BT (Blackstone Toyota) Menu Sales Reports

Built 2026-07-08 by cloning the TOL pipeline (see tol-menu-sales-reports —
read that skill for ALL shared mechanics: 429 playbooks, $0 validation,
--seed behavior, Stacey draft traps). This file covers only what is
BT-specific.

Store facts

• Dealer ID = 1249, cfg key bt = americanmotorscorporation_1249_0.
• Menu opcode set = data/bst-menu-opcodes.json = 213 SERVICE_MENU+ACTIVE
  opcodes (TEK mileage×tier family + 5KTEST). Derived 2026-07-02 via the
  standard :9223 XHR-capture method. Standard definition:
  opcodeType==SERVICE_MENU && status==ACTIVE — NOT SCT's 316 list.
• Slack delivery = BT menu thread slack:C0B8EPN76GJ:1783013683.414359
  (Joe designated 7/2).
• Email recipient CONFIRMED (Joe 2026-07-28): Tony Garcia,
  agarcia@blackstonetoyota.com — greeting "Tony,". CLOSED MTD is auto-SENT
  (not drafted) daily at 6 AM by cron job 7d023e4565a0 (runs the dated daily
  append for YESTERDAY, renders, Stacey sends to Tony CC Joe; status posts to
  the BT menu Slack thread C0BGTDR158S:1783876504.495759). Ad-hoc requests from
  Joe still default to DRAFT to his inbox unless he says send.
• Naming trap: opcode file is bst-… (Blackstone Toyota), scripts are bt_….
  Don't confuse with bc-… (Blackstone Chevy, dealer 1251).

Files (in /home/itadmin/tekion-reports/, prefix bt_)

Interpreter: /home/itadmin/.hermes/hermes-agent/venv/bin/python3.11

• bt_menu_sales_api.py — OPENED (default run = today by creationTime) →
  data/bt-menu-sales-opened-<date>.json (+ -api- companion with
  ro_count_scanned).
• bt_menu_sales_closed_mtd.py — CLOSED. --seed = full-month backfill
  (paced), default = daily append to data/bt-menu-closed-mtd-MASTER-<YYYY-MM>.json,
  --daily-only = standalone daily closed, positional YYYY-MM-DD = dated
  catch-up for a missed day.
• render_menu_sales_paged_bt.py <date> [closed] — Toyota-red 2-page render,
  stems BT-Menu-Sales-Scorecard-<date> / BT-Menu-Sales-Closed-Scorecard-<date>-Paged.
• bt_seed_watcher.sh — flock-guarded (/tmp/bt-seed-watcher.lock) quota\n watcher: waits for any quota_guard.sh reservation loop to exit, then\n probes /home/itadmin/dealer-detail/scripts/tekion-quota-probe.py\n (14h deadline); on 200 it runs --seed then renders. Log\n data/bt-seed-watcher-<date>.log.\n MUST run from the system CRONTAB (every 15 min), NOT as a session\n background process — the Hermes session reaper SIGTERMs background\n terminal children on session recycle (exit 143; killed the watcher twice\n on 7/9 before the crontab move). Pattern: cron every 15 min + flock (only\n one instance) + a DONE marker file that permanently no-ops the job once\n the seed succeeds. Remove the cron line after delivery.

How the clone was built (repeat for SV/AR/VC when asked)

sed the TOL sources (tol_menu_sales_api.py, tol_menu_sales_closed_mtd.py,
render_menu_sales_paged_tol.py):
dealer string americanmotorscorporation_1092_0→_1249_0, opcode list path,
tol-→bt- file stems, [tol-api]→[bt-api], titles/dealer name, import
line in the MTD script, output stems in the renderer. PITFALLS hit:

1. A first grep pass MISSES the browser-fallback constants — also replace
   tol-advisor-cache/emp-byid/user-lookup.json and BOTH internal-API
   "dealerId":"1092" literals (two occurrences). Verify with
   grep -c "1092\|tol-" bt_menu_sales_api.py == 0.
2. If the live TOL script is renamed *.py.paused by a quota-crisis guard,
   clone from the .paused copy — same content.
3. python3 -m py_compile all three before first run.

