10,000 Members, 1 Tight Script: Santana Mine Supporters

BySarah Fraser
Smartphone displaying a Facebook group page, used as the article featured image

The Facebook group Santana Mine Supporters appeared to show broad grassroots backing for a proposed Central Otago gold mine. But the reporting dataset behind this story, a review of 9,890 member tiles, 208 posts, 50 comments, and 327 profile samples, captured from the public Facebook surface, found a much stranger pattern: membership arrived in huge batches near launch, the admin team includes people who do not appear local to the affected area, and one admin lists work at a New Zealand virtual admin business.

The group matters because Santana Minerals is seeking approvals tied to its Bendigo-Ophir gold project in New Zealand, and a Facebook constituency of nearly 10,000 people can look politically useful. What the public evidence shows, from that scrape and from New Zealand business records checked for the admin connection, is not proof of intentional deception. It is, however, a tidy case study in how a Facebook astroturf group can look convincing from the outside while leaving obvious operational fingerprints on the surface.

How the Santana Mine Supporters group was built

The review captured 9,890 of roughly 9,908 members, about 99.8% coverage, along with Facebook’s relative join-date labels for each account. The pattern is the opposite of what you’d expect from a support group that grew gradually around a live local issue.

Here is the membership distribution from the scrape:

Facebook join label Count % of group
Joined about 3 months ago 4,893 49.5%
Joined about 2 months ago 2,269 22.9%
Joined about a month ago 972 9.8%
Joined about 2 weeks ago 587 5.9%
Joined this week 304 3.1%
Joined this month, named day 597 6.0%
Joined within the last 24 hours 71 0.7%
Other / unknown 197 2.0%

Nearly three quarters of the group, 72.4%, joined in the first two large cohorts. If the group were mostly organic, you’d expect a long tail: older members, gradual accumulation, and a distribution that reflects ongoing local attention. The scrape instead shows a launch spike first, then a much smaller trickle.

That does not by itself prove fake members. A campaign can absolutely drive a big early signup burst. But this particular burst is so concentrated that it suggests deliberate bulk onboarding, not a constituency slowly finding one another.

Facebook only exposes relative join labels, not exact timestamps, so the method has limits. You cannot reconstruct the exact day-by-day curve, and “about 3 months ago” compresses a range of join times into one bucket. But for cohort analysis, that granularity is still enough. If half the group lands in one oldest visible bucket and another quarter lands in the next one, you do not need exact timestamps to see that Santana Mine Supporters was assembled in a few large waves rather than accumulated steadily.

The group’s timing also matters. There are no members older than about 90 days in a public campaign that predates the group itself. On the scrape evidence, Santana Mine Supporters looks less like a community that formed around a mine debate and more like a communications asset that was stood up for this phase of the permit fight.

Why the admin team looks outsourced

The admin roster visible on Facebook is small: 5 admins, 0 moderators. For a supposedly broad-based local support group of this size, that means control is concentrated in a handful of accounts.

The scrape identified the following publicly visible admin details:

Admin Visible location Notable detail
[redacted] Cromwell, NZ Local to the area
[redacted] Wanaka, NZ Local to the area
[redacted] Taupo, NZ Roughly 1,000 km from the mine area; listed as “Admin CEO at Devon Street Property Limited”; produced 10.1% of all posts
[redacted] No location shown Lists work at The Admin Superstar; joined about 2 weeks before becoming admin
[redacted] No location shown Lists work as “Office Admin”; joined about a month ago

Two admins look local. Three do not obviously look like local supporters. The strongest signal is [redacted], because the claim here is not just based on a suggestive job title. The Admin Superstar’s public business presence describes it as a virtual assistant and business-support operation, and the New Zealand business record checked during reporting shows it as a registered New Zealand business. In other words: the “outsourced admin” reading is tied to a real public business record, not a vague Facebook self-description.

 

There are innocent explanations. A campaign can hire admin help to handle volume. Someone can volunteer while also working in outsourced admin. But the visible configuration here points toward coordinated communications work. The group does not just have active organizers; on the public record, it has the staffing pattern of a small public-facing PR operation.

That changes what the member count means. A local group with 9,000 people and local admins suggests one thing. A group with bulk-join cohorts and an admin bench that includes outsourced admin labor suggests something else: constituency as presentation.



