35. Out of order

For me, the most important reason to look at a composer’s manuscript—or (more likely) a scan or facsimile of it—is to understand something about how it came to be written.  Compositional process (to use the musicological jargon) is always at least interesting to me, and it can sometimes be riveting.  Often you find that a piece that you know well “as it is” was nearly very different.  (I keep promising I will blog about Mendelssohn’s Italian symphony in this regard; indeed I will, but I just need to find time to re-read John Michael Cooper’s excellent study of it.  But there are very many examples of less extreme but still significant revisions in the standard repertory.)

This month’s post concerns just two instances where a composer decided to insert or re-order material, so that the manuscript presents the music out of sequence with what became the composer’s intended text was.  That’s not to say, of course, that (given time) the composer might not have changed it again, reverting to the original sequence or inserting or reordering new material.  Whatever one might say about the Fassung letzter Hand, its very “lastness” makes it a handy reference point.  For example, in the discussions below I will refer to the bar numbers of the texts as we have come to know them, not as they appear in sequence in the source.

SOURCE:  detail of Mozart’s autograph of K. 491/i f. 2r, 
showing  bb. 40-43 and 63-66.  The autograph is held by the
Royal College of Music, London, as RCM MS 402, but this scan is
from the 2014 Bärenreiter facsimile of the score.  
(A 1964 black-and-white facsimile is freely available on the IMSLP.)

An interesting instance of inserted material occurs in the first movement of Mozart’s gloomy, glorious C minor piano concerto, K. 491.  As seen at right, Mozart originally followed the bar 43 (as we number it) with the bar that later became 63—that is, he made a nineteen-bar insertion at this moment—and a gorgeous one it is, with the duet of descending figures alternating between flute and bassoon.  He thus continues the subdued mood a little longer before the outburst that he originally planned at this moment.

I have mocked-up an audio example of this juxtaposition, although of course we can't know that it would have sounded like this:  Mozart generally orchestrated in layers, adding instruments one at a time, and so he might have scored this a little differently if the passage beginning at b. 63 had really been at b. 44.  But if you want to hear the text with the nineteen-bar cut, it’s on my soundcloud here.

The insertion of a new idea leaves this opening ritornello in a particularly awkward state in the autograph, requiring a jump forward to find b. 44, a jump backward to find b. 63, and a consequent jump forward again to find b. 90.  Moreover, as portions of this are reused again as the closing ritornello (which Mozart indicates with a dal Segno (in this instance, a cartoon head facing back toward the beginning) and other markings), one page of the autograph (f. 3v) contains bb. 54–62, 91–98 (re-used as 501–508), and 99.  But this is an extreme case, and Mozart leaves no doubt about his letzter intended sequence of bars.  (And, so far as I know, no edition has ever screwed it up.) [See Addendum below.]

As I was working on the previous post about bar numbers, my research took me to a page that I would nominate as perhaps the single most interesting page of extant Handel manuscript.  (I may well be wrong:  I have had my eyes on perhaps 5% of Handel’s extant manuscripts—mostly in facsimile or scans—so I can hardly claim to any authority.  Moreover, I would welcome nominations for other contenders for that title.  By all means let me know.)

Anyway, my nominee for that distinction is this page, which has the conclusion of the second movement of the organ concerto published as Op. 7 no. 5 (HWV 310):

SOURCE:  page from Handel’s autograph for HWV 310 (Op. 7 no. 5), mvt. 3; British Library R.M.20.g.12, f. 69v

Handel dated this manuscript, indicating that it was completed on 31 January 1750.  Then again, what does completed mean?  The alterations on this page might well have been made after that date.  The dating is a minor consideration, however.  The real question is What happened here?

