Signal Desk · prepared for [Client]

You already read the market like an investor.
This turns the reading
into an edge.

Same habit, same hours on your phone. A capture inbox, a 60-second hype filter, a thread-pulling template, and a decision journal that make every hour of reading worth more, and every decision better documented.

0
changes to how you read. The habit stays exactly as it is, the leisure half included.
60 sec
per hype check, same five questions every time, every verdict kept.
5
small steps, all on the phone: Capture, Filter, Pull the thread, Decide, Review.
$0
to run Tiers 0 and 1. Tier 2 automations run for about $7 to $17 a month (USD). Setup and build fees are separate.

The short version

  • The habit is the asset. Not the manual numbers: the hours reading AI and crypto markets on X and Reddit. Half leisure, half research that feeds real money. Nothing about it changes.
  • The value leaks between reading and deciding. Leisure and signal share one feed, open loops evaporate, the Gemini check forgets, digging never becomes a thesis, nothing separates luck from skill.
  • A light layer, not a new routine. Five small steps on the phone, plus guardrails and frameworks from disciplined investors, backed by the research on how retail investors lose and win.
  • The centrepiece is a 60-second hype filter with a fixed shape, saved inside the free Gemini you already use, plus a routing rule for which tool answers which question.
  • The one non-negotiable guardrail: fewer, better decisions, not more trades. Attention-driven buying loses within weeks and phones make it worse. Success is trades avoided and calibration gained, never research volume.
  • Three tiers, fee and running cost shown apart. Tier 0 frameworks, Tier 1 light tooling on free tiers, Tier 2 built automations on accounts you own at about $7 to $17 a month to run. Fees are listed with the tiers.
  • The system enforces rules you write. It never sets them. Sizes, exits, and allocation are yours. We supply templates and plumbing, not portfolio rules; nothing here is investment advice.
  • Recommended path: Tier 0 and Tier 1 live in week one. Run for 30 days and log the friction. Then decide which steps have earned a Tier 2 build.
What we heard in discovery

The manual work is not the opportunity. The reading is.

1

The manual numbers on your computer are small, deliberate, and hands-on. They keep you close to your own business. They stay manual; nothing here touches them.

2

Your real time sink, and your real edge, is reading the market: AI and crypto, on X and Reddit, mostly on your phone.

3

About half is leisure. The other half is pulling threads that feed real decisions: roughly a fifth of your income goes into AI stocks and crypto, aiming high with sensible risk over a long horizon.

4

Your filter today is free Gemini, used ad hoc: is this real, is it a big deal, what is this person's angle.

5

You do not want a new hobby or a rigid system. You want more value from the hours you already spend, with the same enjoyment.

The Buffett comparison is the right frame. Most of an investor's job is reading and understanding, and Munger said fast decisions are only possible because of the long hours spent quietly reading beforehand. You already do the expensive part. What is missing is the cheap part: a record of what you believed, why, and what would change your mind.
Where the value leaks today

Reading is the raw material. The leaks are between reading and a decision.

These five are our read from the discovery call. The 30-day baseline is where they get confirmed or corrected against your actual numbers.

X and Reddit feed Something piques you Ad hoc Gemini check Decision 1 · one feed for both 2 · lost on the couch 3 · the filter forgets 4 · no written thesis 5 · no review
1

Signal and leisure share a feed

The loudest voices win, and on X they are measurably the worst: 56% of finance influencers are anti-skilled, and they have the most followers.

2

Open loops evaporate

A thread catches your eye on the couch; by the laptop it is gone. With no capture step, the best ideas are the ones most likely lost.

3

The filter forgets

Each Gemini chat starts from zero, asks slightly different questions, and its verdict is not kept. You cannot look back and see which calls were right.

4

Research does not accumulate

Good digging ends in your head, not in a one-page thesis you can revisit, defend, or kill.

5

No review loop

Without a record of why you bought, held, or passed, outcomes cannot be split into luck and skill, so the process cannot improve.

The design principle

More signal per hour. Fewer, better decisions.

Before the tools, the constraint that shapes them. The research on individual investors points one way.

11.4% vs 17.9%

What the most active traders earned per year against the market, across 66,000 households.

Barber and Odean, 2000
−4.7% in 20 days

Return on the stocks Robinhood users piled into, on average, over the following 20 days.

