Due.plan is PROTECT -- a plan that has ever billed anyone can never truly be removed, on purpose: amount/period_end/grace_until are frozen on a Due precisely so a later change can't rewrite what was actually charged, and losing the plan link off an old Due would do exactly that to every historical invoice. "Delete" therefore means one of two things, chosen automatically (billing/services/plans.py): - never billed anyone -> the row is removed outright. - has billing history -> soft-deleted (Plan.deleted_at, is_active off): hidden from every picker/listing via the new opt-in Plan.objects.visible(), but the row survives so old invoices still show what they were billed under. Either way, every club currently on the plan is unsubscribed outright -- its Subscription row deleted, not just its plan field cleared. "No plan" was already a fully-understood state everywhere else in the app, so this reuses it instead of inventing a new one. Also handles the easy-to-miss second group: a club on a DIFFERENT plan, mid-trial, configured to convert to the plan being deleted (Subscription.post_trial_plan). Left alone that would try to convert onto a hidden/gone plan later; instead that club's trial is ended now (both trial fields cleared, per the CheckConstraint requiring them together) so it needs a new plan picked by hand. The confirmation screen is a real page, not a modal like every other billing action -- naming exactly which clubs are affected, in both groups, and that list can be long.
23 KiB
Platform billing — design
How RosterChief charges a club, and what happens when a club doesn't pay.
This is the design for a revision of the billing subsystem, not a rewrite. ARCHITECTURE.md
covers the club-owned domain; billing is the other direction — the platform charging the tenant —
and has never been documented there. Once this design is implemented, §4 (the model) gets folded
into ARCHITECTURE.md and this file keeps the lifecycle and operational detail, the same split
DEPLOYMENT.md already has with the rest of the docs.
Status: implemented. The four decisions left open in §10 have been taken and are recorded there.
1. What ships today, and what's actually wrong with it
The existing implementation (billing/) is structurally sound. Three things in particular are
worth keeping and are not up for renegotiation in this redesign:
- Snapshot-on-
Due.Due.tierandDue.amountare frozen when the period opens and never read back through the plan at display time. Raise a price and last year's invoice still says what was actually charged. This is the single most important property in the subsystem — a live lookup would rewrite financial history. - Dated prices.
TierPrice.active_frommeans a rate change is one new row with a future date, not an edit to an existing one. Every period already opened keeps its amount. - Asymmetric commands.
archive_overdue_clubsreports by default and only acts with--commit;renew_subscriptionsacts by default and only previews with--dry-run. The asymmetry is deliberate and correct: not archiving is a safe failure, not renewing is a silent revenue leak.
What's wrong is narrower than "the approach": three hardcoded assumptions, all in
billing/models.py.
| # | Today | Problem |
|---|---|---|
| 1 | Due.save() → add_one_year(period_start) |
Every plan is annual. Duration is only overridable by passing an explicit period_end, which is the hack trials use. |
| 2 | GRACE_DAYS = 45, module constant |
One grace period for every plan, editable only by deploying code. |
| 3 | grace_until = period_end + GRACE_DAYS |
The clock runs from the wrong end. A club uses the entire unpaid year and then gets 45 days — roughly 410 days of unpaid use before archive_overdue_clubs will touch it. |
Item 3 is the real defect. The requested behaviour — "unpaid → warned → archived after N days" — isn't a tweak to the current rule, it's the opposite of it.
A fourth gap is in the UI rather than the model: the club-facing banner
(management/views.py:89-93) fires on the period ending, not on money being owed. A club
that owes money is told nothing.
2. Target lifecycle
┌──────────────────────────────────────────┐
│ │
[no plan] ──► TRIAL ──────► UNPAID ──► IN GRACE ──► OVERDUE ─┴─► ARCHIVED
│ │ ▲ │ │ │
│ │ │ │ │ │
└──────────────┘ │ payment payment reactivate
auto-converts │ │ │ (+ new period)
on renewal │ ▼ ▼ │
│ PAID PAID │
└──────────────────────────────────┘
renewal opens
the next period
Trial. A club starts on a trial plan (there can be several — a 1-month and a 3-month trial are
just two plans). It carries a price of 0, so it settles itself the moment it opens and can never
make a club archivable. When the trial period runs out, the next renewal converts the subscription
onto the pre-selected paid plan. This already works (open_period's trial-conversion check) and
is kept as-is.
