Ghost Team Brain

ado-platform

6 min read

✏️ Edit this page on GitHub

ADO Platform (product hub)

Start here for product context. This page is the index + current focus. Spec lives in the repo docs; the backlog lives in Linear; this holds the durable what and why.

What it is

  • An Agent Discovery Optimisation (ADO) analytics platform + public registry tracker: measures how operator capabilities (MCP servers/apps, connectors, skills, callable APIs) are discovered, selected, and invoked by agents across LLM clients. Implementation focuses on Claude (measured today) and ChatGPT (built, unmeasured — flag off), with Codex/Cursor/Gemini as expansion. Built by Ghost Team.

Product knowledge (index)

  • discoverability-framework — the four R’s + the Measure→Score→Fix→Lift loop + the product UI structure (the customer-facing model).
  • ado-score-model — how the ADO score is computed (the four 25-pt buckets + categories prerequisite).
  • architecture — durable invariants + the canonical repo docs to read.
  • roadmap — high-level strategic roadmap (Now/Next/Later) + links to Linear epics. The why/sequencing.
  • ideas — idea-triage funnel (pre-Linear): discuss → accept onto roadmap → or reject with rationale.
  • build-offering — the discovery-optimized BUILD offer: six USPs + the tools-first build philosophy, with per-claim evidence status.
  • product/pricing-packaging — (to add) pricing, free tier, packaging strategy.
  • Backlog / live status → Linear (system of record; never duplicated here — the roadmap links down to its epics).
  • Spec → repo docs (vincentmcleese/discoverability-platform → docs/).
  • design-system — company-wide Ghost Team design tokens for slides, websites, product UI, docs, reports, and apps.
  • naming-and-positioning — unified brand/name criteria, headline candidates, target audience, and website conversion narrative.

Current focus

  • Status (2026-07-07): the platform is live on its staging environment (which is the live env — no separate prod yet): public benchmark per-integration Discoverability Scores, private custom arenas, and a Stripe subscription wall (test mode; Free / Pro $149 / Enterprise). Claude is the measured platform; ChatGPT is built but unmeasured (flag off). A continuous research instrument runs daily (frozen-prediction registry probes → next-day scoring → natural-experiment ledger — see claude-registry-search-decoded).
  • Product UI being rebuilt around the framework (decided 2026-06-23): a score cockpit on entry + tabs Prompts / Discoverability / Fixes / Usage / Registry (Fixes + Usage grayed for v1). Full model + decisions → discoverability-framework; customer-facing mocks are MADE artifacts in the product repo (docs/drafts/ado-loop-architecture.md, docs/drafts/dashboard-overview-mock.html).
  • Partner-log analytics direction (2026-06-22): server logs feed into Ghostteam analytics, then discovery optimization sits on top. Hosting-provider APIs (Alpic/Manufact) can supply logs, but the Ghostteam dashboard should stay standardized rather than splitting the customer experience by provider.
  • Surface: website + tracker rebuilt (Fable 5); tracker UX rolled everything into one search (incl. tools), with the registry tracker split out as a separate top-level feature.
  • Codex plugin tracking (with quan): Codex plugins = app + skills (vs ChatGPT apps = tools; plugins don’t exist on Claude). Codex becomes the third live-check tab (ChatGPT, Claude, Codex); daily morning scrape pipeline ready to merge into backend. Skills DB table TBD.
  • Market tailwind: ChatGPT app review down to 1–2 days, apps submitted daily; organic discovery live in Claude, expected in ChatGPT soon. (Broader market signal belongs in agent-discovery-market — migrate market lines there over time.)
  • #1 build risk: dev velocity — trivial fixes taking 15+ min; mitigations trialed (second Max account, budgeted API spend with hard caps). (Ops concern, not product — candidate to move to a decision/ page.)

Open questions

  • Rebrand (open): Ghost Team may need one short brandable name covering platform + service under a single authority (Alpic cited as the model). Leaning: rebrand Ghost Team rather than run two brands.
  • Score naming (open): headline metric leaning the “[platform] Discoverability Score” (host-scoped) — must stay clearly separate from AEO and leave room to broaden past discovery (discovery and used). Decide what the score is for before locking. See ado-score-model Open questions.
  • Score-surface UX: resolve possible confusion between public tracker benchmark scores and internal/customer dashboard scores; benchmark should likely remain a prominent high-level category health metric inside the platform.
  • Elliot’s ADO flow diagram (research channel) — refine inputs and next steps.

Timeline

  • [2026-06-22] (internal discussion — Alpic/report/product strategy) Product direction sharpened around partner logs and score surfaces: hosting margin is thin, the durable value is server logs → analytics → discovery optimization; partner APIs can feed data but Ghostteam owns the standardized dashboard. Open UX issue: public tracker score vs internal dashboard score could confuse customers; benchmark score should remain a prominent high-level category health view. Navigation feedback: overview/score → analytics → prompts → discoverability, with recommendation engine showing where a customer does not appear, why, and expected rank lift.
  • [2026-06-24] (human correction) Design-system link moved to the related company resources section and retargeted to design-system because the resource is broader than product.
  • [2026-06-24] (code review) Added the reusable design-token resource for Ghost Team apps, extracted from the live marketing frontend.
  • [2026-06-19] (internal discussion — duplicate recordings from Elliot + Vincent) Product/GTM direction for the public website sharpened: lead with a domain/app score funnel and “show up and sell” channel outcome; keep build as an on-ramp rather than the headline. Details live in naming-and-positioning and roadmap.
  • [2026-06-14] (setup) Split into atomic pages: score mechanics → ado-score-model, architecture pointer → architecture. This page is now the hub/index. Roadmap intentionally NOT duplicated from Linear.
  • [2026-06-12] (call — M&B wrap-up; figures as stated by Elliot to the client, from our tracking) ChatGPT app store grown ~200 → ~1,500 apps since early 2026, ~20–30 new apps/day; ChatGPT no-mention app surfacing fully rolled out, multiple apps per prompt; Claude connector usage “exploding” (we hold confidential Claude connector usage data); Claude going consumer (AllTrails, Audible) but submission waits long. Platform pitch: client-defined target prompts run daily + keyword reverse-engineering → recommended metadata changes → resubmit → re-track.
  • [2026-06-09] (internal discussion) Stand-up: core platform pages scaffolded (slow in dev, 5–10s/page — rollups missing); benchmark categories (fixed, every brand lands in one) vs custom categories (user-added). Category creation handed to Elliot. Rebrand discussion opened. Elliot prediction: ChatGPT–Codex merge → organic discovery close → rush; nailing ChatGPT GTM would be huge.
  • [2026-06-09] (call) Codex plugin tracking review with quan: ranking/overview + filters working; plugins to display “Includes”; Codex added as third live-check tab; morning scrape pipeline ready for backend merge.
  • [2026-06-08] (internal discussion) Build slowed by AI rate limiting (20–40 min per heavy request); product “nearly ready to present” but not yet click-through-able.
  • [2026-06-05] (internal discussion) Full platform rebuild: 15 epics / 330 issues in Linear; decided to build the whole dashboard at once with agents (“the hard part is below the waterline” — ~10 versioning tables for score history). Outreach paused until profiles+scores exist (planned: send a company its profile/score → free-tier sign-up → call for more data).

Graph View

Backlinks