Every statement carries fingerprints: how its author refers to themselves, how confidently they hedge, whether it agrees with itself. Veracity Vault measures those fingerprints against the research on what truthful and deceptive language actually look like — then tells you how well you read the room.
Research disclaimer: no single linguistic cue proves anyone is lying. This lab trains pattern recognition on statistically documented tendencies (distancing, hedging, cognitive-load disfluencies, specificity, self-reference) and your own calibration — not to detect real deception. Use it as a thinking tool, not a lie detector.
1 · The Statement
0 characters
Ground truth (optional but recommended)
Enables contradiction checks and a "matches reality" score. Leave blank for a pure cue profile.
50%
2 · Cue Profile
Deceptive-cue density: – (0 = reads like truth-telling language, 100 = dense in deception cues)
truth-telling profilemixeddeception-dense profile
Cues are frequency-normalized per 100 words. Counts are shown for each flag detected in this statement.
3 · Internal Consistency
Run the analysis to extract the factual claims in the statement and check them against each other and against your ground truth.
4 · The Verdict & Your Calibration
–
AWAITING ANALYSIS
Sealed Cases
Each case: a classified case file (what you know is true), one witness statement, and six claims you must rule. The witness is a rule-based generator — truthful cues and deceptive cues are embedded by design, so the lab can measure how well you separate them.
Case
Your case file — verified facts
— Witness statement, recorded on site
Rule each claim (does the witness tell the truth about it?)
About Veracity Vault
Why this project?
The internet's lie-detection tools are all either AI video analysis (unavailable offline, privacy-hostile) or multiple-choice quizzes about a single "telltale". Meanwhile a large body of peer-reviewed research — meta-analyses of 79 linguistic cues across 44 studies, corpus studies of deceptive vs. truthful speech, cognitive-load experiments on fabricated narratives — has documented real, measurable tendencies: liars distance themselves from events, add confidence without evidence, produce cognitive-load disfluencies, and break internal consistency. Almost none of it exists as a hands-on tool you can use on a phone to practice reading language. Veracity Vault turns that research into a training instrument.
What is it?
A forensic linguistic analysis lab with two modes. Forensic Analysis: paste any statement (and, if you know it, what actually happened), state how confident you are that it's truthful, and get a cue profile, an internal-consistency report, and a calibration grade. Cross-Examination: six sealed cases, each with a classified case file of verified facts and a witness statement. You rule each of six claims true or false, and the lab scores your verdict, shows you which cues you missed, and debriefs the witness.
How does it work?
Cue detector: 16 documented cues, each a pattern matched against the text and frequency-normalized per 100 words. Red cues (distancing, non-self-reference, hedges, disfluencies, unsupported confidence, self-serving emotion) vs. green cues (self-reference, concrete specifics, memory hedges, admitted uncertainty, sensory detail). Each cue has a weight from its documented effect size direction.
Consistency engine: extracts structured claims — times ("around nine"), places ("the warehouse"), and actor references ("I"/"we"/"they") — from both the statement and your ground truth, then flags contradictions (different times for the same moment, places that conflict with verified facts, pronoun drift away from the speaker).
Calibration grader: compares your pre-analysis confidence (0–100%) against the computed veracity score and reports your calibration error in the style of a meteorologist's Brier score: were you overconfident, underconfident, or sharp?
Case engine: each case is a deterministic template. The generator plants N deceptive cues in the witness statement tied to specific false claims, and truthful detail in the rest. Your per-claim rulings are compared to the ground truth; the debrief reveals which planted cue backed each false claim.
How to use it
Open Forensic Analysis, paste a statement (or hit Load Sample), optionally paste what you know to be true.
Before running, set your confidence slider — how sure are you it's truthful? Be honest; this is the part being measured.
Press Run Forensic Analysis. Read the cue profile (red vs. green flags with counts), the consistency report (contradictions found), and your calibration grade.
Switch to Cross-Examination, pick a sealed case, read the case file, then the witness statement. Rule each of the six claims True/False and submit your verdict.
In the debrief, compare your rulings to the ground truth and see exactly which linguistic cues supported each lie. Revisit the Field Manual to build the habit.
Press ⓘ anytime for this help. Press Esc to close it.