31 From Dossier to Research Note
This chapter is part of a book in active development and has not yet been through the author’s review. Content may change as the review advances.
The research decision. When you turn your research dossier into a written note, decide which finding has earned the headline and exactly how big each claim is allowed to be given your evidence, then package the note so a stranger can rerun it and reach your number without you in the room. Whatever you present along the way — a poster, a talk, a brief — is one input to that note, never its source.
31.1 Why this decision matters
The decision on the table: which finding leads your written note, and how big each claim is allowed to be once your voice is no longer in the room.
“Show me your results in the order you trust them. If your flashiest number comes from one lucky run and it leads the paper, I already know how this review ends. I do not want your most surprising result first. I want your most defensible one.” — a conference reviewer, on the first page of a paper they actually weigh
When you present, you stand next to your work and fill the silences with your voice. The note has no you standing next to it. Every leap a presentation would let you imply, the prose has to earn, in order, for a reader who will never hear you explain it. The reviewer above is not impressed by a big number. She is asking which finding you would still defend after the room empties, and whether she could rerun your work and watch the same number return. This chapter is how you answer both.
31.2 The concept
Your dossier and your note make the same argument, but they carry it differently. Working logic is how the dossier and any presentation of it speak: compressed records and, when you present, your live voice filling the silences. Example: a poster’s Gap box that reads only “few studies of wait times at peak hours” while you say the rest out loud. Paper logic is communication that must stand alone on the page, where every claim is argued in prose because you are not there to answer. Example: that same gap written as a paragraph naming what prior studies covered, what they skipped, and why the missing piece matters. Turning one into the other is not adding words. It is making the reasoning visible.
A note carries that reasoning on a backbone (Mensh and Kording 2017). The problem–gap–question spine is three linked pieces: the real trouble your work responds to, the specific thing prior retrievable work has not established, and the answerable question that closes exactly that gap. Example: customers give up and leave during the morning rush (problem); published wait-time studies measure only quiet, steady hours (gap); how much does a mobile pre-order option cut the wait during the rush at this shop (question). The spine holds only when the gap is real, and a gap is real only when you show what is already known with evidence density, the share of a paragraph resting on sources a reader could open rather than bare assertion. Example: “as everyone knows” carries no density; a sentence resting on a published figure a reader can look up and recompute carries some.
The back half of the note needs one more term. A results hierarchy is the order you present findings in, best-supported first and fragile last. Example: a median measured over a thousand simulated mornings leads; one unusually fast order waits at the back, labeled exploratory. Your limitations follow the same discipline, stated in proportion to what actually threatens the claim: “measured on a simulated arrival pattern, not a real Saturday rush,” not a generic “more research is needed.”
Finally, a note a reader cannot rerun is half a claim (National Academies of Sciences, Engineering, and Medicine 2019). The other half is a reproducibility capsule, everything a stranger needs to rebuild your number and nothing they must guess: a notebook that passes restart-and-run-all (clear everything, run every cell top to bottom, and the headline number returns), pinned data and tool versions, a fixed seed, a decision log, and your AI-use ledger (Wilson et al. 2017).
31.3 A worked example
You tested a mobile pre-order option for a campus coffee shop. Your claim–evidence table holds the headline row: “38% faster.” The note has to turn that row into a claim you can defend.
Rank the findings. Across a thousand customer arrivals from a seeded simulation of the morning rush, the median wait time dropped about 38%, with a tight interval around it. You also noticed one pre-ordering customer who was served almost instantly, a 99% drop. The surprising number is the single customer; the defensible one is the median over a thousand arrivals. So the median leads, and the lone customer goes to the back, labeled as one observation, not a result.
Size the claim. The honest headline reads: “on this simulated arrival pattern, adding mobile pre-order cut median wait time by about 38% (interval attached); this is a measurement on simulated demand, not a guarantee for a real Saturday rush.” The inflated version you refuse: “mobile pre-order makes the shop 38% faster.” That sentence drops the demand pattern, drops the word median, and promises every customer a shorter wait your simulation never tested.
Package it. Your capsule pins the simulation’s seed, records the machine and library versions, logs that you excluded the first order of the morning and why, and passes restart-and-run-all so the 38% returns on a clean kernel. The note is now readable and runnable.
Carrying the argument on one clear problem-gap-question spine, so the prose stands without you beside it, is the standard structure for a paper (Mensh and Kording 2017).
