Scope note: the three passages below are editorial composites, used to see which kind of basis each might cite, not to allocate roles, responsibility or management style by type.
When the two run an acceptance together, the usual arrangement gives contract terms and timing to whoever knows the contract better, and technical review to whoever knows the system better, with one rule both accept: no signature until the technical review passes, and none until the terms are written. The arrangement follows experience and available time, not type; the same type can perfectly well do the reverse.
Accepting outsourced development: both kinds of basis on paper
(Fictional scene.) A supplier delivers an order-sync module, the demo scrolls rows of data past, the document runs to four pages, and the contract says final payment within thirty days of acceptance. The module has to be integrated for next month's promotion, and a week's delay misses it.
One person may focus on schedule and terms: whether the interface documentation is complete, who handles exceptions, whether final payment can be tied to an observation period, and whether the acceptance form states a remediation deadline. The other may focus on behavioural boundaries: pull the module into their own environment, load last year's historical orders through it, and find that a bulk import above two thousand rows silently skips some records without raising an error.
Both findings have to go on paper, and a formal acceptance record should carry the technical defect and the contractual responsibility together. The defect side states the trigger conditions, the scope of impact, the owner, the remediation deadline and the conditions for re-verification; the terms side states the payment conditions, the observation period, incident response and dispute handling. Record only one and the other half surfaces after signature — if the technology was not verified, an exception becomes "you don't know how to use it"; if the terms were not written, the fix is done and nobody owns it.
On "can demo" and "can reproduce": a smooth demo does not show it will be smooth on your data, so it is worth agreeing the delivery version, the runtime environment, the test data, the expected result and how evidence is kept, and asking them to run it in your environment on your data. But failing to reproduce does not mean the supplier does not understand their own code — it may be an environment difference, different data or a version mismatch, and the conditions should be checked further rather than a conclusion drawn.
Record in three layers: "observed", "reliably reproduced", "cause identified". Mixing the three in one sentence makes accountability impossible afterwards. Test thresholds — data volume, concurrency, network-drop duration — follow the contract and the actual load, not numbers borrowed from elsewhere.
Severe issues and deferrable items are handled separately. Something that threatens an outward commitment, such as a suspected silent drop, is remediated and re-verified before any signature is discussed; something deferrable, such as layout or wording, goes into the observation period as agreed. Applying "any single failure goes into the observation period" to everything merges blocking defects with finishing work.
A cousin comes for a summer internship: real work or not
(Fictional scene.) A cousin joins the team as a summer intern. One person proposes a small hands-on task from the start. Another proposes a period of observation so the intern understands the system first. Both weigh the same time and risk constraints, and either preference could inform either proposal.
The first is counting time: the internship is only six weeks, and two weeks of watching without touching is a waste. The second is counting something else: an intern who has not understood the data structures yet can break something that takes two days to roll back, and will not know what they changed.
This is not about who is more willing to mentor. The first worries the opportunity is wasted; the second worries damage is being planted. A middle path: in week one give a small module that can break without affecting the main flow, expand the assignment when the intern is ready, using approved test data, and before every commit write three lines — what changed, why, and how it was verified.
Counterexamples are common too: some ENTJs will patiently mentor an intern for three months, hands on; some ISTPs will drop a newcomer into the hardest pit because that is how they learned. How someone mentors depends on their own mentoring experience, the risk in the task, the intern's readiness and the team's staffing at the time, with no correspondence to type.