If you missed it, here’s the recap of VC night.
Double last week’s turnout, three hours flat. Roughly: fifteen minutes of “what is this,” fifteen of Q&A, an hour on the new role system, an hour of Conqueror screen-sharing Bifrost, and the rest was BBIG’s CEO. Last week’s recap has the deeper platform explainer; this one is mostly roles and how to actually read a signal.
The short version of why this exists
Conqueror ran the 2021 → 2026 timeline again for the returners: the forum era (“if we could just filter news to what mattered, we’d be really good traders”), the 2022 crowdsourced-quant idea after RSX, and Catalysts — the Discord-thread version that ran ~10 months and was one of the most consistently green stretches trading floor ever had. It died because nobody can research plays every single day forever, him included. That’s why the challenge never got past warm-up. So after monetization the plan became: build the real version.
- Nexus = the 2021 idea, done. Every SEC filing with every exhibit read, FDA, economic data, and now the commercial wires (Dow Jones, Benzinga) we tried to get for years. Scored high/medium/low impact so you scroll only things that matter.
- Heimdall = the flagship. A model built on five-plus years of what this community learned — his notes, calculations, theories, chart drawings. It hunts what we would hunt, and things we never would have (UFG again: a Cayman exemption on Tuesday, a surprise guidance raise on Thursday, stock +50% — nine times out of ten we take that trade; Heimdall said this is fake, they can now sell shares without shareholder approval).
- ~16,000 arcs live at once. Ten reasoning stages, four are LLM. Currently chained — throttled wakes, cheaper models than it was designed for — and still doing this.
Performance, his framing: signals with positions taken on them have been above 80% win rate for a week and a half on trading floor. Heimdall on its own hovers 65–72%, and that’s not the point — it isn’t designed for a high win rate, it’s designed for asymmetry: last week winners averaged +11% in the called direction, losers 2.5% against. Cuts are small, wins are bigger, and that is the whole trick. (He noted the irony of presenting this on the first red day in two weeks.)
The other thing he keeps coming back to: nobody’s feelings are in these plays. “When I open an arc and say this play is fucking dumb, I’m not killing someone’s kid in front of them. I’m saying the bot’s dumb.” People cut Heimdall positions easily because they didn’t build the thesis. Tiddly has been killing it for exactly that reason.
Q&A, part one (joebeewon, back after ~5 years)
Are the old DD threads still around? Archived, all hidden. Heimdall isn’t trained directly on them but they play a part, so they’re proprietary for now. He’ll unhide specific ones case-by-case — joebeewon wants his TSM thread (wrote it at $60–70, “pussied out,” it’s $400 now). Remind him. He also loves the history and is looking at ways to expose more of it — the front-page animation on ascendedtrading.com, maybe a long-form video.
So does anyone write DD anymore, or is it all Bifrost now? Yes and no. Most people are going to work the signals, and that’s what it’s for. But arcs exist without signals — Heimdall creates an arc from meaningful news and then very often decides this isn’t worth speaking up about, which is the hard part of building one of these (“you can make something that speaks up about everything, but that’s a firehose”). Those arcs are still there as a record. If you trade a specific name — biotech people waiting on PDUFA dates, for instance — open the arc and put your commentary on it. One place per storyline instead of twenty AMC threads, and when the arc exhausts, the page turns: old conversation stays readable, next arc starts fresh.
What about J-Pow days — does it hear the speech? Not live. Market state and the economic calendar are in there and Heimdall knows them when it writes a signal, but that thinking isn’t surfaced to you yet. What it does do is circle back: Boeing was called bullish on a contract resolution, the macro print hit, BA drilled, Heimdall woke up and invalidated its own signal — and the dies-if on the invalidation said if it reclaims this level the play’s back on. Working as intended. What he wants to add is a cheaper, active warning (“this person is speaking today”) without waking the model, plus a suite of trading floor notifications that’s about to ship: who’s speaking, what mattered from what they said.
UI status: back end was a year and a half hand-built because he’s “a little conspiracy-driven” about exposing that code to LLMs; the front end, he stopped caring. It’s his style, to his spec, and it needs work. TradingView license is acquired — their charting library is coming into Bifrost for drawing, with the goal of live chart data so you never leave.
