g1t/docs/BILLING_OPERATIONS.md

361 lines21,889 bytesCodeBlame
1# Billing operations: costs, margin and prices
2
3How g1t checks what it charges against what Cloudflare charges it, keeps
4prices at cost plus 20%, and tells staff when the margin slips. Internal.
5Code: `services/billing/src/costs.rs` (reading the bill), `margin.rs`
6(reconciliation, drift, alerts), `pricing.rs` (versions, proposals,
7notice), `keeper.rs` (sandbox and Workers for Platforms measurements),
8`budget.rs` (what g1t pays for itself, and its caps; see
9[Spend caps](#spend-caps)).
10Page: sudo **Costs & margin** (`/costs`).
11
12Several Cloudflare products g1t runs on are new. Artifacts bills
13"operations" from 2026-10-14 without defining them (see
14[ARTIFACTS.md](ARTIFACTS.md), M1). So nothing here hard-codes a product
15list or a unit: every line Cloudflare bills is kept, and how a line maps
16to what g1t sells is data you change from sudo, without a deploy.
17
18## Data sources
19
20| Source | What | Where it lands |
21| --- | --- | --- |
22| Billable usage, `GET /accounts/{account}/billable-usage?from=&to=` | One row per service per day in FOCUS columns: `ServiceFamilyName`, `ServiceName`, `ChargePeriodStart`, `PricingQuantity`, `ContractedCost` / `BilledCost` / `ListCost`. Every product g1t uses appears once it is used: Workers, Workers for Platforms, D1, KV, R2, Queues, Containers, Durable Objects, Artifacts, Browser Rendering, Workers AI, Vectorize, Cloudflare for SaaS, Email. Inside an included amount the cost is 0. | `cost_lines`, source `billable_usage` |
23| GraphQL `artifactsEventsAdaptiveGroups` | Artifacts' own count by `date`, `eventType` and `repositoryName`. Operations are `create`, `fork`, `push`, `pull`, `delete`; errors (`rateLimited`, `serverError`, …) are kept but not counted. | `cost_lines`, source `artifacts_events`; per workspace (from the store key `<workspace>--<repo>`) in `own_counts` as `cloudflare_git` |
24| The ledger | Every charge: its cost at the price book's cost, what it was charged at price, what paid for it. | read, never written |
25| `pending_usage` | Month-end meters (git, storage, scans, embeddings, the cache) as they stand. | snapshotted daily into `pending_days` |
26| `plan_payments` | The plan's $20. | read |
27| repos `git_operations` | Operations customers are charged for, per workspace, counted by repos through its `operation_mapping`. | `own_counts` meter `git_operations` |
28| repos `artifacts_usage` | Every raw meter of the git store (`git.fetch`, `git.receive_pack`, `binding.*`, …) per day and workspace, with repos' `operation_mapping`. | `own_counts` meters `artifacts_<raw meter>`, and `cost_operations` (raw counts × the mapping's `cost_operations`: what g1t expects Cloudflare to bill) |
29
30A meter's slug is Cloudflare's name lower-cased with words joined by `_`
31and the "(First … included)" note dropped: `Workers for Platforms CPU ms
32(First 60M ms are included)` under `Workers` is product `workers`, meter
33`workers_for_platforms_cpu_ms`. Several rows of the same day and meter
34(regions, tiers) are added together before they are stored.
35
36## Credentials
37
38| Secret on g1t-billing | Permissions | Used for |
39| --- | --- | --- |
40| `CLOUDFLARE_BILLING_TOKEN` (optional) | Account: **Billing Read**, Account: **Account Analytics Read**, for the g1t account only | Reading the bill and the Artifacts events |
41| `CLOUDFLARE_USAGE_TOKEN` (exists) | Billing Read, Account Analytics Read, AI Gateway Read | The keeper; also the bill when `CLOUDFLARE_BILLING_TOKEN` is not set |
42
43With neither, the daily run reconciles only what g1t counted itself, and
44the page says the bill cannot be read. Nothing fails. To set the scoped one:
45
461. Cloudflare dashboard → My Profile → API Tokens → Create Token → Custom token.
472. Permissions: Account · Billing · Read; Account · Account Analytics · Read.
483. Account resources: Include · the g1t account. No zone permissions.
494. `cd services/billing && npx wrangler secret put CLOUDFLARE_BILLING_TOKEN`.
505. In sudo, Costs & margin → **Read the bill now**.
51
52Alerts are emailed through the `EMAIL` binding (Cloudflare Email Sending)
53to `COSTS_ALERT_EMAIL` (`hey@flagon.io`). An empty value sends none.
