kathryn-email-voice — EVIDENCE.md additions, 2026-08-17
What this is. Entries to APPEND to EVIDENCE.md in the kathryn-email-voice skill. That skill is synced to the machine from Kathryn's Claude account and exists in no repo, so an edit made in a session container is lost when the container is reclaimed and would be overwritten by the next sync. This is the durable copy.
To apply: open the skill's EVIDENCE.md and paste everything below the rule into it, then bump the frontmatter/QC version string in SKILL.md to 1.2.1.
Source round: Elite double-send, 2026-08-17 — Kelly 1:1 agenda (sent 4:48 PM ET) and Elite BK team prep (sent 5:26 PM ET as a reply on the 8/12 recap). 2 edits by Kathryn.
R5 · Scope limit — a hedge covering uncertainty is not an escape hatch
2026-08-17, Kelly pre-session agenda. One word.
Before: "Send it when you get a minute." After: "Send it if/when you get a minute."
R5 reads as a blanket ban on hedges, and on that reading "when" is the correct hardening. Earlier the same afternoon the line had been "If it's handy, send it along; if not, we'll pick it up later" — cut for exactly that reason after Kathryn: "What do you mean by this?"
The edit separates two things the rule currently conflates:
- A hedge that lowers a bar the recipient could clear → still banned.
- A hedge covering genuine uncertainty about whether the thing exists → hers.
The Mosiman transcript was named in Kelly's email and absent from her attachments. "When" assumes there is a file to send. "If/when" also covers the case where that FIT meeting was never recorded, and keeps Kelly from carrying an obligation she has to track.
Test: does the hedge protect them from a task that may be impossible, or from one they can do? The first is hers. The second is the ban.
R1 · New shape — allocating the session's time before the room does
2026-08-17, Elite BK prep, sent as a reply on the 8/12 recap.
Cut in full, nothing replacing it: "That's where most of our time goes tomorrow: dates for both, and whether the transcript turned up in the SharePoint folder."
The paragraph now ends at "Marc Kayem is what's left — 2 businesses, so 2 blueprints." Two errors in one sentence:
- "That's where most of our time goes" sets the agenda before the room is in it — designing their process, applied to the clock.
- "whether the transcript turned up in the SharePoint folder" is a reminder to Biff, sitting verbatim in his own action item in the recap quoted directly below it.
The reminder problem survives the removal of bullets. The round before, the same email carried 6 per-person bullets (Biff ×2, Brent ×2, Samantha ×2), cut after Kathryn: "what's with the line by line reminders on the bookkeeping agenda? Why not reply to last week's recap?" Removing the bullets and replying to the recap fixed the structure. One reminder still came through as prose. Check the prose after the list is gone.
R4 · The grouped person-line is hers
2026-08-17, same email, prior round.
Her prep emails to this group give each person ONE line carrying their items: "Biff — where Cam's response for Kayem to Brent stands; and confirm Kelly and Nicole have folder and tracker access." Arden split those into one bullet per item, citing R4's "one thought per bullet," turning 3 person-lines into 6 bullets and a prep note into a checklist of the team's outstanding homework.
R4's split rule governs a bullet already doing too much. It is not a licence to atomise her grouped person-line into a per-person checklist. When the grouped form is in the goldens, the grouped form wins.
Resolved structurally rather than by wording: sent as a reply to the recap that already lists every commitment, carried by one line — "Everything else from the recap below, we'll pick up in the session."
R7 · Scope limit — "tomorrow" is hers for a next-day send
2026-08-17. Three sends, zero edits to the relative time.
Sent by Kathryn, unedited:
- Tracy, 1:41 PM — subject "Tomorrow at 10 — quick setup checklist"; close "See you tomorrow,"
- Kelly, 4:48 PM — subject "Tomorrow at 10 — turning the monthly review into a skill"; body "Here's the plan for tomorrow"; close "See you tomorrow,"
- Elite BK team, 5:26 PM — "See you tomorrow at 1:00 PM Pacific."
R7's founding evidence is a case where the ask itself expired before delivery — an opener built on "tomorrow morning" for a meeting that would already have happened. That is the failure the rule exists for. It does not extend to a pre-session note sent the afternoon before, where the subject or the sentence carries the date or the time alongside.
check.py still flags these as REVIEW, which is right — the flag is cheap and the judgment is hers. What changes is the reading: a REVIEW on "tomorrow" in a day-before prep email is expected, not a defect to repair.
Proper Nouns · An attachment is a source; a filename is not
2026-08-17, Kelly agenda. "Mission Hope" was carried from the 8/11 Otter transcript into the subject line, then flagged to Kathryn twice as unverifiable rather than checked. A Gmail search returned only Kathryn's own 8/12 recap, and that was reported as "no written source anywhere in the mailbox."
Wrong. Gmail does not index text inside .txt attachments. The name sat in 2 of Kelly's own attachments the whole time — the Machabee and Olds FIT transcripts, where Elite's own people name it as one of 3 charities the firm supports, with Michelle's Place and Rose Again Foundation. It is a Temecula nonprofit providing transitional housing for single mothers. Not a client.
Two corrections:
- Verify a proper noun by opening the sources, not by searching a mail index. An attachment is a source; a filename is not.
- An unverified proper noun does not go in a subject line. Flagging it to Kathryn is not a substitute for checking it or cutting it.
Downstream: the cascade labelled it "Kelly's client" in the 8/11 digest — sessions/recap-2026-08-11-aos-digest.md, open question: "Is 'Mission Hope' the correct spelling of Kelly's client?" Nobody said "client" in the room. The mis-framing is still in the JSON and digest and will flow into the CPM and project plan unless corrected.
Version history addendum — 1.2.1
2026-08-17. Elite double-send round: Kelly 1:1 agenda + BK team prep. 2 edits by Kathryn, both narrowing rules that were over-applied.
- R5 gains a scope limit: a hedge covering whether the thing exists is hers, not an escape hatch ("if/when").
- R1 gains a shape: allocating the session's time before the room does.
- R4 gains a limit: "one thought per bullet" does not split her grouped person-line into a per-person checklist.
- R7 gains a scope limit: "tomorrow" in a day-before prep email is hers; the rule targets an ask that expires before delivery.
- Proper Nouns gains a verification standard: open the source, do not search a mail index.
Process defects from the same round, no rule change:
- The first Kelly agenda was built from the internal session-agenda file rather than the recap that actually set the session. The recap was only read after Kathryn asked "Did you go through the recap originally when putting together this agenda for Kelly? If you did, I'm very worried."
- A draft was rewritten 17 minutes after the client's reply landed, without re-checking the inbox. Re-check the client's mail before every rewrite, not only before the first draft. A draft is stale the moment the client replies.
- An email stated "the Mosiman transcript didn't come through" on the basis of reading attachment filenames off Gmail metadata. No attachment was opened. The claim held, verified afterwards against the Drive folder, but it was asserted before it was checked.
- Times were reported to Kathryn on the client's Pacific clock. The whole Elite team is Pacific; Kathryn is Eastern. Report times on her clock, name the zone.