Worker T9, 2026-09-26, 00:37 to ~02:30 Beirut. Inputs: 00-MANDATE.md, RULES.md, business-plan-2026-09-25/PLAN-v3.md and its deployed page (site/index.html, sections 01 to 13, read in full including the eight decisions in section 11), the three critics, BRAND-BRIEF-2026-09-15.md, and the wiki synthesis §9. Web research ran on 2026-09-26 through two research sub-agents (pricing, beta precedents) and two direct searches (MiCA, Lowe v. SEC). Grades: A primary source fetched today, B credible secondary or a search snippet of a primary source, C unverified or my inference. Our own stores are A where a script in scripts/ produced the number.
Three things, in this order: (a) find out which parts of the companion a working trader opens every day and which they ignore, so the build plan after week 2 is shaped by use, not by us; (b) produce the first paying members from people who already use it, instead of from cold DMs; (c) produce quotable, named evidence ("I stopped opening five tabs") for the paid launch. It is not a marketing stunt and not a waitlist.
| Criterion | Rule | Why we want it | How we check |
|---|---|---|---|
| Active trader | Trades on-chain at least 3 days a week in the last month | The product is for decisions made fast; a tourist cannot tell us where it is slow | Self-declared on the form; optional wallet check (below) |
| Chain | Trades on at least one chain we cover: Solana, BSC, Robinhood Chain (Base as a secondary if T1/T2 add it) | We cannot learn from someone whose tokens we do not see | Form field |
| Size | Typical ticket $100 or more, or monthly volume $5,000 or more | Below this, a $69 subscription is more than their trading fees (1% of $5K is $50 a month; see §4), so they are not the paying customer we are testing for | Form field; optional wallet check |
| Current tools | Uses at least two of: a terminal (Axiom, GMGN, Photon, Padre, Fomo), a screener (DexScreener, Birdeye), a wallet tracker (Cielo, Bubblemaps), caller channels, X | The value claim is "one view instead of five apps"; they must have the five apps | Form field |
| Voice | Willing to write or voice-note weekly in English (Arabic and French accepted if Thomas reads them) | Feedback is the price | Form field plus the free-text answer |
| Mix target | At most 30% of a wave from any one source (one caller group, one network) | A single community gives us one trading style | Recruiting tag on the form |
| Excluded | Sanctioned jurisdictions; anyone running a paid caller group who wants to repost the feed; related parties unless disclosed (Hayden's team, our own team) | Rights, redistribution, and a clean record | Form fields plus manual review |
Optional wallet check. The applicant signs a message with one public wallet address. We read its last 30 days of swaps from our own tape or a public explorer to confirm "active" and "size". Optional because it is intrusive and many traders rotate wallets; people who do it get priority in the wave. We store the address hashed after the check unless the tester opts in to the "my wallet in context" features later. (Design choice, C.)
Tally allows webhooks on the free plan, with a SHA256 Tally-Signature header and retries (A, https://tally.so/help/webhooks, fetched 2026-09-26). Typeform's free plan has no webhooks (A, https://www.typeform.com/pricing). So: Tally.
Fields, in order (13, of which 9 required, about 3 minutes):
The webhook posts to an endpoint we own and writes beta/applications.jsonl (append-only, one row per submission, raw payload kept per the data retention rule).
Evidence: Linear hand-picked about 10 users a week and shipped a new version per cohort (A, First Round Review, above); fomo opened with a 1,000-user beta cohort (A, PR Newswire, https://www.prnewswire.com/news-releases/fomo-debuts-groundbreaking-social-crypto-trading-app-aiming-to-onboard-10m-users-302447294.html); Kaito invited about 2,000 of a 35,000 waitlist (C); Superhuman's PMF survey gives direction from about 40 responses (A, https://review.firstround.com/how-superhuman-built-an-engine-to-find-product-market-fit/).
| Wave | Opens | Seats | Who | Purpose |
|---|---|---|---|---|
| 0 (friendly) | build day 10 to 14, before the public beta | 5 | Elie, two or three traders Thomas knows personally, one of Hayden's intros | Break it before strangers see it; test the tooling end to end |
| 1 | beta open (see §5 for dates) | 15 | Best-fit applicants, wallet-checked first | First real read on daily use |
| 2 | wave 1 + 7 days | 15 | Fill the gaps wave 1 exposed (other chain, other speed, other ticket size) | Breadth |
| 3 | wave 1 + 14 days | 15 | Same, plus replacements for anyone dropped | Reach about 40 active for the PMF survey |
| Cap | 50 active at once | One human (Thomas) plus Vesper can read 50 people's feedback weekly; more cannot be read properly (C) |
A seat freed by a drop goes to the next applicant in the queue. The queue is never "full"; it is ranked.
Cadence. One weekly check-in (Mondays, due by Wednesday 23:59 tester local time), plus any number of ad hoc "moment reports".
Weekly check-in (in the Telegram bot, 5 questions, about 3 minutes; voice notes accepted and transcribed): 1. How many days did you open Caverio this week? (auto-filled from our access log, tester confirms) 2. Name one trade decision this week where Caverio was open. What did it show, and did it change what you did? 3. What did you still have to open another app for? 4. What would you remove? 5. One thing that confused you or looked wrong (screenshot welcome).
Moment report (any time, one bot command /moment): screenshot plus "what I was deciding" plus "what Caverio showed or missed". This is the highest-value feedback because it is tied to a real decision at a real time, and we can replay the situation from our store at that timestamp.
End-of-month survey (Sean Ellis PMF question, unchanged wording so it compares over time): "How would you feel if you could no longer use Caverio?" Very disappointed / somewhat disappointed / not disappointed; plus "what is the main benefit" and "who else would benefit". 40% "very disappointed" is the published threshold (A, First Round Review, Superhuman).