Why performance roles died
His research into the server’s history: a performance role was the exact moment you should stop listening to someone. They grinded to get it, then became a gatekeeper — commentary on other people’s plays from then on. Exceptions: House, Ridn. And they were impossible to take away; the second you make them “performance-based” they stop being that, because you give them to people you like and never remove them. He got overruled on promotions he thought were terrible traders all the time.
So he built one that runs itself.
The new role system — the rules
- Not about how much money you make. About whether and how often you’re right, and more so what you do when you’re not.
- Every role is a rolling 30-day window. Every night at 12:30 a recalculation decides whether you keep it.
- The bar is Heimdall. Your read precision is measured against it. Agreeing with the bot gets you nothing; you have to beat it.
- No leaderboard. You’re not competing with anyone.
How reads work: the comment box gets three buttons — Confirm / Contradict / Unclear. Change your mind as much as you want; the only stance that scores is the one you’re holding the day before the arc ends.
The Ladder (the old “trading roles,” names kept for nostalgia, meaning changed)
| Role | How you get it |
|---|---|
| New Member | You’re a farmer. Go cultivate some gains. |
| Vanguard | Be active in Bifrost — comment on or watch at least 3 arcs in 4 weeks. |
| Heroes | Vanguard + 10 or more resolved reads that beat Heimdall. No trading required. |
| Champions | Hero + put money on it: 5+ closed entries on your reads, 65%+ win rate, and misses average ≤3% on the underlying. That last one is the point — it exists to make you cut. |
| Legends | 20 resolved reads, 10 vetted entries, 15 replies, and at least one Harbinger call. Reading, entering, and helping people. |
Lose Hero or Champion and it’s gone. Lose Legend and you become Fallen Legend — legends never die; re-hit the bar and you ascend again. From that point on you are always one or the other.
He was explicit that trading is not the point: “Valhalla’s best calls were made by people that likely didn’t trade it, by people that didn’t have a performance role.” Hence:
Contribution roles (no trades required)
- Harbinger — ahead of Heimdall or against it, and right. 2+ Harbinger calls in 30 days with at least half landing. A Harbinger call is any of: first stance on an arc that resolved your way; a watch on a ticker Heimdall had nothing on that got a signal within 5 days (so no, you can’t watch the whole universe); or contradicting an arc that then got invalidated. Sorting out which signals are bullshit is the fastest path. Required for Legend.
- Scout — 15+ cold reads in 30 days: a stance plus a real comment on an arc with no live signal. Purely for people doing the work before Heimdall speaks.
- Scribe — 8+ write-ups in 30 days: a comment long enough to be an argument, or citing a source / filing / number. One arc per day. “Nobody has to like it.”
- Mentor — 12+ counted replies in 30 days (someone other than you liked or replied to it), max 3 arcs/day. Mentors get pinged when arc comments sit unanswered.
Two old ideas that finally shipped
- Play of the Day — one person, one day, automatic. Has to be a Harbinger call (first on the arc, or the contradict); biggest move in your called direction wins it until someone takes it tomorrow.
- Hall of Fame — 25 Harbinger calls. Permanent.
Staying as-is
Valkyries (16 of them, all comped — “being a woman on the internet sucks, being a woman in a trading discord sucks more”) and Demigods (mods, to be sorted later).
The enforcement arm
His words: “The entire purpose is steering people away from bad trading behavior. Getting people to sell, take profit, contribute. They’re shame roles, but really they’re reminding you what you can do better at.” The ladder already punishes these by costing you roles; these make it visible. Hold two at once → Double Dunce Cap. Three → Triple. Everything is measured off the positions/arc ledger, and he expects to tune thresholds once they’re live.
