Skip to content

ADR-0014: A damaged witness cannot establish absence

Status

Accepted.

Date

2026-08-22

Context

saucier diff reported eight sauces the 1909 revision added. All eight are in the 1907 printing, at the same entry number, under the same heading.

1907:  6s— BERCY  SAUCE          1909:  65—BERCY SAUCE
1907:  loi— POULETTE  SAUCE      1909:  101—POULETTE SAUCE
1907:  I22~ANDAL0USE  SAUCE      1909:  122—ANDALOUSE SAUCE

The scanner corrupted the marker rather than the heading. A digit read as a letter, a space inside a number, a tilde for an em dash. The parser never saw the entry, so the comparison saw a concept in one witness and not the other, and called it an addition.

Two earlier releases mended two shapes of the same damage. These are four more, and mending them would leave a fifth shape unmended and the same claim resting on it.

The measurement that settles it: the parser reads 2,679 entries in the scan against 2,963 in the transcription. That is a 284-entry blind spot. "Not in the 1907 printing" and "not found in the 1907 printing" differ by exactly those 284 entries. The diff has been saying the first while knowing only the second.

Decision

Absence is only observable through an instrument that can see everything present. Where the instrument has a measured blind spot, absence is unobservable and is not reported as observed.

A comparison involving an OCR witness reports unmatched. The cause says the diff found no counterpart. It says nothing about what either book contains. The row still names the witness that holds the concept, so a reader loses no information, only a claim that was never established.

added and removed survive between two witnesses of equal fidelity. The rule is about damage, not about comparison. Two proofread texts are an instrument with no measured blind spot, and absence is observable through them.

The blind spot is printed with the counts. Entries read from each witness, and the gap between them, sit in the summary rather than in a note. No reader sees how many rows the diff found without also seeing how much of the source it could not read.

The blind spot is what has to shrink before the claim returns. Restoring added for a scanned witness is not a matter of taste or of a better comparison. It requires reading the entries the parser currently cannot see, and measuring what is left.

This is ADR-0002 and ADR-0012 applied to a third surface. An unresolved parent is recorded as absent rather than inferred. A resolver refuses rather than ranks. An absent preparation is now recorded as not found rather than as absent. One rule, three places.

Consequences

Positive

  • The release states nothing about Escoffier that its evidence cannot carry. No editorial difference between the two printings is confirmed, and that is the finding.
  • A reader sees the instrument's limit beside its output, which is the ordinary standard for a measurement and was missing here.
  • Closing the blind spot now has a number attached to it, so the work has a before and an after rather than a feeling.
  • The eight false additions cannot come back by a route that looks like a feature, because restoring the cause requires the measurement.

Negative

  • The diff says less. Eight rows that read as findings now read as questions, and a reader who wanted a number is given a smaller one.
  • unmatched is weaker language, and weaker language is easy to argue away. A future contributor will read it as timidity rather than as a rule.
  • The rule is coarse. A witness with a one-entry blind spot is treated like a witness with 284, because the alternative is a threshold nobody can defend.
  • The blind spot is measured against the other witness, so a work with only one scanned witness has no way to state its own.

References