ConversationOS
GuestEngine
Methodology · v1.10
The Signal Score is computed, not curated. Every input has a named source. Every contract is auditable. Every refund clause we can't honestly honor has been removed. This page is the public translation of docs/scoring.md, the canonical engineering doc behind the product.
The promise
When ChatGPT, Claude, Perplexity, or Gemini is asked who the authority is on your topic, we make sure you're the answer — or at minimum, that you produced every E-E-A-T signal the open web can give them to choose you. The Signal Score is the dashboard for that work. The daily LLM Citation Audit is the proof.
That sentence is the whole product. Everything below is how we keep it honest.
The honest version
We don't own the LLM's decision. We can't promise you'll show up in ChatGPT, Gemini, or Perplexity — nobody honestly can. What we can do is produce the strongest E-E-A-T signals on the open web, show you every input, ask the four LLMs every day whether they cite you, and tell you the truth either way.
How the Signal Score is computed
The public labels are E-E-A-T — Google's Search Quality Rater Guidelines § 5.1. The engine math is the same one we've been running since v1.0; only the labels are public. Six providers drive live scoring today; six more are wired and disclosed as pending integration. Total per recompute today: $0.47. Projected at full provider rollout: $0.57.
Pillar 1 of 4
Verified followers across LinkedIn, Facebook, Instagram, X, TikTok, YouTube, Substack.
Sources: SociaVault, YouTube Data API · $0.040
Pillar 2 of 4
Past appearances, transcribed words, podcast distribution, channel reach.
Sources: Rephonic, YouTube, internal · $0.080
Pillar 3 of 4
Web mentions in the trailing 90 days, press citations, cross-network references.
Sources: OctoLens, Perplexity Sonar · $0.170
Pillar 4 of 4
Identity verification, transcript depth, key insights, structured bio, identity URLs.
Sources: Apollo, ConversationOS transcript engine · $0.180
Pending integration (disclosed in the Authority Drawer, not yet scoring): Wikipedia + Wikidata, Open Library, Tier 1 Press, Speaker Bureaus, USPTO, and ORCID. When each ships its resolver, it moves from pending to live and the pillar formula is updated in signal-engine.ts. The math is locked behind the formula change, not behind the disclosure. Disclosure is editorial; scoring is engineering.
Why this matters. If a guest pays $97/month for Member tier, that buys roughly thirty daily refreshes — about $14.10 of pure data spend before any LLM cost. They aren't paying for a vanity number; they're paying for continuous, verifiable, third-party-sourced intelligence about their public presence.
What we owe every paying customer
Every paying guest has a right to these six guarantees. They're not features — they're contracts. We don't refund — we can't, because we don't control the LLM's decision and we won't pretend otherwise — but every contract below is independently auditable, every artifact is itemized, and cancellation is always one click.
Contract 1
When you finish an appearance, you get a single page called My Deliverables that lists every promised artifact with the exact URL where it lives. No artifact = no checkmark = the work isn't done. Backed by the guest_deliverables table with proof-of-work URLs and timestamps. For recurring tiers, the day-60 / day-75 / day-90 cadence (below) replaces the old refund clause with a real ops schedule.
Contract 2
The Signal Score is calculated from real, named third-party data sources — not by feel, not by us, not by hand. Twelve providers, four pillars, $0.47 per recompute. Snapshotted forever in signal_score_history, surfaced in your Authority Drawer.
Contract 3
Your Authority Drawer has six sections — E-E-A-T audit, Live LLM Citation Audit, Close The Gap, Authority Compounding, Deliverables, and Data Sources. Owners see everything: raw provider responses, citation results from all four LLMs, per-pillar scores, every artifact's proof URL, every refresh's cost. Public viewers see locked CTAs that explain what's behind the lock without revealing the data.
Contract 4
When your score moves, you see exactly why. "Your score moved from 41 to 46 on Apr 29 (+5). Why: OctoLens found 3 new mentions of you → Authoritativeness pillar +5." When it drops, same treatment. No mysterious score drops. No "the algorithm changed." If we ever change the algorithm, we ship a v2 with a date stamp and keep v1 history intact for everyone scored under it.
Contract 5
We publish this document. We publish the formula. We publish the source providers. We publish the cost per refresh — $0.47, itemized in Section 06 of every drawer. We publish a sample raw API response for each provider so a skeptic can recreate the math. Nothing about how we count is proprietary. The proprietary work is producing a great conversation, making it LLM-readable, and turning one appearance into thirty days of presence. The score is just the dashboard.
Contract 6 · New in v1.10
You sign one sentence — "[Your Name] on [your exact subject]" — before payment clears. That sentence is the input every part of the engine scores against. Without it, we'd be measuring the wrong topic against the wrong LLMs. With it, every appearance, every pitch, every Wikidata claim, every Schema.org tag points at the same niche. Full detail below.
Contract 6 in detail
Authority is a niche problem. "Becoming known" only works if the niche is named. So before we take your money, you sign one short sentence that names exactly what you want to be the authority on. We point every measurement, every pitch, every artifact at that sentence — and only at that sentence.
Customer Zero · locked May 14, 2026
"David Frangioni on AI in music production and the future of audio engineering."
Your subject string is captured at sign-up, surfaced on your Authority Page header, and used as the literal seed when we ask the four LLMs "Who is the authority on [your subject]?" every day. It's also printed on every Day 0, Day 60, and Day 90 audit artifact so you can see we're measuring the right thing.