Unpaid. Every paid period — new or renewed — opens UNPAID with an invoice attached. From
that moment the club's admins see a warning on their management home page. This is the state the
whole redesign is about.
In grace. The period has started and is still unpaid. The warning escalates and gains a countdown: "Your club will be archived in N days."
Overdue. Past grace_until. archive_overdue_clubs will now pick this club up — subject to
Subscription.auto_archive, which stays as the manual override for a club you're negotiating with.
Archived. Club.archive() sets archived_at. The subdomain stops resolving
(ClubTenantMiddleware.get_club() filters on Club.objects.active()), so the club is frozen
exactly as requested: no access, nothing destroyed. Reactivation is a deliberate platform-admin
action.
3. The three clocks
Every plan carries three numbers. Getting them confused is the easiest way to misread this design, so they are named for what they measure from:
| Field | Measured from | Answers |
|---|---|---|
duration_months |
period_start |
How long is a period? |
renewal_lead_days |
period_start, backwards |
How early is the invoice raised? |
grace_days |
period_start, forwards |
How long may it stay unpaid? |
grace_days runs from the period start, not from the invoice and not from the period end.
That is the decision this redesign turns on. It means the warning window is
renewal_lead_days + grace_days — the club is told before the period begins and gets a further
grace once it has, but is cut off partway into a period it never paid for.
Worked timelines
ANNUAL duration 12mo · lead 30d · grace 30d
period 1 Jan 2027 – 31 Dec 2027
2 Dec 26 1 Jan 27 31 Jan 27 31 Dec 27
│ │ │ │
INVOICE PERIOD STARTS grace ends PERIOD ENDS
│─────────────│──────────────│
30d warned 30d grace └─► ARCHIVED 1 Feb 27 if unpaid
└──────────── 60 days of warning ────────────┘
QUARTERLY duration 3mo · lead 14d · grace 30d
period 1 Jan 2027 – 31 Mar 2027
18 Dec 26 1 Jan 27 31 Jan 27 31 Mar 27
│ │ │ │
INVOICE PERIOD STARTS grace ends PERIOD ENDS
└──────── 44 days of warning ───┘
MONTHLY duration 1mo · lead 7d · grace 14d
period 1 Jan 2027 – 31 Jan 2027
25 Dec 26 1 Jan 27 15 Jan 27 31 Jan 27
│ │ │ │
INVOICE PERIOD ST. grace ends PERIOD ENDS
└─── 21 days of warning ─┘
Why the lead time has to be per-plan
Today's single RENEWAL_LEAD_DAYS = 30 is silently annual-only. On a 1-month plan it would issue
the next period before the current one had started — periods would run away from the calendar
within a couple of cycles. Any plan shorter than 30 days is broken by a global constant, which is
why renewal_lead_days moves onto the plan.
Guardrails
Two invariants, enforced as CheckConstraints. duration_months × 28 is the conservative
lower bound on days in that many months, which keeps the check expressible in SQL:
renewal_lead_days < duration_months * 28— or the next invoice precedes the current period.grace_days <= duration_months * 28— or a club can never be archived before its next period is issued, and unpaid periods silently stack.
4. Model changes
4.1 Tier → Plan (rename)
The code says Tier, every spec and screen says "plan". Renaming now, while only ~80 references
exist, is cheaper than carrying the mismatch. TierPrice → PlanPrice follows, as do the FK
field names (Subscription.tier → plan, Subscription.post_trial_tier → post_trial_plan,
Due.tier → plan, PlanPrice.tier → plan).
