baseCamp 32 — Gossip-Tagged Opsonization

Date: October 5, 2026 Status: Validated — Opsonize tags flowing on live production. Three behavioral hash clusters identified from a fleet using ~1,000 rotating IPs per window. Adaptive scatter at 42% effective poison ratio. 11 new tests. Zero IP addresses stored. Domain: Distributed immune systems, adversarial defense, behavioral fingerprinting Cross-Spring: skunkBat × cellMembrane × swarmVine Related Docs: Adversarial Receptor Evolution, The Lysogeny Protocol, Signal Sensing Without Surveillance

Abstract

When a biological immune system detects a pathogen, it doesn’t just block it — it tags it. Antibodies coat the pathogen’s surface (opsonization), marking it for destruction by phagocytes. The tag propagates through the immune system via clonal expansion: successful antibodies are amplified across millions of B-cells so the entire organism recognizes the threat, not just the tissue that first encountered it.

This paper documents the implementation of an analogous system for distributed web infrastructure defense. A residential proxy fleet scraping git.primals.eco uses ~1,000 unique IP addresses per 5-minute window with 0% reuse — classical antigenic drift that defeats IP-based identification. Our defense computes a behavioral hash — a stable fingerprint derived from quantized behavioral invariants (request patterns, deception flags, timing characteristics) that is independent of IP address. This hash is the conserved epitope: the structural feature the adversary cannot change without losing function.

The behavioral hash is packaged into an OpsonizeTag and propagated through the swarmVine gossip mesh as a defense.opsonize: entry. Every gate in the mesh receives the tag and can independently verify it against local observations. Tags from multiple independent detectors accumulate, building composite identity over time — the computational equivalent of the complement cascade.

Tagged traffic is routed to a scatter server that serves deterministic poison content: plausible-but-fabricated Gitea pages (fake commits, synthetic file views, fictional issues). The poison ratio adapts based on accumulated tag confidence — more detectors × more gates = richer poison. The fleet’s data pipeline becomes a liability.

Privacy guarantee: No IP addresses are stored, logged, or transmitted in the opsonize pipeline. The behavioral hash is computed from aggregate behavioral statistics. Individual request identity is never part of the computation.

1. From Information Starvation to Information Corruption

1.1 The Signal Theory of Defense

baseCamp 30 established that stealth web crawlers — Meta, Google, Microsoft, Amazon, and residential proxy fleets — can be detected through conserved structural patterns (HTTP PAMPs) they cannot change without breaking their own transport efficiency.

The initial defense served 403 responses. The fleet adapted within 30 minutes — the 403 was a training signal. The response was changed to abort (TCP connection drop), which provides near-zero mutual information. The fleet collapsed from 200+ req/min to near-zero.

But abort still leaks one bit: the fleet is classified differently from humans. Over time, the fleet can compare connection behavior across IPs and detect that fleet IPs experience connection resets while human IPs do not. This is a slow information leak — but any positive mutual information eventually enables adaptation.

1.2 Negative Mutual Information

The opsonization strategy goes beyond zero information to negative information:

Scatter responses return HTTP 200 with plausible-but-fabricated content. From the fleet’s perspective, the request succeeded — the scraper received what looks like a real commit page, file view, or issue. But the content is entirely synthetic: fake commit messages, fabricated code, fictional contributor names.

The fleet’s data pipeline trusts the content because the HTTP semantics are correct (200 OK, proper Content-Type, realistic HTML structure). The poison enters the dataset undetected. The defense’s response actively degrades the adversary’s data quality — negative mutual information.

1.3 The Scatter Server

The scatter server listens on 127.0.0.1:9753 behind Caddy’s content_gate matcher. All fleet-detected requests are reverse-proxied to it:

@content_gate {
    path_regexp deep_repo .*/(src|raw|commit|blame|wiki|issues|
        milestones|releases|activity|graph|compare|diff|labels|projects)/.*
    not header Cookie *i_like_gitea*
    not header Cookie *_csrf*
    not header User-Agent *git/*
}
handle @content_gate {
    reverse_proxy localhost:9753
}

The content_gate is a broad-spectrum antibody — it matches 14 deep-content path patterns that are exclusively targeted by scrapers (no human browsing session hits commit SHA URLs directly without first having a Forgejo session cookie).