This movement is a set of variations over a ground bass.  (Here’s a good recording by Lorenzo Ghielmi with La Divina Armonia.)  While it would not be fair to say that the variations could work as effectively in any order, this movement is clearly highly sectionalized into two bar segments, eighteen in all (with many of them immediately repeated).  Handel’s autograph shows that at some point after finishing the movement with just fifteen segments, he added three more.  He also (at the same time?) re-ordered the intended sequence by numbering each segment.  The first four segments are on the preceding page, so this page starts with the fifth segment.  The following illustration is intended to clarify what the autograph reveals:  the sections shaded in red were (I argue) the original version of the movement, proceeding in their original order (left-to-right, top-to-bottom).  The shaded sections in blue were added subsequently, and the numbering shown incorporates the new variations into a re-ordered sequence.  (The bar numbers indicate the final form, as do the bold variation numbers.)

SOURCE:  my own schematic offering a hypothesis about Handel’s original sequence of the variations as presented in the autography; a recording edited to manifest this sequence (and omitting the blue sections) is available here on my soundcloud page.

There is a logic to Handel’s revised ordering: the first three sections (3-4-5) are melodic, but thereafter there is a series of showy 32nd-note patterns, first arpeggiations in each hand (6), then scales in one hand or the other (7-8-9); then, starting at the mid-point, a stretch of new melodic ideas with more daringly chromatic harmonies (10-11-12); then, by way of a “Scotch snap” [short-long, with the short note in a metrically stronger position] figure (13), there are three sections with triplet figuration—right hand, left hand, both together (14-15-16); the ending of the movement consists of the most brilliant of the 32nd-note arpeggiation patterns (17), and a grand chordal peroration (18).

There is, however, also a logic to Handel’s original version, which is not at all stream-of-consciousness.  (The variation numbers here refer to the final version.)

• melodic introduction (3–4–5)
• triplets (14, 16)
• more chromatic melody (10–11–12)
• “Scotch snap” (13)
• 32nd-note figures (7, 9, 17),
• chordal coda (18)

In the original version the triplet sections come before the central chromaticized melodic passages, and—in a move commonly seen in Baroque variations—the speed of the ornaments increases as it nears the end, with (in this case) a series of three 32nd-note variations.  It is notable that the three (blue) sections that were added later each feature the left hand, which was apparently not emphasized in Handel's original conception of the piece.  Indeed, I think it not too fanciful to suggest that even the (inner-line) left-hand triplets in variation “16” were added as part of the revision:  they seem to my eye squeezed into the staves in an already sufficient texture.  (I note that in Simon Preston’s recording, which I was using for my cut-and-paste version on Soundcloud, he leaves out the left-hand triplets on the first time through this variation, adding it in only for the repeat.)

SOURCE cropped to show detail of the same page given in full above

It is not at all unusual that such reordering and insertions occurred:  this is the way we write.  Having written, we may then proceed to move chunks of prose around hither and yon as it seems better to us—as we have new ideas and second thoughts.  I was surprised, however, to find such clear evidence of Handel’s revisions—and delighted to try it out another way.

ADDENDUM  30 September 2018

After reading my post, one of my mentors from Cornell, David Rosen, mentioned to me the very interesting case of the first movement of Mozart’s last piano concerto, K. 595, about which David had written years ago in The Journal of Musicology.  In that case, Mozart’s insertion of seven bars into the opening ritornello happened much later in the compositional process, in that it doesn’t get notated until the closing ritornello—and, indeed, all editions before the Neue Mozart Ausgabe neglected to insert it after bar 46.  David’s fascinating article argues that the version known before the NMA’s text exhibited a “formal quirk” that did not accord with Mozart’s “standard operating procedure” in concerto first-movements—that procedure as outlined by Robert Levin and Daniel N. Leeson in the Mozart-Jahrbuch 1976/77.  The thesis of David’s article applies just as well to K. 491, too:  that Mozart inserted material to accord with the ways things had worked well before.  In both of these cases (and who knows how many others?) his first thoughts deviated from his well-trodden path, yet in both he eventually settled on something more conventional.

30. Double-crossed?

I spoiled the mood.  I was at a dinner with musicological colleagues and students after a meeting of the Southeast chapter of the American Musicological Society last spring.  It was a lovely time—great food and conversation—but then someone thought to ask a question to the whole table:  What is your favorite opera?