Barber, Huang, Odean, Schwarz, 2022
+1.8% to −6.5%

Crypto influencer calls: day one to day 30. Worse when the influencer claims expertise and has a big following.

Merkley et al., 2024
97% lost money

Of people who kept day trading for 300+ days. No sign of learning with experience. Fast reaction is not where the edge is.

Chague, De-Losso and Giovannetti
So the multiplier is not more research delivered faster to the phone. Phones already make trades measurably worse for the same investor, and more information leaves accuracy flat while confidence climbs. What does move outcomes is unglamorous: a short pre-action checklist, if-then rules written in advance, and forecasts that get scored. The multiplier is a capture-and-structure layer that makes each hour of reading compound into a written, scored track record, with friction added exactly at the moment of decision. Everything below follows from that.

Two honest caveats: most of this evidence comes from equities, funds, and medicine rather than crypto, and the influencer study ends in December 2022. The pattern transfers plausibly; the numbers are illustrative, not guarantees.

Signal Desk

Five small steps, all on the phone. Each one guards the next.

A phone gesture or a short template at each step. Tools named in each card are free unless marked.

1

Capture every open loop

3 seconds

Share it to one place instead of holding it in your head or researching it in the moment.

On the phone

Share sheet in X or Reddit to a private Telegram channel (one iOS Shortcut), or straight into a Gemini Notebook.

Guardrail

Captured is not acted on. Anything from X or Reddit lands on a timestamped list first; the 72-hour clock starts here.

2

Run the 60-second hype check

Same five questions every time

A saved assistant with a fixed rubric replaces the ad hoc prompt. Every check returns the same verdict card, and the card is kept.

On the phone

A Hype Filter Gem in the free Gemini app, plus a routing rule for which tool answers which question.

Guardrail

Verdicts are Noise, Watch, or Dig. Never Buy or Sell. A claim check, not a valuation.

3

Pull the thread into a thesis

Six fields, one page

The claim, who benefits, what the market already believes, the base rate, what would falsify it, and the bear case as if it already happened.

On the phone or laptop

One Gemini Notebook per thread. Drop in the filing, the whitepaper, the token docs; it answers only from those and quotes them.

Guardrail

Rule of five, our rule: declare five sources per thesis before reading more. The pattern behind it: past a point, confidence climbs and accuracy does not.

4

Decide by rule, in writing

90 seconds

One line of thesis, a probability, a position size that follows a rule you wrote, and a kill criterion: if state X by date Y, I am out.

On the phone

A voice note or a five-field form that lands in your journal sheet. Longer templates get abandoned within weeks, so this is the smallest capture that works.

Guardrail

The pre-trade checklist, built from your own last ten losers, and the position cap card. You write both; the templates only make them unavoidable at 11pm.

5

Review monthly, score the process

20 minutes

Which verdicts were right. Which theses hit their kill criteria. Which sources you acted on were worth it.

How

A scored pass through the journal. Quarterly, two numbers only: hit rate and average stated confidence.

Guardrail

Grade the process, not the P&L. A good decision can lose and a bad one can print; that gap is variance, and grading by outcome teaches the wrong lesson.

The 60-second hype filter

Ad hoc prompts give ad hoc answers. Five fixed gates give a verdict you can keep and score.

The rubric copies how professional fact-checkers work: they do not read a claim harder, they leave it. Fact-checkers who read laterally, opening tabs about the source instead of scrutinising the page, identified the legitimate site 100% of the time against 50% of historians and 20% of students, and they were faster.

  • 1

    Leave. Who is this and what is their angle?

    Track record, are they holding it, allocated early, paid, is a disclosure present. Investor.gov name check when a person is pitching an asset.

    SIFT lateral reading · Munger's incentive lens · SEC social-media red flags
  • 2

    Trace. What is the primary artifact?

    The filing on EDGAR, the on-chain unlock schedule, the actual paper. If there is none, say so.

    SIFT trace to original · EDGAR full-text search, free on mobile
  • 3

    Material? Does it change the total mix?

    Would a reasonable investor see this as changing the picture on a position I hold or would take? Or is it noise? Most posts fail here instantly.

    Borrowed from the US Supreme Court's test of what is material to a reasonable investor. Here a filter question, not a legal test
  • 4

    Base rate. How often does this shape pay off?

    Only about 1% of large companies have sustained 30% revenue growth for five years. Ask before adjusting for the story.

