What notation cannot record

Why can no amount of scholarship recover exactly how a 1600s piece was played?

We have the scores. Many of them are in the composer's own hand, some were printed under his supervision, and the paper has survived four centuries in good order. We also have treatises, ornament tables, instrument-building records, church accounts listing who was paid to sing. By the standards of most historical questions this is an extravagant amount of evidence.

And yet no scholar claims to know how a piece from 1620 actually sounded. Not "we are still working on it" — the stronger claim, that the exact performance is unrecoverable in principle. That is a surprising thing to say when you are holding the composer's own notes. So the question is not what we are missing. It is why more evidence of this kind would not close the gap.

Start by asking what a score is for. Not what it is — what job it does. It is a set of instructions written by one musician for other musicians, all of whom were trained in the same tradition, in the same decade, often in the same building.

Now follow that through, because it has a consequence that is easy to miss. If you and I share a body of assumptions, I do not write them down. Writing them down would be wasted effort and slightly insulting. Notation is compression, and compression works by leaving out whatever the receiver can supply. So the score does not contain the things everybody knew. It contains the things this particular piece did differently.

Which means the score is systematically silent about exactly the material we most want — the shared practice. And that material was never written anywhere, because at the time it was not information. It was competence.

Consider how much sat in that silent category. A bass line with numerals under it — figured bass — specifies which harmonies to build, and nothing whatever about the texture, spacing, rhythm or figuration the keyboard player invented above it, which was improvised afresh at each performance. Ornamentation was expected, largely unwritten, and idiomatic to place and instrument. Whether unequal rhythms were applied to notes written equal was a convention, not an instruction. Even the pitch itself is unfixed: there was no international standard, and A varied by roughly a whole tone across regions and decades. Tuning was not equal temperament, so the character of each key — a real expressive resource then — depended on a choice the score never states.

At this point the natural move is: but we have treatises. Some are superb, and the printed ornament tables — d'Anglebert's of 1689 is the famous example — do tell us what a symbol means. So why is that not enough?

Two reasons, and both are about the channel rather than the diligence of the writers. First, a treatise explains notation using more notation, or using words. Both are the same lossy medium as the score. A table that says a tremblement begins on the upper note has told us something real, and has told us nothing about its speed, its acceleration, or how it varies with tempo and affect — the parameters a player learned by ear from a teacher. Second, treatises are prescriptive documents. People write rules down when practice is contested or when learners are getting it wrong, so a treatise is evidence about the argument as much as about the norm.

Now the decisive point. Suppose an inspired scholar reconstructs a performance perfectly. Could we tell? There is no recording. Sound was not captured until the phonautograms of the late 1850s, and not played back until 1877. So there is no channel through which the original acoustic event reached us at all — only symbolic descriptions of it, written by people who had no reason to describe the obvious. We can rule things out, narrow ranges, and build performances that are defensible. What we cannot do is verify, because the ground truth was never recorded and every surviving trace passed through a lossy encoding on the way here.

That is the shape of the limit, and it is worth naming honestly: this is not a gap in scholarship that funding would close. It is a property of the evidence.

Three refinements.

First, the misconception worth correcting: this is not an argument that historically informed performance is pointless. Constraining the range is a real achievement. Knowing that a work was sung by a dozen voices rather than eighty, at a lower pitch, in a specific temperament, with a continuo player improvising rather than reading, changes the result enormously and is well evidenced. Narrowing is not the same as solving, and the field has largely stopped claiming otherwise.

Second, the unrecoverable part is not uniformly distributed. Pitch content and relative rhythm are encoded well and travel intact; tempo, articulation, ornamentation, timbre and balance are encoded weakly or not at all. So the loss is structured, not total, and knowing which parameters are which is most of the practical skill.

Third, an uncomfortable consequence, argued forcefully by Richard Taruskin among others: what fills the gaps is inevitably modern taste. A reconstruction is a collaboration between the evidence and the present, and honest practitioners treat it as an interpretation with historical constraints rather than a recovery.

• How pitch standards varied by region, and what a semitone of difference does to a singer's line.
• What figured-bass realisation actually asks a keyboard player to invent.
• Whether early recordings from the 1900s show that even nineteenth-century notation was read very differently from how we read it now.

Every term the collection defines is gathered in the glossary.