Cross-project infrastructure spend memo — 2026-08 (Phase 1)

Prepared: 2026-08-02 UTC Scope: Operator-approved Phase 1 of the infrastructure-budget plan — the FIRST cross-project spend ledger covering every vendor across every project (Raxx / TradeMasterAPI, Avenue, Smarter, imhere, moosequest.net shared properties), not just Raxx. Companion file: docs/business/bookkeeper/2026-08-cross-project-spend-matrix.csv (30 vendor rows — the row-level source for every number in this memo) Visibility only. No spending decisions, no cancellations, no tax advice. This is Phase 1 input for the operator's Phase 2 budget-line work.

Hard blockers this session

• Gmail / Google Drive MCP connectors not available in this runtime. No receipts could be pulled, and the MooseQuest Ledger 2026 Drive workbook could not be read or written. Every figure below is sourced from prior in-repo reconciliation memos (last full close: 2026-06-04), in-repo billing-collector config defaults, and one live data point available directly in this session (PostHog MCP — see imhere below).
• Heroku CLI present but unauthenticated (Invalid credentials provided) — no vault-sourced HEROKU_API_KEY in this environment. Per feedback_heroku_key_from_vault_not_keychain / feedback_main_loop_vault_limit, pulling that key is sre-agent territory, not bookkeeper's — not attempted here.
• console.anthropic.com billing dashboard not reachable. Anthropic API metered usage is UNKNOWN for a 4th consecutive reconciliation cycle (April, May, June, and now August).
• AWS Cost Explorer intentionally NOT queried. AWS credentials were reachable in this sandbox, but a parallel sre-agent is pulling the authoritative AWS breakdown this cycle per the task brief. Duplicating that work here would risk two conflicting numbers in the same budget cycle. The CSV carries a placeholder row keyed "AWS: see sre CE breakdown" — re-key it to the sre-agent's per-project split once it lands.
• No repo evidence for Avenue or Smarter as billed projects. Searched this repo's docs tree, the raxx-app GitHub org (single repo: TradeMasterAPI), and the MooseQuest personal GitHub account (27 repos, all pre-existing personal/home-lab tooling — SmartThings, Splunk, git-technique demos — none of them "Avenue" or "Smarter"). Both projects are recorded in the matrix as NO REPO EVIDENCE rows, not as $0 — treat their true spend as unknown, not zero, in any budget rollup.

This memo is therefore a best-effort, repo-evidenced snapshot, not a full reconciliation. Every number below traces to a specific row in the companion CSV.

Finding: an entire Postmark ledger entry was drafted and never landed

While reconciling the Postmark row, git log --all surfaced commit 4c6a2e78b ("docs(books): record 2026-06-25 Postmark charge + lock vendor convention") — a prior bookkeeper session's full write-up of Postmark's first-ever paid charge (~2026-06-25, amount UNKNOWN, NO RECEIPT flag) plus a docs/finance/vendor-conventions.md addition locking Postmark in as Internet & Hosting / Raxx product.

That commit lives only on the unmerged branch origin/worktree-agent-add69720fe5205c30. It never went through a PR (gh pr list --head worktree-agent-add69720fe5205c30 returns nothing) and never reached develop. As a result:

• docs/finance/vendor-conventions.md on develop today still has no Postmark row.
• The backend_v2/config/billing_fixed_costs.yaml postmark_pricing.active_tier: developer ($0) assumption — used elsewhere in this repo, including by the June 2026 run-rate memo — has been stale since 2026-06-25 and nobody has corrected it.
• The June 2026 month-end close never happened at all (the last full close on record is 2026-06-04, which predates this charge).

This memo's Postmark row is corrected to UNKNOWN (not $0.00) to reflect this finding. See the companion CSV's Postmark row for the full citation. This is a process gap, not a dollar-amount gap — the fix is to either resurrect that branch as a PR or redo the entry in the next full monthly close, not to guess the amount.