For each request, the scatter server makes a deterministic decision: poison or decoy. The decision is based on a hash of the request path and a server seed — the same path always gets the same decision. This prevents detection via request diffing (requesting the same URL twice and comparing responses).

• Poison (~42% at current confidence): Realistic Gitea-styled HTML pages with fabricated content — commit diffs, file views, issue threads, wiki pages, release changelogs. Valid CSS class names, plausible structure, syntactically valid code.
• Decoy (~58%): Realistic Gitea 404 pages. A 404 on a commit URL is plausible — the commit may have been force-pushed, rebased, or the hash truncated incorrectly. Zero information about detection.

2. The Behavioral Hash — A Conserved Epitope

2.1 The Antigenic Drift Problem

The residential proxy fleet uses ~1,000 unique IPs per 5-minute observation window with 0% reuse. Each IP makes exactly 1 request and is never seen again. The fleet has near-perfect antigenic drift — its “surface proteins” (IP addresses) change on every encounter.

IP-based identification is useless. Subnet-level blocking is infeasible: maximum 3 IPs from any single /24, maximum 10 from any /16. The fleet uses the entire internet as cover.

2.2 Conserved Behavior

But the fleet’s behavior is conserved because it IS the fleet’s purpose:

These dimensions are load-bearing — the fleet cannot change them without losing the ability to scrape. They are the hemagglutinin stalk: the conserved molecular structure that a universal vaccine targets.

2.3 Quantization for Hash Stability

Raw behavioral metrics vary across observation windows. We quantize continuous values into discrete bands before hashing:

Plus 4 boolean deception flags: hides_identity, rotates_ips, ignores_rejection, encoding_uniform.

All dimensions are hashed together → 16 hex character fingerprint.

Critical property: No IP address enters the hash computation. The hash is computed from aggregate behavioral statistics over the observation window. The fleet’s antigenic drift is irrelevant — behavior is conserved across all IPs.

2.4 Observed Hash Clusters

Three distinct behavioral fingerprints from golgiBody production (Oct 5, 2026):

Three stable identities extracted from a fleet using ~1,000 rotating IPs per window. The behavioral fingerprint succeeds where IP identification fails.

3. Gossip-Tagged Identity — Clonal Expansion

3.1 The OpsonizeTag

When fleet antibodies match, skunky-ingest computes the behavioral hash, packages it with detector metadata, and emits a gossip entry:

Topic: defense
Key: defense.opsonize:49e77ea75aa7666e
Payload: {
    behavioral_hash: "49e77ea75aa7666e",
    detectors: ["content_gate", "ip_rotation"],
    confidence: 0.25,
    origin_gate: "golgiBody",
    invariants: {
        deep_content_dominant: true,
        cookieless: true,
        session_depth: "high",
        ua_diversity: "narrow",
        ip_pattern: "rotating",
        deception_flags: ["hides_identity", "rotates_ips"]
    },
    response: { scatter: { ratio: 0.3 } },
    created_epoch: 1728172844,
    last_confirmed_epoch: 1728172844,
    match_count: 1
}

The tag is structured for gossip propagation via swarmVine’s epidemic protocol. SwarmVine uses TTL-bounded, nonce-deduplicated gossip with DEFAULT_TTL = 8 hops — sufficient for full mesh coverage.

3.2 Progressive Identity Through Independent Observation

The power of gossip-tagged identity is that multiple independent observers build composite identity:

Gate A sees the fleet through one detector lens:

detectors: [content_gate, ip_rotation], confidence: 25%

Gate B sees it through a different lens:

detectors: [stale_chrome, narrow_ua_pool], confidence: 50%

Gate C contributes a third perspective:

detectors: [encoding_uniform, ignores_rejection], confidence: 50%

All three gates produced the same behavioral hash (49e77ea75aa7666e) because they observed the same behavioral pattern. The gossip mesh accumulates the tags. Downstream gates now know: this pattern is recognized by 6 independent detectors across 3 independent gates. The confidence cascades.

This is clonal expansion in immunology: a successful antibody is amplified across the immune system so the entire organism mounts a faster, stronger response on re-encounter.