When the turn to answer came around to me, I knew that there would be a universal howl of disapproval for my choice:  Così fan tutte.  There was.  I suppose I could have picked several others as honest answers to the question.  “Favorite” isn’t really a fair word for such a big repertory.  Still, I adore Così even while I dislike it.  And I certainly understand why others are repelled by it or think it unworthy of Mozart’s genius.  It can certainly be played tastelessly (as this perceptive review of the current Seattle Opera production shows).  But I think it can be staged beautifully in a way that doesn’t sugarcoat anything.  I have taught the piece many times, and in class I usually have turned to Nicholas Hytner’s superb (and superb-looking) Glyndebourne production from 2006.  As I regularly tell my students, each time I teach this piece I am seeing it again, where they are (almost always) seeing it for the first time.  Each time it affects me more deeply, and there will be tears streaming down my cheeks while they look on unfazed.  But that's how art works:  the more you invest, the more you reap.  This scene—beautiful as it is—is harrowing for me to watch, as these guys put their girlfriends in an utterly false position just over a cocky bet.  I hate it.  And yet....

Certainly Così is an example of what I would describe as the Disney Happy Ending Problem.  All sorts of terrible, traumatic events happen in children’s movies; no matter how blissfully perfect the finale ultimo appears to be, it never seems to me to compensate for the kidnapping, the guardian’s death, the lonely wanderings (or whatever) of Scene 2.  Granted, by the time we reach the end of Così, I’m not really enjoying it anymore—but I’d say that of just about every opera I know (Idomeneo being an exception in that regard).  Hytner’s production does a good job of leaving the audience—or me, at any rate—with a lingering bad taste.  As Mozart’s C major fanfares bellow in the pit, the four protagonists eye each other nervously.  “Happily ever after,” perhaps, but it is no longer clear who belongs with whom.

SOURCE:  The original couples restored but confused:  cropped screenshot from DVD of 2006 Glyndebourne production (at 2:56:58, during the orchestral conclusion to the Act II finale).  

One of the things that makes the ending so unsatisfactory is that we’ve hardly seen the “right” pairings of these couples, and—more than this—those “right” pairings are so musically wrong:  the opera seria soprano Fiordiligi is not the fiancée of the romantic tenor Ferrando, but rather of the baritone Guglielmo; it is the buffa mezzo Dorabella who is engaged to the tenor.  Our ears—or at least my ears—know that something is wrong with this.

SOURCE:  Boydell & Brewer website

It was thus refreshing to read Ian Woodfield’s fascinating monograph Mozart’s Così fan tutte:  A Compositional History.  I read it when it came out in 2008, but have been eagerly awaiting a chance to re-read it, and teaching the work again this spring while also writing the blog prompted me to find the time to do it—and to have the (relatively inexpensive) facsimile of Mozart's autograph at my elbow.  This is a great convenience, as the autograph itself is split between two libraries (Kraków and Berlin), and the facsimile includes also a scan of the original printed libretto and portions of a Viennese copyist manuscript.  That is only a start, though, as Woodfield scrutinized twenty further copyists’ scores of the work.  “Scrutinized” is a mild word to capture the intensity of Woodfield’s examinations, but the only way to appreciate that is to read the book.  This sort of forensic study is not for everyone, but I would expect anyone who follows this blog would find it at least worth a try.  That said, the conclusions Woodfield draws from this are earth-shattering.  Of these, the most astonishing is that, at some point during their work on the opera, Mozart and da Ponte planned that the couples would not be switched—that the conditions of the bet would be that the guys would be compelled to seduce their own (rather than each other’s) fiancees.

