Ugrás a fő tartalomhoz

Contracts, Scope, and Boundaries

Basic

Applies to: every engineer who delivers to a customer as a contracting developer · Time to learn: 40 min reading + 20 min to annotate the documents of your current engagement

Practical guidance, not legal advice

This page explains how to work well inside a contract. It does not interpret any specific contract. The signed documents govern. When a clause is unclear or a request touches liability, data protection, or intellectual property, ask your engagement lead, who will involve Lieber legal where needed.

TL;DR
  • The contract protects you as much as the customer. Working inside it is how you stay strong, calm, and paid.
  • Read the whole stack before day one: framework agreement, SOW or order, NDA, DPA, IP and licence terms, AI-usage terms. Note what each one allows and forbids.
  • Know your SOW by heart: deliverables, out-of-scope list, assumptions, acceptance criteria, payment milestones.
  • Fixed price and time and materials reward different behaviour. Know which one you are in.
  • Every change, even a small one, is written down: a change request ("Yes, and here is what it costs.") or an explicitly agreed goodwill item.
  • Write the customer's dependencies down as assumptions, and agree what happens when they slip.
  • Set boundaries politely and early: channels, hours, who can assign work, least-privilege access.

Why it matters​

Contracting work runs on trust, and trust is easiest to keep when everybody knows the rules. The contract is the shared memory of what was promised. When a conversation drifts, "let me check what we agreed" is a neutral, friendly sentence that protects both sides: the customer gets what they paid for, and you are not asked to deliver, for free, things nobody priced.

The Agile Manifesto values "Customer collaboration over contract negotiation" and "Responding to change over following a plan" (see Sources). Those values and good contract discipline fit together. Collaboration is the daily mode. The contract is what makes it safe to collaborate openly, because changes in direction have a clear, quick path: a change request, not an argument.

Related pages in this section: Overview, Customer Communication, Accountability, PoC Expectations → MVP, Upsell Opportunities. Engineering standards that apply on top of the contract are in How We Build.


The document stack​

Before day one, ask your engagement lead for every document that applies and read them. Lieber is an EU company, so expect EU data protection terms in most engagements.

DocumentWhat it isWhat it protectsRead it for
MSA / framework agreementThe umbrella terms between Lieber and the customerBoth parties: liability, payment terms, termination, governing law, how orders are placedWho may request work, how disputes are handled, notice periods
SOW / orderThe specific piece of work under the frameworkBoth parties: a shared definition of what is delivered, when, and for how muchDeliverables, out-of-scope list, assumptions, acceptance criteria, milestones
NDAConfidentiality obligationsThe customer's information, and you from accidental breachesWhat counts as confidential, what you may mention in public or in AI tools
DPA (data processing agreement)Terms for processing personal data on the customer's behalfData subjects, the customer, and you, because it defines what you are allowed to do with the dataDocumented instructions, sub-processors, security measures, deletion or return at the end
IP and licence clausesWho owns code, models, data, and results; which third-party licences may be usedBoth parties: ownership and reuse are clearOwnership of deliverables, pre-existing tools, open-source licence limits
AI-usage clausesRules on AI tools, models, and customer data sent to themThe customer's data and your deliveryApproved providers and tools, what data may leave the customer environment, disclosure duties

What to take from the DPA​

When you process personal data for a customer, GDPR Article 28 requires a binding contract. Among other things, the processor processes personal data only on documented instructions from the controller, binds authorised personnel to confidentiality, applies the required security measures, follows the rules for sub-processors, and deletes or returns the data when the services end. A processor must also tell the controller at once if it believes an instruction infringes data protection law. A processor that decides purposes and means of processing on its own is treated as a controller for that processing (Article 28(3) and (10), see Sources).

