Terminology and Concepts to look out for

observation · narrative · audience

Relevance to Me

For four days you wrote notes for yourself. Today you start writing for somebody who was not in this room, and that is a completely different job.

Board Question

Chromebooks. The four observation forms and the report body form are in the deck, side by side.

• 1) Revisit your observation form from last week. If you need to finish them finish them.
• 2) Pull the MuggsOfPrompts Tool up on the work station PC to refer to.
• 3) Prioritize Tuesday, Wednesday and Thursday's forms from last week.
• 4) If you did all forms for last week, you will receive bonus points for Monday's work.
• 5) I will do another grade run at the end of today and update PowerSchool.

A lab report has three parts. You will finish all three by Friday.

Introduction. What question the lab was trying to answer, and how it was set up.

Body. What happened. Three prompt days — Tuesday, Wednesday, Thursday — written up so a reader who was not here can follow them, one section each. Each section opens with exactly what the three models were sent.

Reflection and Conclusion. What it adds up to, and what your evidence cannot show.

Today the whole shape gets a passage — you can see where each part lives — and then the day's work is the body. The body is the longest part of the report and the part where four days of notes turn into something a reader can use. Get the body right today and Thursday's intro-and-conclusion writing lands on solid ground.

Read this sentence, which a student actually wrote during a lab like this one.

pane 2 was longer and had numbers, pane 1 didnt

You knew what that meant when you wrote it, because you were sitting in front of the screen. Nine days later, a reader who was not there gets nothing out of it. Longer than what. How many numbers. What did pane 1 not have.

That sentence is an observation. It is a note you made for yourself while something was happening, and it did its job. What you are writing today is a narrative, which is an account of what happened told to somebody who was not there. Same facts. Different reader.

That reader is your audience, and today the audience is the whole thing. Every change you make comes from asking one question: would a person who was not in the room understand this?

Here is the same observation rewritten.

Pane 2 returned about three hundred words in six numbered steps. Pane 1 returned about ninety words in one paragraph with no numbered amounts anywhere in it.

Nothing was invented. Everything in the rewrite was already in the notes or on the screen. Three moves did all the work.

Name the thing before you use a shortcut. Pane 2, not it. The 3B model, not the little one.

Give the number and say what it is out of. Two is meaningless. Two out of three is a finding.

Say what happened before you say what it means. Notes mix these together. A report separates them, because a reader has to be able to check your evidence before deciding whether they believe your point.

One rule for today, and it matters. You are revising, not rewriting.

The failure mode from last week was stretching a definition to fill a blank. Nobody's three models produced solid hedges. Some of you wrote "no hedges observed" plainly. Others felt the notes looked thin without something in the hedge column and started calling a comma or a soft verb a hedge. That is not what a hedge is, and that is not what happened.

If a feature was not there, record that it was not there. "Pane 1 zero hedges, pane 2 zero, pane 3 zero" is a real finding. That models ranging from a 3B parameter count up to a 119B mixture-of-experts all skipped hedging on the same prompt — when a hedge costs the model nothing extra per token — is a bigger finding than any single pane. The absence is the data.

Same rule for counts, examples, and anything else. Ugly numbers stay ugly. Empty columns stay empty and get labelled. Do not change a count because you would prefer a different answer now, and do not invent a soft version of a feature to make the notes look more complete.

The same impulse — refusing a plain negative finding — will show up elsewhere this year as "go get a bigger hammer." Small model returned no hedges? Try the big one. Big one still no hedges? Try the cloud one. That is stretching the search instead of the definition, and it is the same mistake. In this lab you cannot rerun, and that is a feature, not a limitation. What is on your screen is what happened. Write it plainly.

Today, the body. The report body form has three sections, one per prompt day: Tuesday's three questions, Wednesday's presentation advice, Thursday's student-loan question with the 120-word limit. Your observation forms stay open on the left; the body is written from them, in that order. You are writing a paper. Complete sentences. If the body comes together fast, read it back as the reader who was not in the room.