54
55## Schedule
56
57The daily cron (`17 4 * * *`, `keeper::DAILY`) runs, in order:
58
591. The keeper's measurements (sandbox seconds, app requests and CPU), each
60 a proposal now, not a direct change.
612. `costs_daily`:
62 1. Read the bill and the Artifacts events. The first run reads the last
63 31 days (GraphQL keeps 31); later runs the last 4, since Cloudflare
64 restates recent days, or back to the last day read after a gap.
65 Lines are upserted on `(day, source, product, meter)`, so a re-read
66 replaces, never adds.
67 2. g1t's own counts for the same days (replaced per day).
68 3. Snapshot `pending_usage` into `pending_days`.
69 4. Reconcile those days into `margin_days` and `workspace_costs`
70 (replaced per day).
71 5. Drift over the last 7 days into `cost_drift`.
72 6. Unit costs over the last 30 days, proposed to the price book.
73 7. Apply price versions whose date has come.
74 8. Open, update and close margin alerts; email new ones.
75 9. Email owners on the plan about rises to come.
76
77**Read the bill now** in sudo (`admin_run_costs`) runs step 2 at once.
78
79## Reconciliation math
80
81Every Cloudflare line goes to one of g1t's products ("buckets") by
82`cost_map`: the row for its product with the longest matching meter
83prefix, `*` last. A line no row claims goes to `unmapped`.
84
85| Bucket | Cloudflare | Paid for by (`revenue_map`) |
86| --- | --- | --- |
87| `sandboxes` | Containers, Durable Objects compute duration | `sandbox`, `self_hosted`, `builds` |
88| `deployments` | Workers for Platforms | `deployments` |
89| `git` | Artifacts operations (and its events, as counts) | `git` |
90| `repo_storage` | Artifacts storage | `storage` |
91| `actions_cache` | R2 | `cache` |
92| `embeddings` | Workers AI, Vectorize | `context` |
93| `security` | (Workers CPU, under `platform`) | `security` |
94| `domains` | Cloudflare for SaaS | `domains` |
95| `models` | not Cloudflare: AI Gateway's settled cost on the ledger | every other task (agent runs) |
96| `platform` | Workers, D1, KV, Queues, Email, Browser Rendering, other Durable Objects | the plan's price |
97
98For each day and bucket:
99
100- **Cloudflare cost** = Σ the bucket's lines' cost. `models` uses the
101 ledger's cost instead.
102- **Own cost** = Σ the ledger's `cost_micros` for the bucket's keys (the
103 price book's cost when charged), plus month-end deltas. A workspace's own
104 model provider is no cost to g1t.
105- **Value** = what customers were charged at price: `-amount_micros` plus
106 what the plan's included usage, a trial, the open-source pool or g1t paid.
107 g1t's own (comped) workspaces are valued at cost plus the margin.
108- **Cash** = what workspaces paid: `-amount_micros`, and the plan's price.
109- **Month-end meters**: a day's figure is that day's `pending_days`
110 snapshot less the day before's, within a month. Their month-end ledger
111 entries are left out, so nothing is counted twice.
112- **Product margin** = (value − cost) / value. **Overall margin** =
113 (Σ cash − Σ cost) / Σ cash.
114- **Quantities**: where a mapping names an `own_meter`, Cloudflare's
115 billed quantity of those lines (or, without one, Artifacts' operation
116 events) against g1t's own count.
117
118**Shared costs to workspaces.** A bucket's cost is shared in proportion
119to, first available: Cloudflare's own per-workspace count
120(`cloudflare_<bucket>`, today the Artifacts events by repository), g1t's
121own count, what each was charged for it, what its usage cost. `platform`
122and `unmapped` are shared by each workspace's share of all usage that
123day. Shares are whole micros that add up to the bill exactly (largest
124remainder).
125
126## Drift (last 7 days)
127
128| Kind | When | What to do |
129| --- | --- | --- |
130| Count | g1t's count and Cloudflare's differ by more than the mapping's `drift_percent` (10%) | Find out what Cloudflare counts: compare its events with `own_counts` `artifacts_*` and `cost_operations`. If it counts more (binding reads, `ls-refs`), either change repos' `operation_mapping` so customers are charged for what Cloudflare counts, or leave it and let the per-unit cost rise (below). |
131| Cost | Cloudflare charged more than `drift_percent` away from the price book's cost of the same usage, with at least `min_daily_cost` | A price is stale: check the proposals. |
132| Leak | Cost of at least `min_daily_cost` and nothing charged for it (never for `platform`), or a meter in `unmapped` | Map the meter (below), or decide it is overhead (`platform`). |
133
134## Prices: versions, proposals, notice
135
136- `price_versions` holds every price ever, never edited. `prices` is the
137 version in force. The daily run applies a version once its
138 `effective_at` has come, and adds the public `price_changes` record.
