One meeting, eight write-ups

Every notetaker in this category turns a recording into a summary. yapyap turns a recording into whatever you point at it, as many times as you like, from the same tape, without recording anything again.

That sentence is easy to write and hard to picture, so here is the whole thing done once. One conversation, eight lenses, outputs shown.

About the sample. The recording below is staged. Three of us read a script into three microphones so that this post could exist without publishing anybody's real client call. The transcript is real output from the pipeline; the conversation it captures was performed. Everything after it is the actual shape each lens produces, field for field.

The recording

Eleven minutes, three speakers, a stalled data migration. The transcript arrives with a turn id and a speaker on every line, which is the format every lens reads and cites back — and getting those names right is most of the work behind everything below.

[t01 | Priya] Right. We're four weeks past the date we gave the board on the migration, and I need to walk out of here knowing why.
[t02 | Dana] The why is the address table. We found the duplicate records on the third of March, during the first dry run, and we've been reconciling ever since.
[t03 | Marco] To be fair, we flagged the address table in the January audit. It's in the report.
[t04 | Priya] It is, and nobody read it. That's on us.
[t05 | Dana] There are about forty thousand rows where the same customer appears twice with different postcodes. We can't just pick one.
[t06 | Marco] You can if you take the most recently updated. It's not perfect and you'd carry maybe a two per cent error rate.
[t07 | Priya] Two per cent of forty thousand is eight hundred wrong addresses. That's eight hundred angry people.
[t08 | Dana] It's eight hundred angry people either way. Doing nothing is also a decision.
[t09 | Priya] Fine. Let's go with most-recently-updated, and we flag anything where the postcodes are in different regions for manual review. Agreed?
[t10 | Marco] Agreed. That's maybe three thousand rows for a human.
[t11 | Priya] Dana, can you get me the count of cross-region duplicates by Thursday?
[t12 | Dana] Thursday's fine.
[t13 | Marco] I'll write up the reconciliation rule so it's documented before we run it. End of week.
[t14 | Priya] And we're not moving the board date again. I'd rather ship with eight hundred known errors than move it a third time.
[t15 | Dana] What do we tell the eight hundred?
[t16 | Priya] I don't know yet. Park it, we'll come back to it.

Summary

Headline, framing line, takeaways, prose, and only the questions the participants actually left open.

address table reconciliation Priya, Dana and Marco settled a duplicate-record rule to unblock the migration without moving the board date.

• Forty thousand address rows contain the same customer twice with different postcodes, found during the first dry run.
• The agreed rule is most-recently-updated, with cross-region postcode mismatches held back for manual review.
• Roughly three thousand rows are expected to need a human.
• The board date is not moving again, and an estimated eight hundred wrong addresses is the accepted cost.

The meeting opened on the four-week slip against the date given to the board. Dana traced it to duplicate records in the address table, found on the third of March during the first dry run […]

open questions

• What to tell the customers whose addresses are wrong after the migration.

Actions

Only things with both a named owner and a doable deed. Deadlines exactly as spoken — "by Thursday" is stored as "by Thursday", never converted into a date nobody said.

action	owner	due	source
Get the count of cross-region duplicate addresses	Dana	by Thursday	t11
Write up the reconciliation rule before it is run	Marco	end of week	t13

Note what is missing. "Park it, we'll come back to it" is not an action. "We flagged the address table in the January audit" is not an action. Neither has a deed somebody committed to doing next.

Decisions

A decision is something the room is now bound to, with the reason only when the reason was said out loud.

decision	why	source
Use most-recently-updated to resolve duplicate address rows, with cross-region postcode mismatches held for manual review	doing nothing is also a decision	t09
Do not move the board date again	shipping with eight hundred known errors is preferable to a third slip	t14

Rationale is null when nobody voiced one. It is never reconstructed, which means an empty "why" column is information rather than a gap.

Brief

The one you forward. Written for somebody who was not in the room.

The one a consultant or agency lead forwards on Friday.

Headline — address table duplicates and the board date

Context — Four weeks past the migration date given to the board, the team met to establish the cause and agree a way through. Duplicate address records surfaced in the first dry run are the blocker.

Outcomes — Cause identified as ~40,000 duplicate customer rows with conflicting postcodes · Resolution rule agreed: most-recently-updated, with cross-region mismatches held for manual review · Expected manual review volume ~3,000 rows · Board date confirmed unchanged

Next steps — Count of cross-region duplicates (Dana, by Thursday) · Reconciliation rule written up before the run (Marco, end of week)

Open items — What to communicate to customers affected by remaining address errors

Chronology

This one is not about the meeting. It is about the events the meeting describes, rebuilt in the order they happened. It is the pass an investigator or a journalist starts from.

when	event	who	source
the January audit	address table flagged as a risk in the audit report	Marco	t03
the third of March	duplicate address records found during the first dry run	Dana	t02
after the dry run	reconciliation work ongoing	Dana	t02
by Thursday	count of cross-region duplicates due	Dana	t11

Time references stay exactly as spoken. "The third of March" is not resolved to a year, because nobody said one.

Quotes

Verbatim, attributed, with a jump back to the moment.

"It's eight hundred angry people either way. Doing nothing is also a decision." — Dana, t08

"Two per cent of forty thousand is eight hundred wrong addresses." — Priya, t07

"It is, and nobody read it. That's on us." — Priya, t04

"I'd rather ship with eight hundred known errors than move it a third time." — Priya, t14

Themes

The analysis pass. Two to six recurring subjects, named in the participants' own words, each with the one line that carries it.

accepting a known error rate — Marco proposed a rule with a two per cent error rate; Priya priced it in human terms and accepted it rather than slip again. "It's eight hundred angry people either way." — Dana, t08

the audit nobody read — The blocker was documented in January and missed. Priya took it without deflecting and the room moved on. "It is, and nobody read it. That's on us." — Priya, t04

what the customers are told — Raised at the end and explicitly parked; the only thread that leaves the room unresolved. "What do we tell the eight hundred?" — Dana, t15

Show Notes

And here is the one that does not work, shown rather than hidden.

Show Notes drafts a listener-facing episode description, a chapter outline and pull quotes. Run against a client meeting it dutifully produces all three, and what comes back is a description of a podcast that does not exist, chaptering a conversation that has no chapters, selling an episode nobody will listen to.

In this episode: a migration four weeks late, forty thousand duplicate rows, and a team deciding what error rate they can live with.

That is not a bug in the lens. It is a lens built for a different input, doing its job on the wrong tape. It exists because podcasters are one of the jobs yapyap does well, and it is in this post because a demo that only shows the flattering seven would be worth less than one that shows all eight.

Why this is the whole argument

Eight outputs. One recording, recorded once, transcribed once, never sent anywhere. Nothing above required going back into the room, re-recording, or re-uploading. Every one of them cites the turn it came from, so every claim in every write-up is one click from the moment somebody actually said it.

And when none of the eight is the write-up you need, the format is open: a lens is a prompt, a declared shape and a view, all editable in the app. The lenses page is where the ninth one starts, and why we built it this way is the shorter argument for the same thing.