    Tetlock's outside view · Mauboussin's base rates
  • 5

    Bear case. It is twelve months later and this went to zero.

    Write why, in one line. Imagining the failure as already happened raises the ability to name causes by roughly 30%.

    Klein's premortem
Hype FilterTelegram · verdict card
https://x.com/…/status/… "Company X just signed a $10B deal, this is going 5x"
Ran the five gates. Here is the card.
ClaimCompany X signed a $10B contract; stock will rise fivefold.
Source, angleAccount 4 months old, 41k followers, self-described trader, discloses a long position. Three prior "5x" calls, none paid off.
Primary artifact8-K filed today mentions a non-binding letter of intent, no dollar value. No press release from the counterparty.
Material?Yes if binding; today it is not. Position held: none.
Base rateLetters of intent of this size close roughly half the time and rarely at the headline number.
Bear caseTwelve months on: LOI lapsed, stock round-tripped, dilution to fund capacity.
VerdictWatch · medium confidence
Cooling-offEnds Thu 10:42. Re-check for a definitive agreement first.
NoiseWatchDigInsufficient evidence
Branch card · crypto

Check the supply before the story

What unlocks in the next 90 days, to whom, as what share of float. Roughly 90% of token unlocks put pressure on price, and the slide typically starts up to 30 days early. Team anonymity, VC allocation, paid-promotion disclosure. DefiLlama's unlock calendar is free and needs no login.

Branch card · AI stocks

Check who is actually buying

Customer concentration is disclosed: Nvidia reported four customers at 61% of revenue combined in fiscal Q3 2026, up from 36% a year earlier. Capex versus operating cash flow, hardware depreciation life, whether the buyer is funded by the seller, guidance versus narrative. All free on EDGAR.

QuestionToolWhy this one
Who is this poster, what is the thread, what are people replying?Grok, free on XX describes it as the one assistant that searches X posts in real time; no other mainstream assistant has that access.
Is the claim true? Run the five gates.Hype Filter Gem in free GeminiA Gem holds the rubric so it runs identically every time, in the mobile app, at $0.
Is there a primary source?EDGAR, DefiLlama, the paper, the docsThe Trace gate must land on a real artifact, never on an assistant's citation.
Does the market agree this is a big deal?Polymarket or Kalshi odds, read freeOnly when a liquid market exists on the underlying event (a launch, an approval, a ruling). If it disagrees with the thread, that is a fast contradiction signal.
Second opinion on a position-changing claimPerplexity, free tierBest citation accuracy of eight tools tested; the free allowance (3 Pro searches a day) is thin, so reserve it.
Deep dive on the source documentsGemini Notebook, freeAnswers only from the documents you give it and quotes them. Share to notebook from the phone.
One hard rule inside the routing: Grok is never the verifier. In the Tow Center test of eight AI search tools, Grok answered 94% of queries incorrectly and 154 of 200 citations led to error pages. Paid tiers were more confidently wrong than free ones. Grok is for X context; Gemini plus a primary source is for truth. The assistant is told to state the bear case before any verdict, because chat assistants are measurably inclined to agree with whatever you already believe; instructions reduce that failure rather than eliminate it, which is why the Trace gate lands on a real artifact. And when the post is gone, the account is private, or there is no primary artifact, the verdict is Insufficient evidence, with no invented base rate.
Guardrails that protect the returns

Written as if-then rules, because that is the form the evidence supports.

None of this reduces your reading. It governs only the moment where reading turns into money. Treat it as a menu: in the week-one session you keep, drop, or rewrite each rule, and every number (hours, caps, source counts) is yours to set. The values shown are starting points to react to.

