a customer
  expects "I disliked Italian" to *exclude* Italian dishes, that
  expectation is not met today. Worth flagging.