ANOMALY NET MIND-RNG EXPERIMENT: PRE-REGISTERED PROTOCOL, VERSION 1
=====================================================================
Registered: 2026-10-15 (UTC), before any experimental data was collected.
Publisher:  Anomaly Net, https://wolfwright.com/anomaly/
Proof:      The SHA-256 of this exact file was timestamped in the Bitcoin
            blockchain with OpenTimestamps (protocol-v1.txt.ots, published
            beside this file). Anyone can verify at https://opentimestamps.org
            that this text existed, unchanged, before the experiment began.
            This file will never be edited. Any change becomes a new version
            with its own timestamp, and results stay tied to version 1.


1. QUESTION
-----------
When many people focus on one shared signal at the same moment, does the
output of a physical random number generator stop behaving like chance?

This is the question behind the Global Consciousness Project (GCP, Princeton,
1998 onward), which reported small but cumulative deviations in a worldwide
network of hardware random number generators during major world events.
Critics argue that the GCP chose its event windows and statistics with too
much freedom after looking at data. This experiment fixes every choice in
advance, uses public random sources that anyone can re-download, and
publishes the results whatever they are, including a null result.


2. THE INTERVENTION
-------------------
The Anomaly Net Beacon: a daily global session at 01:11 UTC lasting 33
minutes, in which participants worldwide play and watch the same
synchronized signal and focus on the same intention (stated on the page:
"we are curious, we mean no harm, and we want to learn").

Session window:  01:11:00 to 01:43:59 UTC, every day.
Control window:  13:11:00 to 13:43:59 UTC, every day (same length, 12 hours
                 later, nobody is asked to do anything).


3. DATA SOURCES
---------------
PRIMARY: the NIST Randomness Beacon 2.0 (chain 2), https://beacon.nist.gov
  - Publishes one 512-bit value per minute ("localRandomValue"), generated
    from hardware entropy sources, signed by NIST and archived permanently.
  - The pulses used are the 33 pulses timestamped 01:11 to 01:43 UTC (session)
    and 13:11 to 13:43 UTC (control), fetched from
    https://beacon.nist.gov/beacon/2.0/pulse/time/<unix-ms>
  - A pulse counts only if the returned timeStamp equals the requested minute
    exactly. Missing or mismatched pulses are excluded and reported.
  - A window counts only if at least 30 of its 33 pulses are valid.
  Note: NIST's value is a cryptographic hash of its entropy sources, so a
  bias in the physical noise itself would not survive into the output. A
  deviation here would require an effect on the outcome, which is the GCP
  hypothesis. That limit is stated now so it cannot be used later to explain
  away either result.

SECONDARY: the ANU Quantum Random Number Generator, https://qrng.anu.edu.au
  - Random numbers measured from quantum vacuum fluctuations.
  - Recorded live by the Anomaly Net server once per minute during the same
    session and control windows (1,024 16-bit numbers = 16,384 bits per
    sample). The public API allows one request per minute, so recording is
    live-only and the raw samples are archived by Anomaly Net and published.
  - This stream starts on the first session after the recorder is switched
    on and is secondary because it cannot be independently re-fetched.


4. MEASUREMENT
--------------
For each 512-bit NIST pulse p:
    k_p = number of 1 bits in localRandomValue (hex decoded)
    z_p = (k_p - 256) / sqrt(128)
Under chance, z_p is approximately standard normal (mean 0, variance 1).

For each 16,384-bit ANU sample the same formula is used with n = 16,384:
    z = (k - n/2) / sqrt(n/4)


5. HYPOTHESIS AND PRIMARY TEST (decided now, used once at the end)
-----------------------------------------------------------------
H1: During session windows the random output is more variable than chance
    (it strays further from 50/50, in either direction).

Primary statistic, over all valid session pulses p = 1..N:
    chi2  = sum of z_p^2          (chance: chi-square with N degrees of freedom)
    Z_primary = (chi2 - N) / sqrt(2N)

This is the per-sample variance measure the GCP uses. It is direction-free,
so effects pushing toward more 1s on some days and more 0s on others add up
instead of cancelling.

Endpoint: after 365 session windows have been scored (planned last session:
          2027-10-15 01:11 UTC; later if windows are lost to outages).
          About 12,000 pulses / 6.2 million bits.
Decision: H1 is supported if Z_primary >= 3.09 (one-tailed p < 0.001)
          AND the same statistic over the control windows lies within
          -2.58 < Z_control < 2.58 (the source behaves normally when nobody
          is focusing). If the control fails, the result is declared invalid
          (a source problem), whatever the session score.
          Otherwise H1 is not supported.
Power:    At the endpoint this detects a ~4% increase in per-pulse variance.

No early stopping. Running results are shown live for transparency, but
the decision is made only at the endpoint. Interim peaks or dips mean
nothing on their own: a random walk wanders.


6. SECONDARY ANALYSES (exploratory, reported but not decisive)
--------------------------------------------------------------
S1  Mean shift: Stouffer Z = sum(z_p) / sqrt(N), two-tailed.
S2  The primary statistic computed on the ANU quantum stream.
S3  Dose-response: Spearman correlation between each session's variance
    score, (sum z_p^2 - n) / sqrt(2n), and that session's number of
    participants as logged by the Beacon.
S4  Sightings: rate of community sighting reports in the 3 hours after each
    session vs. all other hours (already shown on the Beacon page).
With four secondary tests, a single secondary "hit" at p < 0.05 is expected
by chance about one time in five, and will be described that way.


7. WHAT A RESULT WOULD AND WOULD NOT MEAN
-----------------------------------------
A supported H1 would mean: random output during sessions was less random
than chance, at odds of about 1 in 1,000 or less, with a clean control. It
would be a reason for independent groups to replicate with their own
generators. It would not by itself prove consciousness, contact, or a new
force.
A null result would mean: no detectable effect at this sensitivity. It would
be published with the same prominence.


8. VERIFY IT YOURSELF
---------------------
Every pulse used is listed (index, time, bit count, z) in the CSV download
on the Beacon page. To recompute any pulse:

    import json, urllib.request, math
    def z_for(ms):
        u = f"https://beacon.nist.gov/beacon/2.0/pulse/time/{ms}"
        p = json.load(urllib.request.urlopen(u))["pulse"]
        k = bin(int(p["localRandomValue"], 16)).count("1")
        return p["timeStamp"], (k - 256) / math.sqrt(128)
    # example: the 01:11 UTC pulse on 2026-10-16
    print(z_for(1792113060000))

END OF PROTOCOL VERSION 1