Why is my SQL Server query behaving differently on UPDATE than on SELECT?

I’ve written a SQL Server query that updates records to have a sequential number after partitioning on a field. When I run it as a SELECT statement, everything looks great:

DECLARE @RunDetailID INT = 448
DECLARE @JobDetailID INT

SELECT @JobDetailID = [JobDetailID] FROM [RunDetails] WHERE [RunDetailID] = @RunDetailID

SELECT
    [OrderedRecords].[NewSeq9],
    RIGHT([OrderedRecords].[NewSeq9], 4)
FROM
    (
        SELECT
            [Records].*,
            [Records].[SortField] + RIGHT('0000' + CAST(ROW_NUMBER() OVER(PARTITION BY [Records].[SortField] ORDER BY [Records].[RunDetailID], [Records].[SortField], [Records].[PieceID]) AS VARCHAR), 4) NewSeq9
        FROM
        (
            SELECT
                [MRDFStorageID], 
                [RunDetailID], 
                [SortField], 
                [PieceID], 
                [Seq9], 
                [BallotType]
            FROM
                [MRDFStorage]
                    JOIN [BallotStyles] ON [MRDFStorage].[SortField] = [BallotStyles].[Style] and [BallotStyles].[JobDetailID] = @JobDetailID
            WHERE
                [RunDetailID] IN (SELECT [RunDetailID] FROM [RunDetails] WHERE [JobDetailID] = @JobDetailID AND [RunStatusID] <> 0)
        ) Records
    ) OrderedRecords 
        JOIN MRDFStorage ON [OrderedRecords].[MRDFStorageID] = [MRDFStorage].[MRDFStorageID]
WHERE
    [MRDFStorage].[RunDetailID] = @RunDetailID

However, when I make the query into an UPDATE command, it begins skipping even numbers:

DECLARE @RunDetailID INT = 448
DECLARE @JobDetailID INT 

SELECT @JobDetailID = [JobDetailID] FROM [RunDetails] WHERE [RunDetailID] = @RunDetailID

UPDATE
    [MRDFStorage]
SET
    [Seq9] = [OrderedRecords].[NewSeq9],
    [Overlay1] = [OrderedRecords].[NewSeq9],
    [Overlay10] = RIGHT([OrderedRecords].[NewSeq9], 4)
FROM
    (
        SELECT
            [Records].*,
            [Records].[SortField] + RIGHT('0000' + CAST(ROW_NUMBER() OVER(PARTITION BY [Records].[SortField] ORDER BY [Records].[RunDetailID], [Records].[SortField], [Records].[PieceID]) AS VARCHAR), 4) NewSeq9
        FROM
        (
            SELECT
                [MRDFStorageID], 
                [RunDetailID], 
                [SortField], 
                [PieceID], 
                [Seq9], 
                [BallotType], 
                CAST([SpecialProcessing] as Int) StartCount
            FROM
                [MRDFStorage]
                    JOIN [BallotStyles] ON [MRDFStorage].[SortField] = [BallotStyles].[Style] and [BallotStyles].[JobDetailID] = @JobDetailID
            WHERE
                [RunDetailID] IN (SELECT [RunDetailID] FROM [RunDetails] WHERE [JobDetailID] = @JobDetailID AND [RunStatusID] <> 0)
        ) Records
    ) OrderedRecords 
        JOIN MRDFStorage ON [OrderedRecords].[MRDFStorageID] = [MRDFStorage].[MRDFStorageID]
WHERE
    [MRDFStorage].[RunDetailID] = @RunDetailID

I’ve tried specifically focusing on this portion:

[Records].[SortField] + RIGHT('0000' + CAST(ROW_NUMBER() OVER(PARTITION BY [Records].[SortField] ORDER BY [Records].[RunDetailID], [Records].[SortField], [Records].[PieceID]) AS VARCHAR), 4) NewSeq9

Is there some side effect I’m ignorant of?

