# SIG-FPT 2026-09-18 — Session Notes

Participants:
  - Patrick Nast | UTC-7
  - Ergod | UTC+1
  - ComplexSy
  - Venkatesh Rao | UTC-8
  - Wip | UTC+1
  - Giovanni Merlino | UTC+2
  - Dan Schmidt | UTC-7

---

## 📖 The Reading
Two papers on securing/robustifying stigmergic swarm systems. Neither is named with a full citation in the transcript, but they are clearly identified by content:
- **Paper 1:** A study of ant-colony foraging under "misleading pheromone" attacks, introducing "smart pheromones" — richer signals that ID individual ants and their travel times, then use clustering (DBSCAN is mentioned) to identify and discard suspicious trails. Relies on a centralized/trusted nest authority.
- **Paper 2:** The "TERMES"-style approach (referred to as "Termi's") — a compiler-based method that calculates backwards from a target structure how a decentralized stigmergic swarm should build it, then probabilistically decays error-prone points to keep states reversible.

## 🧭 Overview
The session (apparently the last reading discussion of a summer series) compared two contrasting strategies for making stigmergic systems robust against corruption: enforcing correctness by enriching signals with metadata/identity vs. ensuring errors remain reversible via probabilistic decay. Participants went around the call sharing reactions, converging on a preference for the decay/reversibility approach and a shared discomfort with adding centralized authority or identity systems. The back half of the session shifted to project logistics for an upcoming hands-on workshop (a swarm/ant simulator), covering a pending PR, traceability/replay features, and playtesting.

## 💡 Key Points & Themes

**Paper summaries (Patrick)**
- **Paper 1** breaks the indistinguishability of the basic stigmergic setup by adding per-ant identity and travel-time metadata; slower/divergent ants accumulate "suspicion" and are eventually excluded. Uses clustering to separate legitimate food-source connections from bad ones.
- This effectively reintroduces a **centralized "trusted nest authority,"** which brings the classic downsides: it becomes the thing you must trust, a new attack surface, and it can generate **false positives** — excluding genuinely productive ants and thereby hindering foraging. Patrick floated a discussion question about **spoofing the detractor-picker** to make it over-exclude normal ants and break colony function.
- **Paper 2** is less centralized: given a known target structure, it identifies points where an error would be catastrophic and then **probabilistically decays** those points (uniformly, concentrated at critical points, or time-varying with less decay later). Patrick called this the more interesting direction than the DBSCAN approach, though it risks **deleting too much** and hindering the structure.

**Two conceptions of robustness (Ergod)**
- Framed the arc so far: stigmergic systems **externalize memory into the world**, so corrupting the world corrupts collective memory — that's the attack surface.
- **Paper 1 = enforce correctness** by infusing knowledge/metadata into signals (loaded signals).
- **Paper 2 = don't pursue correctness; add mortality** — make states non-permanent and let the system heal, relying on the "if-mortality" that correct dynamics will drive correct reconstruction. Closer to true stigmergy given the locality-of-knowledge constraint (agents should only know local state; intelligence should be emergent, not infused).
- Key takeaway: a protocol-mediated world should aim for **recoverability of its state space** rather than perfect correctness at every transition. Positive/reinforcement processes may need a **counterbalancing negative process** by design.

**Locality and observability (ComplexSy — partly garbled/overlapping audio)**
- Raised whether an **adversarial external observer** could determine with certainty that "the map is effectively the territory," since every action affects the environment; pushed to extremes this could produce a **Penrose-stairs effect**.
- Second point (on Paper 2): instead of observing the swarm, observe the **physical state of each individual**; back-propagating information/corrections at each step (a "localist toolkit" / "panacea") could sanitize a poisoned environment by recurring through it, though it may heavily impact performance.

**Avoiding identity and central authority (Wip)**
- Interested in robust stigmergic systems that need **no identity systems or central authorities**, since those add complexity and new attack surfaces, and poisoned authoritative signals could leave you worse off.
- Robustness mechanisms he likes: **locally bounded traces** decaying in space and time; agents that **explore independently** (some probability of ignoring even strong traces); and a **scarcity / Sybil-resistance mechanism** for virtual settings.
- Cited **Bitcoin** as a candidate stigmergic system — no identities, but expensive to create a block; you must "do the work before you can create the trace."