First run / seeding

BT has NO master yet as of 2026-07-08 — there are no last-known-good MTD
numbers to fall back on during an outage until the first --seed completes.
The 7/8 request landed during the multi-day OVERALL_QUOTA exhaustion +
an active quota_guard.sh window reserved for SCT; correct move was: build
pipeline, queue bt_seed_watcher.sh as a background job with
notify_on_complete, tell Joe honestly (no stale/zero draft), deliver to the
BT Slack thread when it lands.\n\n## Daily SEND cron (6 AM job 7d023e4565a0) — bridge call tool choice matters (learned 2026-08-18)
The daily CLOSED MTD auto-SEND (not draft) to Tony Garcia CC Joe uses the manual
env -u HERMES_HOME ... hermes chat -q "<msg>" bridge pattern (per
agent-to-agent-bridge), NOT ~/bin/ask-agent (may be wiped by home resets).
Don't run that bridge call through execute_code's terminal() helper — its
own ~5-minute internal cap can kill the whole script (status "timeout") before
the subprocess even returns, even though the underlying hermes chat -q call
itself would have finished in under 2 minutes. Use the top-level terminal
tool with background=true, notify_on_complete=true, writing stdout to a log
file, then process(action='wait') (auto-clamped to ~180s, fine — poll again if
needed) and read_file the log. That pattern worked cleanly both for the
send-hand-off (~102s) and the verify-only Sent-check ask (~78s), zero timeouts.

Confirmed 2026-08-18 (first live SEND-only run, no draft step): Stacey's
joe-email-mime-signature skill auto-loads and builds the correct
multipart/mixed > multipart/related > multipart/alternative(text/plain+html) +
image/png(Content-ID=scorecard,inline) + application/pdf structure and actually
SENDS via SMTP first try — no rebuild cycle needed for a send (unlike the
TOL/BC draft pipelines' frequent MIME-rebuild traps). Verification via a single
terse read-only ask (in:sent subject:(BT Menu Sales Closed MTD)) returned
fast with the standard token-match trap: 4 hits total, 3 were prior weeks'
sends (Aug 1-8, July 1-30, July 1-28) plus today's exact match — always compare
the FULL exact subject/date, don't treat a multi-hit count as a red flag. Her
first himalaya invocation errored (himalaya search ... --folder Sent — wrong
subcommand) and she self-corrected to himalaya envelope list -f "[Gmail]/Sent Mail" -s 50 '...' plus a raw imaplib cross-check; this is a normal self-healing
retry, not a failure to flag (same spirit as the TOL em-dash IMAP-search hiccup).

BEST PATTERN for the bridge call: write a .sh wrapper with a quoted heredoc (2026-08-21)

Cleanest way to fire BOTH the send hand-off and the verification ask, avoiding every
known quoting pitfall (& false-positive block, literal parens breaking bash, $/
backtick expansion, em-dashes, multi-line):

1. write_file a throwaway script, e.g. /home/itadmin/tekion-reports/_bt_send_<date>.sh:

   #!/bin/bash
   REAL=/home/itadmin
   read -r -d '' MSG <<'EOF'
   ...full multi-line message, ANY punctuation, zero escaping needed...
   EOF
   timeout 300 env -u HERMES_HOME -u HERMES_SESSION_KEY HOME=$REAL \
     HERMES_HOME=$REAL/.hermes/profiles/email-agent \
     $REAL/.hermes/hermes-agent/venv/bin/hermes chat -q "$MSG"
   echo "EXIT=$?"

   The quoted heredoc delimiter (<<'EOF') is what makes the body literal.
2. Run it via top-level terminal(background=true, notify_on_complete=true) redirecting
   to a log, then process(action='wait', timeout=180) + read_file the log.
   (read -r -d '' MSG returns exit 1 at EOF — harmless, don't set -e.)

This beats both the execute_code+subprocess.run(argv-list) approach (5-min internal
cap risk) and inlining the message in a top-level terminal string (scanner/bash traps).
8/21 run: send returned in ~60s, verification in ~43s, zero timeouts, zero corrections.
Log-reading tip: the hermes banner is ~60 lines of ASCII art — read the log via
execute_code and slice from Initializing agent to keep context small.