Total monthly run-rate

Evidenced total (Raxx product + Shared/parent entity infra only): $252.62/mo floor — $278.62/mo ceiling

The $26/mo spread is entirely the Sentry Team-tier question (see Top-5, below). This total does not include Postmark, whose actual monthly cost is now UNKNOWN rather than confirmed-free as of 2026-06-25 (see finding above) — so this total should be read as a floor that is missing at least one more line item. This total also explicitly excludes:

Read this total as a floor on Raxx + shared-infra spend, not as "the business's total infrastructure spend." Three of the six categories in the task's project list (Avenue, Smarter, and most of imhere) are not represented in the dollar figure at all, and Postmark is now a confirmed gap rather than a confirmed zero.

Per-project subtotals (evidenced only)

Top-5 spend lines (evidenced, by monthly $)

1. Heroku (Raxx product) — $216.62/mo, STALE (62 days, May 2026 invoice). The app fleet has grown materially since then (raxx-eyetok-*, raxx-freescout-prod, raxx-tickets-prod, raxx-support-*, raxx-waf-logs-prod, raxx-rack-*, raxx-analytics-*, raxx-antlers-*, raxx-learn-staging, raxx-status-prod, raxx-iac-state-prod are all now live per docs/ops/runbooks/heroku.md's app roster, none of which existed at May invoice time). The current bill is almost certainly higher than $216.62. Top operator action for this cycle.
2. Sentry (Raxx product) — $0.00–$26.00/mo, UNRESOLVED CONFLICT. The task brief for this cycle assumes "Sentry (Team tier)" — that would be the published $26/mo Team plan rate. The last repo-confirmed status (2026-06-04) was still the free Developer tier. No upgrade or invoice evidence exists in-repo for the intervening ~8 weeks. This is the largest unresolved swing item after Heroku.
3. Google Workspace (Shared) — $22.00/mo, HIGH confidence but STALE (102 days since last billing-dashboard read). Business Plus, 1 seat. Note: a config file (backend_v2/config/billing_fixed_costs.yaml) carries an unfilled $6.00/seat "Business Starter" placeholder that was never reconciled against the real $22/mo Business Plus rate — worth a config cleanup pass regardless of budget impact.
4. Apple Developer Program (Raxx product) — $8.25/mo amortized ($99/yr, paid 2026-04-22, renews 2027-04-22). Confirmed, low-risk.
5. Dreamhost / Oracle Dyn (Shared) — $5.00/mo each, tied. Dreamhost is HIGH-confidence (confirmed billing pattern); Oracle Dyn is MEDIUM (never invoice-confirmed across three prior reconciliation cycles — always estimated). GitHub's $4.00/mo Pro plan is a close #6.

Honorable mention (unranked — amount unknown, not zero): Postmark's first paid charge (~2026-06-25) could plausibly land anywhere from ~$15/mo (Basic tier) to a one-time top-up with no recurring impact — see the finding above. It is excluded from this ranked list only because its dollar amount cannot be evidenced, not because it is small.

Stale items — needs an operator receipt-pull or dashboard check

All figures below are more than 30 days old as of this memo (2026-08-02) and were never refreshed because Gmail/Drive access has been unavailable across the last three reconciliation attempts (April, May, June).

One correction to the task brief's assumptions

The task brief asked for Cloudflare plan tiers across "raxx.app, moosequest.app, getraxx.com, moosequest.net." moosequest.net is not on Cloudflare. Per backend_v2/config/billing_cloudflare_zones.yaml's own comment ("moosequest.net is registered elsewhere — not on this CF account") and the standing decision in feedback_dyndns_stays.md, moosequest.net DNS is served by Oracle Dyn and stays there — no Cloudflare zone plan applies to it. raxx.io (also on Cloudflare, also free tier) was added to the matrix for completeness since it shares the same collector config and vendor history as the other two Raxx domains.