What counts as useful. Each feedback item gets a score 0 to 3 from a rubric, first pass by Vesper (LLM, stated cost below), confirmed or overridden by Thomas in the weekly digest:
| Point | Criterion | Example that scores | Example that does not |
|---|---|---|---|
| +1 Specific | Names a screen, field, token, time, or source | "The flow gauge on $XYZ said inflow at 14:05 but the chart showed dumping" | "The UI is confusing" |
| +1 Grounded | Tied to a real decision or a real session | "I skipped it because the caller had flipped bearish and you showed that" | "Would be cool to have AI" |
| +1 Actionable or verifiable | We can change something, or reproduce it from our store | Screenshot with a timestamp; "move X above Y" | "Make it better" |
A useful item scores 2 or more. Bad feedback means: abusive; spam; generated filler with no reference to anything on screen (the rubric catches this as 0 Specific); or copied from another tester. Bad feedback is not "negative feedback". "This is useless because it misses every Pump.fun launch in the first 3 minutes" is a 3.
Where it is captured. Everything lands in our own store, never only in Telegram history: beta/feedback.jsonl (append-only: testerId, ts, kind [checkin|moment|survey|dm], raw text, voice file path, transcript, screenshot path, rubric scores, reviewer, linked situation id if any). Screenshots and voice notes go to beta/media/ with sha256. This follows the 09-19 retention rule: keep everything received that could be useful.
Each tester has a 28-day cycle from their first login. On day 28 at 12:00 Beirut the evaluator runs:
inputs per tester, over the current cycle:
active_days = distinct days with >= 1 authenticated session on the gate (access log)
checkins = weekly check-ins submitted on time (0..4)
useful_items = feedback items with rubric score >= 2 (check-ins, moments, survey free text)
hard_strike = any of: redistribution of the feed or screens without permission,
sharing the passcode, abuse of other testers or the team,
bad-feedback flag confirmed by Thomas twice in the cycle
extensions = cycles already granted after the first
on_probation = previous result was PROBATION
decide(t):
if t.hard_strike: return DROP_NOW (no re-application for the beta)
if paid_is_open and t.extensions >= 1: return OFFER_FOUNDING (see below)
if t.active_days >= 8 and t.checkins >= 3 and t.useful_items >= 2:
return EXTEND_28D
if t.on_probation: return DROP (polite, may reapply as a paying member)
if t.active_days >= 6 and (t.checkins >= 2 or t.useful_items >= 1):
return PROBATION_14D (one warning, re-evaluate on day 14)
return DROP
cap: at most 3 extensions (four months free), then OFFER_FOUNDING regardless.
Notes on the thresholds (all C, to be tuned after wave 1): 8 active days in 28 is "about twice a week"; a trader who opens it less is not using it for decisions. 3 of 4 check-ins allows one missed week. 2 useful items a month is a low bar on purpose; the bar that matters is honesty, not volume. The warning message goes out automatically on day 21 to anyone trending toward DROP, with the counts, so nobody is surprised (Tesla's FSD Beta published its strike rule in advance and restored access later; B, CleanTechnica 2022-07-09, https://cleantechnica.com/2022/07/09/tesla-full-self-driving-beta-5-strikes-youre-out/).
OFFER_FOUNDING. When paid opens, testers with at least one extension get the Founding price locked for life plus their current free cycle finished before billing. Testers who never extended get the public price and the public trial.
Drop mechanics. Gate passcode revoked; Telegram: banChatMember then unbanChatMember, so they leave the group but can come back later through a new link (A for the API methods, https://core.telegram.org/bots/api; the ban-then-unban pattern is C). A one-line thank-you with their counts. Their feedback stays in our store under the data-rights clause.
| Piece | Choice | Cost (estimate, stated basis) |
|---|---|---|
| Application form | Tally free, webhook to our endpoint | $0 (A, free-plan webhooks) |
| Endpoint | A route on the existing signal-room server or a Cloudflare Worker, writes beta/applications.jsonl |
$0 incremental; under 1 MB/month disk at 200 applications |
| Access | The existing 3047 passcode gate (year cookie, five tries then lockout, device registry): one passcode per tester with an expiry date = cycle end; revocation is deleting the row | $0; needs a small Bolo hook (per-passcode expiry and tester id on each session so active_days can be counted). This is the "trial hook" item already in the 09-25 plan, re-scoped |
| Private group | Telegram supergroup, admin bot. One named invite link per tester: createChatInviteLink with member_limit=1, expire_date 72 h, creates_join_request=true; bot approves join requests only for Telegram ids on the roster (approveChatJoinRequest) |
$0 (A, Bot API) |
| Feedback capture | Same bot: /checkin (5-question flow), /moment (photo plus text or voice), free DMs to the bot captured as dm. Voice via Deepgram (key works per memory) |
Deepgram: 50 testers x 4 voice notes x 1 min = 200 min/month, a few dollars (C) |
| Rubric scoring | Vesper, LLM first pass per item | about 50 testers x 8 items x 1.5K tokens = 600K tokens/month (C); on the Claude subscription route, not metered API |
| Store | beta/roster.json (tester id, handle, wave, cycle dates, status), beta/feedback.jsonl, beta/media/, beta/decisions.jsonl (every evaluator output with its inputs) |
under 1 GB/month with screenshots (C) |
| Evaluator | beta/evaluate.py, runs daily, acts on day-28 testers, writes decisions, sends warnings on day 21; Thomas approves DROP lines in the digest before they execute for the first two cycles |
CPU negligible |
| Weekly digest | Monday 09:00: (1) counts per tester, (2) top 5 issues clustered with quotes and links to the situation ids, (3) what we shipped because of last week's feedback ("you said, we did"), (4) PMF score once a month, (5) pending DROP / PROBATION decisions for Thomas. Internal version to Thomas and Vesper; a short "you said, we did" version posted to the group | Closing the loop publicly is the main thing that keeps testers writing (C) |
The short clauses are drafted in §3. In summary: information not advice, no guarantee of accuracy or uptime, no execution or custody by us; we may use, quote anonymously and act on their feedback without payment (named only with consent); the feed, screens and data are for their own use, no forwarding, scraping, reposting, or resale, screenshots outside the group only with permission; the beta can end or change at any time; the extend / drop rule as published.