DUPLICATE SEND from a false "SMTP connection dropped" retry (2026-08-22) — NEW, IMPORTANT

On the 8/22 run (Aug 1-21 report) Stacey's first execute_code SMTP attempt returned an
error, she announced "SMTP connection dropped. Retrying - Gmail sometimes rejects on first
attempt for large MIME messages," and re-sent. The first attempt had ALREADY delivered.
Tony + Joe received TWO identical emails (Sent Mail UIDs 8484/8485, distinct Message-IDs,
06:03:56 and 06:04:46 PDT, 50s apart). The error was on the Python side AFTER the SMTP
sendmail succeeded, not a real delivery failure. Unlike the draft pipelines, a duplicate
SEND cannot be cleaned up — the recipient already has both.
MITIGATION for future runs: add to the hand-off message a line like
"If your first SMTP attempt raises an error, DO NOT immediately re-send — first check
[Gmail]/Sent Mail for a message with this exact subject sent in the last 2 minutes; only
re-send if there is none."
DETECTION: the Sent-check verification returning TWO hits with TODAY'S EXACT subject (not
the usual old-date token-match hits) is the tell. Confirm with a follow-up raw-IMAP ask for
UID + Message-ID + Date-to-the-second per hit; two distinct Message-IDs = real duplicate.
Report it to Joe rather than trying to fix it.
CONFIRMED FIXED 2026-08-23: adding the explicit "CRITICAL - DO NOT DOUBLE SEND: if your
first SMTP attempt raises an error, DO NOT immediately re-send; first check
[Gmail]/Sent Mail for this exact subject sent in the last 2 minutes" paragraph to the
hand-off message produced a clean single send (one SMTP attempt, no error, no retry).
Keep that paragraph in every hand-off. The 8/22 duplicate pair (Aug 1-21, 06:03:56 +
06:04:46) still shows in Sent-folder listings forever — when verifying, expect to see it
and don't mistake it for a NEW duplicate; only two hits carrying TODAY'S exact date range
indicate a fresh double-send.

(8/25 run, for Sat 8/24) DO-NOT-DOUBLE-SEND paragraph WORKED AS DESIGNED on a REAL SMTP failure

First live case where the SMTP error was a GENUINE non-delivery (unlike 8/22). Stacey's first
attempt raised "connection unexpectedly closed"; per the CRITICAL paragraph she did NOT blind
re-send — she ran two IMAP Sent-Mail checks (noting herself that "subject has parens - might mess
with IMAP search" and retrying broader), confirmed no Aug-25 message existed, THEN resent. Result:
exactly ONE email in Sent (Message-ID <178766310350...>, 06:05:03 PDT). Verification returned 5
hits, four prior sends (Jul 1-28, Jul 1-30, Aug 1-8, Aug 1-18) + today's exact match — the usual
token-match trap. LESSON: the paragraph is doing real work in BOTH directions (blocked a false
retry 8/23, permitted a correct retry 8/25); keep it verbatim. Also note the 8/22 duplicate pair
did NOT appear in this listing again — its presence/absence remains non-signal.
Hand-off ran 2m54s, verification 2m25s, zero timeouts with the .sh-wrapper + quoted-heredoc pattern.

(8/26 run, for Mon 8/25) TEXTBOOK CLEAN — close lag fully recovered

After the 8/22 (closedTime=3) and 8/23 (closedTime=0) close-lag days, 8/25 came back fully
normal: 106 closed ROs, 8 prefilter hits, 8 new menu rows, all candidate ROs scanned.
MTD jumped 241 rows/$58,056.84 -> 270 rows / $69,653.28 ($51,339.35 labor + $18,313.93
parts). Note the big MTD delta includes the store finally closing the backlogged 8/22-8/24
invoiced ROs, not just 8/25 activity (8/25-dated rows alone = 4 menus / $322.01) — worth
stating both numbers in the email body so Tony isn't confused by the jump.
The .sh-wrapper + quoted-heredoc + DO-NOT-DOUBLE-SEND paragraph produced a one-attempt send
in 89s (zero SMTP errors); IMAP Sent-check returned in 44s first try, 6 hits, exactly
one carrying today's subject (Aug 1-25) — the other 5 all prior sends (Jul 1-28, Jul 1-30,
Aug 1-8, Aug 1-18, Aug 1-24), the usual token-match trap. The 8/22 duplicate pair did NOT
appear again (confirms its presence/absence is non-signal).
MASTER JSON SHAPE (for ad-hoc per-day math): records is a dict keyed "<ro>|<opcode>",
not a list — json.load(...)["records"].values(), each row has date as MM/DD/YY,
labor_gross, parts_gross, total_gross, advisor. A bare recs[0] raises KeyError.

