Service 01
Someone has to own the whole picture.
Most orgs are a stack of individually reasonable decisions nobody has read together. We read them together, then write down what they add up to.
- Shape
- Diagnostic sprint
- Duration
- 2 to 3 weeks
- Output
- Costed backlog
01What we do
Five things, and the order matters
Architecture work that stops at a diagram is a document. These five produce decisions somebody can act on this quarter.
- 01
Org architecture
What exists, what calls what, and which of it is load-bearing. Counted from source control, not from memory. The map names the things nobody wants to touch and says why they are like that.
- 02
Data model design
Where a lookup relationship has quietly cost you native rollups, and what it costs to fix. In one org that answer was 14 rollup definitions on declarative tooling, because most parent-child links were lookups and master-detail was not available.
- 03
Automation strategy
Triggers, flows and rollup tooling each have a job. We write down which layer owns what, so the next change has one obvious home instead of three plausible ones.
- 04
Permission and sharing models
Who sees which record, and why. We never reimplement permissions in application code. The platform enforces the calling user’s field-level security and sharing natively, and everything we build inherits that.
- 05
Technical debt audits
Findings anchored to a file and a line, condensed to root causes, then costed. A list of 87 symptoms is noise. Five causes behind them is a plan.
02The flow, rebuilt
A flow with no fault path fails silently, and your salesperson takes the call.
Architecture is mostly deciding where failure is allowed to go. This conversion had eleven create and update steps and no handling on any of them.
- Every step gets a fault path, so a failure names itself instead of stopping.
- The error screen says what failed and what to do next, in the words the user already uses.
- Triggers, flows and rollup tooling each own one layer, written down, so the next change has one obvious home.
- Scheduled work carries a heartbeat and a watchdog, because a job that dies quietly is worse than one that fails loudly.
- Permissions stay with the platform. We never reimplement sharing in application code.
Point agents at your org and see what comes back.
Point agents at your org. Get back a costed, verified backlog.
Every item is analysed against your live data, with real record counts and real field names, then bucketed into fix-now, needs-decision, needs-data, needs-browser and needs-infrastructure. Each item carries the objects it touches, the plan, and the exact login-as click path to verify it, plus a cleanup line. A blockers line names what a human has to unblock. You can hand the file to your own team and act on it without us.
Fixed price. You keep the backlog whether or not you continue.
04How the audit runs
Overnight, then a file you can act on without us
Four stages, and every judgement in the output shows its working.
- 01
Point agents at the org
Parallel adversarial passes across the code, the automation and the data. One run, overnight, nine independent result blocks.
Overnight - 02
Verify every finding against live data
Real record counts and real field names. A finding that cannot be reproduced against your own records does not make the file.
Live records - 03
Bucket by what is blocking it
Fix-now, needs-decision, needs-data, needs-browser and needs-infrastructure. A blockers line names what a human has to unblock.
Five buckets - 04
Hand over the file
Each item carries the objects it touches, the plan, the exact login-as click path to verify it, and a cleanup line. Your team can start on the fix-now list the same week.
Yours either way
05Deliverables
What lands on your side
Four artifacts. All of them are yours, and all of them are useful to someone who has never spoken to us.
- 01
A system map
Objects, automations, integrations and the paths between them, counted from your source control. It names the two parallel ledgers rather than pretending the org has one.
- 02
A costed backlog
Every item verified against live records before it was priced. Bucketed by what is blocking it, so your team can start on the fix-now list the same week.
- 03
An architecture decision record
Each decision with its constraint, the option we chose, and the options we did not. The rejected options are the part that saves an argument in six months.
- 04
A CLAUDE.md for your org
The operating instructions your own team, and your own agents, work from. Deploy footguns are documented at the point of failure, including the ones that silently break things.
Not sure it is an architecture problem yet? Send us the symptom.
Bring us your hardest Salesforce problem.
Tell us what is broken and we give you a written read on it within one business day, before any money changes hands.
- Reply within one business day
- Fixed price, backlog yours either way