Echo Undertow

The question this instrument answers: is this audience about to act — or still just talking? Every other tool measures what an audience says. Undertow measures what it is preparing to do — by listening for the moment a public conversation stops hedging ("maybe", "we'll see", "poate", "mai vedem") and starts committing ("never again", "am decis", "de mâine"). That moment is the hedging collapse, and it precedes aggregate action: churn, switching, boycott — or advocacy.

AGGREGATE DISCOURSE ONLY. SCORES DESCRIBE A CONVERSATION'S COLLECTIVE LINGUISTIC FIELD — NEVER AN INDIVIDUAL. NO PERSONAL PROFILES ARE BUILT, STORED OR INFERRED. EVERYTHING RUNS IN YOUR BROWSER; NOTHING IS UPLOADED.

01 — Feed the corpus

COLLECT public comments about your brand/topic (3+ weeks)
PASTE below or upload CSV — or try the demo first
READ the verdict: go, monitor, or intervene — with next steps

⌨ Paste text

▦ Upload file

Format: YYYY-MM-DD | text per line (RO + EN supported). Without dates, line order is treated as chronological and split into six periods. Minimum ~60 lines across 3+ weeks for a stable read.

00 — Verdict

WHAT TO DO NEXT

02 — Behavioral pressure

Intention drivers — Ajzen, Theory of Planned Behavior

Attitude · evaluative load0

Social norms · proof density0

Perceived control · deontic modality0

02b — Signal provenance

THE SAME COLLAPSE TRACE IS COMPATIBLE WITH GENUINE DECISION AND WITH COORDINATED INJECTION OF COMMITMENT LANGUAGE — BEHAVIOR ALONE CANNOT ATTRIBUTE (ADAM 2026, COORDINATION–CONSENSUS CONTINUUM). THIS SCREEN REPORTS WHAT THE TRACE CANNOT EXCLUDE.

semantic homogeneity (near-dup share)–

burst concentration–

organicity–

03 — Hedging vs commitment over time

HEDGING RATE COMMITMENT RATE PER 1,000 TOKENS · WEEKLY

04 — Drift alerts

05 — Reading

06 — Method & ethics

Pipeline. Bilingual (RO/EN) marker extraction across five behavioral-linguistic sets — hedging, commitment, exit, deontic control, social proof, future orientation — normalized per 1,000 tokens, bucketed weekly. The first 60% of the timeline defines the corpus's own baseline; the trailing window is tested against it. Drivers are scaled 0–100 and combined with the Armitage & Conner (2001) meta-analytic weighting (attitude .40 · norms .20 · control .40), then modulated by hedging-collapse momentum. Lead-time bands are heuristic and labeled as such.

Adam (2026), The Coordination–Consensus Continuum, DOI 10.5281/zenodo.21191675 — the provenance screen · Ajzen (1991), The theory of planned behavior · Armitage & Conner (2001), BJSP · Michie, van Stralen & West (2011), COM-B · Hyland (1998), Hedging in Scientific Research Articles · Tausczik & Pennebaker (2010), LIWC.

AGGREGATE DISCOURSE ONLY. NO INDIVIDUALS WERE PROFILED IN THE MAKING OF THIS SIGNAL. FULL DRIVER DECOMPOSITION, TOPIC-LEVEL DRIFT AND OUTCOME-CALIBRATED LEAD TIMES SHIP IN THE PAID AUDIT — SEND A SIGNAL →