As part of a project I’m working on – I came across an interesting phenomena while using bq console.

consider given SQL:

WITH
  users AS (
  SELECT
    *
  FROM
    my-gcp-proj.my_dataset.users
  WHERE
    MOD(id, 97) != 0),
  teams_events AS (
  SELECT
    *
  FROM
    my-gcp-proj.my_dataset.events
    -- my_dataset.events
  WHERE
    team_id IS NOT NULL),
  helper1 AS (
  SELECT
    users.id AS user_id,
    timestamp,
    event_name,
    event_name = 'Cartoon' AS cartoon
  FROM
    users
  INNER JOIN
    my_dataset.teams
  ON
    teams.id = users.team_id
  LEFT JOIN
    teams_events
  ON
    teams_events.team_id = teams.id
    AND timestamp >= '2023-06-01 00:00:00'
    AND timestamp < '2023-11-01 00:00:00' ),
  helper2 AS (
  SELECT
    user_id AS entity_id,
    timestamp,
    event_name,
    event_name = 'activation' AS activation,
    NULL AS cartoon
  FROM
    my_dataset.events
  WHERE
    timestamp >= '2023-06-01 00:00:00'
    AND timestamp < '2023-11-01 00:00:00'
  UNION ALL
  SELECT
    user_id AS entity_id,
    timestamp,
    event_name,
    NULL AS activation,
    cartoon
  FROM
    helper1
  WHERE
    timestamp >= '2023-06-01 00:00:00'
    AND timestamp < '2023-11-01 00:00:00' ),
  helper3 AS (
  SELECT
    users.id AS entity_id,
    SUM(CASE
        WHEN activation OR cartoon THEN 1
        ELSE 0
    END
      ) AS custom_signal
  FROM
    users
  LEFT JOIN
    helper2
  ON
    users.id = helper2.entity_id
  GROUP BY
    1 )
SELECT
  *
FROM
  helper3

The amount of bytes will be processed is 128.07 MB

But, If I switch the table name being used in line 13, the amount of bytes will be processed changes to 120.2 MB.

Mind that the tables being used are equal!

One variation is using the gcp project as prefix, the other does not.

The 2 SQLs return same output.

Can someone explain how this is possible?
and perhaps what Can I do to avoid such difference for next time?

(I do understand this SQL vague – but just wanted to supply the most accurate example I can)