You asked for a few fun, novel things — some on paper, some out in the world — that would let you practice reading signal rather than just reading about it. This is that list. It pairs with the signals brainstorm; the brainstorm is about where to look, this is about training the eye.
The short answer. A signal is a small thing you notice now that tells you something big later. You cannot get better at spotting them by consuming more — more input is the noise. You get better by writing down what you think will happen, putting a date on it, and going back to check. That loop is the whole skill. Everything below is a way to make the loop fun instead of a chore.
Why this is your skill in particular
This isn't a general self-improvement habit. For you it sits directly on top of things already written down, and three of them say the same thing from different angles.
- Your purpose line is "I empower innovation that improves lives." You can only empower an innovation if you see it early enough to do something about it. Signal-reading is the front door to the purpose, not a side hobby.
- §09 Judgment Over Execution. Machines now do the gathering. What's scarce is knowing what's worth acting on — which is exactly this skill. Every hour spent making your calls more accurate is an hour spent widening the one moat you've named.
- §03 Time Is the Asset. You have a date. A signal that moves the floor, the income stack or the timing is worth ten that are merely interesting. The log makes that difference visible instead of felt.
- §01 already contains the method. "Don't force conclusions. Track themes lightly; let patterns emerge" — and the canon notes the Keys thesis emerged exactly that way. A dated card is that principle with a timestamp on it: a light hold, not a decision.
- §17 Agent-Driven, Human-Shaped. Agents can collect, cluster and summarise. They cannot make the call or take the score. The monthly scoring pass is the human half, and it's the one thing on this page you should never hand over.
The honest version. You already read signal well — the Keys thesis, Barbell the Stack and the Engine Room read all came from this instinct. What's missing isn't perception, it's the receipt. Without dated cards you can't tell a good instinct from a good memory, because memory quietly edits itself to be right.
What makes something a signal
Four tests. A thing that passes all four is worth a card in the log. Three or fewer and it is interesting, not useful.
| Test | Ask | Why it works |
|---|---|---|
| Early | Who else has noticed? | If it is in the headlines, the price already moved. Signal has to arrive before the crowd or it buys you nothing. |
| Specific | What exactly would I see next? | "AI is big" predicts nothing. "Three job ads for an evals lead at one bank" predicts something you can check. |
| Costly | Did someone spend real money, time or reputation? | Talk is free, so talk is cheap evidence. A hire, a lease, a shipped product, a lawsuit — those cost something, so they are harder to fake. |
| Falsifiable | What would prove me wrong? | If no future fact could make you wrong, you are not reading a signal, you are holding a mood. |
The 60-second version. Something catches your eye. Ask: early, specific, costly, falsifiable? If yes four times, write the card. If not, enjoy it and let it go. Most things should be let go — that is the point.
Your principles, read as signal rules
Thirteen of the nineteen invariants already say something about reading signal. None of this is new instruction — it's your own canon pointed at one job.
| Principle | What it says about reading signal |
|---|---|
| §01 Systems & Emergence | Hold themes lightly. A card is a dated hold, not a conclusion. Lock the narrative early and you stop seeing. |
| §02 Human Job-to-Be-Done | Rewrite every card as a job: whose unchanging need just got easier? A signal about a tool fades. A signal about a job compounds. |
| §05 GTD / Trust the Systems | If a signal lives in your head it isn't a signal, it's a worry. Out of the head, into the log, same day. |
| §07 Iconoclast | The best signals hide inside defaults nobody questions. Whenever you hear yourself say "that's just how it's done," write a card. |
| §09 Judgment Over Execution | Delegate the collecting. Never delegate the call or the score — those are the judgment. |
| §10 Research Like a Scientist | Calibrated language is the scientist half: a number, not an adjective. Know which mode you're in when you write it. |
| §11 The Kaleidoscope | One fragment appearing in two domains beats two separate signals. The interesting work is at intersections — so tag the domain on every card. |
| §12 Embrace the Tensions | A tension that refuses to resolve is usually the signal itself. Name it and leave it standing; don't settle it to feel tidy. |
| §14 Don't Make Me Think | One card a day is a ceiling on purpose. A practice that grows becomes the noise it was meant to cut. |
| §15 The Adirondack Test | If the log doesn't work from a phone, on a trail, before coffee — it's the wrong log. Design it for the chair on the beach. |
| §16 Would I Do This For Free? | Pick only games you'd look forward to. The dutiful version of this dies in three weeks and takes the habit with it. |
| §18 No Job, No Build | A card scoring "right" is not permission to build. State the job in a sentence and say yes to it first. The gate still holds. |
| §19 Dream & Ask What-If | The what-if register surfaces signals your analytical filter throws out. Dream first, score later — never in the same sitting. |
Six lenses from Singularity that make a signal count double
These are your existing theses used as filters. A thing that lands cleanly on one of them is worth more than a thing that's merely true — because you already know what to do with it.
