The Growth Signal Stack: Measure the Decision, Not the Available Data

The easiest metric to collect often becomes the strategy by accident. Competitor post counts become “share of voice.” Ranking volume becomes “market demand.” TVL becomes “adoption.” The dashboard looks decisive because the substitution happened before anyone asked what behavior the number actually represents.

Start with the decision and relevant behavior, then define the observable, unit, population, window, missing-data rule, and threshold that would change the decision. A metric is useful only when its source-native meaning survives all the way to the budget, channel, or competitive choice it is meant to inform.

Violet’s 13 concentration profiles demonstrate the risk. X publishing, DappRadar rankings, and DefiLlama catalog and TVL evidence produced valid but incompatible distribution shapes. The analysis does not tell a company which channel or investment to choose. It shows why that choice must begin with the decision rather than the available proxy.

This article combines original Violet analysis of public data with a separately labeled operating framework. The source-native posts, ranked rows, listings, and TVL histories remain attributable to X, DappRadar, and DefiLlama.

At a glance

Growth-leader question	Decision
Can one metric describe the public ecosystem?	No. Each observable has its own unit, population, window, and missingness.
Is competitor publishing the same as audience attention?	No. Post volume measures publishing supply in this dataset.
Can rankings or TVL stand in for adoption?	No. They retain their source-native meanings.
When is a metric decision-worthy?	When its specification is complete and crossing its threshold changes a declared decision.
What should happen to dashboard metrics with no decision?	Relabel, replace, or remove them.

The objective is not to eliminate proxies. It is to stop a proxy from silently inheriting a stronger business meaning.

Signal evidence

The same cohort does not have one concentration score: X’s 16-account eligible-post panel has a top-node share of 16.8% and HHI 0.105, DappRadar’s 50-row all-category 24-hour-volume panel has 37.9% and HHI 0.158, and DefiLlama’s five-product latest-core-TVL panel has 57.7% and HHI 0.470.

The X account corpus describes 7,546 eligible non-retweet posts across 16 verified accounts during the frozen 90-day window. The DappRadar rankings describe source-specific page-one rows and ordering metrics at capture time. DefiLlama describes accepted catalog components and qualifying core-TVL histories for a smaller eligible set.

Observable	Eligible nodes	Top-node share	HHI	Effective nodes	Nodes covering 80%
X eligible posts	16 accounts	16.8%	0.105	9.49	8
DappRadar all-category 24h volume	50 ranked rows	37.9%	0.158	6.31	18
DefiLlama accepted listings	15 matched products	20.8%	0.108	9.29	11
DefiLlama latest core TVL	5 qualifying products	57.7%	0.470	2.13	2

These rows do not disagree. They answer different questions. The growth mistake begins when one row is relabeled as the answer to another.

Growth signal stack

These are source-specific distributions, not market shares: an effective-node count of 9.49 for X posts, 6.31 for DappRadar all-category volume, or 2.13 for DefiLlama core TVL describes only that panel’s observed total.

A metric belongs to the layer its unit can represent, not the layer the team wishes it represented.

Signal layer	Behavior of interest	Evidence required	Convenient but insufficient proxy
Publishing supply	What organizations produce publicly	Comparable published objects	Follower count
Audience attention	What the intended audience actually consumes	Qualified impressions, viewing, or attention behavior	Competitor post count
Discovery exposure	Where qualified people encounter the product	Search, referral, ranking exposure, or attributed discovery	Listing presence alone
Economic activity	Value moving or held under a defined system	Source-native economic units and scope	Social activity
Activation	Whether acquired users reach meaningful product value	Product-defined activation events	Clicks or sign-ups alone
Retention	Whether value persists	Cohort behavior over a declared window	Total registered accounts
Commercial outcome	Whether the motion produces the intended business result	Qualified revenue, pipeline, purchase, or expansion evidence	Reach or engagement alone

The current X data can represent publishing supply. It does not contain audience attention. DappRadar’s visible ranked rows can represent the captured ranking observable, not total discovery or adoption. DefiLlama TVL can represent value locked under its adapter scope, not users, transactions, revenue, or retained product use. DappRadar and DefiLlama therefore remain useful only when their native unit matches the decision.

Decision-to-metric specification

A metric is decision-worthy only when its unit, population, window, missing-data rule, and action threshold are explicit.