Practical consequences for you:

  • Treat the DPA as your instruction manual for data. If a customer colleague asks for something the DPA does not cover, ask for a documented instruction first.
  • Do not copy personal data to your laptop, a test environment, or an AI tool unless the DPA and the AI-usage terms allow it.
  • Agree early how data is deleted or returned at the end, and put it on the project plan.

What to take from the AI-usage clauses​

Customers differ on which coding agents, models, and hosted services they approve. Check the clauses before you open a session in the customer's code or data. Agent Harnesses covers how to choose a harness that fits the customer's rules. If the clause is silent, ask in writing. Whatever you deliver is your responsibility regardless of how it was produced; see Accountability.

Annotate your documents​

Create a one-page "engagement card" in your own notes (not in the customer's repository):

engagement-card.md
Engagement: <short internal name>
Model: fixed price | time and materials | capped T&M
Deliverables: D1, D2, D3 (SOW section)
Out of scope: (copy the list)
Assumptions: A1 ... An (who owns each)
Acceptance: how and by whom, within how many days
Milestones: M1 date / amount trigger
Data: personal data? DPA reference. Where may it live?
AI tools allowed: list, plus data restrictions
Change approver (customer): name
Change approver (Lieber): name
Escalation contact (Lieber): name

Know your SOW​

The SOW is the document you will quote most. Read it for these six parts.

SOW partQuestion to answerExample of a good answer
DeliverablesWhat exactly will I hand over?"A deployed API with documentation and an evaluation report", not "an AI solution"
Out of scopeWhat is explicitly not included?"Mobile client, data migration, 24/7 support"
AssumptionsWhat must be true for the plan to hold?"Customer provides test data by week 1"
Acceptance criteriaHow is a deliverable accepted, by whom, and by when?Measurable, with a named acceptor and a review period
Payment milestonesWhat triggers an invoice?Acceptance of D1, not "end of month"
GovernanceWho decides, and how often do we meet?A named sponsor and a weekly status meeting

If something you need to know is missing or vague, say so in week one, while it costs nothing to fix. See PoC Expectations → MVP for how acceptance criteria work for a PoC.

Fixed price vs time and materials​

Fixed priceTime and materials (T&M)
Who carries effort riskMostly the supplierMostly the customer
What the customer buysA defined outcomeCapacity and effort
Your behaviourGuard scope tightly; track effort against the estimate; raise change requests quicklyKeep timesheets precise; keep the customer informed of burn rate and remaining budget; prioritise with them
Typical trapUnpriced extras eat the marginUnclear priorities lead to surprise at the end
Good habitWeekly "scope vs plan" line in the status updateWeekly "spent vs remaining" line in the status update

Capped T&M works like T&M until the cap, then behaves like fixed price. Ask your engagement lead which model you are in and what the cap rules are. Keep the same clarity either way: the customer should never be surprised by effort, and you should never be surprised by scope.


Change control​

Recognise scope creep​

Scope creep rarely sounds like a big request. It sounds like these:

  • "While you are in there, could you also..."
  • "It would be great if the export also supported..."
  • "Can we just tweak the acceptance test a bit?"
  • "Another team needs the same thing, can you add them?"
  • A new stakeholder who expects a different result than the sponsor signed off.
  • A "bug" report that describes behaviour nobody specified.

A quick test: find the sentence in the SOW that covers the request. If you can point to it, it is in scope. If you cannot, it is a change, and it may still be a welcome one.

The answer: "Yes, and here is what it costs"​

Most change requests are reasonable. Say yes to the idea and attach the price and the schedule effect. This keeps the relationship warm and the decision with the customer.

"That makes sense, and we can do it. It is outside the current SOW, so I will write a short change request with effort and schedule impact. You can then decide whether to add it now or put it on the next phase."

Never do unpaid "small favours" silently. A favour that nobody records becomes an expectation, then a baseline, then a dispute. If a tiny item is genuinely not worth a formal change, ask your engagement lead to agree a goodwill gesture explicitly and write it down, so the customer knows it was a gift and not a precedent.

