Evidence Weigher
Updated
IntermediateWhat you'll learn
~50 min- Ask an AI tool, in plan mode, for a page that lays out every source for one disputed date
- Read its plan for what matters when sources disagree: both dates side by side, nothing that picks one, and your conclusion kept with what still contradicts it
- Weigh each source by its informant, the person who supplied its information, and what that person could have known first hand
- Count copies of one account as one witness, and decide yourself whether an account is independent of the others
- Classify one record twice: first hand for one fact, second hand for another
- Write a conclusion in your own words, together with what still contradicts it
The story, and what you will end up with
The family story says Walt was “lost on the twenty-second”. On 28 March 2026 the casualty file you requested in February arrives: the file of reports and letters kept about Walt’s death and burial. With it, your folder holds eleven records that give a date for the loss of his aircraft, TRAINING-AC-0527. Some say 21 March 1945. Others say 22 March.
The easy ways to choose are to trust the earliest record that names Walt, or to trust the latest, on the idea that later records correct earlier ones. Here both give 22 March. The question this lesson asks of each record is different: not when it was written, but who supplied its information, and what that person could have known first hand.
By the end of this lesson you will have a small page on your own computer, an evidence page, that sets the two dates side by side with every source under the date it gives, keeps copies of one account together, and holds a conclusion you write yourself, with what still contradicts it. You build it by asking your AI tool in plain English, and then you do the weighing.
Before you start
First, if your AI tool is still open, leave it: type /exit and press Enter (if you use Antigravity, press Ctrl+D with nothing typed instead), and wait until you see the $ or % prompt. The next line goes there, in the terminal, not in the AI tool. It makes a folder for this lesson and goes into it:
mkdir -p ~/archive-playbook/05-evidence-weigher && cd ~/archive-playbook/05-evidence-weigherStart your AI tool with the start line you kept from the setup lesson, the one with approval prompts off. Then ask it to set the folder up:
Set this folder up so we can undo any change you make, download https://learn.intellimetrics.net/starters/history/05/loss-date-sources.json into a data folder, and save a checkpoint.Look for it saying three things: the folder is set up with Git (the program that saves snapshots, from setup), the file is saved as data/loss-date-sources.json, and a first snapshot, your checkpoint, is saved. If it does not mention one of them, ask: “Did you save a checkpoint with the file in it?” If the download does not work, download loss-date-sources.json in your browser, check where it went (the browser’s list of downloads shows the folder), and tell your AI tool, for example: “loss-date-sources.json is in my Downloads folder. Copy that one file into a data folder here, and save a checkpoint.” In Windows PowerShell or a desktop app, “my Downloads folder” works as is; only say “my Windows Downloads folder” if you use Ubuntu (WSL) on Windows, which keeps its own Downloads folder apart from Windows.
Everything in the file is fictional, made up for this course.
Open this box once, because it says what each code used further down stands for.
ℹWhat is in loss-date-sources.jsonclick to expand
One disputed fact, DF-1: the date of loss of TRAINING-AC-0527 with Walter O. Engstrom aboard. Then eleven rows, one per record, in the order the records were written:
| Row | Record | What it is | Date it gives | Written |
|---|---|---|---|---|
| 1 | R-19 | Squadron operations log, kept by the operations officer | 21 March | 21 March 1945 |
| 2 | R-18 | Morning report, the unit’s daily list of changes in status | 22 March | 23 March 1945 |
| 3 | R-21 | Signed statement of Cpl. Leonard Dahlke, a crew member who jumped | 21 March | 2 April 1945 |
| 4 | R-20 | Loss report for the aircraft (its date line only) | 21 March | 5 April 1945 |
| 5 | R-24 p02 | Casualty file, page 2: notice to his mother that he is missing | 22 March | 7 April 1945 |
| 6 | R-23 | Kettle Ford Courier, Walt’s home-town newspaper | 21 March | 17 May 1945 |
| 7 | R-22 | Squadron history for March 1945 | 21 March | 30 June 1945 |
| 8 | R-24 p09 | Casualty file, page 9: identification memorandum | 21 March | 11 June 1946 |
| 9 | R-24 p12 | Casualty file, page 12: report of his final burial | 21 March | 22 May 1948 |
| 10 | R-27 | Letter of his sister, Lorraine Brekke | 22 March | 10 June 1983 |
| 11 | R-26 | County honor roll, online index | 22 March | 2009 (year only) |
Row 4 copies only the loss report’s date line. The full report, which you also hold, has a crew list that names Walter; it was compiled afterwards by someone who saw nothing first hand.