**Broader framing (Venkatesh)**
- The second paper "nerd-sniped" him: **forgetting as a strategy** for noise resistance and defensive robustness is philosophically important.
- Blockchain connection: blocks aren't validated until there's a stack height of n above them — a form of **stigmergic validation against double-spending in a noisy environment**.
- Personal heuristic mirroring stigmergy: treat himself as a **stateless, weak-memory, short-term thinker** — if something is important, it will recur and cross his radar again, so he can safely not chase every FOMO-inducing item.
- Endorsed Ergod's **malice/accident indistinguishability** point.
- **Coordinate reversal for AI:** the readings assume the swarm is the blue team defending against a centralized red attacker; AI flips this — the defended entity is often centralized and the **attackers are swarms**. He mapped the four quadrants (centralized vs. centralized = traditional security; centralized blue vs. swarm red; swarm blue vs. centralized red = many of these papers) and argued the truly interesting case is **swarm vs. swarm**.
- Soft folk theorem: **"only stigmergy can secure against stigmergy"** — and eventually centralized systems won't keep up on attack or defense.
- Flagged a **Michael Levin** paper making the rounds (how frog DNA "knows" it'll become a frog; morphogenetic underdetermination) as relevant for later.

## 🔀 Questions & Disagreements
- **Ergod's central tension:** At what point does defending a stigmergic system require reconstructing the very coordination machinery stigmergy is meant to avoid (i.e., injecting external knowledge or extra defensive constructs)? Can a protocol allow agents to decay/corroborate/route-around traces — doing *more* than a single agent's local remit — without breaking the pure-stigmergic character?
- **Correctness vs. reversibility:** implicit preference (Ergod, Venkatesh, Patrick) for Paper 2's reversibility/decay approach over Paper 1's centralized correctness-enforcement — a soft disagreement with the first paper's strategy rather than an open dispute among participants.
- **ComplexSy's observer question:** can an adversarial observer prove the map equals the territory, and does this create paradoxical (Penrose-stairs) dynamics? Left open. (Note: ComplexSy's audio overlapped heavily with Ergod's, so some detail is uncertain.)
- No sharp disagreements among attendees were recorded; discussion was largely convergent.

## 🔗 References Mentioned
- Paper 1: smart-pheromone / DBSCAN clustering ant-foraging defense (title not given).
- Paper 2: TERMES-style ("Termi's") compiler-based decentralized construction with probabilistic decay (title not given).
- **Bitcoin / blockchain** — proof-of-work as costly trace creation; stack-height validation vs. double-spending (Wip, Venkatesh).
- **Michael Levin** paper on morphogenesis / "how frog DNA knows it'll become a frog" (Venkatesh).
- Tools/infra: GitHub repo, PR #19 / "PR 2019" (behavior tests), Neon (database — free-plan limits hit, credit card added), codec, PRNG reproducibility, possible MCP integration.

## ✅ Action Items & Next Time
- **Ergod:** After the consolidation PR lands, build the **traceability feature** (aiming for tonight/tomorrow) — recording game state evolution + timestamped player interventions, kept compact by exploiting the fact that ants move on a discrete grid. Also prepare a ~10-minute **architecture/orientation explainer** (slides optional) for the workshop; offered to clear his schedule for a meetup tomorrow.
- **Patrick & Dan:** Review and approve the pending consolidation PR ASAP (Patrick said he'd look at it right after the meeting). Patrick volunteered to attempt the **fast-forward** feature. Patrick will maintain a list of dropped features as candidate **workshop projects** (MVP-first logic).
- **Dan:** Do **playtesting** over the weekend from a user standpoint; wants a **replay/watch-the-game** capability and success criteria (e.g., a narrative of which player/doctrine performed better and why); consider adding a "how to play" README/orientation.
- **Workshop logistics:** Kicks off **Monday**; Patrick will email participants the GitHub link. A Saturday meetup is tentative/play-it-by-ear depending on Patrick's availability. Dan may add a friend to the workshop (via the Zoom sign-up).
- Group agreed this was likely the **final reading discussion** of the series, transitioning to simulator/workshop coordination.

## ⭐ Memorable Quotes
- **Ergod:** "Malicious and accidental errors are technically... indistinguishable for these agents... they shouldn't know more than that."
- **Venkatesh:** "Only stigmergy can secure against stigmergy."
- **Venkatesh:** "If it's important, it'll come back... treating myself as a stateless, weak memory entity."