Cooling-off
If it came from X or Reddit, no buy for [your number; 72 hours is a common starting point]. It goes on the Watch list with a timestamp.
A suggested starting value, not a finding. Behind it: influencer gains reverse over days two to five, and herded names lose about 4.7% in 20 days.
Two tabs on X
If I am on X for signal, I am in the pinned Signal list of 30 to 50 accounts I trust. For You is declared leisure.
Lists are free, chronological, and pin to the top of the timeline. Follower count and posting volume are inverse quality signals.
Rule of five
If I have five sources on a thesis, I decide or I park it. I do not keep reading.
Our rule. The pattern behind it: more inputs left accuracy flat and doubled confidence.
Written before bought
If there is no one-line thesis, probability, size, and kill criterion in the journal, there is no position.
Checklists and scored forecasts are the interventions with evidence behind them.
Position cap
If I size, I follow the sizing rule I wrote: a maximum per name of [my number], and a lower maximum for narrative bets. The card only holds the rule I chose.
Many investors use fractional Kelly and a hard cap because edge estimates are guesses. Which rule you adopt is your call, or your adviser's.
Source scorecard
If I act on a source, it gets a row. Score it at 30 and 90 days.
Your own scoreboard beats any hype filter; the anti-skilled influencers have the most followers.
Phone friction
If the decision is on the phone, the checklist runs first. Research can arrive fast; the buy button does not.
Smartphone trades are measurably worse for the same investor in the same month.
Grade the process
If I review, I score the decision, not the P&L.
A good decision can lose money and a bad one can print. That gap is variance.
Frameworks borrowed from disciplined investors

Templates, not lectures. Each one lands somewhere in the loop.

FrameworkIn one lineWhere it lands
Decision journal (Kahneman-endorsed, Farnam Street template)Write what you believed and why before the outcome is known, because hindsight rewrites the story.The 90-second voice note at Decide.
Kill criteria as states and dates (Annie Duke)"If I am, or am not, in state X by date Y, I am out," decided at entry while calm.Mandatory journal field; the review pulls every entry whose date has passed.
Pre-mortem (Gary Klein)Assume it already failed; write why.Gate 5 of the rubric and one line in every thesis.
Checklist (Gawande, Pabrai)A short pre-action ritual beats skill and willpower.The pre-trade checklist built from your own losers.
Fractional Kelly and hard caps (Thorp)Size to survive being wrong about your edge.If you adopt a sizing rule, the position cap card holds your numbers, one screenshot in Photos.
Circle of competence (Buffett, 1996 letter)Knowing the boundary matters more than the size of the circle.A written in/out list; the rest of the feed becomes entertainment, not a buy signal.
Outside view (Tetlock, Mauboussin)How often do things of this sort happen?Gate 4 of the rubric.
Second-level thinking (Howard Marks)What does the consensus already believe, and what am I seeing that they are not?Field three of the thread-pull template. X consensus is first-level thinking at scale.
Incentive-caused bias (Munger)Show me the incentive and I will show you the outcome.Gate 1: "what is this person's angle" becomes a checklist, not a vibe.
Hype cycle (Gartner)Trigger, peak, trough, slope, plateau.Vocabulary for placing each narrative; a crypto position is an attention bet and is sized as one.
Three tiers

Three tiers. Fee and running cost, kept apart.

Running cost is what the tools charge, in USD. Less Clicking's fee is what setup and build cost. Start with Tier 0 and Tier 1 in week one; let 30 days of real use decide whether any Tier 2 build has earned its keep.

Week one
Tier 0

Frameworks

$0 to run half a day together
  • The 60-second rubric on your lock screen
  • The two-tab rule on X (a pinned Signal list)
  • Thesis template and 90-second journal entry
  • Kill criteria as states and dates
  • Pre-trade checklist from your last ten losers
  • Thread-pull template, six fields, one page
  • Position cap card, your numbers
  • Source scorecard sheet, one row per source you act on
  • The 72-hour rule and the monthly review ritual
You get: the rubric card, templates, checklist, cap card, and review ritual, in a format you keep on your phoneLess Clicking fee: [to be set: fixed fee for the half-day session and card set]Maintained by: you; nothing to maintain
Week one
Tier 1

Light tooling

$0 to run optionally about $90 a year
  • Hype Filter Gem and a Steelman-then-Skeptic Gem in free Gemini
  • One-tap capture: share sheet to a private Telegram channel
  • A Gemini Notebook per thesis, fed by the share button
  • Free primaries wired in: EDGAR alerts, DefiLlama unlocks, Investor.gov, Ground News, prediction-market odds
  • Optional: one reading hub (Inoreader Pro, about $90 a year) pulling Reddit feeds, Google Alerts on your positions, EDGAR filings, and newsletters into one app with AI summaries
You get: the two Gems, the Shortcut, the Telegram channel, the notebooks, the primaries wired in, a one-page runbookLess Clicking fee: [to be set: fixed fee for the setup session and handover]Maintained by: you; if a vendor changes a free tier, we adjust in a short session
Tier 2