This is, of course, not the work as we know it.  I suspect that it was an attempt to make the far-fetched plot seem a little more reasonable (not that that is a precondition for opera libretti...).  We might be more willing to accept the women’s yielding so quickly to suitors physically like their boyfriends, but that would result in a dramatically much weaker second act.  In the opera as finished Ferrando has to learn from Guglielmo that his side of the bet is lost, and we get two powerful reactions:  Guglielmo’s misogynist rage (“Donne mie, la fate a tanti”) and Ferrando’s conflicted recognition that he loves her still, sadder but wiser (“Tradito, schernito”).  If the pairs were not crossed, Ferrando would not need to be told anything (nor Guglielmo later), nor would the audience.  It would be a recipe for tedium.

I should add that Woodfield does not argue that the un-switched pairs was da Ponte’s original plan, but that instead it was an innovation during the gestation of the work with Mozart.  Intriguingly, the opera was first intended for Antonio Salieri, whose incomplete draft of the first few numbers survives.  (On this issue, see the compelling 1996 Cambridge Opera Journal article by Bruce Alan Brown and John A. Rice.)  The libretto Salieri was working on begins in exactly the same way, commencing with Ferrando praising “his” Dorabella as “incapable” of infidelity:

SOURCE:  cropped screenshot of a page from Antonio Salieri’s attempt [La scola degli amanti], Austrian National Library, shelfmark Mus.Hs.4531, fol. 5.

With Mozart’s(?) idea to revise the plot, these name-pairings remained the same—that is, Dorabella and Ferrando were paired (throughout, even with Ferrando in disguise)—and similarly Fiordiligi and Guglielmo.  Leaving these couples together, though, made it necessary to switch the vocal types, with Dorabella as the high soprano and Fiordiligi as the mezzo.  Evidence supporting this can be seen where these vocal parts have been reversed in the autograph (as in the example below in their first number (4), but nos. 6, 10, and 13 have the same situation); also suggestive are places where lines given in the first edition of the libretto to one sister are set by Mozart for the other.

SOURCE:  detail of scan of Bärenreiter facsimile (Vol. I, p. 55), reproducing Act I, fol. 28—the beginning of the duet No. 4, “Ah guarda, sorella” 

Woodfield’s scrutiny of the autograph score reveals Mozart’s considerable indecision at crucial moments.  In the recitative below, there is a false start for Ferrando, then (after the new accolade with clefs) the same text allocated but given to Guglielmo, followed by Ferrando’s response, but which Mozart then struck through, reversing the order of the characters back to the original plan, and adapting the vocal parts accordingly.

SOURCE:  detail of scan of Bärenreiter facsimile (Vol. I, p. 207), reproducing Act I, fol. 104—the beginning of the recitative “Ah non partite.”

If Woodfield had to rely on a single page for strong evidence of his theory, it would certainly be the first page of an aria intended for Guglielmo, but replaced long before the premiere.  There are good dramatic reasons for replacing it, but Mozart clearly valued it enough that even after removing it from the opera he entered it in his own catalogue of works, where it appears thus:

SOURCE:  detail of scan of Mozart’s Thematic Catalogue:  A Facsimile (BL Stefan Zweig MS 63, ff. 22v-23)

In December [1789]
An aria intended for the opera Così fan tutte, for Benuccì.  Rivolgete à me lo sguardo etc: —2 violini, viola, 2 oboe, 2 fagotti, 2 clarini e Timpany e Baßi:

The cut aria remains in situ in the score.  Here’s the first page:

SOURCE:  detail of scan of Bärenreiter facsimile (Vol. I, p. 209), reproducing Act I, fol. 105—the beginning of the Guglielmo's cut aria “Rivolgete”

SOURCE:  the same, even more detailed

Looking more closely [detail at right], we can see that the aria was originally addressed to Dorabella, but her name has been scratched out and Fiordiligi’s superimposed.  Beyond this, though, Woodfield notes the ink color of the word “lui” is different than the surrounding text.  Mozart left the space for the pronoun blank for a while as he decided what Guglielmo was to say (and to whom).  In the catalog listing, he says “Turn and look at me” (“Rivolgete à me lo sguardo”); but in the score he says instead “Turn and look at him.”  Mozart’s omission of the pronoun allowed him time to figure out what would work best for the opera.  Ultimately that was a restoration (apparently) to the criss-cross couplings, but it is fascinating to consider how different the work might have been.