139 Ledger entries made from the price book carry `price_version` (the
140 version ids, comma-separated), so a past statement is always explained
141 by the prices of its day.
142- Proposals come from the keeper (sandbox seconds, app requests and CPU)
143 and the reconciler (mappings with `scale_to_own`: today git operations).
144 For git operations: Cloudflare's rate per its own operation (the median
145 over charged days of cost ÷ quantity) × (Cloudflare's operations ÷ g1t's)
146 × 1,000. If Cloudflare counts three for each one g1t counts, the per-1,000
147 price triples. At least 1,000 of g1t's operations are needed.
148- Decision (`pricing::decide`): under 2% is noise; more than 4× either way
149 is suspect and waits for staff; within `auto_apply_percent` (25%) it is
150 applied on its own when `auto_apply` is on; anything else waits.
151- Notice: a fall applies at once. A rise applies `notice_days` (14) after
152 the decision, and for a monthly meter (git, storage, cache, domains,
153 embeddings, scans) at the start of the month after that, so no month
154 is charged at two prices. Owners of workspaces on the plan are emailed
155 once per rise (`price_notices`), and the pricing page lists it with
156 "takes effect". Rises are never retroactive; margin protection is for
157 new usage once notice has run.
158- Staff approve or reject in sudo. A rejection needs a note.
159
160## Changing a mapping
161
162In sudo, Costs & margin → **Mappings**: Cloudflare's product and meter
163prefix (as **Cloudflare's lines** lists them; `*` for the rest of the
164product), g1t's product, and optionally:
165
166- **Price meter**: the price book meter the line measures.
167- **Own meter**: g1t's count of the same units (`own_counts.meter`).
168- **Scale to g1t's count**: price one of g1t's units at as many of
169 Cloudflare's as it took (proposals as above).
170- **Drift threshold**.
171
172It applies from the next run; **Read the bill now** applies it at once.
173Every change is in the audit log (`cost_mapping`).
174
175## Which raw meters are operations
176
177There is one mapping, and the repos service owns it: `operation_mapping`
178in g1t-repos' database, one row per raw meter with `cost_operations` (how
179many operations Cloudflare bills for it) and `billable_operations` (how
180many the customer is charged for). Change it with repos'
181`set_operation_mapping` RPC (services only), or
182`npx wrangler d1 execute g1t-repos --remote` until sudo has a form. A
183change applies to counts from then on, never to what was counted.
184Billing keeps no mapping of its own: it reads repos' `git_operations`
185(already mapped) for what customers are charged, and `artifacts_usage`
186(raw counts with the mapping) for `cost_operations`. Migration 0023 drops
187the `billable_units` table 0022 made for this, which was never written.
188
189**Charged at price** (a product's value) is what each day's usage was paid:
190charged to a card or credit, or drawn from the plan's included usage, a
191trial, a pool or a gift. Usage nothing paid for, as in a free period, is
192valued at price (cost plus the margin), since it was given away at its price
193rather than sold for nothing; so are g1t's own workspaces. Runs on a
194workspace's own model provider have no cost to g1t. Every daily run
195reconciles the whole 31-day window from what is already kept, so a change in
196how a day is valued reaches every day sudo shows.
197
198## Alerts runbook
199
200| Alert | Raised when | First steps |
201| --- | --- | --- |
202| Margin under the floor | A product's value against cost under `margin_floor_percent` (10%) for `alert_days` (3) days running, each with at least `min_daily_cost` | Open the product on Costs & margin. Cost up? Check proposals (approve a rise; it waits out the notice). Value down? A mapping or `revenue_map` may have moved. |
203| All of g1t under the floor | The same for money in against every cost, comped workspaces' share left out (their spend is a budget, watched in budget.rs). While less than $1 a day comes in, it says the dollars, not a percentage | Look at which products moved; check `platform` (it has no revenue of its own and grows with traffic). Before launch, with little paid usage, expect it. |
204| Leak | Drift of kind leak | Map the meter, or decide it is overhead. |
205| Drift | Count drift | See Drift above. Cloudflare's definitions change in beta: ask them in writing ([ARTIFACTS.md](ARTIFACTS.md), §7). |
206| Costs more than it pays | A workspace's shared cost over 30 days above what its usage was priced at (`value_micros`, whoever paid: card, trial, gift or included usage) × `anomaly_factor`, at least `anomaly_floor`; not comped workspaces | Shown on Reach out as "Costs more than it pays": its usage is priced below what it costs. Abuse (Abuse & fraud page) or a gap in pricing. Not emailed. A trial or gift paying for usage does not raise it. |
207
208Alerts close on their own when the condition clears. Open ones are
209emailed again weekly. The red bar on every sudo page shows margin,
210overall and leak alerts.