| Product | Gate | Cohort | What was asked in return | Dropped inactive? | Conversion | Grade, source |
|---|---|---|---|---|---|---|
| Superhuman | Waitlist plus intake survey; onboarding call; PMF survey only to users active twice in 2 weeks | Waitlist over 275,000 at $30/month (Feb 2020) | Survey; segmenting to the users who loved it moved "very disappointed" from 22% to 33% to 58% | No rule found | Paid from day one | A: https://review.firstround.com/how-superhuman-built-an-engine-to-find-product-market-fit/ ; B: https://techcrunch.com/2020/02/28/superhuman-ceo-rahul-vohra-on-waitlists-freemium-pricing-and-future-products/ |
| Linear (2019 to 2020) | Waitlist of about 10,000 with "why do you want it, what are your problems" | About 10 hand-picked per week | Weekly email to daily users, in-app feedback; "every week we basically made a new version" | Not stated | About 10% of waitlist became users in year one, about 1,000 DAU at launch | A: https://review.firstround.com/linears-path-to-product-market-fit/ |
| fomo (fomo.family) | Waitlist plus invite friends | 1,000 initial beta, public 6 May 2025 | No formal programme described | Not stated | Fee-based, 0.5% per side | A: PR Newswire release (URL above) |
| Friend.tech | Invite code from an existing user | First airdrop reached about 44,000 (C) | Usage, rewarded with weekly points | No | Points, not subscription | B: https://www.coingecko.com/learn/how-to-use-friend-tech |
| Hyperliquid | Closed alpha; points for the alpha period to 31 Oct 2023 | Not stated | Usage | No | Points, then fees | A: https://hyperliquid.gitbook.io/hyperliquid-docs/points |
| WhatsApp beta (TestFlight) | Capped programme | TestFlight caps: 10,000 external testers, builds expire in 90 days | In-app feedback | Yes: removes inactive testers when full | n/a | B: https://wabetainfo.com/wa-testflight/ ; A: https://developer.apple.com/help/app-store-connect/test-a-beta-version/testflight-overview/ |
| Tesla FSD Beta | Safety score to enter | Not stated | Driving behaviour | Yes, five forced disengagements revoked access, restored later | n/a | B: CleanTechnica 2022-07-09 (URL above) |
Lessons applied above. Screen on entry (Superhuman, Linear) → the form's free-text question and the wallet check. Small steady waves (Linear) → 15 a week. Define "active" before judging feedback (Superhuman surveyed only active users) → active_days is the first test. Prune with a published rule and a way back (WhatsApp, Tesla) → day-21 warning, ban-then-unban. Crypto early-access programmes reward volume with points (Friend.tech, Hyperliquid, Axiom and GMGN referral codes, B/C), which gets farmed → our rubric rewards specificity, not volume. No product I found ties access length to feedback quality (C: absence of evidence after a targeted search).
The gtm critic's bands: warm outreach replies 25 to 35%, cold 10 to 15% (their estimate, C). A free beta is an easier ask than a $69 payment, so I use the top of each band for reply and assume half of repliers apply (C).
| Source | What we already have | Motion | Touches | Expected applicants (C) |
|---|---|---|---|---|
| Caller channels we watch | 5 Telegram channels in callers.json (A, scripts/t9_caller_counts.py, updated 2026-09-25 21:39Z): Pow's Gem Calls (1,070 msgs, 183 calls), Sugar Calls (306 / 154), Columbus Trades (76 / 28), OverDose Gems Calls (45 / 38), CSC 卡神社基金 (60 / 9); 412 calls, 300 priced, lifetime |
Borrowed distribution: Thomas asks each admin to post the beta once, offering the admin a seat and a named credit. Do not DM members scraped from these channels | 5 admins | 1 or 2 admins say yes; a post in a few-thousand-member channel yields 5 to 15 applications, low fit on average |
| Thomas's network | Personal traders; Elie; the Fomo profile Thomas already uses | Direct DMs, variant A | 20 | 5 to 7 |
| Hayden's network | The plan's five contractual intros (still to be written, §5) | Ask Hayden that at least two of the five be active traders, not only launch-stack people; variant A via intro | 5 | 2 to 3 |
| Fomo network | 3,488 handles profiled (09-11 memory figure; data, not permission) | Not a DM list. Thomas posts in-app and replies publicly where Fomo's social features allow; the gtm critic's "cold list with a dossier" warning applies | public posts | 0 to 5 |
| X threads from the public record | X API dark for us (reading), but Thomas can post by hand | One thread per week from the situations log: a real situation, what the companion showed, what happened. Link to the form. Variant C | 2 to 3 threads | 3 to 8, depends on reposts |
| Farcaster | Nothing held | Variant C posted in memecoin / Base / degen channels from Thomas's account if he has one (C: channel names and activity not verified tonight) | 2 posts | 1 to 4 |
| Robinhood Chain community | We run a live book and a complete holdings tape on RH Chain | Variant B to the community admins (Discord/Telegram; which ones is T1/T4 territory, not verified here); we are one of few tools covering it | 2 to 3 admins | 2 to 6, high fit |
Sum of the ranges: 14 to 50. The middle is about 30. The weakest assumption is the admin posts; if two admins decline, the total falls to about 15 and the recruiting takes two weeks, not one. Wave 0 plus wave 1 (20 seats) needs about 30 applicants at a 2-in-3 accept rate.