Change request template​

Keep it to one page. Your engagement lead and the customer approver decide; you provide the facts.

change-request.md
# CR-<number>: <short title>

- Requested by: <name, role, date>
- Related SOW item: <deliverable or section, or "none">

## Description
What exactly is requested? (observable behaviour, not solution detail)

## Reason
Why does the customer need it? What problem does it solve?

## Effort estimate
Engineering: <days> · Testing: <days> · Documentation: <days>
Confidence: low | medium | high, and the main uncertainty

## Impact
- Schedule: <moves milestone Mx by n days | none>
- Cost: <amount or T&M effort>
- Risk or dependency: <what else changes>
- Effect on acceptance criteria: <none | updated criteria>

## Options
A) Do now · B) Swap with item <X> at no extra cost · C) Defer to next phase

## Approval
Customer approver: <name, date> · Lieber approver: <name, date>

Option B (swap) is often the friendliest: the customer keeps the same budget and chooses what matters most.

Change-request flow​

Urgent production issues need a faster path. Agree in advance who can approve emergency work by message, and write the confirmation down afterwards.


Collaboration scope: who owns what​

Delivery depends on the customer as much as on you. Make that visible and agree it early.

TopicYou (and Lieber)Customer
Deliverables and their qualityResponsibleAccountable for acceptance
Requirements and prioritiesConsulted, ask clarifying questionsResponsible and accountable
Access (accounts, VPN, repositories, cloud)Request exactly what you need, in timeGrant, review, and revoke it
Data and test dataHandle per the DPA and NDAProvide, label, and confirm it is allowed to be used
Test and staging environmentsUse and report problemsProvide and keep running
Decisions (design, scope, trade-offs)Prepare options with recommendationsDecide, within agreed time
Feedback and reviewsAsk with a clear deadlineRespond within the agreed time
Change approvalPrepare the change requestApprove or reject
Security rules and tooling of the customerFollowDocument and communicate

R and A are shorthand for "responsible" and "accountable". Adapt the table to your engagement and share it in the kick-off meeting. It is not a blame list; it makes dependencies visible.

Dependencies are assumptions​

Every item in the right-hand column is an assumption of the plan. Write each one down with an owner and a date:

AssumptionOwnerNeeded byIf it slips
Test data deliveredCustomer data leadWeek 1, FridayDependent milestone moves by the same number of days
Staging access grantedCustomer ITDay 3Work reassigned to documented backlog items; delay noted in status
Feedback on design within 3 working daysCustomer sponsorEach design reviewReview treated as approved after the agreed period, if the SOW says so

When a dependency slips, act early and kindly:

  1. Tell the customer the same day: what is late, what it blocks, and by when you need it.
  2. Propose a workaround and what it costs (for example working on a lower-priority item).
  3. Record the effect on schedule in the status update; see Customer Communication.
  4. If the delay affects a milestone, tell your engagement lead and raise a change request for the date.

Waiting silently and absorbing the delay shifts a surprise to the end of the project. Raising it early keeps the end date honest.


Setting boundaries​

Boundaries are part of professional service. A clear boundary tells the customer how to get the best from you.

BoundaryWhat to agreeWhy it helps the customer
Working hours and availabilityCore hours, time zone, holidays, on-call (if the contract includes it)They know when to expect answers and when you are offline
ChannelsOne channel for day-to-day (for example a shared chat), email for decisions, tickets for workDecisions are traceable; nothing is lost in private messages
Who may assign workNamed product owner or sponsor; others forward requests to themThe team works on one prioritised list
Contract firstAn instruction that contradicts the SOW or DPA is clarified before you actYou never put either party in breach by accident
Least-privilege accessAsk for the access the task needs, for the time it needs itA smaller blast radius for mistakes and for incidents
Law, security, licencesDecline requests that would violate themThe customer is protected from legal and security risk

Requests you should decline or escalate rather than carry out:

  • Sharing credentials, or using another person's account.
  • Moving personal data outside the agreed environment or into an unapproved AI tool.
  • Disabling security controls "just for the demo".
  • Copying code or data under a licence that does not allow it.
  • Anything illegal, or anything you are uncertain is allowed. Ask first.