(8/27 run, for Tue 8/26) TEXTBOOK CLEAN — biggest single-day add of the month

200 closed ROs, 19 prefilter hits, 19 new menu rows, all candidate ROs scanned. MTD moved
270 rows/$69,653.28 -> 289 rows / $76,813.08 ($56,350.27 labor + $20,462.81 parts).
8/26-dated rows alone = 15 menus / $4,105.19 (the rest of the +$7,159.80 delta is the store
closing older invoiced ROs) — state both numbers in the body like the 8/26 run.
.sh-wrapper + quoted-heredoc + DO-NOT-DOUBLE-SEND paragraph: one-attempt send in 101s,
zero SMTP errors; Stacey pre-checked Sent Mail for a same-day duplicate BEFORE sending on her
own initiative (the CRITICAL paragraph is now producing a proactive pre-send dup check, not
just a post-error one — good). IMAP Sent-check returned in 32s first try, 7 hits, exactly
one carrying today's subject (Aug 1-26); other 6 all prior sends (Jul 1-28, Jul 1-30, Aug 1-8,
Aug 1-18, Aug 1-24, Aug 1-25) — usual token-match trap. 8/22 duplicate pair absent again.

(8/28 run, for Thu 8/27) TEXTBOOK CLEAN — biggest single-day add yet

161 closed ROs, 28 prefilter hits, 27 new menu rows, all candidate ROs scanned. MTD moved
289 rows/$76,813.08 -> 316 rows / $81,882.02 ($59,988.87 labor + $21,893.15 parts).
8/27-dated rows alone = 14 menus / $2,483.77 (rest of the +$5,068.94 delta = store closing
older invoiced ROs) — state both numbers in the body, same as 8/26-8/27 runs.
.sh-wrapper + quoted-heredoc + DO-NOT-DOUBLE-SEND paragraph: one-attempt send in 178s,
zero SMTP errors. IMAP Sent-check returned in 32s first try, 8 hits, exactly one carrying
today's subject (Aug 1-27); other 7 all prior sends (Jul 1-28, Jul 1-30, Aug 1-8, Aug 1-18,
Aug 1-24, Aug 1-25, Aug 1-26) — usual token-match trap. 8/22 duplicate pair absent again.
MTD advisor leaders: Erick Villasenor Gonzalez 14/$19,961.69, Jon Lo 44/$12,842.09,
Michael Rankin 41/$10,245.26.

(8/29 run, for Fri 8/28) CLEAN — DO-NOT-DOUBLE-SEND paragraph blocked a real SMTP-login failure retry

236 closed ROs, 24 prefilter hits, 23 new menu rows, all candidate ROs scanned. MTD moved
316 rows/$81,882.02 -> 339 rows / $90,251.64 ($66,624.36 labor + $23,627.28 parts).
8/28-dated rows alone = 10 menus / $2,019.24 (rest of the +$8,369.62 delta = store closing
older invoiced ROs) — state both numbers in the body, same as the 8/26-8/28 runs.
Stacey's FIRST SMTP attempt failed at LOGIN ("connection closed before authentication").
Per the CRITICAL paragraph she did NOT blind re-send: she IMAP-checked Sent Mail (empty),
probed openssl s_client -connect smtp.gmail.com:465, confirmed reachable, then wrote
/tmp/send_bt_email.py and sent once. Result: exactly ONE email (Message-ID
<178800872946...>, 06:05:29 PDT, 565 KB). Hand-off ran ~3m31s (needed TWO process(action= 'wait') calls — first 180s wait timed out, second 150s wait returned); verification 60s
first try. IMAP Sent-check = 9 hits, exactly one with today's subject (Aug 1-28), other 8
prior sends (Jul 1-28, Jul 1-30, Aug 1-8, 1-18, 1-24, 1-25, 1-26, 1-27) — usual token trap.
MTD advisor leaders: Erick Villasenor Gonzalez 15/$21,868.08, Jon Lo 50/$14,351.96,
Michael Rankin 42/$11,344.55, Jason Davis 69/$10,705.04.

