What 10b5-1 actually means (and why "EXTREME SELLING" usually isn't).

Tim Cook just sold $25M of AAPL. Bearish, right? Probably not. The Form 4 says SALE; the calendar says he committed to those dates last September, while not in possession of any material non-public information. Reading the chip color matters more than reading the headline.

The Form 4 surface trap

Insider sells one of your holdings. The notification fires, you check the SEC filing, the transaction code says S. Sale, $X million, executed yesterday. Brain pattern-matches: insider knows something, get out.

The trouble is that Form 4's S code doesn't distinguish between two very different stories. Story A: the insider looked at the chart this week, decided this is a good time to sell, instructed the broker, the broker executed. Story B: the insider signed a written plan eight months ago committing to sell N shares on the third Tuesday of every month at any price above $X, can't remember what's about to happen at the next earnings call, and the broker mechanically executed yesterday's tranche regardless of news, sentiment, or the insider's current view of the company.

Story A is a signal. Story B is a calendar event. The Form 4 surface treats them the same.

What Rule 10b5-1 actually is

Rule 10b5-1 is an SEC regulation, in force since 2000, that gives insiders an affirmative defense against insider trading charges if they execute trades under a pre-arranged written plan. The plan has to be set up when the insider is not in possession of material non-public information (MNPI), specify the trade parameters (dates, prices, quantities, or a formula), and be followed mechanically.

The 2023 amendments tightened the rules — a 90-day cooling-off period between plan adoption and the first trade, a one-plan-at-a-time limit for most insiders, and a certification requirement that the insider was not in possession of MNPI when adopting the plan. Plans aren't a tax-evasion gimmick or a fig leaf; they're a structured pre-commitment that lets executives diversify a concentrated position without their selling timing being interpreted as a signal.

The relevant artifact for our purposes: every Form 4 filing now carries a flag, <aff10b5One>, set to 1 when the underlying transaction was executed under a 10b5-1 plan. It's been there since the 2023 amendments. Most retail dashboards don't read it.

Why the distinction breaks the framework's read

The naive insider-selling classifier looks at one ticker, counts the sells, sums the dollar volume, and labels the result. Five sells, zero buys, $25M of selling — that's an EXTREME_SELLING reading by any conventional definition.

Apply that classifier to a mega-cap CEO with a standing 10b5-1 plan and it fires every quarter. Tim Cook sells AAPL in basically every measurement window. So does Sundar Pichai with GOOGL. So do most large-cap executives. Their plans run for years, executing automatically, completely decoupled from anything happening in the business. A classifier that flags every one of those windows as EXTREME_SELLING is correct on the math and useless as a signal.

The fix isn't to lower the threshold. It's to read the 10b5-1 flag and route differently.

The AAPL calibration

A v6.4 audit on AAPL came back with signal_label = EXTREME_SELLING. Nine sales in the 30-day window, zero buys, ~$25M total, well above the EXTREME_SELLING floor. The Whale Confirmation Coach — the AI surface that runs web_search against the framework's own conclusions — disagreed. Pulled the underlying Form 4s, found that all nine sales were marked aff10b5One = 1, returned a CONTRADICTED verdict with the explanation: these are calendar-driven dispositions under a longstanding plan, not a discretionary signal.

The AI was right. The framework's structural read was wrong. v6.5.0a-4 closed the gap by promoting the 10b5-1 distinction from a verification-layer override to a first-class classifier input — read the flag at parse time, compute the share of dollar volume under 10b5-1, route accordingly.

The 80% threshold

The classifier short-circuit looks like this:

pct_10b5_1 = sum(t.value for t in sells if t.is_10b5_1) / total_sell_value

if pct_10b5_1 >= 0.80 and sell_count > buy_count:
    return "SCHEDULED_SELLING"

Eighty percent of dollar volume, not count. A 10-sale window where nine sales totaling $24M are 10b5-1 and one $1M sale is discretionary still qualifies as scheduled — the discretionary sliver isn't a meaningful signal at that scale. A 10-sale window where five sales totaling $13M are 10b5-1 and five sales totaling $12M are discretionary doesn't qualify; the discretionary half is a real read and the chip should reflect that.

The threshold is deliberately conservative on the discretionary side. Buys are exempted from the short-circuit entirely — buys under 10b5-1 do exist but are rare in practice, and an insider buying their own stock under any plan structure is a stronger signal than the equivalent sell. The asymmetry is intentional.

How to read the chip

The card chip used to be binary in tone: red for sells, green for buys, hidden for neutral. Now there's a third state for SCHEDULED_SELLING — a calendar glyph and a cyan tone, distinct from both the red EXTREME chip and the hidden NEUTRAL chip:

📅 10b5-1 sells  ·  Pre-scheduled — insider committed months ago while not in possession of MNPI. Volume reflects a calendar, not a decision.

Same data magnitude as a red EXTREME chip; different interpretation. The chip color is the answer to "is this discretionary or scheduled?" — at a glance, no Form 4 reading required.

When insider sells do matter

The point of the SCHEDULED_SELLING label is not that insider sells stopped mattering. It's that the bearish read narrows to the cases where it actually applies:

• Discretionary sells — Form 4s with no aff10b5One flag, especially after a meaningful price runup.
• Multi-insider clusters — CEO + CFO + Director selling within a tight window, especially when the plans aren't aligned.
• Plan terminations followed by new sells — an executive cancels their 10b5-1 plan and then sells discretionarily within weeks. That's a deliberate move to act on something.
• The buy side — open-market insider buying remains a strong positive signal. It almost never happens for theatrical reasons.

Most of what gets reported in financial media as "insider selling intensifies at $TICKER" is calendar volume on a standing plan. Most of what actually predicts forward returns is the discretionary residual after you strip the scheduled portion out. The chip is doing that subtraction for you.

The real question

"Did insiders sell" is the wrong question. The right question is: was it discretionary or pre-scheduled, and what does that ratio look like compared to this insider's history?

The framework now answers the first part automatically — the chip color tells you which story you're reading without a single Form 4 click. The second part still requires judgment, and the dashboard surfaces the inputs (top transactions, role of the seller, plan vs discretionary breakdown) so the judgment has the right material to work from.

A red 🚨 INSIDER SELL chip means something. A 📅 10b5-1 sells chip means something else. Treating them the same was the false positive. Treating them differently is the fix.

Related: v6.5 release notes — SCHEDULED_SELLING label · Whale Confirmation Coach (where the AAPL contradiction surfaced) · R:R isn't a number, it's a question