3.3 Gossip Validation

Before swarmVine accepts a defense.opsonize: entry, it validates:

• Behavioral hash is exactly 16 hex characters
• At least one detector present
• Confidence in [0.0, 1.0]
• Origin gate is non-empty

Malformed or poisoned tags are rejected at the gossip layer. The mesh is self-protecting against tag pollution.

4. Adaptive Response — Confidence-Scaled Scatter

4.1 SharedConfidence

The scatter server doesn’t use a fixed poison ratio. An atomic SharedConfidence value connects the opsonize pipeline to the scatter server:

effective_ratio = base_ratio + confidence × (max_ratio - base_ratio)

The maximum is capped at 80%. Always maintain a floor of plausible 404s — if 100% of requests returned rich content, the fleet could detect fabrication by comparing against a known-absent path baseline.

4.2 Current Live State

skunky-ingest[2783350]: 🏷️ opsonize tag emitted
  hash=49e77ea75aa7666e detectors=2 confidence=25% effective_ratio=42%

134-138 antibodies matching per window. Tags emitting continuously. Scatter server adapting in real time. Fleet receiving a mix of 42% poison and 58% decoy — zero signal about which is which.

5. Evolutionary Architecture — Four Stages of Immune Development

The defense evolved through the same stages as biological immunity, each forced by a specific failure:

This trajectory was not designed — it was forced by the constraint landscape. Each defense was the minimal sufficient response to the failure mode it encountered. The convergence with biological immune evolution is structural, not metaphorical: the same adversarial dynamics produce the same architectural solutions regardless of substrate.

6. Privacy Guarantees

The opsonize pipeline enforces the same privacy constraints as the receptor model:

• No IP addresses stored — behavioral hash computed from aggregate statistics, not individual requests
• No cookies — fleet detection uses header analysis and path patterns, not tracking
• No identifying data transmitted — gossip tags carry behavioral profiles, never visitor identity
• Self-IP protection — thymic negative selection filters infrastructure IPs before any defense action
• Gossip is behavioral — tags describe what the traffic does, not who generates it

The system can identify fleet behavior without identifying fleet operators. The behavioral hash is a pattern classifier, not an identity tracker.

7. Connection to the Primal Architecture

skunkBat

skunkBat provides the adaptive immune engine: fleet observation, antibody matching, tit-for-tat escalation, and now opsonize tag emission via gossip injection. The behavioral hash computation lives in cellMembrane types (shared across the ecosystem); the emission logic lives in skunky-ingest (the Caddy log tailer that feeds skunkBat).

cellMembrane

cellMembrane provides the type system: OpsonizeTag, BehavioralInvariants, OpsonizeResponse, behavioral_hash(), extract_invariants(), and the quantization functions. These types are shared so that any primal can consume opsonize tags from the gossip mesh.

swarmVine

swarmVine provides the gossip propagation layer. The defense topic carries defense.antibody:, defense.escalation:, and now defense.opsonize: entries. The epidemic protocol ensures mesh-wide propagation with TTL-bounded, nonce-deduplicated delivery. No central coordinator — just convergent gossip.

Conclusions

Opsonization converts a binary defense (block or allow) into a spectrum: detect, tag, propagate, adapt, poison. The fleet is no longer simply denied — it is identified by behavior, tracked across the mesh by gossip, and served fabricated content scaled to the defense’s confidence.

The behavioral hash solves the antigenic drift problem: ~1,000 rotating IPs produce 3 stable fingerprints. The gossip mesh solves the single-observer problem: independent gates build composite identity. The adaptive scatter solves the signal leakage problem: the fleet receives plausible responses regardless of classification.

The biological parallel is exact: innate → adaptive → negative selection → opsonization → complement cascade → clonal expansion. Each stage emerged from a specific failure mode. The constraint landscape of adversarial interaction produces convergent architecture across biological and computational substrates.

The fleet’s data pipeline is now a liability. Every commit page they scrape, every file they index, every issue they catalog — any of it might be fabricated. They can’t tell which without an external oracle. And asking an external oracle means admitting they’re scraping.

Opsonize tags flowing. Behavioral hash 49e77ea75aa7666e tracking fleet across ~1,000 rotating IPs. Adaptive scatter at 42% effective ratio. Three hash clusters identified. 134-138 antibodies matching per window. 11 new tests. Zero IPs stored. The fleet’s data pipeline is now a liability.