| Lens | The question to ask | Why it counts |
|---|---|---|
| Barbell the Stack | Which end — the infrastructure everyone must buy, or the surface where a person feels the value? | Infra signals route to Investing, surface signals route to Ideas and Forge. A signal that sits in the middle of the stack is usually noise. |
| The Keys thesis | Is one person doing something that used to need an institution? | Individuals building without permission is the earliest visible sign of a layer collapsing. You saw this one first; keep looking where you looked. |
| Intelligence to use intelligence | Who just got a tool they can't yet use? | Capability is outrunning the ability to apply it. Every widening gap is both a product and a fairness problem — which is the Caring half of the same observation. |
| Livelihood sovereignty | Does this free someone from a dependency, or deepen one? | Systems thinking shows the lock-ins and the escape routes at the same time. Escape routes are the ones worth a card. |
| Billionaire agent model | What do wealthy people delegate that everyone else struggles with alone? | Each "call my guy" moment that becomes available to everyone is a market opening. You're already the proof of concept. |
| Authenticity premium | As this gets automated, what becomes scarce because it's demonstrably human? | Every automation signal has a mirror signal about what gets more valuable. Most people only read one side. |
Use the lenses second, never first. Notice the thing, write it plainly, then ask which lens it lands on. Reversed, you'll find your own theses everywhere — which is §01's warning about premature narrative lock, and the most likely way this whole practice goes wrong for you specifically.
The journaling practice: the Signal Log
One card per signal. Six lines, two minutes, written the same day you notice it. On paper, in Notes, in a file — the medium does not matter, the date does.
| Line | Example |
|---|---|
| Date | Oct 3, 2026 |
| What I saw | Two trail runners at Green Mountain both wearing carbon-plated shoes on dirt |
| Where exactly | Trailhead lot, Saturday 7am — not a race, just a regular morning |
| So what — if true, then… | Road-shoe tech is crossing to trail faster than the brands admit; expect three more carbon trail models by spring |
| Check on | April 1, 2027 |
| Confidence | 60% |
Four rules keep it honest:
- One card a day, maximum. The limit forces you to choose, and choosing is the skill.
- Write it the day you see it. A memory three weeks old has quietly rewritten itself to be right.
- Never edit a closed card. If you change your mind, write a new card. The old one is the evidence.
- Confidence is a number, not a word. "Probably" cannot be scored. 60% can.
Two optional lines, only when the card feels Singularity-grade. They cost five seconds and make the monthly pass sortable:
| Lens | Barbell · infra end — or Keys, sovereignty, the gap, leave blank |
| Domain | Investing · Ideas · Forge · Caring · Active · Research — where it would route if it turns out right |
Both stay blank on most cards, and that's fine. A card with no lens and no domain is still a card; a card forced onto a lens is a thesis wearing a disguise.
Keeping score — the part everyone skips
Once a month, ten minutes: open the cards whose check date has passed, mark each right or wrong, and tally. The only output that matters is this question — when I say 60%, am I right about 60% of the time?
That is called calibration (your confidence matching reality). The formal measure is the Brier score, but you do not need the arithmetic. Four buckets on a page does the same job:
| When I said… | Cards | Right | Should be about |
|---|---|---|---|
| 50–60% | 8 | 5 | 4–5 |
| 60–75% | 6 | 3 | 4 |
| 75–90% | 4 | 4 | 3–4 |
| 90%+ | 2 | 1 | 2 |
Two patterns show up fast, and both are fixable. If your high-confidence calls miss, you are overconfident — drop every number by 10 and see what happens. If everything sits at 50–60%, you are hedging — you are not making calls, you are describing weather.
The whole system in one line. One card a day, ten minutes a month. If you only ever do that, you will read signal better in a year than someone who reads five newsletters a day for ten.
