Product build roadmap · internal · 2026

The build, sequenced.

The five-layer build order as Now, Next, and Later - written for the founder building solo full-time, and for the founding engineer who inherits it when the gate clears.

Gate-faithful. Now is specs, spikes, and throwaway prototypes; no production code exists before the Phase 0 gate. Columns advance when gates clear, never when a clock runs out. Companion to the plan (operating truth) and the book (product truth). When this document and the plan disagree, the plan wins and this file gets updated.

Section 01

How to read this roadmap.

Now - work allowed before the Phase 0 gate: the MVP spec, the voice and parse spikes, a throwaway demo prototype.
Next - the wedge MVP production build. Entry is gated, not scheduled: R1, R4, and R5 green, plus three or more committed design partners.
Later - layers that only earn their build once the wedge proves out. Each carries its own gate.
Effort sizes (S / M / L / XL) assume one full-time founder building with AI tooling. No calendar commitments anywhere in this document.

This is a plan, not a contract. It gets re-cut at every monthly phase-gate review, and section 07 lists exactly what forces a re-cut.

Section 02

Decisions locked for this roadmap.

Set during roadmap planning, July 4, 2026.

Decision	Call
Roadmap type	Product build roadmap - the five-layer build order as Now / Next / Later.
Gate stance	Gate-faithful. Production layers sit in Next, explicitly blocked on R1 / R4 / R5.
Capacity	Founder full-time through the wedge MVP. The founding-engineer hire stays gated per the plan.
Tracking data source	Manual spine + selective parse for the three highest-value events only: commitment letter, appraisal, inspection response. Parsing breadth is deliberately not the MVP fight.
Voice scope (MVP)	Seller updates only. One recipient, one register. Chase messages are a Later expansion.

Assumed, flagged for confirmation

Assumption	Call	Why
Audit log	In the MVP data model from day one	The compliance file "accrues underneath" only if recording starts with the first deal. Cheap now, painful to retrofit.
Sensitivity tags	Carried as data on every document reference from day one	The matrix is the config table for the document layer. Tags first, enforcement Later.
Paper-tool deep links	In the MVP - link out per deal, no data read	Cheap proof of "integrate, do not duplicate" before real integration.
Seller status link	Deferred to Later	A second surface to design and secure; not needed to validate the wedge.

Section 03

Now.Pre-gate · runs alongside the R1 / R2 / R4 field research

Everything here is disposable or on paper. The point is to make Next start fast and start right.

N1Wedge MVP specEffort MDepends on nothing · start here

A locked data model and screen inventory before the gate clears cuts weeks off the production build, and forces the audit-log and sensitivity decisions early, where they are cheap.

Scope

Data model (deal, milestone, contingency, party, document reference with sensitivity class, audit event) · screen inventory mapped against the dealroom design, with Screen 04 split into operational and regulatory views · sprint plan for X1 to X3.

Done whenThe founding-engineer-ready spec exists and the model doc signs off on it.

N2Voice bootstrap spike (R5)Effort MNeeds one agent's corpus

An onboarding sample plus roughly fifty sent emails trains a profile a top producer cannot reliably distinguish from their own writing.

Scope

One willing agent; train; blind-test drafts against real sent mail. Throwaway harness, no product code. This IS the R5 gate.

Kill-criteriaA red read means the moat is gone - kill, or rebuild the moat around the document layer (plan §11).

N3Selective-parse spikeEffort S–MNeeds sample docs from R1

Commitment letter, appraisal, and inspection response can be detected from documents and email with precision worth a human confirm-queue - never auto-commit.

Scope

Sample corpus from ride-alongs; measure precision and recall per event type.

Done whenWe know whether X4 ships in the MVP or degrades to manual-only.

N4Demo-grade deal room prototypeEffort SNeeds N1 sketch state

A tangible fake-data prototype sharpens the R1 ride-alongs and converts them to design-partner commitments faster than a verbal pitch.

Scope

Static or lightly interactive; reuses the dealroom-design screens; explicitly throwaway.

Done whenUsed in at least three ride-alongs.

Now exit gate = Phase 0 exit gateR1 shows closing-readiness is a top-three pain, and R4 shows the compliance file is acceptable, and R5 (N2) is green, and three or more design partners are committed. Any one fails: stop or pivot before Next begins.

Section 04

Next.Wedge MVP production build · entry gated on Phase 0

Build in this order. Each layer earns the one after it.