UPDATE WITH TABLE DEFINITIONS

CREATE TABLE [dbo].[MRDFStorage] (
    [MRDFStorageID]           INT            IDENTITY (1, 1) NOT NULL,
    [RunDetailID]             INT            NOT NULL,
    [PieceID]                 VARCHAR (15)   NULL,
    [SortField]               VARCHAR (20)   NULL,
    [BallotType]              VARCHAR (100)  NULL,
    [Seq9]                    VARCHAR (15)   NULL,
    CONSTRAINT [PK_MRDFStorage] PRIMARY KEY CLUSTERED ([MRDFStorageID] ASC),
    CONSTRAINT [FK_MRDFStorage_RunDetails] FOREIGN KEY ([RunDetailID]) REFERENCES [dbo].[RunDetails] ([RunDetailID])
);

CREATE TABLE [dbo].[BallotStyles] (
    [BallotStyleID]     INT           IDENTITY (1, 1) NOT NULL,
    [JobDetailID]       INT           NOT NULL,
    [Style]             VARCHAR (20)  NOT NULL,
    CONSTRAINT [PK_BallotStyles] PRIMARY KEY CLUSTERED ([BallotStyleID] ASC)
);

CREATE TABLE [dbo].[RunDetails] (
    [RunDetailID]        INT            IDENTITY (1, 1) NOT NULL,
    [JobDetailID]        INT            NOT NULL,
    CONSTRAINT [PK_RunDetails] PRIMARY KEY CLUSTERED ([RunDetailID] ASC)
);

Not seeing the data you have makes this a bit harder but I managed to reproduce what you see if you see with this:

insert into RunDetails(RunDetailID, JobDetailID) values(448, 1)
insert into MRDFStorage(RunDetailID, SortField) values(448, 'S1')
insert into MRDFStorage(RunDetailID, SortField) values(448, 'S1')
insert into MRDFStorage(RunDetailID, SortField) values(448, 'S1')
insert into BallotStyles(JobDetailID, Style) values(1, 'S1')
insert into BallotStyles(JobDetailID, Style) values(1, 'S1')

The key factor here is the two rows in BallotStyles.

You are joining to BallotStyles in the inner most query and with data like above you will get duplicate rows for each row in MRDFStorage. Since you are not using any of the columns in BallotStyles you are only checking for existence of rows and that can be done with a exists clause instead and that will of course not create the duplicate rows.

UPDATE
    [MRDFStorage]
SET
    [Seq9] = [OrderedRecords].[NewSeq9]
FROM
    (
        SELECT
            MRDFStorageID,
            [Records].[SortField] + RIGHT('0000' + CAST(ROW_NUMBER() OVER(PARTITION BY [Records].[SortField] ORDER BY [Records].[RunDetailID], [Records].[SortField], [Records].[PieceID]) AS VARCHAR), 4) NewSeq9
        FROM
        (
            SELECT
                M.[MRDFStorageID], 
                M.[RunDetailID], 
                M.[SortField], 
                M.[PieceID], 
                M.[BallotType]
            FROM
                [MRDFStorage] as M
-- Remove this join
--                    JOIN [BallotStyles] ON M.[SortField] = [BallotStyles].[Style] and [BallotStyles].[JobDetailID] = @JobDetailID
            WHERE
                [RunDetailID] IN (SELECT [RunDetailID] FROM [RunDetails] WHERE [JobDetailID] = @JobDetailID) and
-- Add this exists check instead of the join
                EXISTS (
                       SELECT *
                       FROM BallotStyles AS BS
                       WHERE M.SortField = BS.Style and
                             BS.JobDetailID = @JobDetailID
                       )
        ) Records
    ) OrderedRecords 
        JOIN MRDFStorage ON [OrderedRecords].[MRDFStorageID] = [MRDFStorage].[MRDFStorageID]
WHERE
    [MRDFStorage].[RunDetailID] = @RunDetailID