Where a card goes once it scores right
Most cards go nowhere. That is the correct outcome and not a failure of the system — the value is in the scoring, not the harvesting. For the few that do land:
| If it landed on… | It goes to | With what gate |
|---|---|---|
| The infra end of the barbell | The Investing watchlist, as a watch item with the card's date attached | Still clears at Checkpoint before any trade, except names already pre-cleared. |
| The presentation end | Ideas — as a theme note, not a project | §18: no job stated and validated, no build. A right card is evidence, not permission. |
| A pattern across two domains | Theories — it's becoming a thesis | §01: give it three more cards before you name it. The Keys thesis took months of light holding. |
| Something that moves the floor or the date | Retirement — straight in, same week | None. These are the highest-value cards you will ever write; treat a right one as urgent. |
| A widening gap between who can use the tools | Caring, or Society | §19 applies: stay in the question before you decide what to do about it. |
Twelve games
Each one trains a different muscle. Do not do all twelve. Pick two you would actually look forward to, run them for a month, swap one out.
| Game | What you do | What it trains | Time |
|---|---|---|---|
| Anomaly run | On a trail or neighbourhood run, name out loud three things that don't belong. No judging, just noticing. | Seeing before explaining — the step most people skip | Free, inside a run |
| Shelf drift | Same grocery or hardware aisle, same photo, first Saturday of the month. Compare to last month. | Change over time in a place nobody watches. Shelves move months before trend pieces. | 5 min/month |
| Parking-lot census | Stand in one lot — trailhead, Home Depot, school pickup — and count one thing for five minutes. EVs, contractor vans, out-of-state plates. | Counting instead of vibing. Your eye lies; a tally doesn't. | 5 min |
| Job-post archaeology | Read ten job ads in one field. Note titles that didn't exist two years ago. | Reading intent. Companies hire for a thing roughly a year before they ship it. | 20 min |
| Two-years-ago audit | Open a newsletter or article from exactly 24 months back. Score it. Who was right, and was it for the reason they gave? | Humility, and spotting which kinds of claims age well | 15 min |
| Byline blind test | Paste five takes on one topic with names removed. Rank them. Then look at who wrote them. | Judging the claim, not the person | 15 min |
| Reverse the headline | Take a headline you believe. Write the opposite and make it sound reasonable in three sentences. | If you can't argue the other side, you have an opinion, not an argument | 5 min |
| Signal supper | Dinner with three or four people. Everyone brings one thing they believe that most people don't — plus what would change their mind. | The "what would change my mind" half is the whole exercise | An evening |
| The stranger question | Ask one person a week: "what's different about your work this year?" Write the answer down word for word. | Ground truth from outside your feed. Ten answers beat any trend report. | 2 min |
| Dead-drop prediction | Email yourself a prediction, scheduled to arrive in six months. Then forget it. | An ambush from your past self — the most honest feedback there is | 2 min |
| Small-stakes market | Put play money, or $20, on a public forecasting market. Watch the price move against you. | Losing is a faster teacher than being told. See the links below. | 10 min/week |
| The 15-year-old test | Explain the signal to someone outside the field in sixty seconds, no jargon. | If you can't, you either don't understand it or there's nothing there | 1 min |
Six more, built from your actual week
The twelve above work for anyone. These six only work for you — they ride inside things you already do, which is the only reason habits survive (§16: you'd do these for free, so you'll actually do them).
| Game | What you do | What it trains | Time |
|---|---|---|---|
| Aid-station read | Next ultra, race, or drive home from one: ask three people "what's different about your work this year?" Write all three answers word for word before you get home. | Ground truth from outside the feed, in the one room where people have stopped performing. The canon already names this as where your people are. | Inside something you're doing anyway |
| The slowest newsfeed | Once a month, open one real dataset nobody posts about and make a single call about next year from it — OBIS species records or NOAA tides & currents. | Reading a primary source instead of someone's read of it. Slow data is the purest practice because there's no crowd to copy. | 20 min/month |
| The ask drift | Keep a running list of what people ask you for. Compare this quarter to last. | What people ask you for changes about a year before the market admits anything changed. It is the cheapest leading indicator you own. | 1 min/week |
| April 2027 dead drop | Write six predictions about your own stack — income layers, what you're building, who you're working with, where you are — sealed until the date itself. | The dead drop aimed at the one transition that matters. Also the only honest test of whether the plan was a plan or a hope. | 20 min, once |
| Coffee at the top of a mountain | The one-person version of signal supper. One person who's earned something real, one hard question, and "what would change your mind?" | Depth instead of breadth. A sparring partner sharpens a call; a room of nodding validates it. | A morning |
| Neighborhood census | Same walk through your own blocks, first Saturday. Count one thing: contractor trucks, for-sale signs, EV chargers, dark storefronts. | Denver tells you about Denver months before anyone writes it up. Local is the one beat where you can genuinely be early. | 20 min/month |
Where to practice against a clock
Journaling scores you against yourself. These score you against everyone else, which stings more and works faster. All four are real places you can open today.