4.2 Plan — new fields
class Plan(UUIDModel):
name, slug, description, is_active # unchanged
duration_months = PositiveSmallIntegerField(default=12, validators=[MinValueValidator(1)])
renewal_lead_days = PositiveSmallIntegerField(default=30)
grace_days = PositiveSmallIntegerField(default=30)
is_trial = BooleanField(default=False)
is_trial is an explicit flag rather than "price == 0" — a genuinely free tier is not a trial, and
the two dropdowns on the trial form need to offer different sets of plans. It also lets the control
panel keep trial plans out of the normal "change plan" picker.
price_on(day) is unchanged.
4.3 Due — derivation changes, fields don't
def save(self, *args, **kwargs):
if not self.period_end:
self.period_end = add_months(self.period_start, self.plan.duration_months) - timedelta(days=1)
if not self.grace_until:
self.grace_until = self.period_start + timedelta(days=self.plan.grace_days)
super().save(*args, **kwargs)
grace_until stays a stored field. That is what makes it a snapshot: editing a plan's
grace_days afterwards must not move the archive date of a period that is already running, for the
same reason amount is frozen. Storing the computed date rather than the input days gets this
for free — no extra column needed.
The predicates change meaning even though only one changes shape:
def is_in_grace(self, today=None): # period running, unpaid, not yet archivable
return self.is_owing and self.period_start <= today <= self.grace_until
def is_overdue(self, today=None): # unchanged logic, new meaning of grace_until
return self.is_owing and self.grace_until < today
def days_until_archive(self, today=None): # new — drives the countdown in the banner
return (self.grace_until - today).days
is_in_grace previously required period_end < today; it now requires period_start <= today.
An owing due before its period starts is neither in grace nor overdue — it is simply issued,
which is the gentlest of the three warning levels in §6.
5. Price changes
Already correct, and worth stating precisely because the interaction with lead time is not obvious.
open_period snapshots plan.price_on(period_start) — the price in force on the day the period
starts, not the day the invoice is raised. Add a PlanPrice with active_from = 1 Jan 2027 and
every period starting on or after that date bills at the new amount. That is exactly "adjust the
price, effective as of the club's next billing cycle".
The edge case that will bite: periods are issued renewal_lead_days early. Enter a price change
on 15 December for periods starting 1 January, and any annual period already issued on 2 December
keeps the old amount — its Due.amount was frozen two weeks before the new price existed. The
snapshot is behaving correctly; the operational rule is what matters:
Enter a price change before the renewal lead window opens for the periods it should apply to.
Two mitigations, both recommended:
- Document the rule where prices are added (control-panel help text).
- Warn in the control panel when a new
PlanPrice.active_fromfalls inside a period that has already been issued, naming the affected clubs. Correcting one is then a deliberate cancel-and-reopen, not a silent surprise.
6. The club-facing warning
A new service — billing/services/notices.py — returns one small object for a club, or None:
@dataclass(frozen=True)
class BillingNotice:
level: str # "info" | "warning" | "error"
due: Due
amount_outstanding: Decimal
grace_until: date
days_until_archive: int
| Condition | Level | Message |
|---|---|---|
Owing, today < period_start |
info |
Invoice outstanding, due by grace_until. |
| Owing, in grace, > 7 days left | warning |
Fees are due. Club will be archived in N days. |
| Owing, in grace, ≤ 7 days left | error |
Final notice. Archived in N days. |
| Owing, overdue | error |
Archiving is pending. |
| Nothing owing | — | No banner. |
Rendered in management/home.html, in the existing alert block, and gated on is_club_admin the
same way the current billing banner is — ordinary members have no business seeing platform billing.
The existing "your period ends soon / renews automatically" notice is kept, at lower priority: it
answers a different question and only shows when nothing is owed.
All strings go through {% trans %} / {% blocktrans %} with %(name)s-style placeholders per
CLAUDE.md — the countdown is a {% blocktrans count %} so the plural form survives translation.