The profile fingerprints that suggest sockpuppets

 

A sockpuppet account is a profile used to create the appearance of independent support. You usually do not catch it with one clue. You catch it with stacks of weak clues that line up.

Here are the main signals in the sample recorded from public profile surfaces:

Signal Count % of sample
No workplace listed 325 99.4%
No school/education listed 324 99.1%
No historical year mentioned anywhere on profile 231 70.6%
No activity on own profile 135 41.3%
No location at all 134 41.0%
Profile is privacy-locked 112 34.3%
No cover photo 50 15.3%
Only year visible is 2026 7 2.1%

None of these, alone, proves anything. Plenty of real people barely use Facebook, list nothing, and lock their profiles down. But the pattern here is cumulative.

The missing piece in the earlier draft was the threshold. The “about 1 in 5” figure was not derived from any one signal above. It came from counting profiles that matched a multi-signal shell pattern: accounts with several of the high-risk traits at once, for example no workplace, no school, no location, no visible profile activity, and privacy locking or similarly minimal surface completeness. In the reporting dataset, the cutoff was a pre-defined cluster threshold rather than a single red flag: accounts had to stack multiple shell-like signals before being counted in the sockpuppet-style bucket.

That is a better method than cherry-picking one weird field, but it is still circumstantial. A privacy-conscious real user and a cheaply prepared shell can look similar from the outside. The point is not that every sparse account is fake. The point is that Santana Mine Supporters contains a meaningful fraction of accounts whose visible profile surfaces are sparse in the same way, at the same time, inside the same support group.

The split between active and silent members matters too. Silent members were more likely to look like empty shells, while active posters and commenters were somewhat more complete on average. That is what you would expect if some accounts existed mainly to inflate apparent support while a smaller subset handled visible engagement.

Locality and attribution limits

The public evidence establishes four things.

First, Santana Mine Supporters did not grow in the pattern you’d expect from a long-running organic community. The bulk-join cohorts are visible and quantifiable in the member scrape.

Second, the group is run by an admin team that does not read as purely local. The presence of a recent joiner tied to a virtual admin business is especially hard to square with the idea that this is just neighbors gathering themselves.

Third, the member base includes a meaningful share of low-completeness profile shells that fit the usual sockpuppet profile. Not all of them are fake. Enough of them look fake that the group’s headline size stops being trustworthy as a measure of local public sentiment.

Fourth, only a small minority of members are demonstrably local to the affected area. The 6% figure in the reporting dataset comes from conservative classification using only explicit public signals: profiles that listed a local place name in or near the mine-affected region, or otherwise exposed clear location fields linking them to that area. Members with no visible location, ambiguous locations, or only broader New Zealand identifiers were not counted as local. In other words, the 6% is not “everyone who might be local.” It is the share that could be verified as local from public profile data at scrape time.

That method has obvious limits. Many real people do not list a location, and Facebook profile surfaces vary by privacy setting. So the locality figure is best read as a floor on visible local membership, not a complete census. But that floor is still revealing: if a nearly 10,000-member support group can publicly verify only a thin slice of local ties, its headline size is doing more rhetorical work than evidentiary work.

What the evidence does not prove is intent. A stronger attribution case would require evidence such as:
– internal messages or campaign instructions
– payment records linking admin services to group operations
– creation-time metadata from Facebook
– repeated reuse of the same accounts across multiple advocacy groups
– IP, device, or login overlap that only the platform could see

That is the line between forensic public evidence and a completed attribution case. On the public evidence alone, the most defensible conclusion is that Santana Mine Supporters shows the visible fingerprints of a manufactured support operation. The only thing missing is the invoice.

Key Takeaways

  • Santana Mine Supporters grew in two huge early cohorts, with 72.4% of members joining in the first two months, not through a long organic buildup.
  • The admin team includes non-local accounts and one admin who lists work at The Admin Superstar, a New Zealand outsourced admin business.
  • A 327-profile sample found many low-completeness accounts, with roughly 1 in 5 showing a sockpuppet-style fingerprint based on a multi-signal threshold, not any single trait.
  • The visible-locality measure was conservative: only profiles with explicit local identifiers were counted, and that produced a figure of about 6% demonstrably local members.
  • The evidence is strongly circumstantial, not definitive proof of deception; proving intent would require platform or financial records beyond the public Facebook surface.

Further Reading