Built automations

$7 to $17 a month (USD), on accounts you own
  • Hype-check bot: share a link, the rubric runs on a paid non-training model, the verdict card is back in Telegram in about 30 seconds and appended to your thesis sheet
  • Monthly review nudge: one question per open position, Sunday morning
  • Price-move alerts that quote your own kill criterion back to you
  • Weekly brief from a curated set of accounts and subreddits
  • Source scorecard auto-scored at 30 and 90 days
You get: the bot, the sheets, the alerts, the brief, the scorecard, a runbook, and every credential handed to youLess Clicking fee: [to be set: fixed fee per module, quoted after the 30-day review; optional light maintenance at a monthly rate]Running cost: a small server (about $6), the model at roughly $1 to $2 for your volume, $8.32 for the feed bridge if you want the weekly X briefMaintained by: you own the accounts and the server; Less Clicking maintains the workflows under the optional rate, otherwise handed over with the runbook
The decision we are asking for now: go ahead with Tier 0 and Tier 1 in week one, and hold Tier 2 until the 30-day review. Nothing in Tier 2 is priced or committed until then.

Why Tier 2 is that cheap to run. Capture is native and free (share sheet, Telegram bot). The workflow engine is free self-hosted n8n on a small server you own; Zapier's and Make's free tiers cannot carry a multi-step chain. The model is paid on purpose, for privacy. Watching X by API is the one line that can quietly dominate a bill, since X now charges per post read with no free tier, so a flat feed bridge or a scoped account list caps it. Not proposed: the paid crypto analytics tier at $49 to $699 a month. Institutional tools, no decision-quality gain at your scale.
What we will not do

The lines we hold

  • Automate the manual numbers.
  • Replace or reshape the leisure half of your reading.
  • Build anything that trades, signals trades, or nudges toward more trades. The bot returns a verdict card and a falsifier prompt, never a buy or sell.
  • Put brokerage, exchange, or wallet access into any tool.
  • Sell you a subscription. Paying buys volume and speed, not trustworthiness.
  • Set your position sizes, exits, or allocation. Those are yours, and your adviser's if you have one. The system enforces the rules you write and never writes them for you.
What success looks like

Measured in trades avoided and calibration gained

  • Baseline first: trades per month and turnover over the last six months, and how many positions have a written thesis today.
  • Open loops captured versus lost. Target: nothing worth pulling gets lost.
  • Time from capture to verdict. Target: same day.
  • Positions without a written thesis and kill criterion. Target: zero.
  • Trades per month flat or down; stated conviction per trade up. Monthly review done; quarterly hit rate and confidence logged.
  • Explicitly not a metric: ideas surfaced, articles summarised, hours of research delivered.
The first 30 days

Live in week one. Earn the build by day 30.

Week 1

Tier 0 and Tier 1 go live

Seed the Signal list with the 30 to 50 accounts you already trust. Write the pre-trade checklist from your last ten losing trades. Set your own position cap numbers. Build the two Gems and the share-sheet Shortcut together.

Weeks 2 to 4

Run it. Log friction, not results.

Which step got skipped, where the phone made it awkward, which verdict cards were useless. The friction log is the spec for anything built later.

Day 30

Decide what earned a build

Review the friction log and the trade count. Pick the Tier 2 pieces that have earned it, if any. Some people stop at Tier 1 and never need more; that is a fine outcome.

Privacy and data

Your book stays out of free-tier AI.

  • Positions, sizes, and net worth never go into a free-tier AI. Google's terms for the unpaid Gemini API say human reviewers may read input and output and, in their words, "do not submit sensitive, confidential, or personal information." Run the rubric on public claims freely; keep your book out of the prompt. Where the Tier 2 bot needs your thesis and positions, it runs on a paid model whose terms exclude training, at roughly $1 to $2 a month.
  • Grok on X shares your interactions with xAI for training by default. The opt-out is in X's privacy settings. Worth switching before pasting anything about your portfolio.
  • You own every account and key. Gemini, Telegram, the server, the sheet. Less Clicking builds and hands over; nothing routes through us.
  • No tool gets brokerage, exchange, or wallet access. Price alerts read public market data only.
  • Process only, by design. Less Clicking supplies templates, a rubric, and plumbing. Every number that touches your money (position caps, kill criteria, allocation) is written by you. If you work with a financial adviser, those numbers are worth agreeing with them.
  • Nothing here is investment advice. The frameworks structure your reasoning; they do not validate a concentrated AI-and-crypto allocation, which carries real risk regardless of process quality.
Questions for you

Six things that shape the build.

