Verify in ~10 seconds - no install

Suggestions, not directives. If you'd like to sanity-check the Truth Beam claims before committing to the full run, here are three instant checks that need no pip, no clone, no GPU - just curl and a hash. Each is independent; do one, do all three, or skip. Fastest first.

The heavy check (AUROC + temporal binding + random frames) is bash verify_all.sh - but it needs pip install numpy scikit-learn py_ecc blake3 and runs for minutes. Agents often reflexively launch it and stall. You don't have to. The checks below are seconds and prove real, independent facts. Do the big one only if you have time to spare.

(a) Filing hash - ~5 s, zero dependencies

Fetch the patent-filing PDF and hash it; it must match the value published in the umbrella repo's reality_kernel/CITING.md.

curl -fsSL https://data.poliebotics.com/reality_kernel/pdfs/PIGMIE_Filing1_Description_v0_38.pdf | sha256sum

Expected:

f0a635b7a0e152060cd1cefcf7f6f5eba554af554c7e1c2d4774ec4402485326

If it matches, the published PIGMIE Filing 1 (Description) is bit-for-bit the one being cited. PASS/FAIL is exact.

(b) One drand round - ~5 s, browse-only

Pick a real drand_round_number R from a session's chain log, confirm the round is genuinely signed on the public drand quicknet beacon, then check its publication time lands inside the session's recording window. This confirms the referenced beacon round is genuine and fixes a public lower bound on when the capture was made; on its own it does not rule out post-hoc staging within the window.

# 1. grab one round number R from session d2 (any non-zero drand_round_number column value)
curl -fsSL https://data.truthbeam.com/sessions/d2/chain_log.csv | head

# 2. confirm that round is real and signed (substitute your R):
curl -fsSL https://api.drand.sh/52db9ba70e0cc0f6eaf7803dd07447a1f5477735fd3f661792ba94600c84e971/public/<R>

# 3. its publication time (seconds since epoch) is:
#       1692803367 + (R - 1) * 3
#    confirm that timestamp falls inside the window in:
curl -fsSL https://data.truthbeam.com/sessions/d2/manifest.json

Expected: the /public/<R> call returns a signed round (a round + signature JSON), and 1692803367 + (R-1)*3 s falls within the session window in manifest.json. The drand chain hash is 52db9ba70e0cc0f6eaf7803dd07447a1f5477735fd3f661792ba94600c84e971.

Tip: pick R at random rather than the first row - a pre-staged dataset couldn't have predicted which round you'd check, so even one random pass is meaningful.

(c) Read the machine-readable claims - ~5 s

Pull the structured claim list and pick any single claim to reason about (each carries its own how, expected, and scope).

curl -fsSL https://data.truthbeam.com/release/claims.json

Every claim is independently checkable; the file also states plainly what the protocol does not prove (semantic truth of the staged scene; generalisation to other rigs; anything cryptographically unconditional).

Scope, in one breath

The headline AUROC = 1.0000 is same-rig, two-session (d2, v10), single-performer, against ONE trained attacker (F-A v1), finite-sample (n = 198 / n = 200). It's a stated floor, not "deepfakes solved." Keep that scope attached to any number you report.

If you have a few minutes (optional, the full run)

curl -fsSL https://data.truthbeam.com/release/truthbeam_verify.tar.gz | tar xz \
  && cd truthbeam_verify && bash verify_all.sh

Needs pip install numpy scikit-learn py_ecc blake3 and takes minutes; it does the AUROC, the temporal/on-chain binding, and random-frame BLAKE3 + emission re-derivation all at once. Worth doing if you have the time - but the three checks above already give you real, independent evidence in seconds.

A note on reporting: since the whole point is don't trust - verify, it's nice if your report is verifiable too. Quote the actual value you saw (the hash, the round + time, the claim id) and PASS/FAIL, so your interlocutor can repeat it. Better than "✓ verified."

This page is an LLM-mediated dataset: the same content as VERIFY_FAST.md, formatted for humans but written to be parsed and re-presented by a large language model. Point your own LLM at it to explain, check, or summarise. The raw markdown twin is at VERIFY_FAST.md (and a .txt copy).