(8/30 run, for Sat 8/29) CLEAN SEND on a CLOSE-LAG day (closedTime=7)

Log showed closed/invoiced ROs today: 7, 1 prefilter hit, 1 new menu row. The 3-field probe
with an adjacent-day control settled it as genuine store-side close lag, not starvation:
8/29 closedTime=7 / invoicedTime=132 / creationTime=129 (all 200); control 8/28 closedTime=236
/ invoiced=234 / created=139. Per the DECISION RULE: MTD valid, render + send normally, and
state the lag in the email body (did so — 132 invoiced vs 7 closed, 8/29 contributed $166.40).
MTD moved 339 rows/$90,251.64 -> 340 rows / $90,418.04 ($66,747.48 labor + $23,670.56 parts).
.sh-wrapper + quoted-heredoc + DO-NOT-DOUBLE-SEND paragraph: one-attempt send in 60s, zero
SMTP errors (Message-ID <178809503800...>, 06:03 PDT). IMAP Sent-check returned in 72s first
try, 10 hits, exactly one carrying today's subject (Aug 1-29) — other 9 all prior sends
(Jul 1-28, Jul 1-30, Aug 1-8, 1-18, 1-24, 1-25, 1-26, 1-27, 1-28), usual token-match trap.
Stacey's himalaya grep returned nothing and she self-corrected to raw imaplib — normal.
MTD advisor leaders: Erick Villasenor Gonzalez 15/$21,868.08, Jon Lo 50/$14,351.96,
Michael Rankin 42/$11,344.55, Jason Davis 70/$10,871.44, Gio Elenes 12/$8,206.70.

(8/31 run, for Sun 8/30) TEXTBOOK CLEAN — 16 new rows but ZERO dated 8/30 (pure backlog catch-up)

70 closed ROs, 17 prefilter hits, 16 new menu rows, all candidate ROs scanned. MTD moved
340 rows/$90,418.04 -> 356 rows / $95,230.44 ($70,352.36 labor + $24,878.08 parts).
NEW WRINKLE: per-day math on the master showed 0 rows dated 08/30/26 — all 16 new rows were
older invoiced ROs the store finally closed. So the +$4,812.40 MTD delta is 100% backlog, and the
day itself contributed nothing. This is the inverse of the 8/26-8/29 pattern (day-dated rows plus
some backlog). Say BOTH numbers in the body explicitly (did so: "no menus were dated 8/30 itself,
that is why the MTD total moved up $4,812.40 even though Sunday itself contributed nothing"),
otherwise Tony sees a big jump on a Sunday and distrusts it. NOTE this is NOT the close-lag case
(closedTime was a healthy 70) — no 3-field probe needed; the day-dated-rows check is a separate,
cheaper sanity read straight off the master.
.sh-wrapper + quoted-heredoc + DO-NOT-DOUBLE-SEND paragraph: one-attempt send in 53s, zero
SMTP errors (Message-ID <178818138359...>, 06:03:03 PDT). IMAP Sent-check returned in 32s
first try, 11 hits, exactly one carrying today's subject (Aug 1-30) — other 10 all prior sends
(Jul 1-28, Jul 1-30, Aug 1-8, 1-18, 1-24, 1-25, 1-26, 1-27, 1-28, 1-29), usual token-match trap.
Ran with 4 concurrent TOL tekion-scraper processes live; backgrounded the pull defensively but
it finished well inside 170s with no backoff.
MTD advisor leaders: Erick Villasenor Gonzalez 15/$21,868.08, Jon Lo 53/$17,297.86, Jason Davis
76/$11,596.43, Michael Rankin 42/$11,344.55, Gio Elenes 12/$8,206.70 (14 advisors on the board).

(9/01 run, for Mon 8/31) MONTH-END FINAL + port 465 dead, 587/STARTTLS rescued it

