roadmap

Roadmap (strategic)

Boundary (keep this true or it drifts into a Linear mirror): this page holds themes, intent, and sequencing rationale + a link to each Linear epic. It does NOT hold status, stories, or sprints — those live in Linear (the system of record). For “what’s the live status of X”, open the linked epic. Ideas enter via ideas triage before they become a Linear epic, which keeps Linear clean.

Now (in flight)

• Scores at scale + categories — done (2026-07-06): public benchmark per-integration scores live; the V2 taxonomy (one-primary rule) shipped and direct-booker was unblocked (recommendations issued 2026-07-02/03). Residual: A/B/C grade bands + bulk-score gut-check view (ado-score-model Open questions).
• ChatGPT score pipeline — prove the ChatGPT tracking/scoring path. Why: organic discovery is coming to ChatGPT; first-mover GTM. Linear epic: TBD — link
• Prediction engine (the measurement moat) (added 2026-07-07) — the daily frozen-prediction registry-probe loop is live (2026-07-05): every registry edit becomes a scored natural experiment; next targets are blind novel-keyword prediction (membership F1 ≥0.90) and the LLM-coded reasoning taxonomy. Why: it’s the evidence engine behind the Fixes tab and the registry-SEO advisory (claude-registry-search-decoded). Linear: BUILD-675 (loop, done) / 680 / 681.

Next

• Codex plugin tracking — third live-check tab (app + skills). Why: expansion surface; ChatGPT–Codex convergence. Linear epic: TBD — link
• Website score funnel + company profiles → outreach loop — shipped (2026-07-06): the marketing funnel is live (domain → score funnel; prod-smoked; apex-domain re-point still outstanding). Remaining: watch conversion, iterate on the profile view. Original why stands: turns the public tracker/score into the GTM engine.
• Post-connection optimization (the “Used” layer) (new, added 2026-06-22) — once a connector is connected, optimize tool calls / invocation / ranking over time from live invocation data; includes intent → invocation reverse-engineering (the Alpic feature, flagged must-have) and memory profiling (important, no method yet). Off-platform needs our own SDK tag; on-platform consumes/enriches the host’s API (manufact / alpic). Why: we sell found → used → sold but have only built found (organic/pre-connection); this is the gap the differentiator depends on, and where metatuner plays. Linear epic: TBD — needs scoping
  • Connected discovery = the enterprise-plan variant (2026-06-23): once a user has connected you, do you still appear for relevant prompts? Requires simulating which other apps the user has connected, then testing appearance. (Distinct from organic connection-picker discovery, which is the current focus.) See discoverability-framework.
• Recommendation engine (the “Fixes” tab) (added 2026-06-23) — turn discoverability evidence into prioritized, suggested changes; placeholder/“coming soon” in v1. Why: the loop’s Fix step; needs intelligence to prioritize across organic/connected/tool-level signals (limited tools/credits). Needs a written “how the system works” approach doc first — the direct-booker workstream. Linear epic: TBD

Later

• Rebrand (see ado-platform open questions) — single brandable authority over platform + service. Why: Alpic-style single-brand model. Linear epic: TBD — link

Open questions

• Which themes map to which existing Linear epics (team Product & Build)? — reconcile and fill the epic links.
• Now/Next/Later vs date-based — keep theme-based to avoid false precision?

Timeline

• [2026-07-07] (maintenance) Now-section refresh: scores-at-scale + categorization marked done (public benchmark scores live, V2 taxonomy, Direct Booker unblocked); website score funnel marked shipped (prod-smoked 2026-07-06, apex re-point outstanding); added the prediction engine as a Now theme (daily loop live 2026-07-05; next: blind-F1 identification + reasoning taxonomy).
• [2026-06-29] (internal discussion — Ghost Team stand-up, Vincent + Elliot, dual recording) Offering now concrete (“build to the discovery ecosystem”) → Elliot drafting a roadmap + financial plan (what we build, when, how it maps to revenue/forecast), target draft ~Thursday before he’s offline in Georgia (Sat→next Sun). Durable planning constraints captured: (1) Vincent is offline ~all of January — the team must stay shippable through it (can’t rely on a just-onboarded engineer or a single founding-engineer/CTO who could vanish), which makes founding-engineer recruitment urgent; Elliot’s personal network isn’t the right channel for that hire, so public job posts on the website + a defined role/comp are needed. (2) HQ / raise geography is open: London (build network, raise pre-seed from London angels) then scale to the US (a US entity already exists), weighed against UK employment-law friction — “all timing” given the ~2-month runway. This-week goals: get the website live on the current domain (don’t block on the unresolved name/branding — positioning thinking decays if not shipped) and the tracker URL→score funnel (“put in your URL → your score”), which is the roadmap “website score funnel” item moving into build.
• [2026-06-23] (call — Vincent + Elliot, recorded by Vincent) Added category re-categorization (one-primary rule + Perplexity enrichment; Direct Booker blocker), the recommendation engine / Fixes tab, and connected discovery as the enterprise variant of the post-connection layer. Framework + UI structure → discoverability-framework; score + category decisions → ado-score-model.
• [2026-06-22] (internal discussion — Vincent + Elliot, website/strategy session) Added post-connection optimization (“Used” layer) to the roadmap — agreed it matters but had never been planned. Driven by the metatuner review and Alpic’s intent→invocation feature. Framed product around two discovery layers (organic/pre-connection = built; post-connection = gap) + conversion (“Sold”). Key insight logged: no canonical optimization rules — models change, Anthropic doesn’t publish golden tips, so optimization must be model-scoped (e.g. Haiku). Loop framing drafted in product repo docs/drafts/ado-loop-architecture.md.
• [2026-06-19] (internal discussion — duplicate recordings from Elliot + Vincent) Website score-funnel idea promoted from generic free-tier/profile thinking: homepage asks for a domain/app, shows visibility and competitor presence, then triages users into optimize vs get-present/build. Product-direction stance: build remains a lower on-ramp; continuous channel monitoring and optimization is the durable value.
• [2026-06-16] (internal discussion) V1 scope confirmed: service definition, platform v1, bug/agentic triage, scaling experiments (safely to ~3 components/day), 100 benchmark categories + custom categories + primitive data. A proper product backlog is a deliberate post-V1 exercise (draft/align/inputs), not now. “Table 5” flagged as a dependency for one component. Ongoing “ship → feedback → update roadmap” treated as a standing workstream, not a discrete goal.
• [2026-06-14] (setup) Roadmap created as the strategic layer above Linear: themes + why + epic links, status stays in Linear. Seeded Now/Next/Later from current brain knowledge; epic links pending a Linear reconcile.