Global Consciousness Project experiment portal

[WanderingIshiki]

I’ve been working on a research platform to help analyze GCP data. This follows on from some discussions last year @ https://forum.mindmatterinteraction.net/t/gcp-2-0-coherence-algorithm/242/21
Treat it as a Work-In-Progress, alpha, beta, what-have-you. Many things need validating.

https://gcp.fp2.dev/

**
Experiment #4: EGG Analysis**

This is something I started back 6 years ago while in quarantine for 2 weeks in a hotel after having moved to Singapore but the project soon got put on the back burner due to other priorities. I wanted a GUI tool that would allow me to easily analyze any period in the 20+ years worth of GCP v1.0 data.

And we have…. http://gcpeggs.fp2.dev/

What it does (AI summary):

• Connects directly to GCP v1.0 data in Google BigQuery (1998-2025) - ingested using code here

• Runs Stouffer Z-score analysis across active eggs for selected window date & time & length

• Interactive time window selection (1 minute to 90 days)
  (I toyed with adding a “Donate me a Coffee / $ / TBC button for this feature: add “/?adv=true” to the end of your URL and you’ll get advanced mode: all glorious 27 years of Window time because the BigQuery processing can cost me some dollars:

  

• Real-time chi-squared deviation tracking

• Filters out broken eggs automatically (TODO: I know there’s a lot of work that can be done on data filtering but for now the biggest issue I found was EGGS reporting back 0-trial sums so I added a feature to exclude those by default, likely due to hardware issues. Seems to be a ‘2nd half of the dataset’ issue)

Cool features (AI summary):

• Web interface with cyberpunk styling (TODO: setup a different forum for design discussions )

• Local caching for faster queries

• Reproduces historical GCP findings, at least verified with 911. Would love to compare others.

• Compare real data vs random baselines

Roger Nelson’s original paper on analyzing the 911 World Trade Centre building collapses:

reproduced (and set as the default when you load the page)

And using one of the features I like the most, easily extended window time analysis, the above image at 30 days not just 4 hrs 10 mins.

It was when I took this screenshot I realized the red p = 0.05 parabolic curve probably isn’t being calculated right, but you can reference the y-axis for actual values.

Whilst not following the GCP’s actual “event selection method”, here’s a couple of exploratory ones I did:

2008 Lehman Bros finance crash

6 hr window:

30 day window:

2011 Japanese Fukushima tsunami & nuclear power plant explosion

4 hour window:

30 day window:

Experiment #6: Finance Correlations (Still Building)

This is the ambitious one - trying to correlate GCP Max[Z] anomalies with financial market movements. Based on Ulf Holmberg’s research from 2020-2022.

What I’m working on…

• Max[Z] extraction from GCP network logs

• Linear regression: r_(t+1) = α + β×Max[Z]t

• Real-time market data integration (SPY, IVV, VOO, etc.)

• Trading signal generation based on consciousness anomalies

• Performance tracking and backtesting

Current status:

• Real-time GCP data collection from global-mind.org

• Market data via Alpaca API

• Basic prediction model in place

• Still tuning the statistical thresholds

[Ulf_Holmberg]

Hi,

This looks really interesting. I’m currently also conducting a pre-registered (OSF) follow up to the 2023 (2024) publication (https://doi.org/10.1108/JES-11-2023-0663) and I’d be happy to discuss any thought you have on how Max[Z] covaries with market data. In summary: it seems like it’s affecting returns through the market sentiment channel.

Best,

Ulf

[WanderingIshiki]

Thanks Ulf.

What is the market sentiment channel?

[Ulf_Holmberg]

Hi,

By “market sentiment channel,” I’m referring to the idea that the observed relationship between Max[Z] and equity returns is driven by collective emotional or attentional states (sentiment).

In my 2024 study, I found that elevated Max[Z] values often covary with the VIX, commonly known as the market’s “fear index.” VIX reflects expected volatility in the S&P 500 based on options pricing, and it tends to spike during periods of market stress or increased uncertainty. As such, the hypothesis I’ve been exploring is that Max[Z] captures large shifts in collective human attention or emotional intensity. These shifts are then postulated to affect market sentiment, which in turn affects risk perception and herd behavior, leading to increased volatility and larger price swings.

So when I say “market sentiment channel,” I am referring to the idea that Max[Z] doesn’t directly cause price changes, but rather correlates with shifts in collective cognitive states that influence how markets behave.

Best,

Ulf

[WanderingIshiki]

Ah, I got you now, you’re just referring to the actual hypothesis you have of the correlation between GCP and market data. I thought the “channel” was something more specific within that context. Thanks for clarifying.

[WanderingIshiki]

Hi Ulf,

Sure it’d be nice to help you out on this research.

Have you tried any other stat methods other than Max[z]?

What securities/indexes did you find the least & most covariance with?

[WanderingIshiki]

All 27 years of GCP data. Looks like a step-wise function. Treat this as alpha-level exploratory. I’m sure there’s dastaset & code/calculation issues that need vetting and tuning.

Without p-curve

WIth p-curve

[nplonka]

@WanderingIshiki This is cool! It’s great that it is publicly available. We are currently working on making GCP 2.0 more publicly accessible through API or our site, so I could imagine a tool coming from that. We auto-calculate the Network Coherence (netvar; chi squared) on our servers so all one would need to do is take cumulative sum over the period.

Also, your step function is probably due to “bad eggs”. Roger Nelson has a list of days on which RNGs malfunctioned in GCP 1 so we typically remove those. Or I believe you can catch it by removing outliers beyond six sigma.

[WanderingIshiki]

Thanks @nplonka

Yeah I figured those big steps were due to a mixture of things like the number of eggs varying greatly over the years but most largely because of bad eggs (I’m sure there’s a rotten egg joke in there somewhere that Roger and his team have used ). Like I mentioned above, the only filtering I’ve done so far is egg values with 0-trial sums so it’s not that sound I know. I do remember some years ago seeing mention to that list of bad eggs you mention that Roger has.

Does GCP 2.0 / HeartMath have something similar? At least a form where egg owners can submit downtimes. I’ve had 2 GCP 2 eggs over the past 3-4 years (1 in Singapore, left it there and gave it to a friend, got my 2nd when I moved back to New Zealand). Every now and then I’ve had to have “downtime” due to things like moving house, and most recently there was a scheduled power outage (which also took the forum offline). Having a way to record those events for future analysis “bad egg” filtering would be invaluable.

[nplonka]

Good point. We do naturally track downtime in the database, but we have on our roadmap to automate this and alert someone if a device is down too long. More critical is the weird values (“rotten eggs”) that can mess with network metrics (vs missing data from a device here or there is no big deal), but this seems to be much less of an issue so far in GCP 2.0, with more reliable tech and a rigorous device testing before shipping it out.

[WanderingIshiki]

I started working on some backtesting using Nautilus Trader with both GCP 1 and GCP 2 data from Feb to July 2024 against some index ETFs. It’s very preliminary and so far the results seem to be in the negative but due to other priorities and commitments I’m not able to allocate much more time to this at the moment.

gcp2-playbox/experiments/7-financial-backtesting at main · vfp2/gcp2-playbox

[MichaelD]

Hi Nachum,

Somewhat unrelated, are you able to share the preprint that is in press with Explore that you referred to in the recent webinar? Great work! Best, Michael.

[nplonka]

Sorry, I haven’t published a preprint, but happy to share when it is published