If a customer's instruction looks like it infringes data protection law, the DPA context above applies: raise it with your engagement lead promptly so the processor's duty to flag it is met.

Polite scripts​

Use them as starting points and adapt them to your own voice. More on tone in Customer Communication.

SituationScript
Out-of-scope request"Good idea. That is outside the current SOW, so I will send a short change request with effort and impact, and you can decide."
Not yet"I can take that on after milestone 2. If it needs to come earlier, we can swap it with item X. Which do you prefer?"
Request from someone who cannot assign work"Thanks for asking. I will check priorities with product owner and come back to you by tomorrow."
Instruction conflicts with the contract"I want to be sure we stay aligned with the agreement. The SOW says Y. Could you confirm in writing, or shall we raise a change?"
Access beyond need"For this task I need read access to the staging database only. I will ask again if that changes."
Security or licence issue"I cannot do that, because it would break the data processing terms. I can do this alternative instead. Shall I proceed?"
Late-evening or weekend request"I am offline until Monday 09:00. If this is urgent, please contact named contact as in our support arrangement."
Unpaid favour"Happy to help. Let me check with my engagement lead how we can fit this in."

Escalation inside Lieber​

Disagreements are normal and usually solved by talking. Escalate early, with facts, and keep the customer relationship calm.

  1. Clarify. Re-read the relevant clause and your written notes. Many disputes are misunderstandings.
  2. Tell your engagement lead the same day, with a short summary: what was asked, what the contract says, what you answered, what you need.
  3. Let the engagement lead speak for Lieber on commercial and contractual matters. Do not promise credits, discounts, or extra scope yourself.
  4. Keep working on everything not in dispute, and keep the status updates factual.
  5. Record the outcome in writing and update the engagement card.

Anything that may involve a legal claim, a data protection incident, or a security breach goes to the engagement lead immediately. The NDA and DPA define the notification duties; follow them.


Pitfalls to design around​

  • Start by reading the full stack, including the DPA and AI-usage clauses, before the first commit. Keep a one-page engagement card.
  • Quote the SOW in your task descriptions, so every task points to a deliverable or a change request.
  • Use one written channel for decisions. Summarise verbal agreements by message the same day: "To confirm what we agreed..."
  • Turn small favours into small change requests, or into explicit, recorded goodwill agreed by your engagement lead.
  • Put customer dependencies in the plan with an owner and a date, and report their status every week.
  • Ask for acceptance in writing, against the criteria, within the agreed review period.
  • Request access just in time and let it be revoked when the task ends.
  • Check the AI-usage terms before using a new tool on customer code or data.
  • Keep estimates visible. Compare effort to estimate weekly, so an overrun becomes a conversation in week two instead of week ten.
  • Escalate early. Telling your engagement lead on day one of a disagreement is cheaper than on day ten.

Checklist​

  • I have read the MSA, SOW, NDA, DPA, IP and licence terms, and AI-usage clauses that apply to my engagement.
  • I filled in the engagement card and know my contract model (fixed price, T&M, capped T&M).
  • I can name the deliverables, the out-of-scope list, the assumptions, the acceptance criteria, and the next payment milestone.
  • I know who approves changes at the customer and at Lieber.
  • Every request outside the SOW gets "Yes, and here is what it costs" and a written change request.
  • Customer dependencies are listed as assumptions with owner, date, and consequence.
  • The responsibility table was shared at kick-off.
  • Channels, working hours, and who may assign work are agreed.
  • My access is limited to what the current task needs.
  • I know which personal data I may handle, where, and how it is deleted or returned.
  • I know my escalation contact inside Lieber and what to send them.

Sources​