Each row has:
- the record’s short code (R-18, R-24 p02 for page 2 of the casualty file), title and reference;
record_date, when the record was written, anddate_given, the date it gives for the loss. They are different things;says, the record’s own words, exactly as written;- three first classifications, made by you earlier and yours to change.
source: Original, the record in its first form or a good image of it, or Derivative, a later version made from it.information: Primary, resting on first-hand knowledge, Secondary, resting on second-hand knowledge, or Undetermined, when you cannot tell who supplied it.testimony, how the informant came by the date: Seen, Told, or Later editorial (written afterwards from other records); informant: your note on who supplied the information and what they could know first hand;account_group: rows that repeat one person’s account share a group, AG-1 to AG-7.
Row 9 also has another_fact, because that page speaks to a second fact, Walt’s burial in 1948. At the end, each account group has independence, which starts at “Not assessed”, and a note; AG-4’s note reads “compiled from AG-1 and AG-3”.
REFERENCE DATA Row 2 excerpt from data/loss-date-sources.json open reference hide reference Reference example of one row in the downloaded JSON file your agent reads. You do not type or edit this JSON by hand.
{ "row": 2, "record": "R-18", "title": "Morning report, 23 March 1945", "reference": "TRAINING-MR-57-450323", "record_date": "1945-03-23", "date_given": "1945-03-22", "says": "Engstrom Walter O TRAINING-SN-20458[?] Sgt fr dy to MIA 22 Mar 45", "source": "Original", "information": "Secondary", "testimony": "Told", "account_group": "AG-2", "informant": "Unit clerk: records the date the crew was carried missing, not the date of loss."}A morning report is a unit’s daily report of changes in each soldier’s status. In this course fr dy to MIA is read as from duty to missing in action, and [?] marks a character that cannot be read.
Switch to plan mode
Your AI tool is already open. Switch to plan mode now: Shift+Tab until it says plan, or /plan (the keys for each tool). In plan mode it usually reads your file, asks its questions and shows you a plan before it builds anything; the order can vary.
The conversation
Each message below is an example. Say it your own way.
Say what you want
Say what disagrees, which file to use, and what you want to do with it:
My records disagree on when Walt's aircraft was lost: 21 or 22 March 1945. Help me build a small tool to lay out every source in data/loss-date-sources.json and write my own conclusion. Ask me what you need first.Your AI tool reads the file and comes back with questions. It may also volunteer a tally: how many sources give 21 March and how many give 22 March. A tally does not decide the question, because copies of one account would count as several witnesses, and what matters is who could know the date first hand. If it offers to put a tally on the page, say: “No count on the page, please.”
Answer its questions
Answer each in a sentence. These are questions AI tools often ask, with one way to answer each. Yours may ask them in a different order or show a list of choices:
| If it asks | You might say |
|---|---|
| A page in your browser, or something else? | A browser page is fine. |
| Should you mark which accounts are independent, or should it just show a table? | I’ll mark independence myself; no counts. |
| Where should your work be saved? | Save my work to a new file; leave my sources file alone. |
| What should the conclusion hold: a date, how confident you are, your reasons? | My own words, plus what still contradicts it. |
| Just this question, or any file like it later? | Just this question for now. |
| Row 9 also gives a 1948 burial date. Show it, or hide it? | Show it on its row, classified separately. |
If it asks several questions at once, answer the ones that matter to your research in a sentence, and for the rest say that its suggestions are fine. If it asks something not listed here, answer it the same way, in a sentence.