211
212## Token usage
213
214The model proxy (`services/models`) reads Anthropic's `usage` from every
215`/v1/messages` answer, streamed or whole, on g1t's models and on a
216workspace's own provider alike (OpenAI-shaped providers are translated
217first). Count-tokens requests are not answers and are skipped. After the
218answer, it calls `record_tokens`, which adds input, output, cache reads and
219cache writes to one row per day, workspace, person, session and model in
220`token_usage` (migration `0030_token_usage.sql`). The person is who the run
221was for, from the model session's `requested_by`; never g1t's agent. A
222report that fails is dropped and never affects the answer.
223
224`token_usage` reads a window (42 days by default, 366 at most) for the
225workspace or one person: totals, every day's tokens and the active days,
226with `costMicros` the window's run charges from the ledger, measured as
227`usage` measures them. These counts are for views only: runs are still
228priced from AI Gateway's logs, never from `token_usage`.
229
230## Tables (migration `0022_costs_and_margin.sql`)
231
232`cost_lines`, `cost_map`, `revenue_map`, `own_counts`,
233`pending_days`, `margin_days`, `workspace_costs`, `cost_drift`,
234`margin_alerts`, `price_versions` (seeded with every current price as
235version 1), `price_proposals`, `price_notices`, `cost_settings` (the
236guardrails, seeded), the `actions_cache` price, and `ledger.price_version`.
237Every create is `IF NOT EXISTS` and every seed `INSERT OR IGNORE`; the one
238`ALTER` is applied once by D1's migration tracking. Migration
239`0023_one_operation_mapping.sql` drops `billable_units` (see above).
240
241## Spend caps
242
243Two caps keep what g1t pays for itself bounded while billing takes no
244real money. Both are measured at **cost** (what Cloudflare and the model
245providers charge g1t), never at price. Code: `services/billing/src/budget.rs`.
246Page: sudo **Costs & margin** → **g1t's own spend** (`/costs#spend`).
247
248### What counts as g1t's own spend
249
250Every charge that settles (an agent run's model cost from `finish_run` or
251AI Gateway's settlement, sandbox time from `record_sandbox`, a build from
252`charge_feature`) is split by what paid for it and g1t's part is added to
253`g1t_spend` (day, bucket, billing account):
254
255| Bucket | What |
256| --- | --- |
257| `comped` | All of a comped account's work (flagon-io) |
258| `trial` | The trial credit's share |
259| `oss` | The open-source pool's share |
260| `given` | A free workspace's overrun past its last bit of trial |
261| `unpaid` | Charged, but with no real money behind it: Stripe's test key, or `FREE_WHILE_BUILDING` |
262
263The plan's included usage and on-demand charges count as revenue only
264with live payments; in test mode they are `unpaid`. A workspace's own
265model provider costs g1t nothing and is not counted. Month-end meters
266(git, storage, scans, embeddings, the cache) are not counted here; the
267daily reconciliation covers them. Migration `0024_spend_caps.sql`
268backfills the current month from the ledger.
269
270### Caps
271
272| Cap | Variable (g1t-billing) | Default | At the cap |
273| --- | --- | --- | --- |
274| A comped account's monthly budget | `COMPED_MONTHLY_CEILING_MICROS`, or the account's own **Limit** in its terms | $150 a month | New work on the account (agents, checks, workflows, builds) is refused with "<name>'s monthly budget for g1t's own agents is used up … Staff can raise it in sudo". Runs already going finish; the per-run cap still applies to them. It lifts when staff raise the budget or the month turns (UTC). |
275| The daily breaker | `PLATFORM_DAILY_SPEND_CAP_MICROS` | $75 a day (UTC) | New agent runs on g1t's hosted models that g1t would pay for are refused until 00:00 UTC. Not paused: agents on the workspace's own model provider, checks and builds, and workspaces paying with live payments on the plan (not given by staff) or an enterprise contract. In test mode that exemption covers no one. |
276
277`0` turns either off. `CLOUDFLARE_FIXED_MONTHLY_MICROS` ($30: Workers
278Paid and Workers for Platforms) is shown on the page only.
