← home

the sequence

what gets built,
& in what order

this changes. it is written down anyway, because a roadmap you cannot argue with is a wish list.

the first three positions were chosen. everything after is sequenced on four questions: is there a buyer who already has budget, is the value measurable, what is the regulatory load, & does it reuse something already paid for. where the answer put something late, the reason is on its row.
wave 0

now

named, and being built. these three are the order of work, not a wish list.

P0 Analogy Explains arithmetic, AI and ML to a child, one analogy at a time. schools and parents, with games. a buyer who already has budget, a result a parent can see in a week, and almost no regulatory load. live P1 Tara ★ humanity's last chatbot. consumer, fully specified, and the only positioning in its category a competitor cannot copy without abandoning their revenue. building P2 Setu ★ Middleware between a coding agent and the model provider. Fewer tokens, fewer wrong turns. the buyer is engineering leadership and the value arrives as a number on an invoice. shortest proof cycle of anything here. building
wave 1

next

each one reuses something wave 0 has already paid for -- the same buyer, or the same infrastructure. the cheapest possible second sale.

04 Socratic Marks the work by asking about it, not by scoring it. the cheapest sale in the whole roster: same buyer as Analogy, same infrastructure, one grade band up. a second product to an existing school beats a first product to a new one. concept 05 Deck Brand design decks, generated to a house system rather than a template. already live and design is the studio's forte. monetise what exists before building what does not. live 06 Grid Critiques a layout against the system it claims to follow. small, pairs with Deck, and the evaluation harness is trivial because the system it checks against is declared. concept
wave 2

the legal wedge

redline and intake carry no advocates act exposure, so they prove demand for bench without carrying its liability. bench follows only once they have.

07 Redline Explains what a contract change actually costs you, in plain language. the legal wedge. informational only, so no Advocates Act exposure and no advocate of record needed. concept 08 Intake Turns an obligation buried in a policy into a dated task with an owner. same wedge, same absence of regulatory load. together these two prove demand for Bench without carrying its liability. concept 09 Bench ★ An agentic law firm: seven agents draft, one licensed human signs. the full firm. sequenced after the wedge on purpose: it needs a signed licensed partner, malpractice cover, and evidence that people will pay before any of that is worth arranging. building
wave 3

needs a partner first

each of these is gated by something that is not code -- a clinical partner, parcel data, a fleet to be wrong on, or a procurement cycle.

10 Design Agent ★ the thread is open on purpose. one prompting surface for every tool design already uses. the personal centre of gravity, and the reason it is not earlier is stated on its own page: it needs thinking, not scheduling. open 11 Consult Turns a consultation into the letter, and the letter into a plan. healthcare needs a clinical partner and DPDP work before a line of code. the compliance bar is the schedule. concept 12 Ward Keeps one patient’s specialists reading the same chart. open, but if it were built it belongs here, same partner, same bar. open 13 Setback Checks a drawing against the building code before the reviewer does. architecture needs code and parcel data partnerships. the agent is the easy half. concept 14 Feasible Site to massing to a number, in an afternoon. follows Setback because it reuses the same jurisdiction data. concept 15 Handoff Decides the moment a machine should stop and ask a person. robotics needs hardware partners and a real fleet to be wrong on. longest lead time in the roster. concept 16 Choreo Schedules a fleet that shares floor space and cannot collide. follows Handoff for the same reason and the same partners. concept 17 Harvest Finds what a future machine would read first, and re-keys it now. enterprise sales cycle measured in quarters, driven by a 2030 deadline rather than by us. concept 18 Triage Reads the week of literature so a lab does not have to. closest to the studio's own habit, the papers already do this by hand. sequenced late because the buyer is a lab, and labs buy slowly. concept
--

not on the road

research is the credibility layer, not a product. the open ones we are not building at all. the devices come last, and the reason is written on their page.

R Closure ★ One theorem: a cycle closes if and only if the ratio of its two rates is rational. research. not launched, and not meant to be, this is the credibility layer the products stand on. research R Sentinel A learning layer above unmodified cryptography, the guard who stays awake. research. research R Critical Line Five plates on one question: does a mirror force the centre? research. research R Kyklos Take an exact structure, perturb one axis, watch what breaks. research. research R Temperament On choirs that sink, pianos that lie, and the wrongness that lets music move. research. research - Cipher A sandbox for classical cryptanalysis, built to teach rather than break. open. we are not building it. open - Replicate Tries to reproduce the result before anyone builds on it. open. we are not building it. open LAST Set0 & SetX the devices the handset is the last chapter, not the first. the agent layer runs on phones people already have, and earns a device only if it becomes the thing people open. draft