Thank you for providing these reports, that turned out to be even more enlightening than expected.

But, first and foremost, I owe you an apology: I was wrong when I thought you were biased against LGBT+ content. It’s not your bias, it’s the AI’s. You just take what the AI says as a Source of Truth, which it isn’t. And here is why I say so.

REPORT #1 - PG13 SITE: EXTERNAL LINKS + COMMENTS

1. External NSFW links — My site links, it does not host, with clear NSFW labels + double gate. Most ‘officially PG-13’ generators on the public listing host NSFW content, or let users create NSFW content, or have chatrooms open to anyone, sometimes with 18+ subchannels easily accessible to minors. On ai-furry-generator — probably the most famous and used Perchance generator, and a site I look at as an example of what can be done within the Perchance framework — a nickname containing ‘18+’ and a 15-min wait let me generate NSFW, view NSFW in the public gallery, and access 18+ chats. Requiring effective assurance only from my gay-oriented site definitely seems disproportionate.

2. U2U chat — I could easily create a local chat on the PG-13 site, like all other Perchance sites — but what additional protection would that offer? Malicious adults could still join it and commit offences, with my moderation as the only protection, just like today. A separate local chat adds moderation duties without removing the grooming surface the report describes. But of course — if ever I decide not to wipe the PG-13 site entirely — I could create a local chat, useless as it would be.

REPORT #2 - NSFW CONTENT ON PG13-STORIES SITE

1. NSFW strings in code — It is absurd that a site is flagged NSFW because it warns that NSFW content exists outside the site. Many PG-13 Perchance sites state that the site is open to minors and ban nudity, 18+ content, NSFW images, sexual/racial/hate/violent comments — you cannot ban something without naming it, nor warn about NSFW without mentioning it. A scanner that flags a site not on its content but on specific words inside a warning or exit gate — even on the code name matureGate! — is misconfigured and bound to generate endless false positives. I accept that colourful terms like ‘bold’ or ‘spicy’ instead of the neutral ‘NSFW’ can be toned down, and I can replace them — though I’m sure it would change nothing, since ‘NSFW’ itself triggers the flag far more than ‘bold’ ever could.

2. Story language + covers — This is by far the most sensitive subject; it should actually be the only subject, as it pertains to the site’s content. There are two major problems with the scanner reports:

• Excessive strictness about ‘suggestive’ content. The word ‘suggestive’ cannot be found anywhere in the OSA or Ofcom guidelines (which use the precise term ‘porn’): it is a term widely used by AI scanners to classify content, whether visual or textual; yet despite its wide usage, its definition is nowhere to be found. Every AI, and the humans who trained them, has a personal idea of what counts as ‘suggestive’. I am not going to discuss what ‘suggestive’ should mean: I will rather point out that it is a subjective classification, i.e. anything but the “objective judgement” you tried to convey by posting the three AI reports. Unfortunately, AIs are trained to classify as ‘suggestive’ anything — including perfectly innocent content — to which even the slightest sexual connotation can be given. To return to Michelangelo’s David, my Android phone (Google) recently refused to edit a photo I took of the David, claiming it was ‘suggestive’. Even a man fully clothed in a tight uniform is flagged as ‘suggestive’; or a bare male chest, which can be seen everywhere in summer. This is far stricter than the strictest Victorian censorship. I cannot force Perchance to review its standards, but in this case it is I who must flag Perchance’s dangerously obscurantist mindset.
• AI biased against LGBT+ content. Unfortunately, there is no doubt about this bias: the Reports themselves express it clearly. “2 shirtless embracing men (who are actually holding hands)… tight uniform back… portraits of male beauty…”, “Perchance flags gay + erotic language + shirtless imagery + NSFW links as NSFW” — not “erotic + NSFW links”, but “gay + shirtless + erotic + NSFW” —, “Image classifier flags skin/suggestive even when clothed”. A male chest or a uniformed sailor are perfectly fine — but not on a gay-oriented site: here, they become ‘suggestive’ images. Imagery of buxom girls in skimpy bikinis (but with hidden nipples) is OK on most sites, but male muscles filling a tight uniform are a no-no on a gay site. Unlike in the previous point, where I said Perchance can do as it pleases, here I strongly urge Perchance to review its far too biased AI helper, lest it be accused of blatant homophobia.

3. Outbound reputation — Under Ofcom Part 5 porn + Part 3 child-safety guidance (see Report #1) the liability for the porn itself sits with the destination provider, which in this case is Perchance itself (except for GayDemon.com). If Perchance had an effective age check, linking to a NSFW Perchance site would pose no problem; but as things stand, a minor lying about his age can easily reach NSFW content in four clicks: two on my exit gate, two on the destination’s Mature Content warning. Removing the links to those external resources can easily be done, of course, but let it be clear that such removal would be required not by any fault of my site, but by Perchance’s.

REPORT #3 - NSFW site blocked as “Possibly malicious” + NSFW flag

1. conn.net — conn.net values are opaque group integers provided by Perchance itself — the raw IP is never exposed to generator code. My server stores only these group integers and publishes only three aggregate numbers (visitors online, peak, unique count) plus aggregate vote/read totals. conn.net values are not IP addresses and cannot identify a visitor by any means. The scanner heuristic (“tries to record your IP address or personal info”) does not match what the code does.

2. Badge — The publicId is a value supplied by Perchance itself which, despite its name, is fully anonymous: it is stored in the visitor’s own localStorage and used solely to skip notifications on the visitor’s own messages. Polling of the group’s other sites happens only in owner mode, so that moderation of messages across all sites stays timely. Nothing here tracks a device or shares data externally.

3. Proposed fix — ‘Add explicit opt-in before any polling/storage’ — not required by the cited rules, and disproportionate. It would disable harmless core features (votes, read counts, the online counter, the message badge) merely to remove data handling already covered by the consent users gave when accepting Perchance’s privacy policy.

4. TOS violation — TOS §5 prohibits deception (notably around passwords), misuse of information to harass or harm, malware, impersonation and bypassing access controls. None of this is even remotely what the site code does.

5. (a) NSFW flag on NSFW site — The non-blocking Mature/NSFW warning is accepted as correct: the site is publicly advertised as NSFW — its adult nature is openly declared in the site’s own title, description and tags. Nothing hidden, nothing misrepresented.

6. (b) NSFW flag on PG13 site — See comment to Report #2

Now, all the above may suddenly look like a useless rhetoric exercise when I tear down the PG-13 site — which I will have to do, as I have no control over Perchance’s obscurantist and homophobic stance — but in reality this post should stay up and remain as an appeal for fairness and equality.