Deposition compare

One witness gives an account. Several witnesses give a set of accounts that overlap, differ, and leave holes between them, and the differences are where a matter turns. This skill lays the accounts side by side, fact by fact, so the team can see for each point who said it, who said otherwise, who was never asked, and what the documents say. It states the differences; it does not decide who is right.

1. Inputs

• The testimony databases: depositions/<witness>-testimony.csv for each witness, with depositions/index.csv, as the deposition-summary skill writes them. If a transcript has no database yet, run that skill on it first; never compare from the summaries alone, and never from memory of a transcript.
• The archive (case-brief): to check statements against documents and to cite them.
• The matter's topic list (topics.md from the summaries, merged): one list for all witnesses. Where two summaries used different topic names for the same thing, map them once in topic-map.csv and use the merged names everywhere.
• The question, when the user has one: a date, a decision, an alarm, a conversation. The comparison then starts there and widens; without one, it covers every topic that two or more witnesses reached.

2. Build the fact matrix

The unit of comparison is a fact: a short, neutral proposition that a witness could affirm, deny, qualify, or not know ("the alarm sounded before the pump stopped"; "the service visit on March 4 included a software update"; "the risk file was reviewed after the first complaint"). Facts come from the testimony rows, not from the pleadings.

1. Collect from all databases the rows under each merged topic. Read them together, witness by witness.
2. Write the facts: for each point two or more witnesses addressed, one line in facts.csv: fact id · topic · fact (neutral proposition) · document evidence (what the archive says, with [D-0042 p.7], or none found with the search terms) · note.
3. Fill the matrix: matrix.csv, one row per fact, one column per witness, each cell one code with its cite:
4. A affirms — A [Smith 44:12–45:3]
5. D denies — D [Jones 102:7–11]
6. Q affirms with a qualification — Q [Lee 77:1–9] (only "to my knowledge")
7. E estimates or is unsure — E [Lee 80:2–6] ("probably", "I think")
8. K does not know — K [Smith 50:3]
9. R does not recall — R [Jones 110:14]
10. N not asked
11. X instructed not to answer / objection, no answer Plus a documents column with the same codes for the archive (A [D-0042 p.7], D [D-0051 p.2], none), and a status column: agree (all who answered gave the same code), differ (A against D or Q), uncertain (only E/K/R), one witness (only one answered), untested (all N).

The matrix is the product. Everything after it is a reading of it.

3. Find what the matrix shows

Work through the matrix and write the findings into tables, each line cited:

• Differences (differ rows): the fact, each witness's position with the verbatim words, the document evidence, and the nature of the difference — a different date, a different actor, a different sequence, presence against absence, knowledge against no knowledge. Order them by the weight of the fact in the matter (the user's question first).
• Witness against document: every cell where a witness's code differs from the documents column. Cite both. Say what the document is (a contemporaneous log, a later report, an email) so the reader knows what kind of evidence it is.
• Internal shifts: a witness who gave two different answers on one fact (the summaries flag these; collect them here with both cites).
• Agreement: agree rows, briefly; counsel needs to know what is settled as much as what is open.
• Vantage points: for the event itself, a short table of where each witness was, what they could see or hear, and from when to when, each cell cited. A difference between two witnesses who stood in different places is a different kind of difference from one between two who stood together.
• Knowledge and time (manufacturer or hospital witnesses): for each fact about a problem being known, who said they knew, when, from what, with cites, in date order — the raw material for the device-timeline skill's knowledge dates.
• Not asked: untested and one witness rows, grouped by topic; and the facts that only one side's counsel explored.

4. Verify

• Every cell cite opened against the testimony row and, for key rows, against the transcript page. A cell with a wrong cite is a wrong finding.
• Every differ row read again in full context (the question, the answer, three lines each side). A difference that disappears when the question is read is not a difference; delete it.
• Every document cite opened.
• The fact wording checked for neutrality: if a fact reads as one side's theory, rewrite it until both sides would accept it as the thing in dispute.

5. Write the output

comparison.md:

1. Scope: witnesses compared (name, role, date, pages), databases and archive used, the question if there was one, merged topic list.
2. In one page: the ten to fifteen differences that matter most, each in two or three lines with cites, and the facts on which the witnesses agree.
3. The matrix, by topic, as tables (fact · one column per witness · documents · status). Codes with cites; verbatim words in a footnote line under the row when they carry the difference.
4. Differences in detail: one short section per differ row of weight: the fact, each account in the witness's words, the documents, the nature of the difference.
5. Witnesses against the documents.
6. Vantage points and Knowledge and time tables.
7. Not asked, and questions for the next witness: by topic, written as questions, with the witness for whom each is intended and the cite that prompts it.
8. Sources.

Beside it: facts.csv, matrix.csv, topic-map.csv, differences.csv (fact id · witnesses · codes · cites · document evidence · nature), questions-next.csv (topic · question · for whom · prompted by).

Citation form: [Lastname 112:4–113:2] for testimony inside this document, [Depo Lastname 112:4–113:2] in anything that leaves it; [D-0042 p.7] or Bates for documents.

6. Refreshing

When a new deposition is taken: run deposition-summary on it, add its column to the matrix, re-run sections 3 to 5, and open comparison.md with a What changed section — facts the new witness affirmed, denied, or moved from one witness to agree or differ; questions from the previous Not asked list that were now asked, with the answers.

7. Rules

• Compare databases, not memories, and not summaries alone.
• A fact is neutral or it is rewritten.
• One code and one cite per cell; verbatim words where the difference lives in the words.
• A difference survives only if it survives the full context.
• No judgment of credibility, truthfulness, or motive, and no conclusion about which account is correct. The matrix shows the accounts and the documents; counsel and the finder of fact decide.
• Say what was not asked. The holes are findings.