The Raptor Audit · Step 2: Delete the Part

Every dollar and motion is a part.

Full inventory of Tare's recurring dollars, crons, workers, flows, and weekly motions, from live bank data, live Cloudflare, live Klaviyo, and live sales, pulled July 12. Each part is tagged with what breaks if it stops. Nothing on this page has been cancelled or deleted — every kill needs your approval.

BUILT JUL 12 2026 · BANK DATA APR 1–JUL 12 · SALES DATA 30D
~$8,400
per year deletable or replaceable (dollars section)
12
true zero-sellers in 90 days (corrected Jul 12 — see item 19)
~2,900
GitHub Action runs per month from ONE workflow
identical label-refresh routines firing every Monday
1
cron slot freed (of your capped 5) if approved
Section 1 · Recurring dollars

The subscription stack

Ground truth from the Pinnacle transaction log (shadow bookkeeping), April 1 to July 12. Software plus professional overhead runs about $930/month. Roughly $700/month of it is deletable or replaceable. Rent, loan, utilities, payroll, and inventory vendors are excluded — those are the engine, not parts.

#Part≈ MonthlyVerdictWhat breaks if it stops
1
CPA — Hunter Feinberg
4 payments · $434 avg · Legal & Professional
$579Replace Nothing that isn't already planned. Shadow bookkeeping runs nightly; the CPA replacement plan exists. Needs a year-end tax filing answer before the final cut. Biggest single dollar on this page: ~$6,900/yr.
2
Webflow (legacy)
not on Pinnacle — annual or other card, verify source
est. $23–49Cancel Nothing. Site fully on Cloudflare Pages since the migration; "kept temporarily" was the memo, and temporarily is over. Verify billing source, export any leftovers, cancel.
3
SmartCredit
$29.99 × 4
$30Cancel Credit-score monitoring stops. Free monitoring (bank, Credit Karma) covers the same need; pull a fresh report only when a lender asks.
4
Manychat
$16.46 × 4 · IG auto-DM
$20Cancel IG keyword auto-replies stop — but the social content system was retired June 25. A subscription serving a deleted system. Verify no live flow, then cancel.
5
Squarespace
$17.18 × 4 · likely domain registrations
$21Replace Identify what's on it. If it's domains, transfer to Cloudflare Registrar (at-cost renewals, everything already lives there) and close the account.
6
Wispr
$15.00 × 3 · voice dictation
$15Decide Dictation app stops. Only you know if you actually use it — if you hesitate to answer, that's the answer.
7
OpenAI
$19.60 × 7 · icon-gen uses gpt-image-1
$42Decide Check what this is. If a ChatGPT subscription: cancel, keep pay-per-use API credits for icon-gen. The 327-icon almanac library is already built; new icons are occasional, not monthly.
8
QuickBooks Online
$41.71 × 3
$42Keep · next domino Books of record, keep until the CPA replacement lands. Shadow bookkeeping was built to make this deletable in a later pass.
9
Klaviyo
$102.43 × 3
$102Keep · but use it This is the retention engine and retention is the constraint. The offense here isn't the bill, it's paying it while the First Visit Retention flow sits at 0 sends ever. Turn the flow on or the $102 buys nothing.
Keepers
Canva $22 · Spotify $14 · Cloudflare $14 · Apple ~$29
~$79Keep Brand kit work is active, the store needs music, Workers Paid runs the whole stack, Apple is mixed personal. Not worth the motion.
If 1–5 are approved: roughly $650–680/month (~$8,000/year) comes off, most of it from executing the CPA replacement you already planned. Items 6–7 add up to ~$57/month more pending your answer. SOURCE: TXN_LOG, APR 1–JUL 12
Section 2 · Machinery

Crons, workers, and flows — the invisible part count

The stack runs 4 Cloudflare cron triggers, 16 GitHub Actions schedules, 2 cloud routines, and 11 workers. Most earn their keep. These don't.

#PartVerdictWhat breaks if it stops
10
Duplicate label-refresh routine
two identical cloud routines · both Mon 6:13am
Delete one Nothing — the April 23 and June 30 routines are the same job. Today it double-runs weekly and would email you twice on failure. Delete one at claude.ai/code/routines (I can't delete routines, only you can). Clear the 8 spent one-shot tests while you're there.
11
bintender-threshold-sweep
GitHub Action · every 15 min · ~2,900 runs/mo
Cut to hourly Threshold alerts arrive up to an hour later instead of 15 minutes. Bins don't empty minute-by-minute; hourly deletes ~96% of the runs and most of the Actions-minutes bill risk on a private repo.
12
o3-report-check workflow
GitHub Action · 6×/day weekdays
Delete Nothing. The worker already runs the same idempotent check twice daily (midnight + 9:15am safety re-run). This is a third belt on two suspenders.
13
mwco-api daily cron
Cloudflare · 5am daily · mwco isn't launched
Delete trigger Daily studio scaffolding for a site that isn't live stops. Bonus: frees 1 of the account's 5 cron slots — the exact cap that forced 20+ jobs to pile onto tare-api's midnight cron.
14
tare-reviews-sync
worker · pending Google OAuth since Jun 15
Fix or delete It's been fetching nothing for a month — the reviews grid is frozen either way. Do the one-time OAuth this week or delete the worker. A zombie is worse than either.
15
Idle workers & databases
tare-dashboard · tare-hq worker · dickson-log · littermap (+2 D1s) · mwco-subscribe
Verify → delete tare-dashboard untouched since Mar 11, tare-hq worker since Feb 25 (HQ serves from Pages), littermap has two orphan databases. I check each for live traffic first; anything at zero gets deleted with a D1 export kept.
16
Dead cron branch in worker.js
*/2 narrative-drain · not in wrangler.toml · never fires
Delete code Nothing — it's unreachable. Pure part-count reduction.
17
Klaviyo dead drafts + legacy flows
2 unconfigured drafts · 1 draft from 2023 · 2 live check-in flows
Clean + audit Deleting the three dead drafts breaks nothing. The two live legacy flows (Quarterly + Yearly Customer Check In) predate the retention work — I pull their send/conversion numbers and anything at zero impact gets archived.
18
Daily content drips
x-daily-prompts · story-studio · new-product-notify · bintender-digest ×2/day
60-day test Each sends you something daily. The question for each: when did its output last change what you did? Answer honestly per item; anything you can't recall acting on in 60 days gets switched off. Expect to miss (and re-add) about one in ten — that's the sign you cut deep enough.
Section 3 · The store