The block below runs the arrivals and prints the ranked findings: the median with its interval first, the single lucky customer last.
import numpy as np, pandas as pd
SEED = 464
rng = np.random.default_rng(SEED)
n = 1000
service = rng.gamma(2.0, 55.0, size=n) # ordinary counter service
preorder = rng.gamma(2.0, 36.0, size=n) # mobile pre-order
med_drop = 1 - np.median(preorder) / np.median(service)
boot = [1 - np.median(rng.choice(preorder, n)) / np.median(rng.choice(service, n))
for _ in range(2000)]
lo, hi = np.percentile(boot, [2.5, 97.5])
luckiest = 1 - preorder.min() / np.median(service)
print(f"median wait, ordinary service : {np.median(service):.0f} s")
print(f"median wait, mobile pre-order : {np.median(preorder):.0f} s")
print(f"median drop : {med_drop*100:.0f}% "
f"[{lo*100:.0f}%, {hi*100:.0f}%]")
print(f"the single luckiest customer : {luckiest*100:.0f}% faster")
print("\nthe headline is the first number with its interval. the last one is")
print("one observation, and it goes at the back of the note labelled as such")31.4 An AI failure case
You paste your discussion into your AI and ask it to make the writing sharper. It hands back cleaner prose, and it runs through your spell-check without a flag. Buried in the polish, one sentence has changed size. Your original said “median wait time dropped about 38% on this simulated arrival pattern.” The rewrite says “mobile pre-order makes the shop roughly 38% faster.” The word median is gone, the demand-pattern boundary is gone, and a measurement on simulated demand now promises every customer a shorter wait. The sentence reads better and claims more.
You catch it by diffing the rewrite against your original sentence and against your printed result. The number came from a median over a seeded simulation, not a mean, and not a real Saturday rush, so the general claim is not licensed. You keep the edits that clarify and reject the one that inflated. Fluent prose is not evidence. You verify the claim, not the polish.
31.5 It is your turn
You are working inside Studio 9: Write, bound, and disclose. Keep what you write here; the studio’s milestone chapter is where it joins the other lessons’ pieces into one artifact you can defend.
Your project already has a dossier — its contract versions, evidence registry, analysis and claim ledgers — plus whatever you present, and the working folder behind them (Studio 11 will make it rerunnable by strangers); this step turns that summary into prose that stands on its own page.
The hands-on half of this section lives in the chapter’s companion notebook: open it in Colab with the badge at the top, and work the steps there.
Commit your own answer first, then delegate. Writing with an AI tool is a loop, not a single ask: you send the prompt, read the output, push back on the part that is wrong, and run it again. Keep the loop pointed at your own text. The moment it starts producing sentences you did not think of and cannot check, you are no longer editing, you are being edited. Each prompt below is a checkable job, not a request for a verdict.
Three calls stay yours no matter how fluent the tool is. Which finding is your headline, how big that claim is allowed to be, and which sources are real are never delegated. An AI can sharpen a sentence, but a sharper sentence is not a bigger finding, and clean prose is exactly how an overclaim sneaks in. You own the words you put your name on, their boundary, and their uncertainty.
Write your problem–gap–question spine as three short paragraphs. Name the trouble, name what prior retrievable work has not established, and name the question that closes exactly that gap. Every sentence in the gap paragraph rests on a source a reader can open.
Locate real sources for the gap.
Act as a literature assistant in service operations. List peer-reviewed or official sources that measure customer wait times or the effect of pre-ordering on queues, in a table with title, authors, venue, year, and one-line finding. Only include work you are confident exists; mark anything uncertain. Do not write my gap paragraph.After running, verify: retrieve each source yourself before it enters the note; a source you cannot open in seconds is fabricated until proven otherwise. Counters confident fabrication (an invented citation arrives as confidently as a real one).
Order your findings, best-supported first. Whatever you can defend with the most data leads. Anything that rests on one observation or one lucky run goes to the back and wears the word exploratory.
Write your headline claim in one sentence, with its boundary inside the same sentence: the setting, the sample, and the population it does not cover. Then write the inflated version you refuse, so you can recognize it if it sneaks back in.
List every over-reach to verify.
Here is my discussion paragraph and the cell that printed my median wait-time drop with its interval: [paste both]. List every sentence that claims more than the printout supports: a bigger number, a wider setting, or "faster for everyone" where I measured one simulated demand pattern. List them; do not rewrite them.After running, verify: check each flagged sentence against your printout and keep only edits that shrink a claim to fit the evidence. Counters silent scope change (a sharper sentence that upgrades “on this arrival pattern” to “for everyone” has quietly answered a different question).
Size your limitations to what actually threatens the claim. Cut anything that reads like “more research is needed” and keep what a hostile reader would raise first.
Assemble the full note, not only its spine: methods (what you did, so a stranger could follow), results with their uncertainty, a discussion sized to the claim, references, your AI-use disclosure placed where the venue expects it, and the pointer to the package Studio 11 will freeze.
Attach your folder to the note and confirm the two agree: the number in your lead sentence is the number a cold restart-and-run-all prints.
Red-team the capsule.
Here are the key lines of my reproducibility capsule: [paste them]. Playing a cold replicator who has only these lines, list every input, version, or by-hand step you would need to rerun my number and might not find here. Then name one problem this line-scan can never catch that only a person running it cold would hit.After running, verify: match each gap against what you can actually see missing, and confirm the “cannot catch” problem is real by rereading your own lines. Counters illusion of completeness (a capsule can look complete and still omit the one input a stranger needs).
Log the drafting round in your AI Research Ledger, and verify at least one output with a named method from the Verification Guide. An AI reviewer may run the check with you; the decision to accept or reject stays yours.
Milestone next. This was the last lesson of Studio 9. Milestone 9: Your bounded claims is where the lessons’ pieces become the studio’s versioned artifact. Produce it before you move on.