n.V2-3.10| WHAT IT RUNS ON
01| Everything in this book part assumes the technology works for the people who have to use it. That assumption is not free, and it is not satisfied by industry standards — because industry standards are calibrated to users who do not resemble ours in any relevant way.
02| Four realities set the specification. Connectivity: intermittent access, low bandwidth when there is access, metered data plans where every megabyte has a price, and a lot of usage on shared connections in cafés and community centres. Devices: phones two to four generations behind current models, limited storage, limited memory, small screens, and shared handsets that make persistent login a bad assumption. Digital fluency: many Maveriqs have little prior app experience, are unfamiliar with the interface conventions developers treat as universal, and carry real anxiety about making a mistake that locks them out. Accessibility needs around vision, motor function and cognition are common here rather than exceptional. Context: noisy environments where audio cues fail, interrupted sessions that need to resume cleanly, low privacy where other people can see the screen, and lighting from bright sun to a dim room.
03| These are not edge cases to accommodate after launch. They are the specification. A platform optimized for good phones on good connections excludes precisely the Maveriqs the platform exists to serve, and builds inequity directly into the technology while looking excellent on the metrics.
04| Which sets the standards that matter. Offline capability is not a feature; previously accessed content has to be readable without a connection, work done offline has to sync without loss, and reconnection has to return people to where they were. Load times need tiering — fast on good connectivity, tolerable on poor, and on minimal connectivity a genuine loading indicator, because an unexplained blank screen is how you lose someone permanently. Data consumption has a budget per session and per module. Battery consumption matters because charging access is not guaranteed. Graceful degradation matters more than the uptime percentage: what the platform does when part of it fails determines whether a broken component is an inconvenience or a lost day of work.
05| Trust is downstream of all of it. A platform that crashes, loses work, or humiliates someone for tapping the wrong thing does not get engagement, and no amount of social architecture compensates. Reliability is not a technical concern sitting beside the mission. It is the mission expressed in technical terms.
06| The second half of what the platform runs on is artificial intelligence, and here we should be precise about what is claim and what is bet.
07| The claim is straightforward: each Maveriq gets an assistant. Mosungi works alongside the individual, learning how she works and what she needs, while giving the platform leading indicators on the state of her team. Mosungi Solo serves the person; Mosungi Pueblo serves the team. Out of that comes a performance profile that lets learning journeys be tailored — showing where someone is weak in mindset, in skills, or in agency, which is a far more useful diagnosis than a course completion rate.
08| This connects to how contribution is valued. Tasks on the platform are public. Good execution is visible; so are delays. Performance ratings feed what a Maveriq earns, and the platform supplies the data she needs to assess herself honestly rather than optimistically. Productive errors are part of learning and cost nothing. Errors from lack of commitment carry a cost. Impact Buyers are not looking for perfection; they are looking for solid execution, delivered.
09| Now the bet. The cost model assumes AI absorbs a large majority of operational load — onboarding, training delivery, compliance scanning, marketplace coordination, data aggregation — at the quality the platform requires and at a cost that lets marginal cost fall as the network grows. Without that, the platform either scales its headcount proportionally and becomes unaffordable, or holds headcount flat and loses quality. There is no third option, and the figure we plan against is a designed target rather than a validated one. If it is materially wrong, the economics change, and everything in this book part about reaching the scale of the problem needs revisiting.
10| The commitment we make alongside the bet is a limit on it. AI does not take the relational layer. Trust-building, conflict resolution, judgment about people, and peer training stay human regardless of how capable the models become. This is not caution about the technology. It is that a system whose entire thesis is peer-to-peer transmission cannot outsource the peer relationship without ceasing to be the thing it claims to be.
11| The honest summary is that AI is our most promising factor and our least proven one. The question is not whether it matures but when — and the platform's cost curve, its ability to keep a personal assistant beside every user, and its capacity to run at low marginal cost across thousands of barrios all sit on the far side of that question.
