Grad school · Immunology · 事例
When the committee asks why you used Protocol C instead of B, the sentence you wrote that day comes back word for word
これでできることIn a workspace holding 900 literature notes and 400 experiment records, asking "every judgment I wrote about Protocol B, in order" returns the sentence written two years ago word for word, along with which note it came from and which line.
Four months to the defense. In the mock committee session, her advisor asked: "Protocol B was right there. Why did you go to C?" She had no answer. She changed it one day two years ago, and there was definitely a reason, but the reason is gone. That night she opened 400 experiment records by date hunting for 'protocol B' and gave up at 2am. Page 3 of her introduction may cite two papers back to back that say opposite things about the same cytokine — she has never checked. One of her committee members wrote one of them.
This person is a fourth-year immunology PhD student. Over four years she has accumulated 900 literature notes, 400 experiment records, 150 lab meeting notes, and 60 excerpts from her advisor's emails, all in markdown. The material is all there. It's just that "why I judged it that way back then" is scattered somewhere across hundreds of files. What follows shows what you can ask, and what comes back, once that material is in a Consilience workspace. It isn't a real customer story — it's a hypothetical scenario we thought through.
The material: about 1,500 notes, four years' worth
The files are all on disk. She's on the diligent end of note-takers. The problem is that the folders are split only by year and date, so there's no way to walk a single topic — "protocol" — in time order across four years of files.
- 900 literature notes — one markdown file per paper. The abstract summary, the sentence worth quoting, and her own comments like "this doesn't match our data" all sit mixed together in one file.
- 400 experiment records — date, protocol, conditions, results. A single line like "switching to C starting today" is just buried between result tables.
- 150 lab meeting notes — talk content and comments jumbled together. Who said what mostly went down without a name attached.
- 60 advisor email excerpts — blobs of text cut and pasted out of email. Some carry a date, some don't.
- 3 drafts of the dissertation — the introduction's citation list differs a little in each one.
How it works today
Defense prep starts with a list of likely questions. She writes down "why Protocol C" and opens the experiment records folder to find the answer. She opens the whole 2024 folder and eyeballs the filename dates to guess when the protocol changed. Grepping 'protocol B' hits 37 files, and most of them are just records that she ran B that day — not a judgment about why.
So she ends up writing the answer first — "it was probably the cell viability" — and hunting backwards for data to support it. You can't say "probably" in front of a committee, so the basis gets bolted on somehow. The contradiction in the introduction has no way of being checked at all. To find which two of 900 papers say opposite things about the same target, she'd have to read all 900 again, and that's not happening in four months.
I definitely wrote the reason down somewhere in an experiment record two years ago. I just don't know which of the 400 files it's in.
How you set it up
Make a new workspace and drop the four years of folders in as they are. The original markdown never changes, so the dissertation drafts folder can go in whole too.
Drop the five folders in as they are
Put the literature notes, experiment records, lab meeting notes, email excerpts, and dissertation drafts into the workspace and kick off extraction. About 1,500 files won't finish in one go, so start it and go do something else. A note that's done shows a "Key facts" badge in the document sidebar.
Register the immunology vocabulary first
Tell the app this lab's words up front: Treg, ILC2, flow panels, Protocol B/C. Across 900 literature notes the same molecule shows up as both "IL-10" and "interleukin-10", and the same cell as both "Foxp3+ Treg" and "regulatory T cell". Skip this step and one topic piles up split across several branches.
Confirm the different names for the same thing
The app proposes that "IL-10" and "interleukin-10" are the same thing. Nothing merges automatically — it only lands once a person confirms it. How far immunology notation variants get caught at 900-paper scale has never been checked, so walk the proposal list yourself.
Confirm the protocol decision pair
In the titlebar press "Open knowledge graph", then choose "Revisions" in the left rail. Put the protocol into "Search decisions…" and the pair shows up: NEW = "switching to C starting today" from 2024-03-11, REVISES = "we'll use B as the standard for now" from 2024-01-08. Press "Confirm" and the 2024-01-08 decision reads "Superseded" across the app.
What you ask
"@Protocol B — every judgment I wrote about it, in order."
→ 3 decisions · 2 open questions · 4 claims. All my own sentences, verbatim.
2024-03-11 decision: "Viability dropped into the 40s, so switching to C starting today."
↳ [[2024-03-11 T cell isolation]] · line 12
2024-02-27 open question: "Is it B's digestion time, or a reagent lot problem?"
2024-01-08 decision: "We'll use B as the standard for now."
↳ struck through, with a "Superseded" badge
"The @IL-10 papers I cited in the intro — do any of them say opposite things?"
→ 6 claims. Other people's words arrive with a speaker attached.
"IL-10 suppresses gut inflammation" — said by Ogawa 2021 · cited on intro p.3
"IL-10 blockade actually reduced lesions" — said by Lindqvist 2022 · cited on intro p.3
Two papers cited in the same paragraph. Whether that's a contradiction is my call.
"What did my advisor say after I switched to Protocol C?"
→ 2 lab meeting notes · 1 email excerpt. Speaker = advisor.
"State the viability basis for C explicitly in the methods section" (2024-04-02)The 2024-01-08 decision is struck through because that pair got confirmed in the "Revisions" panel. The decision itself isn't deleted — it stays. If a committee member asks "didn't you say B was the standard?", both that sentence and the 2024-03-11 sentence that replaced it are on screen, and the "Superseded by {note}" link opens the revising note. On the entity page, the "In your words" section shows the same sentences grouped into "Decisions", "Open questions", and "Claims", each one linked back to its source note.
What changes
| Before | Now |
|---|---|
| Reconstruct why you switched from memory, then hunt backwards for data | "Viability dropped into the 40s, so switching to C starting today" comes back verbatim from line 12 of the 2024-03-11 note |
| Walk 400 experiment records by date and open 37 grep hits one at a time | The @Protocol B entity page's "In your words" already has the decisions, open questions, and claims sorted by kind |
| January's decision and March's decision both surface in search, and you judge from memory which is current | The 2024-01-08 decision is struck through, and "Superseded by {note}" opens the 2024-03-11 note |
| Submit without knowing whether the two papers cited side by side contradict each other | The @IL-10 entity page puts both sentences side by side with their authors' names |
The work of the next four months doesn't shrink. The introduction's citation list still has to be fixed by hand, and whether the two papers actually contradict each other still has to be read and judged by hand. Revision pairs still only land once you press "Confirm". What changes is where the answer comes from. Answering one committee question means quoting line 12 of the 2024-03-11 note instead of walking 400 files.