RELEASE27 April 2026 · 14 min read · v6.4.0

Hidden Tape v2 ships: AI-verified whale flow + structural classification

Three of the four Hidden Tape v2 modules went live today. v6.3 cross-checks framework whale signals against external sources via AI + web_search. v6.5.0a surfaces SEC Form 4 insider activity per ticker. v6.4 classifies the current bar's structural state (TRAP / FADE / CHASE / CLEAN / NEUTRAL) and gates entries with an action verdict (OK / WAIT / AVOID). Calibration on 8 tickers shows zero hallucinations, 37.5% AI override rate, and a $435M insider-selling pattern on PLTR the framework alone would have missed entirely.

The Founding 100 program is now open. $14.50/mo (50% off Pro forever) for the first 100 paying subscribers, locked in for as long as you stay subscribed. When slot 100 fills the program closes and Pro returns to $29/mo for new subscribers. We'll cover what shipped, what the calibration evidence showed, what's still missing, and the honest pitch for why $14.50/mo is the right way to support a one-person dev shop building this.

What "Hidden Tape v2" actually means

The original Hidden Tape Coach (v1, shipped in v5.8.8) was an inference layer. It took the framework's hidden_tape sub-score (Point 11 of the 11-point indicator framework) plus the whale_sentiment sub-score (Point 5), wove them together with OBV trend and volume ratio, and narrated what the framework thought institutional flow looked like. It worked. Users liked it. But it had a structural problem: it was a closed loop. The framework computed its scores from public price/volume/options data, then v1 narrated those same scores back. There was no external check.

The reframe was simple to state, hard to ship: turn Hidden Tape from an inference panel into a verification panel. Wherever the framework says "whales are accumulating," an outside source has to back that up before it counts as tradeable. Whatever the framework misses, an outside source has to be able to surface. The user's actual decision question at trigger time is "are the whales doing what the framework thinks they're doing?" — and a closed loop can't answer that.

Hidden Tape v2 is four modules, each answering one question:

  1. v6.3 Whale Confirmation Coach — are the whales doing what the framework thinks? AI cross-checks against SEC EDGAR, FINRA, public dark-pool aggregators.
  2. v6.4 Trap & Structure Coach — are we in a gap trap, a fade, a chase, or a clean breakout? Where exactly should the entry go?
  3. v6.5 Critical-gaps batch — relative strength, gamma walls, float/short, R/R enforcement. (v6.5.0a Form 4 module shipped early; rest in May.)
  4. v6.6 Entry/Exit Optimizer — single TRIGGER OK/WAIT/AVOID call stacking all the above. (Shipping end of May.)

Three of the four are live as of today. The fourth ships in May. Founding Members shape what gets prioritized and how the calibration thresholds are tuned.

v6.3 in production: what the AI actually catches

Real result on PLTR, run live during calibration: 

🐳 WHALE CONFIRMATION · PLTR · PARTIALLY_CONFIRMED · 72% agreement · high conf

AI VERIFICATION · 6 SOURCES CHECKED
  ✗ Heavy insider selling: 227 sales vs 0 buys in 6 months;
    Peter Thiel sold 4M shares in 90 days through April 22
    → SEC EDGAR Form 4 (StockTitan, Yahoo Finance, Quiver Quantitative)

  ○ Mixed institutional: BlackRock +4.8M shares (+2.6%) Q4 2025;
    Cardano Risk -5.9M shares (-89.4%) Q1 2026
    → SEC 13F (Quiver, Fintel, Daily Political)

  ✓ 92% of tape volume routed through dark pools/TRF; 63.3% ASK-side
    aggressor lean indicates institutional buying pressure
    → Databento consolidated SIP + FINRA TRF (provided tape)

  ✓ 19 analysts rate Buy consensus; avg target $186.47 (30% upside)
    → Public.com, 24/7 Wall St., Daily Political, Seeking Alpha

  [+ 2 more findings...]

FLAGS
  [OVER_READ]   hidden_tape (2/10) too low: Databento shows 92% dark
                pool routing with 63.3% ASK lean — institutional buying
  [MISSED]      Framework missed insider selling pattern (227:0 sells)
  [UNDER_READ]  Insider selling correctly identified but not contextualized:
                Thiel's March 2 sales were 10b5-1 plan adopted Nov 2024 —
                pre-arranged, less informative than discretionary

RECOMMENDATION  CONFIDENCE HIGH
  Extreme insider selling is CONFIRMED but contextualized as 10b5-1
  mechanical. Framework's whale_sentiment=5 is partially supported but
  hidden_tape=2 is too low given dark-pool tape behavior.