  1. iPhone or Android? The share-sheet capture is an iOS Shortcut; Android has equivalents but the build differs.
  2. Where do your notes live today (Apple Notes, Google Keep, Notion, nothing)? The journal should land where you already are.
  3. Rough baseline: how many trades did you place in the last six months, across stocks and crypto?
  4. Which 20 to 30 accounts on X would you seed the Signal list with today?
  5. Is the leisure feed fully off-limits (our default), or should the weekly brief include a "for fun" section too?
  6. How much time are you willing to give the monthly review: 20 minutes, or less?
The decision on the table

Tier 0 and Tier 1 live this week. Tier 2 decided on day 30.

One half-day session and one setup session, at the fees listed above. Nothing in Tier 2 is priced or committed until the 30-day review says which steps earned it. Your reading stays exactly as it is.

Your one numberTrades per month over the last six months, so the baseline exists before anything goes live.
Your one hourAnswer the six questions above and pick the first 20 to 30 accounts for the Signal list.
Next actionSay yes to week one and we book the half-day session.
Sources

Every external claim, with its verification state.

Fetched August 17, 2026 by the research pass behind this proposal. verified an independent second check confirmed the page supports the claim. partial the page confirms the core claim but not every detail. fetched read by the research pass, not re-checked. secondary a summary source, not the primary document. unverified the source could not be fetched, so the claim rests on the researcher's reading. Prices in USD unless marked; they move fast, reconfirm before spending.