| Role | Trigger |
|---|---|
| Bag Holder | Still in a position after Heimdall says the arc is over. |
| Married to the Play | Position red for 30+ days. |
| Divorced | You finally sold it. 7 days. (Yes, stacks with the above → Double Dunce.) |
| Hopium Dealer | Most confirms on arcs that went on to invalidate, over 30 days. Only one person can hold it. |
| Inverse | Read precision under 35% with 10+ calls. Statistically, people should do the opposite of whatever you say. |
| Heimdall’s Yes Man | Confirms only, zero contradicts, precision exactly Heimdall’s. You’re sucking the bot off and contributing nothing. |
| Pied Piper | Rushed to be first and led 3+ members astray. |
| Exit Liquidity | Last one into an arc before it reverses. (Tiddly: “I pray to God no one gets that title.”) |
| Domed by the Top | Bought within 1% of the high. |
| Power Bottom | Bought within 1% of the low and it ran 10%+. |
| Paper Hands | 5+ exits within 1% of entry in 30 days. |
| Revenge Trading | Back into the same ticker within 24h of getting it wrong. |
| Definitely Has a Job | Active in Bifrost 9:30–4 ET every trading day of the month. Miss a day. Take a vacation. |
| Gang Bang | 4+ members enter the same arc within a 15-minute window. Held 48h. “Probably the most frequently awarded.” |
| Nothing Burger | You somehow find arcs that go on to do absolutely nothing in 30 days. |
| Missed Generational Wealth | Closed, then the ticker ran another 15%+ your way before the arc expired. (Tiddly: “we’re gonna have a lot of those.”) |
| Catching a Knife | Entered a ticker whose arc invalidated 24h+ ago with no new signal. |
| Chart Grabber | 50+ charts pulled in 30 days, zero stances posted. He has already named the first recipient. |
| Dunce Cap | The town dumbass, so everyone knows. Put in the work and it comes off and you become Reformed — but you have to make Hero to get it off, i.e. be right more often than Heimdall 10+ times. |
Vote roles (the community gives these)
Trust Me Bro (theories, never a source) · Options Flow Astrologer (unusual whales, no thesis) · Crayon Eater (dumb, but not Dunce Cap dumb) · Touch Grass · Guest Speaker (shows up occasionally to talk about some big play and dips) · Emo (won’t shut up about the one that didn’t work) · Dick · Asshole (when Dick isn’t enough) · Village Idiot (“that will definitely be Duke’s”) · Zero Bitches (“the way that one came out fucking slays”) · Simp · Wife’s Boyfriend (when trading floor decides to listen to one person and won’t stop) · Tinfoil (still thinks GME was a conspiracy) · Perma Bear · NPC (authentically can’t tell if you’re a bot) · Karen · Main Character · Stage Hook (slow mode, for gun-shy moments)
Sparrow: “There’s no way you didn’t have specific people in mind.” Conq: “I have someone in mind for every single one.” Tiddly filed a motion for an honorary Holi role. Under consideration.
These go live next week.
The Bifrost walkthrough — how to read a signal
Bifrost is “some tryhard bringing you the best DD of their life whenever there’s an opportunity, all day every day.” Anatomy of an arc, top to bottom:
Signal. Heimdall saying there’s an opportunity here. Theoretically the smartest part of the loop — it sees everything and when you see a signal cite a cohort or regime, it’s pulling a trend from elsewhere in the universe. The OPEN put he and srslywhyumadbro took came from exactly that: “Existing home sales print on September 10th will confirm a sub-4M transaction volume stall, forcing downside repricing in volume-sensitive real estate platform OPEN where IV is discounted relative to 20-day realized.” Green from entry to the print, underlying sank ~10% more after. “One of the easiest plays I’ve had in years” — and the research behind it was every home-sales filing, earnings and story in the system, already distilled.
Signals have a category. Catalyst (a meaningful event coming); Positioning (Heimdall thinks the market is wrong-footed — high short interest names land here); IV Misprice (theta — options cheap or rich; he doesn’t trade these and they’re unvetted); Second Order (the fallout of a catalyst, e.g. what happens after they announce dilution); Transition (phase change, typically once earnings drop); Invalidated (Heimdall resurfacing that a live call is dead — the Boeing one).
Why now. The plain-English version. He’s considering promoting it above the dense text. Everything isn’t translated yet purely for cost.
Cites. Where the primary basis came from — but it sees all evidence on the arc when it decides.
Dies if. The triggers for Heimdall waking back up: price levels, IV, or specific language in a filing that doesn’t exist yet (Nexus checks every new filing against active watches and wakes the arc if one is covered). Secondary use: they are your support/resistance. Dell’s dies-if was 425, it bounced at 430. WOR sold off ~$20 today and stopped right at the level Heimdall named. “There is very little technical analysis inside Heimdall, but it is authentically good at price levels.” Use them to invalidate your own positions.
Expires. Heimdall decides how long a signal is good for. Most are here-and-now; making them more forward-looking is active work.
Projection. The lower part of the model working out what happens next, and what happens if it doesn’t. Each step has % event (chance it happens) and % move (chance it actually moves the stock — plenty of events land and do nothing), plus magnitude: drift (slow), step (a gap, date unknown), binary (a gap on a known date), terminal (going to zero — added because RSX broke every other description when he back-tested on our history). Timeframes: take them with a grain of salt — getting LLMs to reason about time is still being calibrated.