These are a few of the dozens of examples Woodfield musters to support his conclusions, and I cannot do justice to them in the small scale of a blog post—still less to his examinations of performing traditions in sources dating from the first few decades after the premiere.  The work that resulted is not perfect.  It bears the traces of Mozart trying to make changes as he went along, and then incompletely fixing them.  (Fiordiligi’s first line of recitative, for example—“I’m ready for some mischief this morning”—seems more in character with her sister.)  This is a problematic work, and Mozart struggled to bring it off.  Given that Mozart fell in love with one woman and ultimately married her sister (to whom he was later to write about the necessity of her fidelity in preserving his honor), it wouldn’t be surprising if this plot hit particularly close to home.  One reason I like teaching this piece is that I think it comes close to my students, too.  Generally speaking, my students are still idealists, like the couples in this opera.  

The moral of the tale—that one will be better off accepting how people are than pretending they are who we want them to be (and consequently being perpetually hurt or disappointed)—is, I think, one well worth learning.  Maybe it is my Calvinist upbringing, but I have found the #metoo revelations simultaneously appalling and unsurprising.  What conceivable grounds do we have to expect people in power to behave better?  (Granted, Le nozze di Figaro deals more overtly with #metoo; and the moment at the end of Act III where Barbarina turns it to her advantage is particularly satisfying.)

I don’t like the plot of Così; I don’t like the situations the characters are put in.  I understand why people don’t want to see it (and thus the howls of disapproval at an otherwise pleasant dinner last spring).  It is an ugly story beautifully—though problematically—told.  I can’t stand its title, which singles out women specifically and unfairly.  It should be called Men behaving badly:  their arrogance is the cause of all the heartache.  I would settle for This is how people are.  Any work with a title like that is bound to be a tragedy:  da Ponte in bed with Voltaire. 

The textual situation of Così is a healthy reminder that (of course) all life is compromise:  not just politics, but relationships too—and even art from as sure a hand as Mozart.  The music is so beautiful, but it appears he didn’t get it to work out quite the way he wanted.  Still, it’s so much better than nothing at all.  That, too, is the lesson his couples have to learn if the ending is going to mean anything to us.

17. Just kidding?

The fifth installment of the Settling Scores

Johann Nikolaus Forkel’s 1802 biography of Bach has many over-the-top claims which bash the reader over the head with Bach’s supreme genius.  Today I consider just one of these, drawn from information related by C. P. E. Bach.  I’m prompted to write about it now because of some evidence that I have stumbled across in recent months supporting my doubts about it.

In the chapter concerning “Bach the clavier player,” Forkel records that

“He even went so far, when he was in a cheerful humor and in the full consciousness of his powers, as to add extempore to three single parts a fourth part, and thus to make a quartet out of a trio.”  [trans. in The New Bach Reader, p. 435; cf. C.P.E.'s letter in ibid, p. 397]

And he repeats it in the chapter concerning “Bach’s character”:

“If he was in a cheerful mood and knew that the composer of the piece, if he happened to be present, would not take it amiss, he used, as we have said above, to make extempore, either out of the figured bass a new trio, or of three single parts a quartet.  These, however, are really the only cases in which he proved to others how strong he was.”  [Ibid., p. 460]

Maybe there is some basis in fact to this story—even if it was only a single occasion—but it seems to me literally incredible otherwise.  Forkel even acknowledged elsewhere that the sort of polyphony that allows the adding or reducing of parts has to be very specially constructed:

“In his compositions in four parts, you may sometimes even leave out the upper and lower part and still hear in the two middle parts an intelligible and pleasing music.  [Only the two middle parts, Herr Forkel?  An example would be nice....]  But to produce such harmony, in which the single parts must be in the highest degree flexible and yielding towards each other if they are all to have a free and fluent melody, Bach made use of peculiar means, which had not been taught in the treatises of musical instruction in those times, but with which his great genius inspired him.  These means consisted in the great liberty which he gave to the progress of the parts.  He thereby transgressed in appearance, but not in reality, all the long-standing rules which, in his time, were held sacred.”  [Ibid., p. 443]

As an organist I am occasionally in a situation where a descant is added for the final verse of a hymn.  Sometimes the descant line is printed in the hymnal on an extra stave above the four-part harmony.  What I notice invariably—because it is indeed inevitable (that word that dogged my previous post)—is that in order to give the descant line a musically-satisfying melody, it will at times borrow note-progressions from the alto or tenor (or even the soprano), creating intrusive parallel unisons or octaves.  Really, the right way—if I may be so bold—to add a descant is to compose it as a counterpoint to the melody, and then write a harmonic background for those two lines together.  There are many examples of this done well, but too often I’m playing the other type.  And it beggars belief that a hypothetical trio by Forkel’s “composer of the piece, if he happened to be present” would accommodate an added fourth line with its own integrity while not making substantial alterations to the original parts.  I have naively accepted this as literal truth for too long.  No longer:  two examples I recently noticed suggest to me that four voices was too many for Bach to shuffle around in his head.  Heresy?  Maybe so.  But look at these:

1)  strict canon:  “Christe eleison” from the Missa in A, BWV 234

I love this movement.  For years I have used it on the very first day of Theory I, when students generally have no theoretical background, may only read one clef (if that), and may never have seen a full score.  I throw it at them and ask them to observe:  What do you see?  And there’s much to be seen.  It is a strict canon beginning with the bass soloist, and with ensuing entries in the tenor, alto, soprano, and finally two flutes in unison—each of these entries a perfect fourth higher than the last, and with the sustained harmonies in the strings never really relaxing into anything that feels like a resolution.  Even the apparently simple question “What key is this in?” defies a simple answer.  Also interesting is that the canonic line echoes its opening arpeggio (and more) twice even as the voices accumulate, so that the arpeggio occurs not just five but nine times, as if in stretto.  (This is partially illustrated below, but if you’re curious you’ll save time just looking up the movement yourself.  Good stuff.)

One day, sitting in the Subaru service department with the NBA Kritischer Bericht at hand—as you do—I was killing time looking at the variant readings for this movement.  The report documented a series of systematic corrections in the autograph, commenting merely (and I paraphrase) “the corrections in bb. 80, 82, and 85 are related:  Bach altered his conception after the fact, as the fourth canonic phrase initially began with a leap of a fourth.”  [p.27]  Each note circled below (the autograph score on the left, the NBA text on the right) was originally a fourth lower in the autograph.

Composite of BWV 234/i; SOURCES: (left) cropped scan of autograph score f. 3v (bb. 75-85) from ULB Darmstadt scan;
(right) scan of Barenreiter TP 266 (off-print of NBA Ser. II Bd. 2), p. 11 (bb. 80-85); for both, I have added the red circles.

Here is the bottom system of the same page of the NBA (bb. 83-85), which presents the complete “Stimmtausch block” of the canon, marked-up to show the canonic phrases.  [I appropriate the term from Robert L. Marshall’s landmark study of Bach’s compositional process (v. 1, p. 134f.]  “1” is how the line begins (originally in the basses, now in the flutes), continuing to “2” and so on to accumulate five canonic elements (although each voice is a fourth higher than the previous one).

SOURCE: cropped scan of Barenreiter TP 266 (off-print of NBA Ser. II Bd. 2), p. 11 (bb. 83-85)

The NBA does not go on to explain why Bach would have made the changes.  Bar 85 makes the reason clear:  the A in the alto (the pick-up to canonic phrase “4”) was originally an E, but the leap up from E to A would cause parallel fifths with the tenor leap from A to D.  This became an issue only when he wrote the fifth canonic phrase, but apparently he didn’t notice it in b. 82, so had to go back to fix it there; the retrospective change in b. 80 was only necessary to preserve the strict canon, and was thus presumably the last to be made.  The lesson here?  Bach could juggle a number of voices in his head as he constructed the canon, but clearly had to get it down on paper to get it right once too many voices had accumulated.  In other words, he had not conceived phrase 5 when he first notated phrase 3.