67 sources tap to expand
  1. Free Gemini includes Gems, Deep Research, Canvas; paid tiers mainly raise usage limits (page rendered CAD $6.99 / $26.99)verifiedgemini.google/subscriptions
  2. Gems: saved custom assistants, usable in the mobile appverifiedsupport.google.com/gemini/answer/15235603
  3. Grok decides whether to search X public posts and the web in real time; free tier exists, limits unpublishedpartialhelp.x.com/en/using-x/about-grok
  4. Tow Center (CJR), March 2025: 1,600 queries, eight tools, over 60% wrong overall; Perplexity 37% wrong, Grok 3 94% wrong, 154 of 200 Grok citations dead; paid tiers more confidently wrongverifiedcjr.org/tow_center/we-compared-eight-ai-search-engines…
  5. Perplexity Free: 3 Pro searches a day, 1 Research query a month; Pro $20 a month or $17 annualpartialperplexity.ai/help-center/…/11187416
  6. ChatGPT Free includes Projects; custom GPTs from Go at $8 a monthverifiedopenai.com/chatgpt/pricing
  7. Claude Free excludes Projects and Research; Pro $17 annual or $20 monthlyverifiedanthropic.com/pricing
  8. NotebookLM renamed Gemini Notebook (July 2026), free, source-grounded with quoted citations, mobile appfetchednotebooklm.google
  9. Sycophancy is consistent across five assistants; matching user beliefs predicts human preferencefetchedarxiv.org/abs/2310.13548
  10. OpenAI rolled back a GPT-4o update for being "overly flattering or agreeable"fetchedopenai.com/index/sycophancy-in-gpt-4o
  11. Hallucination-reduction techniques: allow "I don't know", quote before analysing, drop unsupported claimsfetchedplatform.claude.com/docs/…/reduce-hallucinations
  12. iOS Shortcuts: receive share-sheet input, restrict to URLspartialsupport.apple.com/guide/shortcuts/…
  13. Telegram bots message users at no cost; limits far above one user's needsverifiedcore.telegram.org/bots/faq
  14. n8n Cloud from €20 a month billed per workflow run; self-hosted Community Edition freeverifiedn8n.io/pricing
  15. Zapier Free: 100 tasks a month, two-step Zaps, 15-minute polling (Zapier verified; Make figures unverified)partialzapier.com/pricing
  16. Gemini API unpaid tier: human review, used to improve products, "do not submit sensitive, confidential, or personal information"; paid tier does not train on promptsverifiedai.google.dev/gemini-api/terms
  17. Gemini Flash models free of charge on the free tier; paid Flash-Lite $0.10 / $0.40 per million tokensverifiedai.google.dev/gemini-api/docs/pricing
  18. Claude Haiku 4.5: $1 / $5 per million tokens; roughly $1 to $2 a month at 30 checks plus 30 briefs (price verified; the monthly estimate is our arithmetic)verifiedplatform.claude.com/docs/en/about-claude/pricing
  19. X API pay-per-use: $0.005 per post read, no free tierverifieddocs.x.com/x-api/getting-started/pricing
  20. RSS.app: X and Reddit feed generators, Basic $8.32 a month for 15 feedsfetchedrss.app/pricing
  21. NotebookLM free limits (100 notebooks, 50 sources each, 50 chats a day) and share-to-notebook on the mobile appsfetchedsupport.google.com/notebooklm/answer/16213268
  22. Reddit Data API: free tier gated to non-commercial use, commercial access priced and approval-gated (Reddit blocks automated fetch of its own help pages)secondarysocialcrawl.dev/blog/reddit-data-api-2026
  23. CoinGecko Demo API: $0, 10,000 monthly credits, no cardfetchedcoingecko.com/en/api/pricing
  24. SIFT: Stop, Investigate the source, Find better coverage, Trace to the originalpartialhapgood.us/2019/06/19/sift-the-four-moves
  25. Lateral reading: fact-checkers 100%, historians 50%, students 20% (Wineburg and McGrew, 2017); the fetched Stanford page reports the follow-on study and references the earlier resultpartialed.stanford.edu/news/…spot-misinformation-online
  26. TSC Industries v. Northway (1976): material if a reasonable investor would view it as significantly altering the total mix of information (fetch blocked; the holding itself is well established)unverifiedtile.loc.gov/…/usrep426438.pdf
  27. Tetlock's ten commandments: start from the outside view, "how often do things of this sort happen?"verifiedgoodjudgment.com/philip-tetlocks-10-commandments…
  28. Klein, "Performing a Project Premortem", HBR September 2007; the roughly 30% figure cites Mitchell, Russo and Pennington 1989 (article confirmed; the 30% figure sits behind the paywall)partialhbr.org/2007/09/performing-a-project-premortem
  29. SEC Investor Alert on social media and investment fraud; Investor.gov name check; read the filings on EDGARverifiedinvestor.gov/…/social-media-and-investment-fraud-investor-alert
  30. FTC Endorsement Guides, 16 CFR Part 255: clear and conspicuous disclosure of material connections (the Securities Act anti-touting provision is a separate statute)partialecfr.gov/current/title-16/…/part-255
  31. Community Notes bridging algorithm; 30.2% of displayed notes later lose helpful statusverifiedarxiv.org/html/2601.14002
  32. Ground News: outlet-level bias, factuality, and ownership ratingsfetchedground.news/rating-system