Read the plan
Next it shows you a plan: what it will build and how it will check it. Read it for six things:
- Every source is shown, with its own words, the date it gives, when it was written, and your informant note. Row 9’s second fact, the burial in 1948, is shown on its own line.
- The two dates sit side by side, every source under the date it gives.
- Nothing picks a date: no winner, no score, no count of sources per date.
- Copies of one account stay together, and whether an account is independent is your mark, with a place for your note on each account group. The page never marks one itself.
- Your conclusion is yours, and it is not kept unless you have also written what still contradicts it.
- Your file is read and left as it is.
A first plan often leaves out the side-by-side dates, or lets a conclusion be saved alone. Say so in a sentence, in your own words, for example:
Show the two dates side by side, and refuse a conclusion without what contradicts it.When the plan has all six, approve it: if your AI tool offers a choice to go ahead, pick it; otherwise say “Yes, go ahead and build it.” In Codex, press Shift+Tab first to leave plan mode. In Copilot, saying it brings up a choice to go ahead; pick that. Let it build. It may test what it built and tell you how to open it. When it has finished, ask it: “Did you install anything, and where?” Nothing, or something inside this lesson’s folder, is fine for these lessons. If it names anywhere else on your computer, ask it what that is for, and to remove it if you do not want it there.
Look at what it built
Ask it to open the page:
Open the evidence page in my browser.Your AI tool may open the page straight from its file, or start a small program on this computer that serves the page at a local address, one ending in a number, such as http://127.0.0.1:8000/ (127.0.0.1 is an address that means this computer); either is fine. If it gives you an address without opening it, copy the address into your browser’s address bar and press Enter. If it says the address is already in use, say: “Something else is using that address. Don’t stop it. Use a different address for my page.” Leave the program running while you work; when you are done, ask your AI tool to stop it. If your AI tool does not take a message after opening the page, wait a minute. If it still does not, close the terminal window, open a new one, type this lesson’s folder line again, start your tool, and say: “Start the page again so it keeps running while we talk.” The page and your saved work are files in the folder, so nothing is lost. Your page will be laid out in its own way.
Before you change anything, check that both dates are on the page, each with the rows that give it, that all seven account groups say “Not assessed”, and that the place for your conclusion and the place for what still contradicts it are both empty.
Weigh the sources yourself
Mark each account group yourself, in the steps below. If the page has no place for something below, such as a note for an account group, tell your AI tool in a sentence, for example: “There’s no place to write a note for each account group. Add one.”
-
One account, three copies. Find AG-1: rows 3, 6 and 7. Row 3 is Dahlke’s own signed statement; he was a crew member who jumped from the aircraft and saw it go down. Row 6, the Kettle Ford Courier (Walt’s home-town newspaper), quotes him, and row 7, the squadron history, retypes him, changing “about 1340” to “1340 hours” on the way. They are three records but one witness: the newspaper does not show a second person who saw the aircraft go down. So the 21 March column holds seven rows but fewer witnesses. AG-1 is the account group, the label the file gives to rows that repeat one person’s account. Your page may show the mark for it as a menu, a button or a box; if you cannot find where to set it, say: “Where do I mark an account group as independent?” Set AG-1 to Independent, with the note “Dahlke’s own observation from the aircraft; his statement cites no log”. Set AG-3, row 1, the operations log written on 21 March itself, to Independent too: “written on 21 March, while Dahlke was still missing”. (He walked out and returned on 1 April.)
-
A careful summary is not a new witness. Row 4, the loss report, was compiled by someone who saw nothing, from the operations log and Dahlke’s statement. Set AG-4 to Dependent (resting on accounts already on the page), or whatever your page calls not independent.
-
One record, two facts. Row 9 is page 12 of the casualty file, the report of Walt’s burial at Norrlid Cemetery on 22 May 1948. The funeral director, Arne Solheim, writes “I was present at the interment”, so for the burial the page is Primary and Seen. Its date of death, 21 March 1945, is copied from page 9, so for the date of loss it is Secondary and Later editorial. Give AG-5 the note “p12 copies p09’s date of death” and leave it Not assessed: page 9 does not say what its date rests on, which is also why row 8 is Undetermined.