What we do not do. No DMs to handles we only know from scraping. No paid posts. No "limited seats" scarcity copy (the gtm critic: scarcity at zero customers reads as desperation). No word "alpha", "smart money", "edge", "signal", "guaranteed", "safe".
Written short, first person, the way Thomas talks in voice notes: direct, a bit informal, no hype. Placeholders in [brackets].
Variant A: DM to a trader Thomas knows, or via an intro
Hey [name], quick one. I've been building a thing for how I actually trade: one screen that pulls what I'd normally check across five apps, where the money's moving, which callers are on a token and whether they've flipped, which wallets are in, volume, flow, all on one page for each token. It doesn't tell you what to buy, it just shows you the picture faster.
I'm opening it to a small group first. Free for a month, and the deal is simple: you use it when you trade and tell me what's useful and what's rubbish, once a week. If the feedback's real you keep it free another month. If you go quiet you drop out, no hard feelings.
Want in? Two-minute form: [link]
Variant B: to a group or community admin (caller channel, Robinhood Chain community)
Hi [name], I'm Thomas, I run Caverio. We read your channel as one of the sources in a trader tool we're building (it shows what callers, wallets and money flow are doing on a token, side by side, no buy/sell calls from us).
We're opening a closed beta: free for a month for traders who give honest weekly feedback. Would you be up for posting it once to your group? You'd get a seat yourself and a credit on the builders page. I'm happy to show you what it shows about your own calls first, so you know what you're pointing people to.
Form's here if you want a look: [link]
(Note for Thomas: variant B tells the admin we read their channel. That is honest and I think necessary, and it is also the moment the admin decides whether they are fine with it. T1 covers whether reading and displaying their calls needs permission; if T1 says it does, this message is also the ask.)
Variant C: public post (X or Farcaster), attached to one real situation
I trade memes and I got tired of running five tabs to make one decision. So we built one screen: money flow, callers and whether they've turned, wallets in, volume, all per token, as it happens. It shows, it doesn't pick.
Here's what it showed on [token] at [time], and what happened after: [image]
Opening a small beta. Free month for honest feedback, weekly. Useful feedback keeps you in, silence drops you. Apply: [link]
Drafts, not legal text. They are short on purpose so a lawyer can mark them up in one sitting. Placeholders: [LEGAL PERSON] (the entity or named sole trader from decision 4 of the 09-25 plan), [JURISDICTION].
Caverio beta terms. Caverio is operated by [LEGAL PERSON] ("we"). By using the beta you agree to these terms. 1. What it is. Caverio shows market information gathered from public blockchains and third-party sources, organised per token. It is a publication of general information. It does not execute trades, hold funds, or connect to your wallet for trading. 2. Beta. The service is unfinished. Features, data and access may change, break or end at any time without notice. Access is free during the beta and is personal to you. 3. Access rule. Your access runs in 28-day cycles. It is extended when you use the service and give feedback as described on the beta page, and ends if you do not, or if you break these terms. The rule is published and applied the same way to everyone. 4. Your use. For your own use only. You may not share your passcode, forward, scrape, republish or resell what you see, or use it to build a competing product. Screenshots outside the beta group need our permission. 5. Feedback. You give us a free, perpetual, worldwide right to use, copy and act on anything you send us (text, voice, screenshots). We may quote it without your name. We will only use your name or handle with your consent. 6. No warranty, limited liability. Provided "as is". We do not promise it is accurate, complete, timely or available. To the extent the law allows, we are not liable for any loss, including trading losses, arising from your use of it. [Cap: amounts you paid us in the last 12 months, which during the beta is zero.] 7. Ending. You can stop any time. We can end your access any time. 8. Law. These terms are governed by the law of [JURISDICTION]. [Disputes: courts of X / arbitration, lawyer to choose.]
Information, not advice. Caverio shows what is happening around a token: prices, flows, wallets, what callers and accounts are saying. It does not tell you to buy, sell or hold anything and does not take into account your situation or goals. Nothing on it is a personal recommendation. Crypto tokens, especially new and small ones, can lose all their value in minutes, can be manipulated, and can be impossible to sell. Data can be late, wrong or missing. Past behaviour of a token, wallet or caller says nothing reliable about the future. Decide for yourself and only risk money you can lose. We and people connected to us may hold or trade tokens that appear in Caverio; see the disclosures page.
Two wording rules that go with it, from the 09-25 ops critic and still binding: no "max safe size" or any sizing language; no green/red verdict wired to a push. In the companion, "a situation lit up in flow and callers" is an observation; "buy zone" (layer 3) is where this statement gets tested hardest, and it should go to the lawyer before layer 3 ships (§6).
Privacy. For the beta we collect your Telegram id and handle, email, the answers you give on the application, feedback you send (including voice and screenshots), and when you log in and what you open. If you choose to sign a message with a wallet, we check its public activity and keep only a hash of the address unless you opt in. We use this to run the beta and improve Caverio. We do not sell it. We keep it while the beta runs and for up to 24 months after, unless you ask us to delete it (email [address]); feedback already used in the product stays. [If EU/UK testers are admitted: GDPR/UK GDPR basis = contract and legitimate interest; controller = LEGAL PERSON; right to access and erasure. Lawyer to confirm.]
Covered by 3.1 items 2, 3 and 7. The extend / drop rule in §1.6 is published verbatim on the beta page, so the term "as described on the beta page" has something concrete behind it.
Sources. Caverio combines our own records of public blockchain activity with data from third-party providers. Some values we show are calculated by us from those sources rather than copied. Third-party data remains subject to its providers' terms; you may not extract it from Caverio. [Attribution line per provider where their terms require it, e.g. "Price data from CoinGecko".]
