Week 2 Workshop: Hiss as a signature

Which shared features are useful evidence of a common intermediate copy?

3 March 2027

Which shared features are useful evidence of a common intermediate copy?

Learning target. Distinguish a feature two copies inherited from a copy they both descend from, from one they share for some other reason entirely.

Before the workshop#

Open the week 2 pack and read feature-table.csv and features.txt. Ten minutes.

Reading. Peter Manuel, Cassette Culture: Popular Music and Technology in North India (University of Chicago Press, 1993), introduction and chapter 3, on duplication as a distribution system and what it cost in fidelity. No linkable copy exists; the lecture states the specific claim used.

The specimen#

One source recording and four copies of it — W, X, Y and Z. Six features have been identified across them: two dropouts and a click at fixed timestamps, a raised noise floor, a high-frequency rolloff, and a door closing.

You are not given the recordings. The evidence is the table, and the inference runs on the table. Supplying clips would let your hearing affect your answer, and this task is not about hearing.

Everything in the pack is an authored teaching scenario. Nothing was measured from a real tape, a real deck or a real recording, and every file says so.

Last week, and this week#

Week 1 measured one copy at a time against a source. That told you a copy had changed; it told you nothing about which copies are related to which.

This week two copies share something. The whole question is what that licenses.

Investigate — 40 minutes#

Tabulate the features (10 minutes). For each of the six, write down where it sits and which of the five recordings carry it. Work from the table; do not assume a feature is important because it has a dramatic name.

Compare against the source (15 minutes). For every feature shared by two or more copies, check first whether the source has it too. A shared feature that is in the source tells you only that both copies were made from this source. Rule those out before going further, and record that you did.

Propose one pair, then attack it (15 minutes). Choose two copies you think share an ancestor that the others do not descend from. Name the single feature that supports it. Then find the strongest feature that supports a different pairing, and say why you are not persuaded by it.

You will find at least one feature that groups two copies and contradicts your answer. That is deliberate. Deciding between them is the work.

Evaluate — 30 minutes#

Build the evidence table you are depositing: feature, location, which copies carry it, your proposed explanation, and your confidence. Then answer three questions in a paragraph each.

1. What makes one shared feature better evidence than another? Not “it is rarer” — say what property of the feature makes independent occurrence implausible.
2. What would change your mind? Name an observation that would overturn your proposed pair.
3. What does your answer not establish? You may have shown two copies share an ancestor. Say what you still cannot say about them.

The construction log is in the pack. Open it after you have your table and your three paragraphs.

What you should find#

Four explanations compete whenever two copies share something, and only one of them is ancestry.

It is in the source. The door closing is in the original recording, so all four copies have it. Sharing it means nothing.

Every copy acquired it separately. The raised noise floor is absent from the source and present in all four. Every dub does that. It separates copies from the source and separates nothing among the copies.

They were made on the same equipment. The high-frequency rolloff groups two copies — and frequency response is a property of a machine, not of a history. Two dubs made on the same worn deck share it without being related at all. This is the one that looks most like evidence and is not.

They share an ancestor. One 40-millisecond dropout, in one channel, at one timestamp. Two dubs producing that independently is not a plausible coincidence. That is the feature that carries the claim.

The general lesson is not “shared means related”. It is that a shared feature is evidence in proportion to how hard it would be to produce twice by accident — and that the strongest evidence here is the smallest, least dramatic thing in the table.

Deposit#

Your evidence table goes into the audio collection, with the relationship you propose recorded as inferred, separately from what is known. Week 3 keeps this source recording and stops asking what happened to it, asking instead how you would test a claim about what happened to it.

The five-minute check#

A fresh feature table. One pair of copies shares a local dropout; another pair shares a raised noise floor. Which better supports a proposed relationship, and why?

Assessment#

This is the reasoning The Stemma is marked on, four weeks early and with four copies instead of nine. Its largest criterion is whether each proposed relationship names the specific shared feature that supports it, under a stated assumption. Practise writing that sentence now.

For The Chain, note which of your measurements would still mean something to someone who had not watched you take it.