# Record Validation Rules One of the key metrics tracked by Telmai is the “**Correctness**” metric. It is defined as the percentage of records which are valid based on a corresponding rule or combination of rules. Correctness is - For example, for a table with 1M rows; if 50,000 records are invalid, correctness is then 95%. The row validity checks (data expectations) are defined by users via **Correctness Rules.** ## Correctness Rules (Record Validation) Correctness Rules are evaluated at row level. A single rule may validate one or more attribute values (plain or nested object) against a condition. Rules are created following SQL-like syntax that uses variables and functions to perform the validation. The main key features for rules: * Eases the process of applying the same rule against multiple tables through templating * Rules DSL (domain specific language) allows for version control and management in a simple human-readable format * Optimized algorithms for unlimited number of rules for faster and more efficient processing ### Building a Correctness Rule Creating a correctness rule requires performing the following steps: 1. Define variables that will be used in the rule * Variables can be assigned to static lists of values, range, entered manually or uploaded, from files; etc. 2. Map variables to attributes used in the expression 3. Write rule expression #### Define and Map Variables Users can define variables to be used in their rule. Supported variable types: * Array of strings * Constant string value * Range (date-time or integer) * Reference to an attribute Example: We can define the following two variables (`accepted_country_list, var_1)`: * `accepted_country_list = [USA, Canada]` * `var_1 = country_name` User can then write a rule like following **`VALIDATE`**` `` ``var_1 ` **`EXPECT`**` ``value`` `**`IN`**` `` ``accepted_country_list ` This rule checks if a value from the attribute *country\_name* is in the list of accepted values. First it defines an expression *var\_1*, which is a trivial expression, which doesn’t add any transformation to the original value. And after that it sets an expectation that the value of the expression should be in the list to which the previously defined variable is assigned. #### Writing Rule Expression A correctness rule consists of three parts:
WITH <transformation>

Variable transformation:

Optional section. It allows the user to create a transformation of one or more attributes to be validated and reused within the expression multiple times

Example:

WITH concat(country, '|', state) AS loc

This creates a new variable “loc” that can be used in validation

VALIDATE

<evaluated_expression>

Validated expression:

This could be simply validating an attribute as is, or adding transformation or case clause.

Example:

VALIDATE

case

when loc = 'US|CA' then country IN ('Alameda', 'Contra-Costa')

when loc = 'UAE|Dubai' then country IS NULL

else false //default output

End

EXPECT

<expectation>

Expected results

Expectation for the expression above. After the “EXPECT” keyword, the validation begins.

Example:

EXPECT is_true

this means the value of validated expression is expected to be true, and anything else is a violation)

### How to get alerted for rule violation? Once a rule is created, users can assign the rule to a policy and specify the threshold for alerting. ## Variables transformation This step allows defining variables from combining existing attributes to be used at a later step of validation. This can simplify writing the evaluated expression especially in case when multiple checks are applied. To define a transformation, you can follow the following syntax: `WITH AS ` Telmai uses standard Spark SQL syntax for those transformations. ## Evaluated expression This is a Spark SQL expression. It allows users to define what will be evaluated, and exposes a wide range of SQL functions to allow for conditional checks. ### Scenario 1: Validate attribute value as is, without transformations ``` VALIDATE my_column// Some code ``` This validates “`my_column`” without any transformation ### Scenario 2: Apply standard functions to attribute value ```sql VALIDATE foo(my_column) ``` This validates “`my_column`” after applying the “`foo`” function to it. ### Scenario 3: Cases ```sql VALIDATE case when my_colA <> 0 then my_colB else my_colC end ``` This validates the rule by using `my_colB` when `my_colA` is not 0; otherwise, it uses `my_colC` ## Expectation Expectation is defined through the context and the expected value. ``` EXPECT [] ``` ### Expectation context The expectation context (``) is how the value gets evaluated. It can be any of the following:
ContextDescription
COMPRESSED_PATTERNCompressed pattern of the expression value
DATE_TIME_VALUEExpression value as Datetime. Refer here for datetime patterns
EXPANDED_PATTERNExpanded pattern of the expression value
FREQUENCYFrequency of the expression value
LENGTHString length of the expression value
NUMERIC_VALUEExpression value as number
SPACE_COUNTCount of spaces in the expression value
SPEC_CHAR_COUNTSpec characters count of the expression value
VALUEString representation of the expression value
### Expectation operator The expectation operator (``) is an optional field. It’s the condition operator for the context and the expected value. It can be from the following list: | **Operator** | **Description** | | -------------- | ----------------------------------------------------------------------------------- | | `IN` |

Value is in the list.