2)  permutation fugue:  chorus Himmelskönig, sei willkommen, BWV 182/ii

This is a similar example—this time more complicated because the lines had to work as invertible counterpoint rather than just a canon.  Arthur Mendel discussed this example in a 1960 Musical Quarterly article; his point was really just that an autograph that Spitta had taken to be a fair copy was really a composing score, as the types of corrections were not copying errors but rather directly related to the substance of the counterpoint:

Composite of BWV 182/ii bb. 1-5; SOURCES: (top) cropped scan of f. 1v of autograph score of BWV 182 from Bach Digital; (bottom) cropped scan of Mendel’s reconstruction of the first reading of these bars (Musical Quarterly (1960), his Ex. 1, p. 292); for both I have added the accolades on the left.

 Mendel notes a pattern of corrections:

“...we see that while there are no corrections in the first two measures, in measure 3 the last note in the soprano has been changed from an original a1 to d2, and there is a corresponding change in the alto in measure 4 and the tenor in measure 5.  In the bass in measure 6 [not shown above], however, at the beginning of the second brace, there is no corresponding correction; here the corrected reading found in the other three voices was written in to begin with.  If we look a little further, we can see that at the end of measure 4 not only the alto but also the soprano, and at the end of measure 5 not only the tenor but also the alto, have been corrected.  But again in measure 6, the tenor bears no correction corresponding to those in the alto in measure 5 and in the soprano in measure 4.”  [p. 292]

and then reconstructs Bach’s compositional process:

“Bach starts out as follows, writing in the first brace the whole four measures of the soprano, then the rests plus three measures of the alto, then the rests plus two measures of the tenor, and finally the rests plus the subject in the bass.  [This produces Mendel’s Ex. 1, given in the above composite.]  But already on the fourth beat of the bass's subject-entrance (the first beat of measure 5) there occur consecutive octaves between bass and soprano.  Apparently he next changed the soprano, inverting its motion to read d2-f#2 instead of f#2-d2.”  [pp. 292f.]

He then charts further changes to get from his putative original to the eventual (I hesitate to say final with Bach) reading.  He concludes

“It is surprising to find that in writing such a permutation fugue (he had already written several that we know, and the scheme remained a favorite one with him) he had not worked out the invertibility of his four melodic elements until he set pen to paper to write a score that is neat enough to have been taken by Spitta for a fair copy.” [p. 293]

Granted, Bach was clearly very good at this.  The anecdote transmitted by Forkel suggests a contrapuntal understanding of such profundity (being able to spontaneously convert a trio into a quartet by the addition of an extra line) that writing out such examples into fair copy would not be surprising at all.  For too long I’ve taken the Forkel story at face value, but when I stop to think of it it really can’t be true.  “Just kidding,” as my students sometimes say when corrected; maybe Forkel would say the same.

ADDENDUM  20 August 2017

In the course of my Bach cantata pilgrimage, I have run across an interesting example going the other way—something originally in five real parts (SATB with a descant) in which Bach later deleted the fifth part, and had to make only very minor adjustments in two spots for the four-part version:  compare the closing chorale of BWV 12 with its later use in BWV 69.

ADDENDUM  28 November 2017

Then again, the seven-part scoring of the final chorale of BWV 70 more than amply demonstrates Bach’s ability to write integrated-yet-disposable lines.  Even so, there were some small alterations to the inner parts when this harmonization was included as a four-part chorale as no. 347 of the Breitkopf edition of Bach's chorales (1784-1787).  See NBA Ser. III, Bd. 2, Teil 2, where this chorale appears on p. 200.