Cross-reference T1 (T1-SOURCES-AND-RIGHTS.md): the field-to-permission table there decides which fields are raw, which are derived only, and which cannot be shown at all. Until T1's table exists, the rule from the 09-25 ops critic stands: every displayed price since 23 Sep is DexScreener-sourced and DexScreener is the largest exposure, so a free beta still shows derived state only. Free does not fix redistribution: showing a provider's data to 50 strangers is redistribution whether or not they pay. This is why the rights file stays blocking before the beta, not only before paid.
What I could establish tonight, and what the lawyer must answer:
| Question | What I found | Grade |
|---|---|---|
| Is a per-token information screen "investment advice" in the US? | The Investment Advisers Act excludes bona fide publications of general and regular circulation that give impersonal advice not attuned to a client's needs (Lowe v. SEC, 472 U.S. 181, 1985). A companion with personal watchlists, "my wallet in context" and push alerts drifts toward personalised; whether tokens are securities at all is a separate question | B: https://supreme.justia.com/cases/federal/us/472/181/ (search snippet, not fetched) |
| EU? | MiCA Art. 3(1)(24): "providing advice on crypto-assets" means personalised recommendations to a client on crypto-asset transactions; doing it commercially needs authorisation as a CASP; ESMA says the MiCA advice perimeter is broader than MiFID II's | B: https://www.mica.wtf/mica/title-i-subject-matter-scope-and-definitions-art.-1-3/article-3 ; https://www.esma.europa.eu/publications-and-data/interactive-single-rulebook/mica/article-81-providing-advice-crypto-assets (search snippets) |
| UK? | The cryptoasset financial-promotions regime (in force since October 2023) may catch marketing a service that promotes specific tokens to UK consumers | C: not checked tonight |
| Where is the legal person? | Undecided (09-25 decision 4). Thomas is in Beirut; tax residence, entity location and governing law all follow from that choice | C |
| Admit US / EU / UK testers in the beta, or geofence? | Lawyer's call. My default: admit, keep the product impersonal (same screen for everyone, no "for you" recommendations, no sizing), exclude sanctioned jurisdictions | C |
| Layer 3 "buy zone / sell zone" | The single feature most likely to read as a recommendation under all three regimes, even worded as "historically, tokens in this state..." | C |
Ask the lawyer for: (1) a yes/no on the publisher posture for the companion as designed; (2) the list of features that would break it (personal alerts, watchlist-based ranking, zones); (3) governing law and entity location; (4) GDPR minimum if EU testers are admitted; (5) the refund and sanctions clauses for paid open.
| Product | Category | Plan | Price / month | Annual | Included (short) | Audience | Model | Source | Grade |
|---|---|---|---|---|---|---|---|---|---|
| GMGN | Terminal / bot | Single | $0 | n/a | Full terminal, 7+ chains | Meme retail | 1% per trade | https://docs.gmgn.ai/index/gmgn-fees-settings | A |
| Photon (SOL) | Terminal | Single | $0 | n/a | Terminal | Meme retail | 1% per buy and sell | https://pies-organization.gitbook.io/photon-trading/photon-on-sol/photon-fees-sol | A |
| Axiom | Terminal, spot + perps | Cashback tiers | $0 | n/a | Terminal, perps, cashback ladder | Meme retail to pro | 1% base, 0.75% at top tier | https://axiompedia.com/compare (2026-08-14) | B |
| Padre ("Terminal") | Terminal | Single | $0 | n/a | Terminal, cashback | Meme retail | about 0.8 to 1% (sources conflict) | https://coincodecap.com/padre-review | B |
| Trojan | Telegram bot | Single | $0 | n/a | Bot, copy trading | TG retail | 1%, 0.9% with referral | https://coincodecap.com/trojan-bot-review | B |
| fomo | Mobile app | Single | $0 | n/a | Mobile meme trading, SOL/Base/BNB/RH Chain | Mobile retail | 0.5% per side, $0.95 minimum on SOL | intercom help article (secondary) | B |
| BullX | Terminal | n/a | n/a | n/a | Trading disabled 1 Jun 2026 | n/a | was % fee | https://crypto.news/bullx-pauses-meme-trading-tool-but-keeps-wallet-access-open/ | B |
| DexScreener | Screener | Free for users | $0 | n/a | Projects pay: Enhanced Token Info $299 one-off, boosts, ads | Everyone | Sells visibility to token teams | https://marketplace.dexscreener.com/product/token-info ; https://docs.dexscreener.com/boosting | A |
| Birdeye | Screener | Free / Pro | $0 / about $45 | n/a | Pro: advanced filters, more alerts, export | Retail | Subscription + data API | https://blog.pumpparade.com/reviews/birdeye-review-2026-fees-features-verdict/ | B |
| DEXTools | Screener | Standard / Premium | hold 1,000 / 100,000 DEXT | n/a | No ads, wallet tracker, alerts | Retail, holders | Token-hold gate | https://info.dextools.io/discover-all-our-plans-full-defi-power-at-your-fingertips/ (2024 post) | A (old) |
| Nansen | Analytics | Free / Pro | $0 / $69 | $49/mo billed annually | Pro: all premium features, 2,000 API credits | Retail to pro | Subscription + credits; $99+ and $999+ plans retired | https://docs.nansen.ai/getting-started/credits | A (docs) |
| Arkham | Analytics | Free core | $0 | n/a | Labels, dashboards, alerts | Retail / pro | Freemium; paid tiers unpublished | aggregator | C |
| LunarCrush | Social analytics | Individual / Builder / Scale | $90 / $300 / $900 | $72 / $240 / $720 | Galaxy Score, sentiment, API | Traders to builders | Subscription | https://lunarcrush.com/pricing | A |
| Santiment | Analytics | Free / Pro / Max | $0 / $49 / $249 | $529 / $2,700 per year | Free = 30-day lag; Pro = real time | Retail to quant | Subscription | https://app.santiment.net/pricing | A |
| Glassnode | On-chain | Advanced / Pro | $49 / configurable | annual | 24 h resolution, 10 alerts | Macro BTC/ETH | Subscription + credits | https://studio.glassnode.com/pricing | A |
| Messari | Research | Basic / Enterprise | $0 / custom | reported $6K to $34K per year | Lite and Pro retired | Institutions | Enterprise | https://docs.messari.io/user-guides/welcome/plan-deprecation-faq | A (retirement) / B (price) |
