What is the Chladni hash chain?

An append-only chain of behaviorally anchored segments. Each segment bundles micro-blocks of engine proofs, is hashed (SHA-256), and signed with a non-extractable device key (ECDSA-P256). The chain grows with interaction, not wall-clock time, so pauses do not break it. It stays entirely on the user's device.

Does a hash prove a real-world identity?

No. A hash chain proves continuity: that the present behavior extends a specific local history, signed by a specific device key. It never proves a real-world identity by itself, and EDNA ID does not claim it does.

What happens when the user takes a break?

Nothing breaks. Because the chain is anchored to interaction rather than fixed time intervals, a closed laptop simply means no new segments. On return, resume is decided by matching fresh aggregated behavior against the stored reference vault (toleranced overlap, default threshold 0.85) — never by the last hash alone. A failed match raises a challenge instead of silently continuing.

Can the user delete everything?

Yes. Revocation destroys the device private key, the vault, the memory model and the local log. Nothing is re-derivable afterwards. The memory belongs to the user: local, exportable, revocable.

How does the Response Experiment work?

A response controller translates consented interaction dynamics (tempo, pauses, channel shifts, corrections, message length) into a cautious, overridable response mode for an AI assistant: FLOW, FOCUS, EXPLORE, EXECUTE or REFLECT. The humanness confidence score is never an input — a human at 99.9% deserves the same answers as one at 90%. Copy the generated context into any GPT chat and compare the same question with and without it.

EDNA ID · Chladni Chain · Main Live Production

A persistent, privacy-first human continuity layer for the age of AI agents.

Not just another bot detector. One adaptive behavioral hash chain — grown organically by human interaction, anchored to a device key, stored only on the user's device — gives AI systems three capabilities at once: perceiving how a human evolves, verifying humanness continuously across the whole session, and recognizing the same human again across days and weeks.

Capability 1

01Human Sensoric for AI

AI reads not only how the human is right now, but how the human changes — the evolution of the interaction pattern over a session and over weeks. What is the person going through?

Capability 2

02Continuous Bot Detection

Not a gate at the door like a CAPTCHA — verification must be re-earned with every organic segment of the chain. A bot that passes once has not passed: it must keep growing an organic chain for the entire session.

Capability 3

03Same-Human Recognition

A device-bound EDNA ID emerges from the chain: a handle, a device key, and a self-confirming behavioral pattern. The human is compared with their own history — never with a universal template.

Honest limits — what this demo proves and what it does not

99%+ re-recognition is a research target, not a shipped fact — it requires long-term studies with real participants across sessions.
The chain alone does not stop mimicry bots. Behavioral analysis and challenge-response carry the detection; the chain makes evasion visible and expensive.
A hash proves continuity, never a real-world identity. The chain shows that present behavior extends a specific local history signed by a specific device key — nothing more, and EDNA ID claims nothing more.
Manipulation resistance holds as long as the local history and its trust anchors are not fully replaced by an attacker; everything is verifiable via verifyIntegrity().
Privacy: keys and chain stay local, consent is explicit, chain events are transparent (inspectable below), and revocation is real — destroyable, not just disabled.

Layer 1 · Seconds

Live Memory

The emergence of recognition, signal by signal: micro-blocks of engine proofs accumulate into segments only while 100% human verification stands. The confidence path below is drawn here.

Layer 2 · Hours & Days

Session Memory

Continuity and change within an anchored rolling window: deltas between consecutive segment patterns — which channels moved, how the interaction mix shifts across the session.

Layer 3 · Weeks & Months

Long-term Memory

The individual pattern against the user's own genesis anchor: drift, stability, and the slowly strengthening EDNA ID. Historical truth stays append-only; the memory model can evolve.

Interactive Chain Demo

This runs the real edna.js (v1.4.2-main.1) and edna-chain.js (v2.1.0-draft) in your browser — the same files as in the repository. Move, type and scroll on this page: every proof the engine emits is fed into the chain. Nothing is transmitted; open your devtools network tab and watch.

EDNA ID Account · Login

Enter your handle and click Log in. Then simply move your mouse and type naturally — EDNA matches your live pattern and logs you in automatically after 3 × 100 % in a row.

Live Memory — emergence of recognition

Current proof: —
Open micro-blocks: —
Emerging segment: —

Continuous Verification

closed segments · 0 cryptographic checkpoints

Head hash: —
Sessions: —
Paused for: —

Session Memory — deltas

pattern deltas in the anchored rolling window

Latest delta: —

Long-term — vs your own past

Stability: —
Drifted channels: —
Identity confidence: —

Genesis required. Measured curve — research target, not a promise.

Transparent chain events — every segment, inspectable

— create an EDNA ID to start the chain —

EDNA Response Experiment

From human detection to human-AI interaction. The controller below translates consented interaction dynamics — tempo, pauses, channel shifts, corrections, message length — into a cautious, overridable response mode for any AI assistant. It is a live bridge to GPT via copy & paste: EDNA observes you on this page, you carry the generated context into your AI chat.

FLOW

interact with this page to generate guidance…

Hard separation of concerns: the humanness confidence score is never an input to mode selection — a human at 99.9% deserves the same answers as one at 90%. This is steering input, not command: the model and you keep full control, you can override the mode at any time. No emotional or medical diagnosis — not what it means, only what was observed.

— interact, then copy —

GPT · Without EDNA

Standard answer — the baseline.

Pick any question you care about (a real work question works best).
Ask it in your AI chat plainly, with no extra context.
Save the answer.

GPT + EDNA · With Guidance

Adapted answer — same question, steered context.

Interact with this demo page for 1–2 minutes so the engine sees your real dynamics (fast clicking, careful reading, corrections — whatever is truly yours).
Press Copy both (A/B protocol) above.
Paste it into a fresh chat and then ask the identical question.
Compare the two answers side by side — length, structure, initiative, tone.

Same question, different context steering — the effect is immediately experienceable. Everything the controller uses stays on this device; the copied text contains only the mode, style parameters and observed interaction features you see in the preview.