Complete this before accepting a metric into a dashboard or strategy memo:

Social-channel decision

Field	Specification
Decision	Choose whether to expand a social channel
Relevant behavior	Qualified audience response
Observable	Not established by the current post-count corpus
Unit	Requires attention and response data
Eligible population	Named target audience
Window	Declare before collection
Missing-state rule	Unavailable remains unavailable
Decision threshold	Define the result that would change channel investment
Refused proxy	Competitor post count
Owner	Named growth owner

Aggregator decision

Field	Specification
Decision	Evaluate an aggregator
Relevant behavior	Qualified discovery and downstream action
Observable	Requires referral and product-path evidence
Unit	Qualified referrals and actions
Eligible population	Users arriving from the source
Window	Declared campaign or cohort window
Missing-state rule	Unattributed traffic remains unknown
Decision threshold	Define the result that would change participation
Refused proxy	Listing or ranking presence alone
Owner	Named channel owner

Economic-signal decision

Field	Specification
Decision	Review an economic ecosystem signal
Relevant behavior	Source-native economic activity
Observable	Accepted scoped economic observable
Unit	Exact source-native unit
Eligible population	Exact eligible protocols/products
Window	Fixed snapshot or history
Missing-state rule	Uncovered is excluded, not zero
Decision threshold	Define the result that changes the economic thesis
Refused proxy	Social attention or user count
Owner	Named strategy owner

The first two examples deliberately leave outcome observables unavailable because the current research does not contain them. The specification exposes the missing evidence rather than promoting post count or ranking presence into the empty field.

Use this sequence:

1. Name the decision.
2. Name the behavior that would inform it.
3. Select the closest observable without changing its unit.
4. Freeze the eligible population and window.
5. Define true zero, missing, unavailable, and source-not-covered states.
6. Set the threshold that would change the decision.
7. Name the tempting proxy that must be refused.
8. Assign an owner and review date.

Signal missingness

True zeroes remain inside an eligible universe, but unavailable records do not become zero: all 16 X accounts are observed, DefiLlama’s listing panel contains 15 matched products and excludes one source-not-covered product, and its core-TVL panel contains only the five products with qualifying histories.

That rule changes strategy honestly. A source-not-covered product in DefiLlama is not evidence of zero adoption or economic activity. A featured row without a normalized ordering value in DappRadar cannot enter a numeric sensitivity analysis. The missingness policy belongs in the metric specification before results are visible.

If unavailable records would materially change the decision, the action is to obtain the missing evidence or decline the decision. Zero-filling is not an analytical shortcut; it is a change to the claim.

Concentration method

Violet calculated five measures independently for every eligible observable: top-node share, HHI, Gini coefficient, effective node count, and the minimum nodes needed to cover 80% of the observed total.

Figure 1. Thirteen independent profiles. Horizontal position encodes concentration inside the declared panel; it does not encode adoption, market power, or growth opportunity.

Figure 2. Separate Lorenz curves for three incompatible units and eligible universes. Similar curve shapes do not make the underlying objects equivalent.

These are source-specific distributions, not market shares: an effective-node count of 9.49 for X posts, 6.31 for DappRadar all-category volume, or 2.13 for DefiLlama core TVL describes only that panel’s observed total.

The calculations remain useful for questions about the shape inside a declared universe. DappRadar and DefiLlama cannot be combined simply because both yield a concentration statistic. The method follows the metric specification; it cannot repair a mismatched business question.

Signal review

Every dashboard metric should earn its place by naming the decision it can change.

Run this review for each metric:

1. What decision owns this metric?
2. Which behavior and signal-stack layer does its unit represent?
3. Is the population and window complete enough for that decision?
4. How are zero, missing, unavailable, and uncovered states treated?
5. What result would change the decision?
6. Which stronger meaning has the label implied but the data not established?
7. Who owns the response when the threshold is crossed?

Keep metrics with a valid decision and complete specification. Relabel honest proxies. Replace metrics whose units cannot represent the required behavior. Remove metrics with no decision. When the necessary outcome input is unavailable, record the measurement gap rather than substituting a more flattering number.

What this does not show

The concentration analysis does not establish monopoly power, product adoption, customer count, economic output, market size, audience reach, causal advantage, or which product is winning. It does not support a channel or investment recommendation on its own.

The signal-stack framework improves decision discipline. It does not turn a proxy into causal evidence or supply the missing attention, activation, retention, or commercial outcomes.

Method and source disclosure

Violet used the complete 16-account X eligible-post corpus for the 90-day window ending September 10, 2026; 10 DappRadar page-one category views containing 453 normalized ranked rows; and accepted DefiLlama evidence containing 24 listings across 15 matched products and core-TVL histories for five products.

Every concentration profile freezes one source, metric, unit, window, eligible universe, and missing-state policy. The release harness independently recomputed the metrics through a second code path. The Growth Signal Stack, Decision-to-Metric Specification, and dashboard review are Violet’s operating framework, not observed growth outcomes from those source panels.