ADJUSTMENT: hidden_tape=5 (not 2); whale_sentiment=4 (not 5);
            insider_selling_flag=HIGH; volume_ratio=0.80 (not 0.63)

The AI didn't just rubber-stamp the framework. It found the insider-selling pattern, but then added the 10b5-1 nuance — pre-arranged sales aren't the same signal as discretionary insider conviction. That's the synthesis layer earning its keep. The framework alone wouldn't make that distinction; it doesn't have a Form 4 fetcher. v6.5.0a closes that gap natively (more on that below).

v6.4 in production: TRAP / FADE / CHASE / CLEAN

The user's actual pre-trigger checklist (verbatim from a conversation that drove the v6.4 spec):

"Last few things I ask myself: are we in a gap trap or liquidity sweep? Or chasing the gap up and falling knife scenario is imminent? Should I buy the fade and stay away from the gap up? I check and tie in RSI to if stock is exhausted or not. I use S/R to see where I should place my entry or exit price levels at and use whale tape to pull or not pull the trigger."

v6.4 codifies that workflow. The panel renders below v6.3 when the Whale Confirmation panel is showing, when Entry/Exit Coach is active, or when RSI is at extreme. It ingests:

Auto-rule heuristic produces a first-pass classification. AI confirms or overrides with reasoning. Real result on NVDA, where the AI overrode: 

🪤 TRAP & STRUCTURE · NVDA · CHASE · WAIT · high conf
                              auto-rule said NEUTRAL → AI says CHASE

GEOMETRY READ
  Price has extended 4.5% intraday from $208 close toward weak $220
  resistance, sitting $5.72 above primary support at $210.89 (1.00× ATR).
  RSI(14) is 76.3 (overbought) with 2 consecutive days above 65, signaling
  momentum exhaustion despite rising momentum direction. This is a
  momentum-only chase into weak resistance with no structural retest or
  clean breakout.

PRICE LADDER
  $230.00  2.33× ATR   round_number [psychological]
  $220.00  0.59× ATR   round_number [psychological]
  ── $216.61 PRICE — ATR(14) $5.74 ──
  $215.94  0.12× ATR   vp_lvn [volume_void]
  $214.19  0.42× ATR   vp_lvn [volume_void]
  $212.44  0.73× ATR   vp_lvn [volume_void]
  $210.89  1.00× ATR   round_number +5 more [psychological · volume_void
                                              · swing_pivot · day_pivot]
  $207.19  1.64× ATR   vp_lvn [volume_void]

WHY WAIT
  CHASE state confirmed by price extension away from primary support
  ($210.89) toward weak resistance ($220) on elevated volume (1.21×) but
  with exhausted RSI (76.3). R/R unfavorable: stopping at $210.89 yields
  $5.72 risk vs $3.39 reward to $220 (0.59:1, far below 2:1 threshold).
  Overbought RSI + weak resistance target + no structural confirmation
  = WAIT for retest or pullback.

BETTER ENTRIES
  $210.89 limit  R/R 2.5:1  Retest of primary support (very strong: round
                            + 5 sources). Risk to $205.43 (1.95× ATR);
                            target $220. Cleaner structure on reclaim.
  $215.94 limit  R/R 1.8:1  Volume void pullback. Tighter stop $210.89.
                            Marginal but less exhausted.
  $199.90 stop   R/R 3.2:1  Secondary target (very strong: prior day low
                            + 2 sources). If $210.89 breaks, this is the
                            structural target.

NEWS RISKS
  [MEDIUM] Hyperscaler earnings on April 29 with capex updates
  [LOW]    NVDA earnings scheduled for May 20, 2026

cost = $0.031 · 1 tool call · auto-rule said NEUTRAL → AI says CHASE

Look at what the AI did that the auto-rule couldn't:

Cost: $0.031 per call. Latency: ~25s. We don't claim it'll be right every time. We do claim it'll show its work, cite specific levels, and refuse to pretend certainty when the geometry is ambiguous.

v6.5.0a in production: Form 4 insider chips on every card

Calibration on the v6.3 module surfaced a pattern: insider selling was the #1 signal AI consistently caught that the framework missed. Every single calibration ticker had the AI flag insider selling via web_search; every single one was [MISSED] from the framework's perspective. That's a real product gap.

v6.5.0a closes it. A free SEC EDGAR Form 4 fetcher pulls the last 90 days of insider transactions per ticker, classifies into EXTREME / HEAVY / MODERATE_SELLING vs BUYING vs NEUTRAL, and surfaces the result two ways:

  1. Per-card chip visible on every ticker card — color-coded with $-shorthand. 🚨 INSIDER SELL ($435M) red on PLTR, ↓ INSIDER SELL ($172M) red on NVDA, hidden on MSFT (NEUTRAL signal — only 1 sale of $5M, normal background).
  2. Whale Confirmation prompt now includes the structured Form 4 aggregate as primary evidence — the AI sees Peter Thiel's 4M-share sale BEFORE deciding whether to web_search for it. Saves tokens, prevents hallucination, and eliminates the [MISSED-insider] flag from v6.3 results.