Crosses to. Where the chain connects to another arc. If Dell’s if/then fires, it links to Broadcom, and those arcs start sharing news.
Evidence / Related. Stories magnetize to arcs; Nexus pulls out what mattered (“expects Q4 revenue of $34.8B, missing $35B”). Related means the story landed on other arcs too.
Connected. First pass at surfacing clusters in the UI: which macro arcs and clusters hold this ticker (Broadcom → AI compute demand, DRAM supply tightening, AI capex cycle) and which companies it shares evidence with.
Wake log. When Heimdall next looks: scheduled wakes (the furthest booked is three years out), random wakes, and momentum wakes — if the stock runs and no evidence explains it, Heimdall does a demand retrieval: goes out to the internet, finds the reason, attaches it.
Meta Calendar. Earnings, PDUFAs, lockups, economic data, presidential schedule, signal expirations — one calendar. Confession: pre-event wakes (Heimdall waking before earnings, which is what produced the Dell call) had been unintentionally disabled. Fixed today.
Clusters and macro arcs, explained better than he’s managed before
Clusters are arcs magnetizing to each other and being reasoned over together — thematic (MAG7 earnings cycle, tariff refunds, tariff refunds in a specific sector). That’s where patterns emerge: every projection has conditionals, and when the wrong ifs start populating across every projection in a cluster, that is the indicator. “It’s a group project, and there are thousands of group projects.” Current read: early signs of AI weakening — profit-taking, bad earnings in specific AI sub-sectors — showing up through the conditionals, not through any one headline. Also cited: a Costco signal that came out of a cluster citing Walmart’s earnings; the oil pullback foreseen before the pullback (Hormuz tensions easing).
Macro arcs are clusters of clusters; the most expensive thing in the system because a wake processes an entire chunk of the universe. They run once a week for now (he wants daily) and produce a standing thesis. AI compute demand pulling forward is currently “tightening” — power grid constraints as the primary bottleneck, headwinds from AI-slowdown narratives — with a dated projection: Oct 1–19 hyperscaler software multiples compress (MSFT/AMZN/GOOG/META), then Oct 14–Nov 14 capex reallocates toward nuclear/baseload PPAs (CEG, SO, DLR, EQIX named). His own layer on top: Heimdall flagged AMAT gapping up post-market and selling off intraday three days running, then a projection about semi-cap names taking profit while design leaders hold — and he connected it to the Anthropic IPO in October pulling capital out of AI to put into AI. “I didn’t do any work on this. That’s the beauty of it.”
Signal vs projection: the signal is here-and-now (Accenture is running, fine), the projection is forward (their earnings will still suck). Both can be true.
Projections are currently chained to ~2 orders out. Chains are graded — there’s a learning loop for both projections and signals. He hasn’t run it yet. Everything so far is off-rip.
Workspace
The thinkorswim grid, in our software. Swap any panel, click a ticker and every panel focuses to it, momentum scanner hits open straight into the arc, comments in a widget so you never leave. Layout lives in the URL — bookmark it and it’s yours. “The view I use most.”
Calls and patterns he walked through
- Dell — one of the Hall of Fame calls. The Sept 1 projection read: “despite headline AI server backlog expansion, gross-margin compression driven by DRAM cost inflation will trigger a post-earnings sell-the-news flush, driven by the institutional decision-delay window.” That last clause is the earnings theory he and Navi worked out years ago — institutions don’t always buy right away — now living inside the model. The regime at the time was theorizing DRAM expensive → storage expensive → margin compression at server makers, so 16 arcs were watching Dell’s print for margin data. It landed with FY27 revenue guidance +70%; Heimdall’s read: “fuck the margin.” The other 16 still mined the release for what they wanted. Signal stayed green the whole run — the “generational wealth” call.
- ASST — earnings, stock dipped a little, then a random wake six days later: same price, hasn’t moved, very high SI, they’re going to cover. Called ~$12.50, ran to 27. What it taught him: the delay is the pattern — stocks run days after earnings and we always said “we should have known.” Heimdall wakes up and remembers.
- Dilution, three times last week, before it hit. The best one: company press-releases $500M in “funding,” follows with the mandatory 6-K disclosing it’s an illiquid shitcoin they awarded themselves; Heimdall woke at 4 a.m. — they are about to sell shares on you — and you had until the open to short it before it halved on the sale.