279
280The checks are cheap: `reserve` reads today's total (one indexed sum) and,
281for a comped account, its month's comped rows. Refusals come back as
282`paused`, which the compute gate honours for every plan, internal and
283enterprise included (`packages/contracts/src/compute.ts`). The runner tells
284billing whether an agent run is on hosted models (`hostedModel` on
285`reserve`); a caller that does not say is treated as hosted.
286
287### Alerts
288
289All to `COSTS_ALERT_EMAIL` (`hey@flagon.io`), through the `EMAIL` binding:
290
291- **Comped budget**: at 50, 75, 90 and 100%, once each per account and month
292 (`budget_alerts`), checked every 15 minutes. A jump past several levels
293 sends only the highest.
294- **Breaker**: at once, from the charge that trips it; if that email fails,
295 the 15-minute cron sends it (`spend_breaker.told_at`).
296
297While the breaker is open or a comped budget is used up, every sudo page
298shows a red **Spend cap** bar.
299
300### Raising and lifting
301
302- **Raise a comped budget**: sudo → the workspace → Billing → **Terms**, set
303 **Limit $** to the new monthly budget (blank goes back to the default),
304 with a note. It applies to the next start; nothing to deploy. The change
305 is in the account's audit log.
306- **Lift the breaker for today**: sudo → Costs & margin → **g1t's own
307 spend** → **Lift for today**, with why (`admin_lift_breaker`; audit action
308 `breaker_lifted`). It resets by itself at 00:00 UTC.
309- **Change a default**: edit the variable in `services/billing/wrangler.jsonc`
310 and deploy g1t-billing.
311
312## Stripe
313
314Billing keeps what it needs from Stripe so reads never wait on it, and
315hears of changes three ways (`webhooks.rs`, `stripe_sync.rs`).
316
317**The webhook.** A destination made in Stripe's dashboard (Developers →
318Webhooks → Add destination) with the endpoint URL
319`https://api.g1t.sh/stripe/webhook`, in the mode of billing's key (at
320launch, make one in live mode and put its secret). Its signing secret
321(`whsec_…`: the destination, Signing secret, Reveal) is the billing
322Worker's secret:
323
324```sh
325cd services/billing && npx wrangler secret put STRIPE_WEBHOOK_SECRET
326```
327
328Without it every event is refused with 400. After rolling the secret in
329Stripe, put the new one; during the roll Stripe signs with both, so there
330is no gap. The event list need not be exact: billing adds any event it
331handles that the destination does not send (daily, or **Fix destination**
332in sudo → Stripe), and enables it again if Stripe disabled it. It never
333changes the secret. sudo → Stripe shows whether the secret is set, the
334destination and its status, missing events, and the latest events.
335Events are claimed once each in `stripe_events`; a handler that fails
336forgets its claim, and Stripe retries.
337
338**What is kept, and how it stays current**
339
340| Kept | Where | Refreshed by |
341| --- | --- | --- |
342| Saved card (brand, last 4, expiry) | `accounts.card_*`, `card_synced_at` | `customer.updated`, `payment_method.*`, `setup_intent.succeeded`; forgotten when g1t sets a default card, so the next read asks once; the daily pass for cards older than 7 days |
343| Plans | `subscriptions` | `customer.subscription.*`, `invoice.*`; a read asks Stripe at most hourly once a period is over; the daily pass for rows older than a day |
344| Payments, refunds, disputes | ledger, `checkouts`, invoices | their events |
345
346**Every cron run (every 15 minutes)** replays missed events: Stripe's event
347list from an hour before `stripe_sync.through`, oldest first, through the
348same once-only claim. `through` moves to 5 minutes before now when all were
349handled, back to the first failure otherwise, and stays when more than
3501,000 events were listed. Claims stuck at `handling` for 10 minutes are
351dropped so the replay retries them. The first run reads 3 days back.
352
353**Daily** (`keeper::DAILY`): the destination at billing's address is
354enabled again if Stripe disabled it and given any missing event, audited as
355`stripe`/`webhook`; then up to 25 stale cards and 25 stale plans are read
356again.
357
358**What still calls Stripe on a request**: starting a payment page, a plan
359or a card check; opening the billing portal; settling a page the person
360came back from; renaming a workspace (the customer's name). Nothing a page
361view reads.