Explicit per-project budget-line placeholders (Phase 2 input)

The CSV's Operator Budget Placeholder (monthly USD) column is intentionally blank for every row. That column exists for the operator to fill in Phase 2 — this memo does not propose budget numbers; it only reports what's evidenced. Suggested Phase 2 starting points based on what Phase 1 surfaced:

• Raxx product: needs a fresh Heroku invoice pull AND the Postmark charge confirmed before any budget line is set — the current $216.62 Heroku anchor is stale and the fleet has grown, and Postmark now has an unquantified gap.
• Shared / parent entity: the $36.00/mo evidenced total is low-risk and stable; could reasonably anchor a budget line as-is pending the Oracle Dyn confirmation.
• imhere, Avenue, Smarter: no budget line can be responsibly proposed from this repo's vantage point. Each needs its own vendor inventory before Phase 2 can attach a number — flagging this as the single biggest gap in Phase 1's coverage.
• AWS (cross-project): defer to the sre-agent's CE breakdown before setting any budget line; that breakdown should also resolve the Raxx/Avenue AWS attribution problem this memo could only flag, not solve.

Open items for CPA

None specific to this cycle beyond what's already open in prior reconciliations (see docs/business/bookkeeper/2026-06-04-monthly-runrate-estimate.md's CPA-flags section, which remains unresolved as of this memo) plus the two CPA questions already drafted — but never landed — in the orphaned Postmark entry (cash-basis treatment of a possible top-up credit; PA sales-tax treatment on SaaS services). This Phase 1 deliverable is a visibility document, not a reconciliation of the Expenses tab — no new Deductible or Entity classifications were made.

Open items carried forward

1. Gmail/Drive MCP access. Unresolved for three consecutive reconciliation attempts (April, May, June) and now this cross-project pass. Every "stale" item above traces back to this same root blocker.
2. Heroku vault credential. sre-agent territory — needs to be dispatched separately if a fresh Heroku invoice pull is wanted without waiting for Gmail/Drive access.
3. AWS Cost Explorer breakdown (parallel sre-agent). This memo's AWS row is a placeholder pending that work landing — re-key it once available.
4. Anthropic API billing. Now unresolved across 4 reconciliation cycles — the longest-standing open item in the ledger's history.
5. Avenue / Smarter / imhere vendor inventories. These need to come from the operator directly (or from those projects' own repos/ledgers, if they exist) — this repo's vantage point genuinely has nothing to offer beyond the one PostHog data point for imhere.
6. Cloudflare moosequest.app zone. Config gap — needs a zone_id entry in backend_v2/config/billing_cloudflare_zones.yaml before it can ever show up in an automated collector run.
7. Orphaned Postmark branch (origin/worktree-agent-add69720fe5205c30, commit 4c6a2e78b). Never became a PR, never merged. Contains a fully-reasoned vendor-conventions.md addition and a drafted (amount-pending) Expenses row. The next bookkeeper session should either open a PR from that branch or redo the work fresh — do not silently cherry-pick it without review, per feedback_worktree_branch_contamination.
8. June 2026 month-end close never happened. The last full close on record is 2026-06-04. This is now a two-month-plus gap in the monthly cadence described in docs/finance/README.md.

Top actions for operator

1. Pull the July/August Heroku invoice — this is the single highest-value pull; the current $216.62 anchor is 62 days stale and the fleet has grown since.
2. Confirm current Sentry plan/invoice — resolves a $0–$26/mo swing that the task brief and the repo's own history disagree on.
3. Confirm the Postmark charge amount from ~2026-06-25 and decide whether to resurrect origin/worktree-agent-add69720fe5205c30 as a PR or redo the entry — this closes both a dollar gap and a documentation gap simultaneously.
4. Add moosequest.app to backend_v2/config/billing_cloudflare_zones.yaml with its zone_id so it stops being invisible to the billing collector.
5. Supply vendor lists (or repo/ledger pointers) for Avenue, Smarter, and imhere's non-PostHog infra — Phase 1 cannot see past this repo's boundary, and three of the six requested projects have effectively zero coverage as a result.
6. Pull Anthropic API billing for the last 4 months (Apr–Aug) from console.anthropic.com > Billing — longest-open item in the ledger.
7. Confirm CPA engagement status from the 2026-05-27 outreach round (Kregel & Co, SDO, Dimov, Pilot, Wipfli) — needed before any CPA fee line can be budgeted.
8. Schedule the overdue June 2026 month-end close — the cadence has slipped to two-plus months; the next close should also formally absorb the orphaned Postmark entry.