Week 7 Workshop: When ancestry becomes hard to read

What happens to ancestry inference as copying errors increase?

14 April 2027

What happens to ancestry inference as copying errors increase?

Learning target. Describe how a simulation’s assumptions limit its conclusion.

Before the workshop#

Download the week 7 pack: a 24-symbol synthetic ancestor, copied forward for 40 generations at three fixed error rates (low, medium, high), with three fixed seeds run at each rate. settings.json names the alphabet, the ancestor, the rates, the seeds and the exact copying method used to produce every number in the pack, so any run can be reproduced from it. This is not a text or an audio recording — it has no content beyond the symbols P, Q, R and S, and no claim of biological origin. Do not open reveal/ yet.

Check — 5 minutes#

Read trend.txt’s low-rate paragraph only, without opening plot.svg or results.csv yet. State a one-sentence prediction: as the error rate rises from low to high, will ancestor agreement at generation 40 rise, fall, or stay about the same?

Receive — 15 minutes#

Read settings.json in full. Confirm you can state, in your own words: what one generation’s copying step does to a sequence (per position, with probability equal to the error rate, swap in a different symbol); what “ancestor agreement” measures (the fraction of positions that still match generation 0, not the immediate parent); and why three seeds are run at each rate rather than one.

If you want a feel for compounding error before you read settings.json’s numbers cold, try the chain simulator. Read it as an analogy only: it corrupts individual characters in ordinary text, not the four-symbol alphabet this week’s synthetic ancestor is built from, so it cannot stand in for the actual simulation — it can only build intuition for what “copying a copy” does over many generations.

Investigate — 40 minutes#

Predict a trend (10 minutes). Before opening results.csv or plot.svg, write what you expect to happen to ancestor agreement over the 40 generations at each of the three rates, and why a fixed per-position error rate should produce that shape.

Compare the prepared repeated runs (20 minutes). Open plot.svg (or results.csv and trend.txt if you are using the accessible route). For each rate, read off the mean agreement at generations 1, 5, 10, 20 and 40 across the three seeds, and note how much the three seeds disagree with each other at the same rate and generation. Do not average away that disagreement — record it.

Identify a conclusion the model cannot justify (10 minutes). This pack has no selection and no population structure: every substitution survives equally, and one lineage is copied forward with no branching. Write one sentence naming a conclusion someone might be tempted to draw from this plot that the model’s assumptions do not support — for example, any claim about a universal error-rate threshold, or any claim about a real biological sequence.

Evaluate — 30 minutes#

Write up your prediction, your reading of the three rates’ trends including the seed-to-seed spread, and your named unjustifiable conclusion — before opening reveal/.

Deposit — 30 minutes#

Open reveal/analysis.md (ask your tutor for the link once your conclusion is written) and compare it against your own limitations note and unjustifiable-conclusion sentence. Where they differ, write one line on whether your reasoning was still defensible from what the pack gave you. Deposit your prediction, your rate-by-rate reading with the seed spread, and your final limitations note into the Sequence model study.

Accessible route#

Plot data as a table plus a prose description of the trend: results.csv and trend.txt give the same information as plot.svg without requiring the image, and every generation used in the plot is also a row in the table.

Assessment#

This workshop’s limitations paragraph — naming what a simulation’s assumptions do and do not support — is direct practice for The Stemma, which asks the same question of an inference method rather than a simulation: what does your evidence justify, and what does it not rule out?