Outflow Build Log: Signals Worth Acting On

What separates a market signal worth acting on from noise — and why everything downstream in Outflow inherits the quality of the signal at the top.

The Cost of Acting on Noise

Markets produce movement constantly — hires, launches, announcements, leadership changes, category shifts. Most of it is noise for any given sender: true, public, and irrelevant to a specific reason for contact. The failure mode is not seeing too little; it is acting on too much.

Acting on noise is not free in outbound. Every weak contact spends something: the recipient’s attention, the sender’s reputation, the operator’s review time at the approval gate. Volume arithmetic says send anyway; reputation arithmetic says each mis-targeted touch is a small withdrawal against a balance that is hard to rebuild. Signal selection is where that cost is decided, before research, before drafting, before the brief ever reaches a person.

The Signal Classes We Track

Outflow’s Observe stage watches several classes of market movement, each answering part of the question a brief has to answer: why this account, why now:

• Hiring. A team posting roles is declaring a priority in budget, not in press. What is being hired reveals what is being built, and where internal capacity is short.
• Funding. A budget event changes what an account can afford to do and how fast it is expected to move. It is timing information.
• Product movement. Launches, deprecations, positioning changes: a surface that just changed has new gaps, new neighbors, and new reasons to look closer.
• Intent movement. Behavior that suggests active evaluation — an account circling a category rather than merely existing in it. The strongest class and the hardest to observe honestly.

No single class answers the question alone. A funding event says money moved; it does not say the account needs what you sell. The classes compose: a hiring pattern plus a product change plus intent is a reason; any one of them is a headline.

Actionable Versus True

The distinction that took the longest to encode: a signal can be true and still be noise. The working tests a signal has to pass before it is worth acting on:

• Verifiable. Can we point to the evidence — the posting, the announcement, the observed change? A signal that cannot be cited inside the brief cannot be trusted outside it.
• Recent. A stale signal is history. What made it a reason has a half-life, and acting on expired movement reads as exactly what it is.
• Specific. Does it distinguish this account from its peers? If the same signal applies to every company in the category, it is context, not a trigger.
• Connected. Does it map to a reason for contact the recipient would recognize as honest? If the link between signal and outreach needs to be explained at length, there is no link.

Fail any test and the signal is noise for action purposes — worth logging, not worth a brief. The working view exists to make that legible: category change found — a reason to look closer; evidence checked — within your rules; next action — ready to shape.

What Did Not Work

The first version treated signal volume as coverage. More sources, more classes, more briefs, and the approval queue flooded. The human gate is the system’s control surface; drowning it in marginal briefs raises the cost of every review and degrades the judgment the gate exists to apply. Recall was the wrong target. Precision at the top of the funnel is what keeps the rest of the loop honest.

The second lesson was about public signals. An announcement every sender can see produces weak reasons: if everyone in the market received the same trigger at the same time, citing it is coincidence dressed as context. The signals that survive the four tests tend to be the ones that require work to notice: patterns across classes, or movement visible only if you were already watching the account.

This is the sense in which signal quality gates everything downstream. Research depth, personalization, the operator’s attention: each stage inherits the input’s quality and can only subtract from it. A mediocre signal does not produce a mediocre brief; it produces a well-researched, well-written reason that should not exist.

What This Does Not Claim

The signal taxonomy is a heuristic, not a predictor. We do not claim that acting on these classes produces replies, meetings, or pipeline — Outflow is an operational experiment, and the taxonomy itself is part of what is being tested. The agent filters and ranks; the judgment that a signal justifies contact stays behind the human gate, where consequential action waits for explicit approval. What this log records is the mechanism (how actionable is distinguished from merely true) and where the evidence currently stops.