Questions worth a second opinion.

Do not treat title and pay as direct proxies for career value. For the next two years, the larger company is probably stronger if the manager will give you visible ownership, the hours are genuinely predictable, and the role does not trap you in narrow execution. The director role wins only if it comes with real hiring, budget, and decision authority—not title inflation. Compare both offers on the evidence they create for your next role: scope, sponsorship, outcomes you can own, and sustainability at home. Negotiate the missing conditions before choosing.

Push once for a measurable plan, but put a time limit on ambiguity. Ask which promotion rubric you are missing, what observable behavior would demonstrate it, who must see that evidence, and when the decision will be made. Propose a 60–90 day assignment that tests the gap and get the plan in writing. At the same time, quietly benchmark internal and external roles. If your manager cannot translate ‘executive presence’ into specific evidence or sponsorship, treat that as information about the path—not proof that you are unqualified.

The equity is not enough by itself to justify the move; treat it as uncertain upside and decide whether the job is a worthwhile career and life experiment even if the options end up worth zero. Before accepting, verify the option type, vesting, strike price, fully diluted percentage, exercise window, runway assumptions, and what turns a pilot into revenue. Then price the household cost of the pay cut and a six-month job search after failure. The move can be rational if the learning, ownership, and mission are valuable at that downside—not because 1.2% sounds large.

Pause broad acquisition, not all acquisition. A 9% week-4 rate means new volume will mostly buy faster leakage unless the retained 9% represents a valuable, repeatable segment. Keep a small diagnostic channel running, segment retention by use case and team type, and interview both retained and recently churned accounts around the moment value breaks. Set a retention or repeated-use gate before restoring spend. The exception is a narrow segment with clearly better retention and credible unit economics; acquisition there can continue while the core product is repaired.

You have evidence of a recurring workflow, not yet of a business. Keep pursuing the market only through a time-boxed willingness-to-pay test. Ask the seven weekly users to choose among a paid pilot, a specific procurement path with a decision date, or stopping service; do not trade more roadmap work for verbal enthusiasm. Track who owns budget, what current workaround costs, and whether usage survives when the product is no longer free. Continue if several users cross a precommitted commercial threshold. Otherwise narrow the buyer or problem before building more.

Use a simple base plan with included usage and explicit overage or tier limits. Pure per-seat pricing is easy to explain but quietly subsidizes the customers who create the most cost; pure usage pricing makes buyers fear an unpredictable bill and can discourage adoption. Anchor the package to a customer outcome or workflow, include enough document volume for normal use, and make the next tier predictable. Test price and packaging with real quotes, not preference surveys. The goal is not perfect cost pass-through—it is understandable value with bounded margin risk.

Six bookings do not yet distinguish weak demand from a broken trust and transaction layer. Map the funnel from urgent need to completed care: response time, matching quality, identity checks, pricing, payment, cancellations, and what happens when a worker is unavailable. Then run a concierge test for one narrow care need and neighborhood, manually matching both sides and guaranteeing follow-through. If families complete and repeat when coordination is reliable, the marketplace has an execution problem. If they still avoid paying or repeatedly delay, the stated need is not translating into demand.

First separate candidate-generation failure from ranking failure on a representative labeled set. If the relevant document is already in the top 30–50 most of the time, a lightweight reranker is the more direct fix for weak top-5 quality; recover its latency through smaller candidate sets, caching, batching, or a cheaper first-stage scorer. If relevant documents are absent from the candidate set, reranking cannot help, so improve embeddings, chunking, metadata filters, or query rewriting first. Do not choose from average recall alone—measure recall at the exact rerank depth and by query class.

Microservices will not fix missing idempotency or unclear transaction boundaries; they may make both harder to reason about. Fix the billing invariant first: stable idempotency keys, a uniqueness constraint at the write boundary, transactional outbox or equivalent handoff, safe retry semantics, and alerts plus a reconciliation path. Then decide whether billing deserves a service boundary based on independent scaling, ownership, deployment, or regulatory needs. Extracting it can be sensible later, but only after the same correctness contract works inside the monolith and is covered by failure-injection tests.

Ship a primary model with deliberately simple, tested failure paths unless your evaluation shows a second model catches important errors the first model does not. Add timeouts, bounded retries, cached safe responses, retrieval-only fallback for supported cases, explicit abstention, and a manual escalation path. A router adds policy, prompt, evaluation, observability, and incident complexity that six weeks may not absorb. Design the interface so a second provider can be added later, and run it in shadow on a small eval set before routing live traffic.

The pilot supports operational promise, not a causal claim of effectiveness. Volunteers may be more motivated, the labor market may have changed, and historical averages may differ in season, demographics, or job mix. Report the 18% change as descriptive, with uncertainty and attrition, then use the next rollout to create a credible comparison—randomized access if feasible, or a phased rollout with matched eligible participants and predeclared outcomes. Scaling a limited tranche can be reasonable for learning, but ‘proven’ overstates what this design establishes.

The data shows an advancement gap, not that remote work causes it or that office attendance will fix it. Role type, tenure, location, manager, project visibility, prior performance, and who chooses remote work can explain part of the difference. Model promotion within comparable roles and managers, inspect changes for people who switched arrangements, and audit how high-visibility work and sponsorship are allocated. If remote employees are equally productive but less visible, the policy target may be evaluation and access—not attendance. Test the proposed mechanism before imposing a company-wide remedy.

With a real three-year move possibility, renting is likely the safer default unless buying has substantial nonfinancial value to you. Compare rent with the full ownership cost: interest, HOA, taxes, insurance, maintenance, opportunity cost of the down payment, and buying and selling costs. A short horizon makes transaction costs and price uncertainty unusually important; a large down payment does not remove them. Buy only if you would be comfortable holding or renting out the condo after a move and can absorb that responsibility without relying on appreciation.

Do not enroll until you test whether the degree is a requirement or merely one possible signal. Sample target roles across government, consulting, and nonprofits; ask hiring managers and recent alumni which gaps actually block candidates with delivery experience. Apply now to a few plausible roles and short fellowships, and look for policy work inside your current projects. The master’s is justified if it provides a specific credential, work authorization, network, or recruiting channel you cannot obtain more cheaply—and if graduates with your background reliably reach the roles you want.

Do not reduce this to income versus loyalty. Sequence the decision around the constraints that are hardest to reverse: your mother’s care plan, immigration and work eligibility, the pension value you would forfeit, and whether your partner’s offer can tolerate a staged move. Price the after-tax household gain after travel, care, and career interruption. A six-month split-location trial or negotiated start may reveal more than a permanent move now. Relocate if care is resilient, your own pathway is credible, and the relationship cost of waiting exceeds the option value of three years.