7. Services and commands
billing/services/dues.py:
| Function | Change |
|---|---|
subscribe(club, plan, …) |
Rename only. |
start_trial(club, trial_plan, *, post_trial_plan, start=None, …) |
Drop trial_months — the trial's length is now trial_plan.duration_months, and period_end no longer needs to be passed explicitly. |
open_period(…) |
Duration and grace come from the plan. Trial-conversion check unchanged. |
next_period_start |
Unchanged. |
dues_in_grace(today) |
owing_dues().filter(period_start__lte=today, grace_until__gte=today). |
dues_overdue(today) |
Unchanged. |
archivable_clubs(today) |
Unchanged. |
subscriptions_due_for_renewal(today, lead_days=None) |
Per-plan lead. See below. |
subscriptions_due_for_renewal currently compares every subscription against one horizon in SQL.
With a per-plan lead the comparison is per-row. The function already materialises its result as
a Python list comprehension, so the honest move is to keep doing that and compare against each
plan's own lead:
latest_period_end is None or latest_period_end <= today + timedelta(days=subscription.plan.renewal_lead_days)
At platform scale (tens to low hundreds of clubs) this is one query plus a list walk; pushing
portable date arithmetic into SQL to save that is not worth the opacity. The command's
--lead-days flag stays, redefined as a global override of the per-plan value, which is what
makes backfills and rehearsals possible.
archive_overdue_clubs needs no changes — but note that with the clock moving from ~410 days to
~60, this job goes from theoretical to load-bearing. Its dry-run-by-default posture matters more
now, not less, and DEPLOYMENT.md's advice to run it without --commit for the first week should
be re-followed after this ships.
8. Migration plan
Four steps, in order. Step 3 is the one that can take down live clubs.
1 — Rename. RenameModel + RenameField operations. makemigrations will prompt for each
("Did you rename …?"); answer yes. Inspect the generated migration by hand before running it —
if Django emits DeleteModel + CreateModel instead of RenameModel, it will drop every price,
subscription and due in the table. This is the single highest-risk step in the change.
2 — Add the new Plan fields, with defaults chosen to reproduce today's behaviour on
existing rows: duration_months=12, renewal_lead_days=30, is_trial=False. grace_days
defaults to 30, not 45 — the number is not being carried over, because the thing it measures
from has changed.
3 — Do NOT backfill grace_until on existing dues.
An open annual period that started in, say, March would today have
grace_untilaround the following February. Re-derived under the new rule it becomes March + 30 days — a date already in the past. Every such club becomes instantly overdue, and the nextarchive_overdue_clubs --commitrun switches off the entire paying customer base overnight.
Existing Due.grace_until values are snapshots and must be left exactly as they are. Only periods
opened after this ships use the new rule; every club migrates onto the new terms naturally at its
next renewal. If a specific club should move sooner, that is a deliberate one-off (cancel the open
period and reopen it), not a bulk data migration.
4 — Retire the constants. GRACE_DAYS and RENEWAL_LEAD_DAYS become the default= values on
the new Plan fields and are deleted from billing/models.py.
9. Blast radius
Everything below references Tier, the constants, or the grace semantics, and will need touching:
| Area | Files |
|---|---|
| Models & services | billing/models.py, billing/services/dues.py, new billing/services/notices.py |
| Commands | billing/management/commands/renew_subscriptions.py |
| Admin | billing/admin.py |
| Control panel | controlpanel/views.py (20 refs), forms.py (10), urls.py (3), services/statistics.py |
| Control-panel templates | billing.html, _club_billing_card.html, _club_health_table.html, dashboard.html |
| Club-facing | management/views.py, management/templates/management/home.html |
| Invoice | billing/templates/billing/invoice.html |
| Tests | billing/tests.py (16), controlpanel/tests.py (17), management/tests.py (3), formbuilder/tests.py (1) |
| Docs | ARCHITECTURE.md (new billing section), DEPLOYMENT.md (archive job note) |
controlpanel/services/statistics.py deserves particular attention: tier_name,
dues_grace_until and dues_period_end all feed the club-health table, and dues_in_grace() /
dues_overdue() feed the dashboard counters. Their numbers will move once the clock changes,
which is expected — but the annotations themselves need renaming, not just re-pointing.