-
Write your conclusion. Write it in your own words. Any conclusion here must contain three things: the date you conclude, the records it rests on, and what still contradicts it. Here is one example that the rows support:
21 March 1945, from the operations log of that day (R-19) and Dahlke's statement (R-21), with the loss report (R-20) listing Walter in the crew. 22 March is the date the crew was carried missing (R-18), repeated by R-24 p02.Before you save, name one limit. The operations log names the aircraft, not its crew, and Dahlke did not see who was aboard at impact. What places Walt aboard is the full loss report’s crew list, compiled afterwards and second hand (your file’s row 4 holds only its date line), and the morning report that carries him missing with the crew. Put that limit in your conclusion or in what still contradicts it.
Type your conclusion, leave the place for what still contradicts it empty, and save. The page should refuse, and nothing should be saved. Then, in the place for what still contradicts it, list the records that disagree with you and save again. For the example above, that is:
R-18 and R-24 p02 (explained above), R-27, R-26.
If saving downloads a file, tell your AI tool its name and that it is in your Downloads folder, for example: “I saved loss-date-review.json; it’s in my Downloads folder. Move that one file into this folder’s data folder.” Use the name your page gave the file.
Notice what the order of writing tells you
Ask:
Add a way to show the sources in the order the records were written, oldest first.Read down that order before you read on. Row 1, the operations log of 21 March, names no one aboard. The first record that names Walt is row 2, the morning report of 23 March, and it gives 22 March. Its informant note says why: the unit clerk, who was not aboard, recorded the day the crew was carried missing, not the day the aircraft went down. So row 2 is an Original record carrying Secondary information. Row 4, the loss report, was typed fifteen days later than the loss and gives 21 March. Row 10, the 1983 letter of Walt’s sister Lorraine, gives 22 March again: she was at his homecoming in 1948, but second hand for everything overseas.
The earliest record that names him gives the wrong date for the loss, a later one gives the right one, and the latest give the wrong one again.
Closeness to the event is weight, not a verdict; a later record can be right, and an original record can carry secondary information.
Check it for these
- Before you changed anything, the 21 March column listed rows 1, 3, 4, 6, 7, 8 and 9, the 22 March column rows 2, 5, 10 and 11, and nothing on the page counted them, picked one or scored one.
- Rows 3, 6 and 7 sit together as AG-1, one of seven account groups for eleven rows, and every group started as Not assessed.
- Row 9 shows Secondary and Later editorial for the date of loss and, apart from that, Primary and Seen for the burial on 22 May 1948.
- In the order of writing, R-18 comes before R-20, R-27 after both, and R-26, dated only 2009, last.
- Saving with nothing in the place for what still contradicts it was refused, and your saved work keeps what still contradicts it.
- When you ask “Has my sources file changed since the checkpoint?”, your AI tool compares it with the checkpoint and says no.
If something on the list is not true, tell your AI tool in a sentence what you see and what you expected, for example: “Row 9 shows only one classification. Show its burial date on its own line, with its own Information and Testimony.”
To see what you saved, ask: “Show me my saved conclusion and what still contradicts it, exactly as stored.”
One real run
Here is part of what one AI tool said when asked that, with Claude Code on 10 November 2026. Yours will word it differently; check your page against the list above, not against this.
One thing to check against your records: your conclusion says the lossreport (R-20) lists Walter in the crew. The sources file's entry for R-20only says "Date of loss: 21 March 1945" and has no crew list. If you tookthat from the full report, the file just doesn't record it.Fictional training example.
Your AI tool is right about the file: row 4 holds only the loss report’s date line. The crew list is in the full report, and it is the limit you named in step 4. Tell it where the fact comes from, for example: “The crew list is in the full loss report, not in this file.”