- Good Judgment Open — free public forecasting tournaments, properly scored, with a leaderboard. The closest thing to a gym for this skill.
- Metaculus — community forecasts on long-horizon questions (AI, science, policy). Free, and it keeps a permanent track record for you.
- Manifold Markets — play-money prediction markets. Fastest feedback of the four; you can also write your own question and watch strangers price it.
- Kalshi — real-money event contracts on a US-regulated exchange. Real money, so treat it as a trade: clear it at checkpoint.pwc.com before you put a dollar in.
Two short reads that make the scoring click: Good Judgment's Ten Commandments for aspiring superforecasters, and Philip Tetlock's Superforecasting — the source of almost everything on this page.
Six things that feel like signal and aren't
| Volume | Lots of people saying it. By then it's priced in — you're reading the crowd, not the edge. |
| Confidence | A certain voice. Certainty is a personality trait, not evidence. |
| A round number | "$1 trillion market by 2030." Round numbers are marketing, not forecasts. |
| Agreement | It confirms what you already wanted to be true. Those are the ones to write a card on — so you can be scored. |
| Unfalsifiable claims | "Everything changes from here." Nothing could prove it wrong, so nothing is being said. |
| Algorithmic novelty | It felt rare because a feed chose it for you. Ask where it came from before you call it early. |
The hardest cards: the ones about your own theses
Cards about the world are easy — you have no stake, so you can be honest. The cards that actually move you are the ones written about things you already believe, because those are the beliefs doing the steering. Four to write this week, each with a real check date and a number:
| Belief | If true, then… | What would make you wrong |
|---|---|---|
| Barbell the Stack | Infrastructure names keep capturing the margin while the surface layer keeps capturing the users and the play. | An app-layer company holding fat margin while compute prices keep falling — the middle refusing to hollow out. |
| Judgment is the moat (§09) | The things people ask you for shift from doing toward deciding, measurably, within a year. | Count your last ten asks today and ten more in six months. If the mix hasn't moved, the moat was a comfort. |
| The April 2027 timing (§03, §08) | By then at least two layers of the income stack are actually paying, and together they clear the $4,124/mo floor — with no new layer having to appear from somewhere. | Any quarter where a layer underdelivers and the gap gets explained rather than closed. |
| Agent-driven, human-shaped (§17) | A real and growing share of a normal week runs without you in the loop at all. | Log one honest week. If the agents produced work you had to redo, the share is zero, not high. |
No editing these. If one of the four comes back wrong, that is the single most valuable card you will write this year — it found a belief that was steering you while being untrue. §12 applies: don't resolve the tension in a hurry. Sit with it, then rewrite the thesis rather than the card.
Start this week
- Today: write one card. Anything. Put a check date on it.
- This week: pick two games. Suggested first pair — anomaly run (free, fits inside something you already do) and dead-drop prediction (two minutes, pays off in six months).
- This month: open a Good Judgment Open account and forecast five questions. Don't research. Just guess with a number, so you have a baseline to beat.
- First Sunday of November: ten-minute scoring pass. Four buckets. That's it.
- Standing invite: put a signal supper on the calendar for sometime in the next six weeks.
- Before April: write the four thesis cards above, and the April 2027 dead drop. Those five are worth more than the next hundred cards about the world.
Deliberately not here
Where to look for signals — which sources, feeds, datasets and people — is the signals brainstorm, not this page. Tooling to automate collection is a Building question. Investing-specific reading (what's already priced in, how to size a position) lives in the Investing domain and carries its own rules.
Rich Price · primer · October 3, 2026 · Finding signal in the noise, the useful 20%