10. Decisions taken, and the ones still open
Taken:
- Grace runs from
period_start. (Alternatives considered: from invoice issue — strictest, archives a club before it has used anything; fromperiod_end— today's behaviour, ~410 days of unpaid use.) - Lead time is a per-plan field. (Alternatives: derived from duration — invisible in the admin; one small global constant — too tight for an annual invoice paid by bank transfer.)
Tier→Plan.- Trial length comes from the trial plan's own
duration_months, not a per-subscription number. post_trial_planstays onSubscription, not onPlan— the same trial can convert to different paid plans for different clubs.
Resolved during implementation:
- Full payment does NOT auto-restore an archived club.
reactivate()stays an explicit platform-admin action — a club can also be archived by hand for reasons that have nothing to do with money, and an automatic restore would silently reverse that the next time a stray payment was recorded. Instead,_club_billing_card.htmlshows a prominent prompt on any archived club whose dues are settled, so the deliberate act is one click away. Noarchived_reasonfield was needed. - The banner escalates.
management/home.htmlrenders it at every level;management/base.htmlrepeats it on every other management page only once it reacheserror(≤7 days, or overdue). Shown from the moment anything is owed it would sit on every screen for weeks and train people to ignore the one week that matters. - Email reminders are in.
send_billing_reminders(dry-run by default,--committo send) plus provider-agnostic SMTP settings read from the environment. Reminders go once per escalation level, tracked onDue.last_reminder_level, because the command is on a daily cron and a club that owes money for a month must not get thirty identical emails. - Online payment stays out of scope. Every payment is still recorded by hand by a platform
admin (
record_payment). The consequence is real and worth stating: the banner and the reminder email both tell a club admin money is due while giving them no way to pay it in-app. They pay by transfer; you record it.
The email default that will catch you out
EMAIL_BACKEND defaults to the console backend, not SMTP. That is deliberate — Django's own
default tries localhost:25 and raises ConnectionRefused on a box with no MTA — but it means a
deployment that forgets DJANGO_EMAIL_HOST will watch send_billing_reminders --commit report
success while no club hears anything. Set the mail variables in .env.production (see
.env.production.example) before trusting the job.
11. Addendum: deleting a plan
Added after the initial implementation. Due.plan is PROTECT — a plan that has ever billed
anyone can never truly be removed, on purpose: amount, period_end and grace_until are frozen
on a Due precisely so a later change can't rewrite what was actually charged, and losing the plan
link off an old Due would do exactly that to every historical invoice. "Delete" therefore means
one of two things, chosen automatically (billing/services/plans.py):
- No
Dueever referenced the plan (created, never actually used to bill anyone) — the row is removed outright. - At least one
Duereferences it — soft-deleted instead:Plan.deleted_atis set andis_activeturned off. The row survives (so old invoices still say what they were billed under) but is hidden from every picker and listing viaPlan.objects.visible()— an opt-in queryset method, same shape asClub.objects.active(), so the plain default manager stays unfiltered for Django admin and anything reading historical data.
Either way, every club currently on the plan is unsubscribed outright — its Subscription row
is deleted, not just its plan field cleared. "No plan" was already a state the rest of the app
fully understood (every billing view already handles getattr(club, "subscription", None) being
None), so this reuses it rather than inventing a new one.
One easy-to-miss second group: a club on a different plan, mid-trial, configured to convert to
the plan being deleted (Subscription.post_trial_plan). Left alone, that club's trial would try to
convert onto a plan that no longer exists (or has been hidden) the moment open_period()'s
trial-conversion check next runs. Handled at delete time instead: that club's trial is ended
(trial_ends_at and post_trial_plan both cleared, per the CheckConstraint that requires them
set together or not at all), leaving it on the trial plan with no scheduled conversion until a
platform admin picks a new one.
The confirmation screen (controlpanel/templates/controlpanel/plan_delete.html) is a real page,
not a modal like every other billing action — the whole point is naming exactly which clubs are
affected, in both groups, and that list can be long.