Physical parts: SKUs and vendors

This is where deletion pays most. Every SKU is a label, a reorder check, a shelf-life watch, a cleaning motion, and a catalog row. Every vendor is a cheat sheet, an email template, an invoice quirk, and an ordering-day decision.

Corrected July 12 (evening): the first version of this page claimed 226 zero-sellers and a 45% dead catalog. That was the archived-items trap — 199 of those "zero sellers" were Square items you already archived, counted as unsold because they can never sell. Against the true active catalog of 477 over a fairer 90-day window: 12 zero-sellers, 29 with nothing sold in 30 days, 28 barely moving — about 8% dead, not 45%. The shelf is much healthier than the old report said, and the fix is an afternoon, not a project. Full list: zero-velocity report. SOURCE: /ANALYTICS/SLOW-MOVERS?DAYS=90, ARCHIVED EXCLUDED
#MoveVerdictDetail
19
90-day zero-velocity cut
Run it Done — and it shrank. The 90-day report (archived items excluded) found 12 true zero-sellers, not ~100: full report with the 30-day tab. Clearance-basket the 12 (the 4 chicken cuts first need a freezer-stock check), then delete labels + catalog rows. The 28 barely-movers are the watch list for the next pass.
20
Archive dead vendors
Archive 4 now Butterfly Bakery, Jamie's Farm, Raw Love, Sipima — last invoices June 2024. Archive the vendor rows and delete their cheat sheets, templates, and reminders. Then adopt the standing rule: no order in 120 days → auto-archive (about 10 more vendors are approaching it; supply vendors get a longer leash).
21
Vendor row dedup
Hygiene Duck River Honey ×2, Sunnyside Eggs ×2, plus junk "Tare/TARE" rows. Merge so invoices and reminders stop splitting across twins.
Section 4 · Paused & proposed parts

Kill or ship — stop carrying

A blocked project isn't paused, it's a part you're paying attention-rent on. Each of these gets one of two outcomes this week: a shipping decision or a formal park (memory line + review date, out of the working set).

#PartVerdictThe decision
22
Weekly interview loop
blocked on 2 UI fixes · 4 test routines already run
Fix or kill It's been blocked on your two fixes. Either they happen this week and the loop ships, or it's killed — with the Weekly Compass proposal covering the same need with less machinery.
23
Website mobile pass
.pb-pill fix committed, HELD
Ship or drop A committed fix sitting unshipped is the worst state — all the risk of change, none of the benefit. Standing rule says live = flawless on mobile, so the default is ship to staging, verify, publish.
24
Formal parks
Refill Ops L2 · LitterGame · Mantel · hardware vision · agent swarm
Park with dates Per the CEO brief: freeze Refill Ops L2, slow-burn Refill Map. Side quests and the idea board get park lines with review dates. Proposed parts follow the same rule as installed ones: the default answer is no part.
25
Brief kill list (already agreed)
awareness gauge · CPA comparison report · 30-min stragglers batch
Execute Named in the July 11 brief; this pass is the execution vehicle. The broken TOC awareness gauge (dead Meta metric) comes off the HQ card instead of lying to you at 512.
Not on the kill list, and the reason this page exists: the First Visit Retention flow. Deletion isn't the goal — attention is. Everything removed above is attention returned to the one part that pays ~$38k/quarter when it turns on. And the standing #1 from the brief still leads everything here: the /gmail/send auth hole gets fixed before any of this — 15 minutes, and it's the only item on this page that's a risk rather than a cost.
Section 5 · Standing rules

The algorithm, Tare-sized

The five steps in order — the order is the whole point. We've historically entered at step 5. These rules make steps 1 and 2 permanent.

  1. Every requirement gets a name. No recurring dollar, cron, or motion exists because "that's how it's done." If nobody can say who needs it and why, it's not a requirement — it's a habit with a bill.
  2. The 60-day rule. Any automation whose output hasn't changed a decision or a dollar in 60 days gets switched off. If we don't end up re-adding about 10% of what we delete, we didn't delete enough.
  3. Optimize only survivors. No polishing, refactoring, or redesigning anything until it has survived a delete pass.
  4. Accelerate the constraint only. Right now that's retention. Speed anywhere else is decoration.
  5. Automate last. New automation requires naming the part it replaces or the constraint it serves. Otherwise it's a new part — and the best part is no part.
Section 6 · Your move

Approve by number.

Nothing has been cancelled, deleted, or changed. Reply in chat with the numbers to execute — e.g. "approve 1–5, 10–13, 16, 19, 20" — or "all", or ask about any line. Items 6, 7, 14, 18, 22, 23 need a one-word answer from you first. I'll execute approved kills, keep a written undo trail (exports, configs, cancellation notes), and log the whole pass to memory.

DELETE PASS · JUL 12 2026 · DATA: PINNACLE TXN LOG · CLOUDFLARE · KLAVIYO · GITHUB ACTIONS · CLAUDE ROUTINES · D1