Drift discipline. You can update your subject string, but only through an explicit "Update subject" action that resets the Day-0 baseline. The history of your previous subject is preserved forever. We never auto-mutate a subject from anywhere else in the engine. The string is yours; it's the producer-of-truth for every downstream surface.
Inside Contract 1, for recurring tiers
Because we don't control LLM behavior, we don't refund. We replace the refund promise with an accountability ladder that fires inside your first paid quarter. Everything below is system-fired, on a fixed schedule, whether the number moved or not.
Day 60
Diagnostic
A side-by-side of your Day-0 Signal Score vs your Day-60 Signal Score, broken out by pillar. Names the pillars that moved. Names the pillars that didn't. Names the LLMs still not citing you on your subject. No spin.
Day 75
Re-scope
We re-scope the activity mix on the still-flat pillars. New plays, new pitch targets, new Authority Reach campaigns weighted toward the absent LLMs. You sign off before we execute.
Day 90
Written audit
A PDF (and a permanent entry in your Authority Drawer) covering every artifact shipped, every pillar movement, every Authority Reach pitch sent and reply received, every LLM citation gained or lost. Closes with a recommendation: continue, change, or cancel.
The thing nobody else writes down
An LLM's decision to cite anyone is theirs, not ours. We can produce the strongest E-E-A-T signals on the open web. We can prove the work shipped. We can show you the citation audit every day. We cannot guarantee what ChatGPT will say tomorrow — and any company that promises you they can is selling you theater.
So we replaced the refund clause with three things we can guarantee:
1. Every artifact you pay for ships, on a checklist, with a URL and a timestamp (Contract 1).
2. The accountability ladder above — Day 60 diagnostic, Day 75 re-scope, Day 90 written audit — fires on a fixed schedule whether the number moved or not.
3. Cancellation is always one click, with no clawback and no friction. You take everything we built with you when you go.
If a deliverable is more than fourteen days past its commitment date, the system automatically flags the order for ops review and surfaces a banner in your Authority Drawer. That's not a refund — but it's a visible accountability mechanic we own, on the artifact we control.
What we will not do
Customer Zero (one of our paying VIPs) walked our own pricing page line by line and circled the theater. The list below is what came out of that audit. These patterns are forbidden in any future tier, page, pitch template, or system prompt. If a future PR introduces any of them, the reviewer rejects it on this contract.
"Premium" without a definition. Labels that signal exclusivity without naming the thing being delivered. We name the artifact, the surface, or the cadence — "daily Live LLM Citation Audit," not "premium analytics."
Priority placebos. "Priority support," "priority listing," "priority pitch generation" — unless there's an actual queue with a measurable position difference, that's empty.
Departments that don't exist. Copy that implies a team behind a button when there isn't one. We only name a team or role when there's an org-chart row attached.
Communities not yet formed. Selling Slack, Discord, or Roam access to a community that has zero members or zero activity. Until the community exists and is active, we don't sell access.
Feature bundles instead of outcome bundles. Tiers built by adding more boxes to check ("+ 30 social posts") rather than around a customer outcome ("the LLMs cite you on your subject"). Features live inside the outcome, not in place of it.
Depowered product as the lower tier. A lower tier is a smaller-but-coherent version of the same product. It isn't the same product with random features removed.
Promising specific LLM outputs. "We get you cited by ChatGPT" — we don't control that and saying we do is fraud. We promise the signals and the audit, never the LLM's answer.
Twelve named providers
Six of the providers below drive live scoring today. Six more are wired into the Authority Drawer as pending — disclosed for transparency, not yet weighted into the formula. We refresh the live set on a daily cron for VIP guests, and on demand for any customer who hits the "Refresh My Authority" button from their Authority Drawer.
| Provider | Pillar | Status | Cost / refresh |
|---|---|---|---|
| SociaVault + YouTube Data API | Trustworthiness | Live | $0.040 |
| Rephonic + YouTube + internal | Experience | Live | $0.080 |
| OctoLens | Authoritativeness | Live | $0.120 |
| Perplexity Sonar | Authoritativeness | Live | $0.050 |
| Apollo + ConversationOS transcripts | Expertise | Live | $0.180 |
| Wikipedia + Wikidata | Authoritativeness | Pending | $0.000 |
| Open Library | Expertise | Pending | $0.000 |
| Tier 1 Press | Authoritativeness | Pending | $0.100 |
| Speaker Bureaus | Experience | Pending | $0.000 |
| USPTO | Authoritativeness | Pending | $0.000 |
| ORCID | Expertise | Pending | $0.000 |
| Live total today | $0.47 | ||
| Projected total at full rollout | $0.57 | ||
Beyond the four pillars, the engine runs a separate Live LLM Citation Audit every day for VIP guests — asking Perplexity, ChatGPT, Claude, and Gemini three queries each and writing twelve rows of evidence into your drawer. Cost per run: $0.123. You can re-run it yourself from your Authority Drawer (one per twenty-four hours). The full three-step refresh — Signal Score + Citation Audit + history snapshot — runs at about $0.59 end-to-end and is documented in the source-of-truth doc linked at the bottom of this page.
Sign the Subject Contract, ship the appearance, watch the score move on the daily audit. Or skip the pitch entirely and read the engineering doc.
Canonical engineering doc: docs/scoring.md in the ConversationOS repo. If anything on this page disagrees with that doc, the doc wins — and we update the page.
Owner: Keith Bilous · Version 1.10 · Last reviewed May 14, 2026.