| Kaito Pro | Narrative AI | Monthly / annual / enterprise | not shown; enterprise seat $833 | 25% off annual | Community, MCP | Funds, pro traders | Subscription | https://www.pro.kaito.ai/pricing | A |
| Token Terminal | Fundamentals | Free / Pro | $0 / $350 | $297.50/mo annual | Dashboards | Analysts | Subscription | https://tokenterminal.com/pricing | A |
| Dune | Query | Free / Analyst / Plus | $0 / $75 / $399 | $65 / $349 | Credits | Analysts | Subscription + credits | https://docs.dune.com/learning/how-tos/credit-system | A |
| Cielo | Wallet tracker + bot | Free / Pro / Whale | $0 / about $59 / about $199 | 1, 6, 12 month | Free 250 wallets (10 on SOL); Pro 1,000; wallet discovery | Copy traders | Subscription + 0.6 to 1% trade fee | https://docs.cielo.finance/wallet-tracking/subscribe (tiers A; prices C) | A / C |
| Bubblemaps | Holder clusters | Free / BMT holders | $0 | n/a | Premium gated by holding BMT | Due diligence | Token-hold gate | https://wiki.bubblemaps.io/bubblemaps-v2/premium | A |
| DeBank | Portfolio | VIP | $15 | $150 | Label, extras | EVM DeFi | Subscription | snippet | C |
| Zerion | Wallet | Premium | about $8.25 | $99 | Halves swap fee | Wallet users | Fee + subscription | zerion.io blog snippet | B |
| Paid caller groups (Whop / Discord / TG) | Signals, community | Examples | $60 to $225, cluster at $99 | lifetime about 5 to 8x monthly | Blackroom Alpha $120 or $600 lifetime (A, https://www.blackroomalpha.com/); Ocula $99; 3M Trading $99; Wealth Group $225 | Meme retail | Subscription | https://whop.com/blog/crypto-trading-whops/ (2025-02-18) | A / B |
What traders actually pay the terminals (DefiLlama fees API, fetched 2026-09-26, data through 24/25 Sep; A): 30-day fees GMGN $52.9M, Axiom $52.9M, fomo $36.1M, Maestro $2.0M, Trojan $1.1M, Photon $0.46M. The top eight took about $145M in 30 days, which at about 1% implies about $14 to 15B of monthly volume (C, arithmetic on A data). GMGN, Axiom and fomo take about 97% of it.
The 09-25 plan (v3, deployed at caverio-plan.pages.dev) was written for the ladder: it sold one rule's windows on the Now feed, alerts on green windows, and a public record of that rule's calls. Thomas's 00:28 decision takes the ladder out and makes the companion the build. The plan's commercial scaffolding was mostly product-agnostic and survives; the parts that were about what is sold and when do not.
| 09-25 item | Verdict | What changes and why |
|---|---|---|
| Wedge: Founding Member subscription | Keep | Still the right first wedge for the reasons all three critics gave: one counterparty class, no availability contract, no related party. What it grants changes: the companion, not Now + green-window alerts |
| $69 / $600, no seat cap | Keep | §4.2: the companion strengthens $69. No seat cap on paid; the 50 cap applies only to the free beta |
| $39 fallback at Gate 1 | Change | To $49 (§4.2: nothing credible sits at $39; $49 is the analytics floor) |
| Free 7-day trial | Change | Replaced by the 28-day feedback-for-access beta for the first 50; the 7-day trial returns at paid open for the public |
| $2,500 pre-launch scan | Keep, as a parallel line | It needs no build and no SLA, and it is the only line that can bring non-equity cash before the companion is ready. The companion makes the scan better (the money brief and discovery layers are most of what a scan reads). Owner: Vesper writes, Elie sells to a top-20. It competes for Vesper's time with the beta digest; that is the real constraint (C) |
| Launch room $6,000, monitoring $1,500/month | Keep, deferred further | Still waits for a stranger payment and Gate 1. Now also waits for the companion's situation room, which is most of its build. The ops critic's 12 to 20 days for the launch room overlaps the companion build rather than adding to it, if the situation room is designed to be pointed at one token for 48 hours (C) |
| Hayden's launch: free, contractual (publish rights, five named intros, disclosure) | Keep, one addition | Ask that at least two of the five intros be active traders for the beta (§2.1). The related-party disclosure goes on the record page and the disclosures page the risk statement links to |
| 14-day reliability gate, 7 measures, restart on incident | Keep the definition, change what it measures and when it closes | The measures (ingest gap, gap share over 10 min under 2%, crossing-to-alert p95 from provider time, missed events by replay, watchdog fires, print-gate bypasses, credit runway) were written for one price stream. For the companion they apply per source the beta displays (chain tapes, price broker, caller channels, Fomo, wallet feeds). "Crossing-to-alert" becomes "event-to-screen" (provider time to the moment the situation shows it). The gate must close before paid, as before. For the free beta it only needs to be running and disclosed ("data can be late, wrong or missing"), because no one is paying for an SLA. Clock starts when ingest for the beta's sources is live (earliest about 4 Oct), so it closes around 18 to 22 Oct, before paid in both timelines |
| One public record: B view, net-sized, start 23 Sep 15:34, blocked calls and losers, conflicts, no login | Keep the principle, change the object | The companion "does not judge; it shows", so a record of one rule's calls is no longer the product's proof. Proposed: two records. (a) The deep-pullback record continues unchanged as R&D evidence and stays public, because the promise was made and the rule keeps running (B-minus keeps the paper engine). (b) A new situations log: every situation the companion opened, when, from which sources, which lights came on, and what the token did after on the forgiving-win definition (T6). Same rules: one basis, one start date, no edits, no login, losers included. This is the companion's honest record and is what beta testers see first (§1.8). The market critic's warning still binds: freeze basis and start date before anyone sells |