- Trial failure, then the real call. Stock fails a trial, sells ~15–20%. Heimdall: not enough — this drug was their entire reason to exist. Dropped at 1 p.m.; five minutes after close it was down 60%. Options went thousands of percent. He’s tracking these emerging strengths for the weekly roundup, and deliberately not hunting them — “if we start hunting specific things, they’ll stop working.”
- The boring put. One member’s put on a float-expansion call dated Oct 6–7 sat green and did nothing for weeks (a bug — the signal didn’t make the date explicit enough), so he cut it out of boredom. It’s now down ~$6. “Good problem to have.” Same member’s account went from ~730 to ~960 in under two weeks on small cuts and steady wins.
- The rate hike — called about a month before it hit mainstream coverage.
FAQ — what got asked
Are these signals bullish or bearish? How do I read one? (Mysterious, on Grocery Outlet) You read it right: bullish unless price breaks $11 or they file another guidance cut. Category + dies-if tells you direction and where it’s wrong.
Does it use prediction markets (Polymarket, Kalshi)? (Ridn) Not yet, considered. He suspects it would just mirror CME’s published odds. Bigger reason to hold off: commercial news has been a nightmare — Benzinga/Dow Jones will randomly repost a year-old earnings recap with no flag, and Heimdall was processing those as current earnings. He’s been filtering hard the past few days.
Social sentiment — Reddit, X? Planned, not started.
What’s IV Misprice actually telling me? (localwoman) An opportunity to sell options, typically. He doesn’t run theta so it’s unvetted. The exception: when options are cheap it’s a directional hint — check the projection for bias (e.g. a biotech with cheap IV into a PDUFA).
Stop losses? con_dad’s eternal problem — 2% gets chopped out, 10% is a lot. Ridn runs 10% personally. Conq: the fix for price levels being wrong at signal generation is found, and decay at price level isn’t working correctly yet — both in the pipeline.
Isn’t there other stuff like this? Every time. Nothing retail-accessible at this scale; the only comparable systems are institutional. The picture he showed with the connections is 3–4% of the universe; the “full universe” one is old and the current one is six times that size. Unchained, it hits Google’s commercial rate limits.
Two fixes from today, because the last two days of signals felt off and he pulled the stats after close: (1) the model had stopped waking before events, so calls were all hindsight; (2) an outdated-package bug meant a woken arc was missing the last 15–30 minutes of news. Both fixed. It can now also backfill SEC filings on its own when it’s missing context.
What he asked of us
Same as last week, louder: get in in the morning and filter. Heimdall isn’t designed for a high win rate, it’s designed to surface well-evidenced opportunities; “short squeeze” means shorts will be forced to cover, not moon — magnitude is on us to find, through the dies-if levels and through talking. The reason the roles all point at reading, contradicting, and replying is that the filtering is exactly what produced the 80%.
And the cautionary tale from this week: a call went out on trading floor the old way someone saw something, people piled in, it went sideways, and instantly it was why? it’ll come back, this is temporary. Held on emotion. Meanwhile multiple people cashed out slightly-red Heimdall trades without a second thought because there are more plays tomorrow. “I’ve seen the thing that used to catch us happen all over again. That’s why this is the direction.”
Sunny’s version: arguing against an AI is easier than arguing against a person who spent all night on a DD. Conq’s: the worst part of the old way was people who put their heart into a play that worked and got no exposure because they didn’t have the role. Gone.
Next VC Night is on the Events calendar. 8:30 stays because that’s what won the vote — “you people are freaks” (Tiddly). Only ~150 people get the pings today; there are still 2,000 in the server.
Roll call: JB is here, technically, in spirit — Tiddly saw him yesterday, joebeewon the other day, and then he died tragically per Conqueror. Beaker got a PC. Holi is remembered yet again for selling the car — and, per Tiddly, he only had it made because Conq told him to sell. Kevin Chen was the Pfizer expert who was never actually in Pfizer. Legacy “Legends” DD were usually a pep rally that was red the whole time. BBIG’s CEO is in jail and for all that time was spending the BBIG pump money on gay blowjobs. Cryptyde dividend nostalgia. Tiddly gets zero bitches (official). It’s Holden Tudois — French Canadian — and Conq should have known, being 1.25% French.