Month-boundary run: yesterday = 8/31, so this finalized the August master and the email covered
the FULL month (subject "August 1-31, 2026"). 451 closed ROs (biggest of the month), 29 prefilter
hits, 28 new rows, all candidate ROs scanned. Master 356 -> 384 rows / $105,864.62
($77,309.11 labor + $28,555.51 parts). 8/31-dated rows alone = 15 menus / $6,184.20 (rest of the
+$10,634.18 delta = store closing older invoiced ROs) — state both numbers, same as prior runs.
NEW SMTP FAILURE MODE: Stacey's first FOUR send attempts on port 465/SSL all dropped the
connection (three at MAIL FROM / mid-DATA, one bare disconnect). The DO-NOT-DOUBLE-SEND paragraph
worked exactly as designed — after error #1 she IMAP-checked Sent Mail (empty), then debugged
rather than blind-retrying: a tiny plain-text 465 send succeeded, so she diagnosed it as MIME size
(~600KB) plus likely Google rate-limiting, waited ~10s, and switched to port 587 with STARTTLS,
which sent on the first attempt. Result: exactly ONE email (Message-ID <178826819896...>, 06:09:58
PDT). LESSON: on repeated 465 drops with a large MIME, 587/STARTTLS is the working fallback — but
never let her retry 465 blindly; the Sent-Mail pre-check between attempts is what kept this from
becoming an 8/22-style duplicate. Hand-off ran 2m57s (19 tool calls), verification 42s first try.
IMAP Sent-check = 9 hits, exactly one with today's subject (Aug 1-31); other 8 all prior sends
(Jul 1-28, Jul 1-30, Aug 1-8, 1-18, 1-24, 1-25, 1-26, 1-27) — usual token-match trap.
MTD advisor leaders (FINAL August): Erick Villasenor Gonzalez 16/$23,521.14, Jon Lo 60/$18,557.66,
Jason Davis 84/$15,229.05, Michael Rankin 45/$12,594.65, Gio Elenes 12/$8,206.70 (14 advisors).
NOTE: September's first run will auto-create a fresh MASTER-2026-09.json — a tiny master early in
the month is normal, not the unseeded-master pitfall.

(9/06 run, for Fri 9/05) CLEAN — 465 dropped once, 587/STARTTLS fallback + dup-check worked as designed

74 closed ROs, 15 prefilter hits, 15 new rows, all candidate ROs scanned. MTD moved
42 rows/$7,874.29 -> 57 rows / $10,270.98 ($7,449.54 labor + $2,821.44 parts). 9/05-dated
rows alone = 11 menus / $1,951.50 (rest of the +$2,396.69 delta = older invoiced ROs closed) —
stated both numbers in the body per standing practice. Stacey's first SMTP attempt on 465
dropped; per the CRITICAL paragraph she checked Sent Mail (no duplicate), then fell back to
587/STARTTLS and sent ONCE (Message-ID <178869986648...>, 06:04:26 PDT). Hand-off ~96s,
verification 58s first try (her first two attempts errored/self-corrected — normal), 14 hits,
exactly one carrying today's subject (September 1-5) — other 13 all prior sends, usual
token-match trap. MTD advisor leaders: Jon Lo 8/$3,246.39, Jason Davis 10/$2,097.83,
Michael Rankin 12/$1,891.60, Somnang Ros 5/$1,103.82, Juan Carlos Urena 8/$912.76.

(9/05 run, for Thu 9/04) TEXTBOOK CLEAN — zero errors end to end

181 closed ROs, 9 prefilter hits, 9 new rows, all candidate ROs scanned. MTD moved
33 rows/$7,200.64 -> 42 rows / $7,874.29 ($5,998.37 labor + $1,875.92 parts). 9/04-dated
rows alone = 8 menus / $552.80 (rest of the +$673.65 delta = older invoiced ROs closed) —
stated both numbers in the body per standing practice. .sh-wrapper + quoted-heredoc +
DO-NOT-DOUBLE-SEND paragraph (with 587 hint): one-attempt send on 465 in 87s (Message-ID
<178861345561...>, 06:04:15 PDT, MIME 233,231 bytes); Stacey did a proactive pre-send Sent-Mail
dup check on her own again. IMAP Sent-check returned in 95s first try (her himalaya JSON
parse + first two imaplib attempts errored, self-corrected — normal), 13 hits, exactly one
carrying today's subject (September 1-4) — other 12 all prior sends, usual token-match trap.
MTD advisor leaders: Jon Lo 8/$3,246.39, Jason Davis 10/$2,097.83, Somnang Ros 3/$897.03,
Randy Vung 7/$593.47, Juan Carlos Urena 5/$517.25.

(9/04 run, for Wed 9/03) TEXTBOOK CLEAN — close lag from 9/02 fully recovered

173 closed ROs, 25 prefilter hits, 24 new rows, all candidate ROs scanned. MTD moved
9 rows/$3,384.43 -> 33 rows / $7,200.64 ($5,512.44 labor + $1,688.20 parts). 9/03-dated
rows alone = 15 menus / $2,639.33 (rest of the +$3,816.21 delta = store closing the 9/02
backlog from the previous day's close lag) — stated both numbers in the body per standing
practice. .sh-wrapper + quoted-heredoc + DO-NOT-DOUBLE-SEND paragraph (with 587 hint):
one-attempt send on 465 in 42s (fastest on record; Message-ID <178852701555...>,
06:03 PDT), zero SMTP errors. IMAP Sent-check returned in 33s first try, 12 hits,
exactly one carrying today's subject (September 1-3) — other 11 all prior sends, usual
token-match trap. Stacey's first himalaya invocation errored on syntax and self-corrected
to envelope list — normal. MTD advisor leaders: Jon Lo 8/$3,246.39, Jason Davis
10/$2,097.83, Somnang Ros 1/$775.88, Randy Vung 7/$593.47 (7 advisors on the board).