| Alerts: Telegram push on every green window | Change | Green/red verdicts were already cut. Alerts become observations: "situation on $X lit up: callers + flow in the last 15 min". Whether these are personal (watchlist-based) is a legal question (§3.6) |
| Record page as Bolo's first deliverable (2 Oct) | Change | Bolo's first deliverable is the companion build per MASTER.md. The deep-pullback record page is a small job and can still ship early if Thomas wants it for credibility in recruiting variant C; the situations log ships with the situation room |
| Rights file (5 sources, field-to-permission table, derived only) | Keep, widen, blocking before the beta | Widened to every source T1 lists (callers, X, Fomo, wallets, flow sources). Blocking before the first stranger logs in, not only before the first payment (§3.5: free is still redistribution) |
| Legal person and wallet custody | Keep, re-timed | Legal person is needed before the beta (terms need a party; §3.1). Wallet custody and sanctions screening at payment are needed only before paid open |
| Terms, risk, privacy, refund live and human-reviewed | Keep, split | Terms, risk, privacy, beta clause: before wave 0 strangers (wave 0 friends can run on a draft). Refund mechanics: before paid |
| "Max safe size" removed; green/red as observation | Keep | Applies to the companion, and to layer 3 zones most of all (§6) |
| Conflict disclosure (live book, Hayden, referrals) | Keep | Also: callers or admins who get a beta seat for posting it are disclosed on the builders page |
| Caverion knock-out in writing; caverio.world | Keep | Unchanged and cheap; the beta form and group should use the Caverio name, so it is needed before recruiting starts |
| Gate 0 (2 Oct): legal, wallet, terms, rights, record, Hayden note | Change | Becomes "before wave 1": legal person, terms, rights file, Hayden note. Wallet moves to paid open |
| Gate 1 (23 Oct): 3 paying strangers + 25 logged replies of 75 conversations, or 1 scan paid | Change | The beta changes the funnel: the people we talk to are using the product for a month before we ask for money. Proposed Gate 1 in two parts. 1a (end of wave 1's first cycle): Sean Ellis survey, at least 40% "very disappointed" among at least 25 active testers; if under 40%, the digest names what the "very disappointed" segment uses and the next cycle builds only that (the Superhuman method). 1b (paid open + 21 days): at least 8 paying members with no equity, at least 5 of them from the beta, or one scan paid. Miss 1b: $49, 75 more conversations, no new surfaces |
| 25 Oct: paper book survives 30 days | Keep, as R&D | Unchanged; it governs the deep-pullback record, not the companion |
| 22 Oct strategy kill date at n=30 | Keep, as R&D | Unchanged |
| 30 Nov: $2K MRR + 3 scans + 1 launch room | Change | $2K MRR (30 paying) moves to paid open + 6 weeks: 19 to 27 Dec, realistically January. 3 scans by 30 Nov can stand (no build). Launch room by 30 Nov: drop, it cannot be signed before Gate 1b |
| 24 Dec: $10K MRR or the wedge is wrong | Change | Move to end of February 2027 (same 90-day distance from paid open as the original had from 25 Sep, roughly). Keep the public "we say so" clause |
| Track 3: labelled outcomes for agents ($199/month) | Keep, parked | The situations log plus forgiving-win outcomes is a better labelled dataset than one rule's calls; still gated on provenance item 9 and a reference customer |
| Crypto decommission B-minus | Keep, one change | B-minus keeps chain tapes, price paths, Now writer and deploy, v2 paper engine, paper monitor. With the ladder out, the Now writer and its deploy move from "keep, it is the product" to "keep until the companion replaces it, then retire"; the chain tapes and price paths become the companion's backbone. T2 should confirm |
| Scope freeze until Gate 1 ("build only against a paying stranger") | Change | Superseded: Thomas has decided a build. The freeze principle survives in a narrower form: after the companion ships to the beta, new surfaces only from beta evidence (a digest item with at least 3 testers asking, or the PMF segment) until Gate 1b |
| Bolo in 30 days: record page, trial hook, gate bugs, provenance item 9 only | Change | Bolo's scope is now the companion build. The trial hook becomes the beta hook (§1.7, per-passcode expiry and tester id on sessions); provenance item 9 stays needed for the situations log |
| Elie: list cut and one-pager, no outbound before Gate 1 or first stranger payment | Keep | Unchanged |
| Thomas: 15 conversations a day from 2 Oct | Change | From beta recruiting start (a week before wave 1): about 5 a day plus the admin asks, then the weekly digest and the group. Fifteen a day was sized for a cold paid funnel |
| # | 09-25 decision | Status after the companion decision |
|---|---|---|
| 1 | Founding Member is the wedge; scan is stranger cash; launch room waits | Survives as is |
| 2 | Hayden free and contractual | Survives, plus "two of five intros are traders" |
| 3 | $69 / $600, 7-day trial, no cap, $39 fallback | Changes: beta replaces the trial for the first 50; fallback $49 |
| 4 | Legal person and wallet by 2 Oct | Survives, re-timed: legal person before wave 1; wallet before paid |
| 5 | Public record, B view, 23 Sep start, no login | Survives for the deep-pullback record; add the situations log with its own frozen basis and start date |
| 6 | Build only against a paying stranger; scope freeze to Gate 1 | Superseded by the build decision; replaced by "new surfaces only from beta evidence until Gate 1b" |
| 7 | Cloudflare DNS token, caverio.world, Caverion clearance, rights file | Survives; rights file widened and moved before the beta |
| 8 | Decommission B-minus | Survives; Now writer becomes "keep until replaced" |
Derived by scripts/t9_timeline.py from a build start of Sun 27 Sep (the day after Thomas's clean-up), wave 0 three days before build end, wave 1 two days after, paid open at the end of wave 1's first 28-day cycle, Gate 1b 21 days after paid open. Calendar days, because Bolo works every day; if Bolo's range is working days, add about a third.
