Tutorial: The first weeks with a new tenant

The 15-minute onboarding tutorial walks you through jin init, your first extract, and your first build. This one is its sibling — what happens in the four weeks after init lands. It’s the honest version, written for when the customer asks “when do we see results?” and you need a defensible answer.

Read Setup is the analysis first if you haven’t. It’s the why behind this how.

Time scales here are illustrative — they depend on pack maturity, source-system quirks, customer data hygiene, and your familiarity with the analytical domain. Take them as orders of magnitude.

Week 1 — Bytes in, KLS out

Goal: the first jin make succeeds end-to-end on real source data.

What you do:

• Receive a representative source data drop from the customer — real, not synthetic
• Write or adapt pipeline.yml — SHA-256-pin every file, declare expected sheets and columns, log to extract_log.jsonl
• Adapt the source-system dispatch macros where the customer’s column names diverge from the pack default
• Run jin make --extract-only until extraction is clean
• Run jin make end-to-end; expect failures in the Bronze → Silver → Gold layer when assumptions don’t match
• Validate the Gold counts are within an order of magnitude of what the customer expects (case count, billing event count, person count — whatever the pack’s headline entity is)

What you don’t do this week: tune signals, write new theses, capture Subject Matter. Those come later. This week is about getting bytes into Gold.

End-of-week test: jin make produces a KLS, jin explore opens it, the Gold counts look approximately right to the customer.

Week 2 — Gold becomes real

Goal: the Gold contract reflects what this tenant actually cares about.

What you do:

• Walk the Gold tables with a domain expert from the customer. Confirm what each row means, what each column means, what’s missing.
• Add tenant-specific Gold columns where the pack default doesn’t capture something material to this tenant. (Tenant overrides Gold via the AFS; the pack isn’t touched.)
• Adjust the Silver-layer validation rules — what counts as is_valid = true in this customer’s context?
• Capture the first batch of Subject Matter — every “oh, that’s weird” the domain expert says is a Subject Matter entry waiting to happen. Start with Level 0 (Statement); add Level 1 Checks later
• Build, explore, present the Gold tables to the customer for sign-off

End-of-week test: the customer signs off on Gold as “yes, this is what we have.” That’s the analytical foundation.

Week 3 — Signals become honest

Goal: signal findings actually mean “look at this.”

What you do:

• Run the pack’s signals against the real Gold. Note which produce 0 findings (the data doesn’t match the pattern), which produce 200K findings (the threshold is wrong for this tenant), and which are meaningful
• Adjust thresholds, weights, and severity bands per signal
• Disable signals that don’t apply to this tenant’s source system or business model
• Write 1–3 tenant-specific signals where the pack doesn’t have one that fits
• Capture the second batch of Subject Matter — by now the domain expert has explained things you couldn’t anticipate

End-of-week test: finding counts per signal are honest. A “high severity” finding is something a person should actually look at. The signal registry no longer carries dormant noise.

Week 4 — Theses that matter HERE

Goal: one confirmed thesis the customer acts on.

What you do:

• Sit with the executive or sponsor. Ask: “what question are you actually losing sleep over?” That question is your thesis.
• Compose evidence from the calibrated signals
• Write the verdict — the root cause and recommendation
• Build, explore, present to the customer
• Iterate based on their pushback. (There will be pushback. That’s the analysis happening — it isn’t done yet.)

End-of-week test: one confirmed thesis, one verdict, one recommendation the customer agrees with. That’s the first real analytical output.

Weeks 5+ — Steady state

After the first thesis lands, the cost curve flattens. New analytical questions cost 2–5 days each, not four weeks, because the Gold + signals + Subject Matter infrastructure is now in place.

The pattern repeats per question:

1. Identify the question (with the customer)
2. Compose evidence from existing signals — or write one new signal if needed
3. Write the thesis + interpretations
4. Build, explore, present
5. Capture any new Subject Matter that surfaces

A mature tenant carries 5–10 confirmed theses, 30+ signals fine-tuned, 20+ Subject Matter entries documenting the institutional reality, and an ongoing rebuild cadence (weekly or monthly).

What changes after week 4

• The customer trusts the system. They’ve seen it answer a question they care about. The next question is easier to scope.
• The Subject Matter catalogue is load-bearing. Every weird thing the expert explained is now durable; the next analyst on the tenant doesn’t need to re-discover it.
• The Gold contract is signed. Any new signal or thesis you write reads from a known shape; you’re not re-validating Gold every time.
• The build is fast and boring. Weekly rebuild becomes routine.

A note on parallel paths

If you’re onboarding multiple tenants of the same pack in parallel (typical for partners), week 1 of tenant N becomes much faster because:

• The source-system dispatch macros are often reusable across tenants on the same ERP
• The Subject Matter catalogue grows pack-wide knowledge that helps every tenant
• The pack itself evolves — improvements found in one tenant ripple back to the pack and forward to others (the Steward — Sense 24 mechanism)

What you learned

• Setup is the analysis. The 15-minute init is the doorway, not the room.
• Each layer (sources, Bronze, Silver, Gold, signals, theses, Subject Matter) gets its week. Skip a layer and the next one will tell you, loudly.
• The first month is the expensive month. The work compounds.
• A confirmed thesis the customer acts on is the goal of week 4 — not the start.

Related

• Setup is the analysis — the philosophy behind this tutorial
• Tutorial: Onboarding a new tenant — the 15-minute init walkthrough that opens this path
• Tutorial: From finding to verdict — the analytical chain you’ll write in week 4
• Tutorial: Capturing expert knowledge — how to write Subject Matter entries efficiently
• The Cycle — make, explore, evolve, publish — the loop you enter every day after the first month