How real archives differ
The records, the account groups and the Testimony field belong to this course; the words Original, Derivative, Primary, Secondary and Undetermined come from a published model.
- The evidence-analysis model taught by Elizabeth Shown Mills defines primary information as “information based on firsthand knowledge” and secondary information as “information based on secondhand knowledge”, and adds that, without knowing who supplied the information, “the quality of our information remains undetermined” (Evidence Explained, QuickLesson 17). That model has three classes of source, not two: “Sources may be an original record, a derivative record, or an authored narrative”. A compiled history such as row 7 may be better called an authored narrative, a work its writer composed from other records; this page records only its retyped Dahlke passage, as Derivative. Seen, Told and Later editorial are this course’s own shorthand, not part of the model.
- The same model also sorts evidence as direct, indirect or negative, and warns that “negative evidence is not the same as negative findings” (same page). The evidence page stops at sources and information; that third question stays with you.
- The Board for Certification of Genealogists lists five components of the Genealogical Proof Standard, among them “4. Resolution of conflicting evidence” and “5. Soundly written conclusion based on the strongest evidence” (BCG, ethics and standards). A conclusion that names what still contradicts it is a step toward that, not the whole standard.
Keep going
Small changes do not need plan mode. Say each in a sentence:
- “Give me a way to add a source to this list.” Before you rely on it, try it with one more made-up row.
- “Keep every conclusion I save, with its date and time, instead of replacing the last one.”
- “Give me a plain-text copy of the sources, my marks and my conclusion that I can print.”
- “Save a checkpoint.” Your AI tool saves a snapshot you can return to (people who use Git call this a commit).
When you have your own records that disagree, weigh them in the finished page. It runs on your real records by itself, with your AI tool closed. Do these steps in order; the first two happen now, while this folder still holds only practice data:
- Go through the data check, the questions in setup about what the material is, who decides how it is used, and which AI tool would see it. Go on only if it allows this use.
- While this folder still holds only the practice files, ask your AI tool: “How do I open the evidence page myself on a file in another folder, and save my conclusion there too?”
- Write down what it tells you.
- Ask your AI tool to stop the page’s program, then leave the tool: type
/exitand press Enter (if you use Antigravity, press Ctrl+D with nothing typed instead). - Keep your real files in a folder of their own, outside this course folder, and do what you wrote down. Your conclusion is saved beside your real files, not in the course folder.
Try another case
Before the quiz, write two or three sentences for yourself. Two records give a great-grandmother’s birth date. A death certificate from 1931, whose informant was her son, says 3 May; a church register written the week she was baptized says 30 April. A cousin says the certificate is official, so it wins. What would you ask of each record before choosing, and what would your conclusion keep?
A good answer asks who supplied each date and what they could know first hand (the son could not know his mother’s birth first hand), does not let “official” decide alone, and keeps the other date in what still contradicts the conclusion.
Key takeaways
- Ask briefly, and let your AI tool ask back. Say what disagrees, which file to use, and what you want; its questions fill in the rest.
- Read the plan for the research, not the code: both dates side by side, nothing that picks one, your marks your own, your file left alone.
- Ask who supplied a date, and what they could know first hand: neither the earliest nor the latest record decides.
- Copies of one account are one witness: three records quoting Dahlke are one account, not three agreements.
- Classify information per fact: one page can be first hand for the burial and second hand for the date of death.
- A conclusion says what still contradicts it: the records against you stay in the write-up, with your reason for setting them aside.
A town fire is dated by three newspapers in three towns, all giving 14 May. Each article quotes the same telegram. A shopkeeper's diary, written the evening of the fire by a man who watched it, gives 13 May. A cousin says: 'Three papers against one diary. The papers win.' What is the careful reading?
What’s next
The loss report lists Walter O. Engstrom with a service number. A draft card, two enlistment index entries, the morning report and the honor roll carry identifiers too: not always the same number, and not always the same Walter. Next you build Same Name, Same Person?, which compares records by identifier, name and household, notes every character you cannot read for certain, and leaves the decision to link two records with you.