Expected value must be in the list.

| | `IN_RANGE` |

Value in range including boundaries.
Expected value must be in the range.

| | `NOT` | Boolean negation operator | | `NOT_IN` |

Value is not in the list

Expected value must not be in the list.

| | `OUT_OF_RANGE` |

Value out of range

Expected value must be outside of the range.

| ### Expected value Expected value (\) can be set using a static value or the boolean-functions belo | **Static value** | **Example** | | ---------------- | ----------------- | | List | `["a", "b", "c"]` | | Range | `(0,100)` | | String | `"abcd"` | | Variable | `var_1` |
Boolean functionDescription
CONTAINS_PIITrue if expression contains PII data
CONTAINS_PII_CREDIT_CARDTrue if expression contains credit card number pattern
CONTAINS_PII_IP_ADDRESSTrue if expression contains IP address
CONTAINS_PII_PHONE_NUMBERTrue if expression contains phone number pattern
CONTAINS_PII_SSNTrue if expression contains social security number pattern
IS_ALPHAValue is only from alphabetic characters
IS_DATEValue is UTC date pattern
IS_DATE_TIMEValue is UTC date-time pattern
IS_EMAILValue is email
IS_FALSEValue is false (case insensitive string comparison, and not boolean)
IS_NUMBERValue is numeric value
IS_TRIMMED_STRINGValue is a trimmed string
IS_TRUEValue is true (case insensitive string comparison, and not boolean)
REGEX(pattern)Value satisfy regex pattern
Examples: ```sql VALIDATE Var_1 expect is_date # Check if Var_1 is a valid date. VALIDATE Var_1 expect contains_pii # Check if Var_1 contains Personally Identifiable Information (PII) VALIDATE Var_1 expect numeric_value in_range (1, 20) # Ensure Var_1 is a number within range 1 to 20 VALIDATE Var_1 expect numeric_value out_of_range (50, 60) # Ensure Var_1’s numeric value is outside a length range VALIDATE Var_1 expect expanded_pattern IN ['D', 'DD'] # Validate Var_1 using an expanded pattern VALIDATE Var_1 expect compressed_pattern in ['D', 'L'] # Validate Var_1 using a compressed pattern VALIDATE Var_1 expect value IN ['USA', 'Germany', 'India'] # Check if Var_1 value is in the accepted country list VALIDATE Var_1 expect value='Telm' # Ensure Var_1 exactly matches the value Telm VALIDATE concat(Var_1, '.', Var_2) expect value='Telm.ai' # Validate full name format as Telm.ai, where Var_1=Telm and Var_2=ai VALIDATE Var_1 expect length<>5 # Ensure Var_1 value is of length not equal to 5 VALIDATE Var_1 expect frequency=1 # Ensure Var_1 has exactly one occurrence VALIDATE Var_1 expect regex('^[a-zA-Z]+$') # Validate Var_1 contains only letters (no numbers/special characters) ``` ### Combining Expectations User is able to combine multiple expectations using AND/OR operator: ``` EXPECT AND OR ``` Examples: ```robotframework VALIDATE case when Status='Active' and LastLogin < current_date - 30 then 'Inactive' else 'Valid' end expect value='Valid' # Mark users as inactive if they haven't logged in for 30+ days, otherwise expect 'Valid' with concat(Country1, ' | ', Country2) AS loc VALIDATE loc expect length<20 # Ensure concatenated location (Country1 | Country2) has a length less than 20 VALIDATE Password expect spec_char_count>=1 AND space_count=0 AND length>=8 # Ensure Password has at least 1 special character, no spaces, and is at least 8 characters long VALIDATE Password expect (spec_char_count>=1 AND space_count=0) OR length>=20 # Ensure Password has at least 1 special character and no spaces, or is at least 20 characters long ``` ## Rule Templates Rule Templates is a predefined rule where the expectation is defined, then applied to 1 or more attributes. Users are recommended to use rule templates when the same check is applied on multiple tables. This centralizes the definition of the rule, and allows managing the rule from a single place. To use a rule template, the user only maps attributes within their table to variables within the rules. ## Defining a Rule The following flow diagram describes the steps to create a new rule or rule template: 1. Define rule info: * Rule name * Rule description * \[Optional] Template being used (this is only if creating a rule) 2. Set static variables: * These are variables that are validated against, and are not dynamic in nature; ex: list of allowed countries 3. Set input variables: * These are variables that are mapped to attributes in the table * In case this is a rule, user will need to map the attributes to variables 4. Write expression

Creating a Rule

Note; In case user is using an existing template, only input variables need to be mapped to variables in the table ## Creating an Expression-Based Rule 1. Navigate to **Alerting Monitors** and click **New Monitor** 2. Choose **Record Validation Rule** as the monitor type 3. Enter a monitor name and optional description 4. Write your rule expression following the [steps](#creating-a-rule) above 5. Validate and save the rule {% hint style="info" %} Threshold will be applied on each rule individually If policy is used in DataBinning, all rules are checked when binning the data {% endhint %}