X1Layer 1 · TrackingEffort LDepends on N1

Every live deal on one screen, sorted by what needs attention now, makes design partners stop keeping a backup checklist.

Scope

Book-of-deals home screen · manual fast-entry for deadlines, contingencies, milestones · audit log recording from the first deal · sensitivity classification carried on every document reference · deep links out to the mandated paper tool, per the R2 landscape.

Success metricShare of partner deals run without a backup legal pad.

X2Layer 2 · Closing-readiness gridEffort MDepends on X1

A live four-track state per deal - financing, documents, contingencies, disclosures - answers "what is blocking each close, and what do I chase today" at a glance.

Success metricPartners cite the grid unprompted in weekly touchpoints; chase actions originate from the grid.

X3Layer 3 · Voice drafting (seller updates)Effort M–LDepends on X1 + N2 green

Deal events trigger seller-update drafts in the agent's own voice that the agent edits and sends rather than rewrites - and the profile sharpens with every edited send.

Scope

Onboarding bootstrap (sample plus imported sent mail, per N2's proven recipe) · draft-on-event · edit-capture feeding the profile · acceptance-rate instrumentation built in from the first draft.

Success metricVoice-draft acceptance rate holds or rises across each partner's deals.

X4Selective parse, productionizedEffort MDepends on N3 green + X1

Auto-detecting the three make-or-break events with a human confirm-queue beats manual entry, without inheriting Folio's parsing-breadth fight.

Scope

Commitment letter, appraisal, inspection response only. Confirm-queue, never auto-commit. If N3 was red, this drops out and the MVP stays manual - an accepted outcome, not a failure.

Success metricDetection precision at or above the N3 spike; partner confirm time trivial.

X5Design-partner instrumentationEffort S · ongoingDepends on X1

The plan's weekly cadence made real: deals tracked, blockers, voice acceptance rate, the backup-checklist question. A founder dashboard, not a product surface.

Next exit gate = Phase 1 exit gatePartners run live deals through Mantle without a backup checklist, voice acceptance holds, and partners retain across multiple closes.

Section 05

Later.Earned, not scheduled

ID	Layer	Gate to start	Effort
L1	Document-sensitivity active handling - store-vs-link enforcement, access scope, retention auto-purge keyed to time-decay, breach-notification triggers	Phase 1 exit; tags already accumulating from X1	M
L2	Seller read-only status link	Phase 1 exit; partner pull for it	S–M
L3	Voice register expansion - chase messages to attorneys, lenders, the buyer's agent	X3 acceptance proven on seller updates	M
L4	Paper-tool read integration (which tool per R2) - checklist and document status pulled read-only	Phase 2 entry; R2 landscape settled	M–L
L5	Hardened secure-handoff for the catastrophic few: wire instructions, proof of funds, closing disclosure	R7 green - legal and technical feasibility; may need a partner	L–XL
L6	Broker compliance and admin tier	Phase 2 exit; audit-ready files stacked inside one brokerage	L

Deliberately parked, per the model's firm boundaries: capture (3D tours, floor plans - a marketing-asset surface for a later era), paper of record and signatures, CRM and lead pipeline, brokerage back office, and any parsing-breadth arms race with Folio.

Section 06

The dependency spine.

N1 spec ──────────► X1 Tracking ──► X2 Readiness grid
N2 voice spike ───────────────────► X3 Voice drafting ──► L3 Chase register
N3 parse spike ───► X4 Selective parse
X1 (audit log + sensitivity tags accruing) ──► L1 Doc handling ──► L5 Secure handoff (R7)
X1..X3 on live deals ──► L2 Seller link · L4 Paper-tool read
Closed audit-ready files × one brokerage ──► L6 Broker tier

The one number that matters stays the plan's: Next is partners on live deals; the first half of Later is paid retention; L6 is brokerage rollout.

Section 07

What re-cuts this roadmap.

Any Phase 0 research track coming back red (see the plan's kill-criteria) - Next does not start.
N3 red - X4 drops out, and the MVP ships manual-only.
R2 revealing broker-mandated lock-in that blocks solo adoption - the whole wedge re-scopes to a broker-first sale, and this document is void.
Folio or a Zillow-backed dotloop shipping per-agent voice - re-run R6 before starting the next Later item.
The founder-full-time assumption slipping - the Now column survives unchanged; Next stretches.