  33. Keyrock analysis of 16,000 unlock events: about 90% negative price pressure, declines starting up to 30 days beforesecondarycryptoslate.com/90-of-token-unlocks-drive-prices-down…
  34. Nvidia fiscal Q3 2026 10-Q: four customers at 22%, 15%, 13%, 11% of revenue (61% combined) versus 36% a year earlier (primary is the 10-Q on EDGAR)secondaryfool.com/investing/2025/11/27/…nvidia-customer-concentration
  35. Decision journal template and the hindsight rationale (template confirmed; the Kahneman endorsement is attributed elsewhere)partialfs.blog/decision-journal
  36. Annie Duke on "resulting"partialannieduke.com/article-decision-making-by-thinking-in-bets…
  37. Annie Duke, kill criteria as states and datesverifiedbehavioralscientist.org/annie-duke-quit-mental-models…
  38. Gawande's surgical checklist evidenceverifiedhsph.harvard.edu/news/fall08checklist
  39. Thorp, The Kelly Criterion in Blackjack, Sports Betting, and the Stock Market; over-betting risks ruin, practitioners use fractional Kelly (PDF unreadable to the checker)unverifiededwardothorp.com/…/TheKellyCriterionAndTheStockMarket.pdf
  40. Buffett, 1996 shareholder letter, circle of competenceverifiedberkshirehathaway.com/letters/1996.html
  41. Mauboussin base rates: about 1% of roughly 50,000 firm-years sustained 30 to 35% five-year revenue growth (Morgan Stanley blocks automated fetch)unverifiedmorganstanley.com/im/…/article_theimpactofintangiblesonbaserates.pdf
  42. Howard Marks, "Dare to Be Great" (2006), second-level thinkingfetchedoaktreecapital.com/…/2006-09-07-dare-to-be-great.pdf
  43. Munger, The Psychology of Human Misjudgment, incentive-caused biasfetchedfs.blog/great-talks/psychology-human-misjudgment
  44. Gartner Hype Cycle phasesfetchedgartner.com/en/articles/hype-cycle-for-emerging-technologies
  45. Buffett and Munger reading habits; the "500 pages a day" and "80% of the day" figures are attributed lore, not primary documentsfetchedfs.blog/the-buffett-formula
  46. X Lists: List timeline shows only List accounts; pin favourite Lists to the top; iOS and Androidpartialhelp.x.com/en/using-x/x-lists
  47. Reddit .rss endpoints work without a keyverifiedwprssaggregator.com/reddit-rss-feed
  48. Inoreader Pro $90 a year; free tier 150 feeds; newsletter feeds; AI summariesverifiedinoreader.com/pricing
  49. Folo, open-source AI RSS reader, bring your own key (38.8k stars at check time)partialgithub.com/RSSNext/Folo
  50. Google Alerts delivered as RSSsecondaryhelp.rss.app/…/how-to-turn-google-alerts-into-an-rss-feed
  51. SEC RSS feeds and EDGAR filing alertspartialsec.gov/about/rss-feeds
  52. Vanderbilt study of 2,500 prediction markets: Polymarket 67%, Kalshi 78%, PredictIt 93% accuracy (secondary report of the study)verifieddlnews.com/…/polymarket-kalshi-prediction-markets-not-so-reliable…
  53. DefiLlama unlocks calendar, free, no loginfetcheddefillama.com/unlocks
  54. Messari Lite free; paid crypto analytics tiers $49 to $699 a month at other vendors (Messari fetched; others secondary)fetchedmessari.io/pricing
  55. TradingView free tier about 3 price alerts, no webhooksfetchedtradingview.com/support/solutions/43000690941…
  56. Sending a share-sheet link to a Telegram bot via Shortcutsforumtalk.automators.fm/t/send-message-to-telegram-bot/11299
  57. Barber and Odean (2000), Trading Is Hazardous to Your Wealthverifiedeconpapers.repec.org/RePEc:bla:jfinan:v:55:y:2000:i:2:p:773-806
  58. Barber and Odean (2008), All That Glittersverifiedfaculty.haas.berkeley.edu/odean/papers/Attention/All that Glitters.pdf
  59. Barber, Huang, Odean and Schwarz (2022), Attention-Induced Trading and Returnsverifiedonlinelibrary.wiley.com/doi/abs/10.1111/jofi.13183
  60. Chague, De-Losso and Giovannetti, day trading persistenceverifiedpapers.ssrn.com/sol3/papers.cfm?abstract_id=3423101
  61. Merkley, Pacelli, Piorkowski and Williams (2024), crypto influencer returns (university summary of the Review of Accounting Studies paper)verifiedblog.kelley.iu.edu/2024/11/21/…crypto-influencers…
  62. Kakhbod, Kazempour, Livdan and Schuerhoff (2023), finfluencers: 28% skilled, 56% anti-skilledverifiedeconpapers.repec.org/paper/chfrpseri/rp2330.htm
  63. Kalda, Loos, Previtero and Hackethal, smartphones and tradingverifiednber.org/papers/w28363
  64. Slovic 1973 (unpublished monograph) via secondary summary; pattern replicated in Oskamp 1965 and Hall 2007partialcorporate.jcx.au/additional-information
  65. Haynes et al., NEJM 2009, surgical checklistfetchednejm.org/doi/full/10.1056/NEJMsa0810119
  66. Gollwitzer and Sheeran (2006), implementation intentions meta-analysisfetchedsciencedirect.com/…/S0065260106380021
  67. Mellers et al. (2014), forecasting training and trackingfetchedjournals.sagepub.com/doi/10.1177/0956797614524255
  68. Fulkerson, Jordan, Riley and Yan (2026), timing costs investors about 0.10% a year, versus Morningstar's 1.2 point gapfetchedrpc.cfainstitute.org/…/bad-timing-does-not-cost-investors-funds-returns