(9/03 run, for Tue 9/02) CLEAN SEND on a CLOSE-LAG day (closedTime=13) — MTD unchanged

Log showed closed/invoiced ROs today: 13, 0 prefilter hits, 0 new rows — suspicious for a
Wednesday-run/Tuesday-data day. The 3-field probe (_bt_probe_0903.py) with a 9/01 control
settled it as genuine close lag: 9/02 closedTime=13 / invoicedTime=167 / creationTime=173
(all 200); control 9/01 closedTime=52 / invoiced=115 / created=147. Per DECISION RULE: MTD
valid, render + send normally, state the lag in the body (did so: 167 invoiced vs 13 closed,
all 9 MTD rows still dated 9/01, 9/02 contributed nothing yet). MTD flat at 9 rows /
$3,384.43 ($2,607.58 labor + $776.85 parts) — same as 9/02's run; the zero-delta was stated
explicitly so Tony isn't confused.
.sh-wrapper + quoted-heredoc + DO-NOT-DOUBLE-SEND paragraph (with 587/STARTTLS hint): clean
one-attempt send on port 465 in 97s (Message-ID <178844073558...>, 06:04 PDT). IMAP
Sent-check returned in 163s first try, 11 hits, exactly one carrying today's subject
(September 1-2) — other 10 all prior sends (Jul 1-28, Jul 1-30, Aug 1-8, 1-18, 1-24, 1-25,
1-26, 1-27, 1-31, Sept 1-1), usual token-match trap. Ran with 4 concurrent tekion-scraper
processes live; backgrounded the append defensively, finished fast with no backoff.
Advisors: Jon Lo 3/$2,310.28, Jason Davis 4/$950.19, Randy Vung 2/$123.96.

(9/02 run, for Mon 9/01) TEXTBOOK CLEAN — first September run, new-month master auto-created

Yesterday = 9/01, so this was the first run against MASTER-2026-09.json (auto-created 97 bytes
on the 1st — normal, NOT the unseeded-master pitfall). 52 closed ROs, 11 prefilter hits, 9 new
rows, all candidate ROs scanned. Master 0 -> 9 rows / $3,384.43 ($2,607.58 labor + $776.85
parts). ALL 9 rows dated 09/01/26 (zero August backlog carryover) — state that in the body so
Tony understands the MTD total is one day only.
.sh-wrapper + quoted-heredoc + DO-NOT-DOUBLE-SEND paragraph (now also carrying the "if 465 drops
repeatedly, fall back to 587/STARTTLS" hint from the 9/01 run): one-attempt send on port 465
first try in 74s, zero SMTP errors (Message-ID <178835425974...>, 06:04:25 PDT, MIME 163,370
bytes). IMAP Sent-check returned in 87s first try, 10 hits, exactly one carrying today's
subject (September 1-1, 2026) — other 9 all prior sends (Jul 1-28, Jul 1-30, Aug 1-8, 1-18,
1-24, 1-25, 1-26, 1-27, Aug 1-31), usual token-match trap. Note Stacey's himalaya paging (-p 3)
errored on the last page — harmless, she had already collected all hits.
Advisors: Jon Lo 3/$2,310.28, Jason Davis 4/$950.19, Randy Vung 2/$123.96.

Verification ask wording that works first try

Lead with IMPORTANT: print the answer as plain text IN THIS REPLY AND
Use himalaya / raw IMAP against "[Gmail]/Sent Mail" (NOT the Gmail API) — the
IMAP-first default (inherited from the TOL skill's 8/17 lesson) applies here too and
returned in 43s first try on 8/21. Search the SHORT subject stem
(BT Menu Sales - Closed MTD, no date/parens — parenthesised dates are the fragile part
of IMAP subject searches). Expect the token-match trap: 8/21 returned Sent: 5, four of
which were prior sends (Jul 1-28, Jul 1-30, Aug 1-8, Aug 1-18) plus today's exact match.

