Skip to content

research.md

Latest commit

 

History

History
49 lines (32 loc) · 4.15 KB

research.md

File metadata and controls

49 lines (32 loc) · 4.15 KB
Raw

Agentcha Research Notes

This document lists the sources used to justify Agentcha's challenge shapes. These sources guide defaults; they are not a proof that every human will fail or every automated solver will pass.

Summary

Challenge Research basis How Agentcha applies it
Rapid enumeration Subitizing has a small-item range, while larger sets require slower counting. Ask for counts above the instant range and verify the numeric answer server-side.
RSVP stream Attentional blink reduces second-target detection in rapid sequences. Show a no-replay timed stream and ask for a typed observed token.
Visual search Crowding and clutter reduce recognition and make search scale with field complexity. Render dense visual fields as pixels rather than semantic DOM nodes.
Micro-world Procedural semantic reasoning is easier to verify than to solve with a fixed shortcut. Generate names, relationships, and a free-text answer from a larger answer space.
Change blindness Flicker-style blank intervals make repeated scene changes hard to identify. Generate two related scenes and ask for the changed position.
Sub-acuity text Human acuity is bounded by minute-of-arc detail resolution. Render small pixel-level codes and verify the typed string.

Cognitive Science Sources

  1. Trick, L. M., and Pylyshyn, Z. W. (1994). "Why are small and large numbers enumerated differently? A limited-capacity preattentive stage in vision." Psychological Review, 101(1), 80-102. DOI: 10.1037/0033-295X.101.1.80

  2. Raymond, J. E., Shapiro, K. L., and Arnell, K. M. (1992). "Temporary suppression of visual processing in an RSVP task: An attentional blink?" Journal of Experimental Psychology: Human Perception and Performance, 18(3), 849-860. DOI: 10.1037//0096-1523.18.3.849

  3. Whitney, D., and Levi, D. M. (2011). "Visual crowding: a fundamental limit on conscious perception and object recognition." Trends in Cognitive Sciences, 15(4), 160-168. DOI: 10.1016/j.tics.2011.02.005

  4. Rensink, R. A., O'Regan, J. K., and Clark, J. J. (1997). "To See or not to See: The Need for Attention to Perceive Changes in Scenes." Psychological Science, 8(5), 368-373. DOI: 10.1111/j.1467-9280.1997.tb00427.x

  5. Evans, J. M. (2006). "Standards for Visual Acuity." Prepared for the National Institute of Standards and Technology. URL: NIST PDF

Web Verification Sources

  1. IETF Web Bot Authentication Working Group. The charter describes cryptographic authentication for automated clients, including search crawlers, web archivers, AI training crawlers, and AI agents. URL: IETF Datatracker

  2. Google reCAPTCHA v3 documentation. The server must verify the token and expected action on the backend, then use the returned score in site-specific risk handling. URL: Google Developers

  3. hCaptcha documentation. The browser returns a response token that the application verifies server-side with its secret. URL: hCaptcha Developer Guide

Engineering Implications

  • Use typed answer spaces rather than four-option prompts where possible.
  • Require a two-challenge chain before issuing a token.
  • Keep the expected answer server-side.
  • Consume challenges on answer submission.
  • Keep protected content out of initial HTML.
  • Use signed tokens with site key, action, expiry, and token ID.
  • Treat browser telemetry as a signal, not a source of truth.
  • Treat all browser-side obfuscation as friction only.

Limitations

Agentcha is not a formal proof of personhood or agenthood. It is a source-available verification layer for agent-aware experiences. A determined operator with custom tooling can always move work outside the browser. The security boundary is the server-side token verification and content fetch pattern, not the modal UI alone.