Calibration evidence on 8 tickers (snapshot from 27 April 2026):

Ticker Signal Sells $ Value Top seller
PLTREXTREME_SELLING60$435MThiel (Director) [10b5-1]
GOOGLEXTREME_SELLING95$410MGV (10% Owner)
NVDAEXTREME_SELLING62$172MPuri (Officer EVP) [10b5-1]
METAEXTREME_SELLING88$107Mvarious
VRTEXTREME_SELLING52$104Mvarious
MUEXTREME_SELLING23$43Mvarious
AAPLEXTREME_SELLING9$25Mvarious
TSLAEXTREME_SELLING23$21Mvarious
MSFTNEUTRAL1$5Mbackground activity

Total cost: $0. SEC EDGAR is free public data; the only requirement is a User-Agent header per their fair-access policy. The integration is local-first like everything else — your dashboard pulls Form 4s on the audit cycle, caches them 24h, and renders the chip without any cloud dependency.

The honest scope notes

Three things to be straight about before you sign up.

v6.3 needs BYOK keys to be at full strength. The Whale Confirmation Coach uses your Anthropic API key (BYOK; we never proxy AI calls). Without Polygon or Databento configured, the AI verifies via web_search alone — still useful, but you lose the live tape grounding. The hybrid Polygon Starter ($29/mo) + Databento PAYG (~$0–6/mo) combo gets you ~80–90% of what Polygon Advanced ($199/mo) would deliver, for ~$30/mo. We documented the trade-offs honestly on the pricing page.

v6.4's latency is ~25–30s per call. The spec was optimistic about <8s. Anthropic's web_search tool unavoidably adds round-trip time for the news-risk discovery feature. We could turn off web_search to hit the latency target — but losing news-risk awareness is a bad trade. We chose the user-visible value over the spec gate.

v6.6 is not yet shipped. The synthesis layer that stacks v6.3 × v6.4 × v6.5 into a single TRIGGER OK/WAIT/AVOID call is still in build. Today you have to read v6.3 and v6.4 separately and mentally combine them. v6.6 is the matrix that does that for you. Estimated ship: end of May 2026.

Founding 100: $14.50/mo, locked forever, only 100 slots

The Founding 100 program is now open. Pricing page here. Key terms:

Why the Founding 100 program exists at all: shipping software to a real audience is fundamentally different from building it in private. The first 100 paying users aren't just early adopters — they're testers whose feedback shapes which Hidden Tape v2 verification signals fire, what counts as a "trap," and how the AI gets calibrated. The discount is the cost of admission to a small advisory community. We don't expect every Founding Member to use every feature; we expect them to tell us when something's off.

What's in the box today

If you sign up right now, here's what you get on first dashboard load:

Plus everything in the roadmap as it ships.

The pitch

Most retail tools sell certainty. Swing Deck sells discipline. The framework forces you to read 11 sub-scores and 13 risk pillars before you trade. Hidden Tape v2 forces you to verify what the framework is telling you against external sources. v6.4 forces you to look at structural risk before you pull the trigger. v6.5.0a forces you to acknowledge that the CEO sold $171M of stock in the last 90 days. None of this is fun. All of it makes the trade better.

If you're already running a similar discipline by hand — checking SEC EDGAR every morning, scanning 13F filings, building S/R levels in TradingView, watching for liquidity sweeps — Swing Deck is just the dashboard that does it for you. If you're not, this is the wrong product. We're not trying to convert anyone; we're trying to find the 100 traders who already think this way and just want better tooling.

Claim a Founding Slot — $14.50/mo

Questions, feature requests, or just want to see a screenshot before you commit? Reply to contact@swingdeck.app. Real human, usually under 24h.

Disclosure: Swing Deck is built and operated by one person. The product is local-first; positions, broker tokens, and journal entries never leave your machine. AI features use your own API keys (BYOK). We don't proxy your data through our servers. Pricing as of 2026-04-27. Past performance is not indicative of future results. Nothing in this post is investment advice; it's a description of what software does.