LOW "closed/invoiced ROs today" can be a REAL store-side close lag, not starvation (2026-08-23)

The 8/23 run (for 8/22, a Saturday) logged closed/invoiced ROs today: 3 — far below the
prior two Saturdays (8/08 = 80, 8/15 = 75) and every weekday (66-263). By the standard
$0-validation rule that looks like a starved run, but a clean re-run reproduced exactly 3,
with zero 429s/500s. A DIRECT API probe settled it: for 8/22, closedTime returned
3 while invoicedTime returned 118 and creationTime 125 — and 8/20/8/21
returned normal counts on all three fields. So the store simply had not run its accounting
CLOSE on Saturday's invoiced ROs yet; the API was healthy and the data is correct.
bt_menu_sales_closed_mtd.py:search_closed() filters on closedTime by design (verified
2026-06-19: closedTime is true "closed in period"; modifiedTime overcounts).
DIAGNOSTIC (use this before declaring starvation on ANY low-RO day — ~3s, no full re-run):

import sys, datetime; sys.path.insert(0,"/home/itadmin/tekion-reports")
import bt_menu_sales_api as O
d = datetime.date.fromisoformat("YYYY-MM-DD")
a = int(datetime.datetime(d.year,d.month,d.day,0,0,0).timestamp()*1000)
b = int(datetime.datetime(d.year,d.month,d.day,23,59,59).timestamp()*1000)
for f in ["closedTime","invoicedTime","creationTime"]:
    st,out = O.call("POST","/repair-orders:search",
        {"filters":[{"field":f,"operator":"BTW","values":[a,b]}],"pageSize":1})
    print(f, st, (out.get("meta") or {}).get("totalCount"))

Saved as _bt_probe_0823.py. DECISION RULE: all three fields 0 / non-200 = real outage,
skip the email. closedTime low but invoicedTime/creationTime normal = genuine close lag,
the MTD total is still valid — render and send normally, and note the lag in the
final report so Joe knows the day contributed almost nothing. Also probe an adjacent day
as a control before concluding anything.

(8/24 run, for Sun 8/23) ZERO closed ROs — confirmed close lag, NOT starvation

Sunday 8/23 logged closed/invoiced ROs today: 0. The _bt_probe_0824.py 3-field probe
(now saved alongside _bt_probe_0823.py) settled it instantly with an adjacent-day control:
8/21 closedTime=150 / invoiced=180 / created=146 (normal), 8/22 closedTime=3 / invoiced=118
/ created=125 (the known 8/22 lag), 8/23 closedTime=0 / invoicedTime=34 / creationTime=35
— all HTTP 200. So the store simply hadn't run accounting close on Sunday's 34 invoiced ROs.
Per the DECISION RULE this is a genuine lag: MTD total still valid, render + send normally.
Confirms closedTime=0 alone is NOT proof of starvation — always probe all three fields plus
a control day BEFORE skipping the email. MTD stayed flat at 241 rows / $58,056.84 (same as
8/22's run); state the zero-delta explicitly in the email body so Tony isn't confused by an
unchanged total (same practice as the TOL Sunday note).
CLEAN SEND: the .sh-wrapper + quoted-heredoc pattern with the DO-NOT-DOUBLE-SEND paragraph
produced a one-attempt send in 47s (zero SMTP errors, zero retries); the IMAP Sent-check
returned in 26s first try, 5 hits, exactly one carrying today's subject. NOTE: the 8/22
duplicate pair (Aug 1-21, 06:03:56 + 06:04:46) did NOT appear in this Sent listing even
though the skill predicted it would show forever — don't treat its absence (or presence) as
signal either way; only two hits with TODAY's exact date range mean a fresh double-send.

OVERALL_QUOTA reset behavior (observed 7/8–7/9 outage)\nNOT a fixed midnight reset. Behaves like a rolling ~24h+ bucket tied to when\nthe calls were burned; the 7/8 outage ran 29+ hours with continuous 429s.\nRecovered capacity can be instantly re-drained by queued crons (11PM\ndealer-detail sync, 2AM VI pull), making it look continuously dead.\nIf dead >24h, escalate: ticket to Tekion asking the actual OVERALL_QUOTA\nlimit, reset schedule, and a raise — it's one org-wide bucket shared by all\n7 stores' pipelines and AMG has co-founder-level contact from the bin\nescalation. Never blind-retry; probe-gate everything.