| Milestone | 09-25 plan (ladder) | "A few days" (4-day build, Thomas's expectation) | Bolo range, fast (12 days) | Bolo range, slow (20 days) |
|---|---|---|---|---|
| Build ends | n/a (no build) | 30 Sep | 8 Oct | 16 Oct |
| Legal person, terms, rights file, Hayden note | 2 Oct | 1 Oct (tight: the rights file depends on T1 plus reading 5 to 10 terms pages) | 7 Oct | 12 Oct |
| Wave 0 (5 friendly) | n/a | 27 Sep | 5 Oct | 13 Oct |
| Recruiting starts (admin asks, DMs, first thread) | 2 Oct conversations | 27 Sep | 3 Oct | 11 Oct |
| Wave 1 (15) | 2 Oct trials open | 2 Oct | 10 Oct | 18 Oct |
| Reliability gate clock | 25 Sep to 9 Oct | starts when beta sources ingest; about 28 Sep, closes 12 Oct | about 4 Oct, closes 18 Oct | about 8 Oct, closes 22 Oct |
| Wave 2 / wave 3 | n/a | 9 / 16 Oct | 17 / 24 Oct | 25 Oct / 1 Nov |
| Gate 1a (PMF survey, end of wave 1 cycle) | n/a | 30 Oct | 7 Nov | 15 Nov |
| Paid opens | 9 Oct | 30 Oct | 7 Nov | 15 Nov |
| Gate 1b (8 paying, 5 from beta, or 1 scan) | 23 Oct (3 paying + 25 replies) | 20 Nov | 28 Nov | 6 Dec |
| First scan paid (no build, independent) | 16 to 30 Oct | 16 to 30 Oct | 16 to 30 Oct | 16 to 30 Oct |
| $2K MRR (30 paying) | 30 Nov | about 11 Dec | about 19 Dec | about 27 Dec, realistically mid Jan |
| $10K MRR or say the wedge is wrong | 24 Dec | end Jan 2027 | mid Feb 2027 | end Feb 2027 |
My view: plan on the slow column and be pleased by the fast one. The 09-25 ops critic's 12 to 20 days was for the launch room, which has fewer inputs than the companion; the companion's three layers have more (T1 and T2 will say how many sources are actually ready). The "few days" column is only possible if the beta opens on layer 2 alone with the money brief as a strip (§6).
The problem in one line. The 09-25 plan sold one wedge (a feed of windows, alerts, and a record). The companion is three layers: a market brief, a discovery-to-situation room, and later strategy zones. Three layers is three products' worth of sources, correctness work, and presentation, built by one builder and sold by one founder.
What the critics said about scope (all on 09-25, all still relevant). - Ops/build: five rebuilds in seventeen days; "five build days is fiction", realistic 12 to 20 for a room with four inputs of which three did not exist yet. The companion has more inputs, not fewer. - Go-to-market: "two clocks, one builder, one selling founder: parallel, neither reaches proof." Three layers is three clocks inside the product. - Market: the daily tools are free; the only thing that is not free is something they cannot get elsewhere. Of the three layers, the market brief (layer 1) is the most available elsewhere (DefiLlama, Artemis, CoinGecko, Nansen, all free or cheap), the situation room (layer 2) is where our own chain tape, caller stance tracking and wallet context combine into something a free terminal does not show, and zones (layer 3) is the most differentiated and the most legally exposed.
The argument for shipping the situation room first, with the brief as a strip.
1. Layer 2 is where the decision happens (buy / no-buy / exit / size); that is the moment the companion claims to save, and the moment beta feedback is richest (moment reports tie to a token and a time).
2. It uses what we already own and run: chain tapes on three chains, the caller pipeline (5 channels, 412 calls; A, scripts/t9_caller_counts.py), Fomo, the provenance programme. Layer 1 needs new aggregate sources.
3. A strip (one line of the brief: stablecoin supply change, chain of the day, BTC vs alts, the hottest cap band, with today / week / month toggles) gives the "environment" feel at a fraction of the build and tests whether traders look at it at all before we build the full page.
4. It makes the "a few days" timeline plausible and gets real users sooner, which is what the beta is for.
5. It keeps layer 3 out of the beta until the lawyer has seen it (§3.6).
The argument against. 1. Thomas's product is the three layers together; a situation room alone is closer to "a better screener", which is exactly the category the market critic says is priced at $0. The brief is what makes it a companion rather than a terminal. 2. Layer 1 is the cheapest to make correct (aggregate numbers, slow cadence, fewer phantom-print traps) and the most shareable (a daily brief is good content for recruiting variant C and for X), so leaving it as a strip wastes the easiest marketing asset. 3. Beta feedback on a partial product can mislead: testers will ask for what the brief would have given them, and we will build it anyway, one request at a time and in a worse shape. 4. The price case for $69 (§4.2 point 5) is weaker on layer 2 alone.
My recommendation. Ship the situation room plus the brief as a strip to the beta; build the full brief page during the beta if testers open the strip (measured by the access log, not by opinion); keep zones out until Gate 1a and the lawyer. Price at $69 only if both layers are in the paid product at paid open; otherwise $49. This keeps one clock (the situation room) while keeping Thomas's shape visible.
scripts/t9_caller_counts.py (caller channel counts, reads projects/crypto-2026-09/data/watchtower/callers.json), scripts/t9_timeline.py (the date columns in §5.3). No other number in this document comes from our stores.Status: DONE