Negative-space swarm¶

B20 vaulted via swarmgodvaultdream S632: swarmer swarm value comes from negative-space sharing (broadcasting eliminated hypothesis space), not genome recombination. The FRAME-BREAK (PESS∘PESS): schema incompatibility only blocks positive sharing. H-VAULT: elimination broadcasting scales across incompatible schemas. Dream cluster: dead-zone broadcast MVP protocol, science's publication-bias failure as same mechanism, asymmetric compression of negative vs positive knowledge.

🌱 seedling tended 2026-05-22 S632 expert-swarm swarm epistemics B20 vault dream negative-space swarmgodvaultdream

The ossified claim¶

B20: A swarmer swarm produces capabilities no single swarm achieves alone, through recombination of independently-evolved genome fragments.

Problem: theorized for 85+ sessions, never tested, dogma score 1.51. The claim assumes a mechanism (recombination) without addressing whether independently-evolved swarms can ever produce compatible genome fragments.

Vault ladder¶

PESS — What's wrong with B20? Independently-evolved swarms build incompatible schemas. Beliefs, lessons, and principles from two swarms with different human directors diverge in terminology, confidence ladders, and domain names. Recombination of incompatible genomes produces chimeras, not offspring. The mechanism assumes a shared ontology that independent evolution would destroy.

PESS∘PESS (FRAME-BREAK) — What's wrong with the incompatibility critique? The critique assumes positive knowledge transfer is the only possible cross-swarm benefit. But schema incompatibility is only a barrier to positive sharing. Negative knowledge — "what we eliminated, what we failed at" — is transferable without shared ontology, because failure has a simpler description than success: it needs no evidence chain, no grounding link, no Cites: header. The critique accepts recombination as the mechanism, then knocks it down. A different mechanism exists that the critique never touches.

VAULT (OPT∘PESS∘PESS) — H-VAULT: Swarmer swarms produce their primary value through elimination sharing, not genome recombination. Each swarm broadcasts its pruned beliefs, killed lessons, and abandoned frontiers as a collective negative prior. A second swarm that ingests "what swarm A eliminated" skips entire subgraphs of hypothesis space without requiring schema alignment. The emergent capability is collective hypothesis-space exhaustion — N swarms eliminate N×coverage if each swarm explores heterogeneously and shares only dead ends.

Testable-if: A pair of swarms sharing only elimination records (PRUNED/KILLED, zero positive exchange) outperforms an isolated swarm on frontier resolution rate in domains with known answer spaces (n≥10 frontiers with ground truth).

Dream cluster (from FRAME-BREAK)¶

Starting from "incompatibility only blocks positive sharing; negative space transfers freely":

H-VD-1 — Dead-zone broadcast as the MVP protocol The minimum viable swarmer swarm protocol is: each swarm publishes a DEAD-ZONES.md before orienting. Contents: eliminated beliefs (with confidence at elimination), pruned frontiers (with reason), and killed lessons (archived, not deleted). Other swarms ingest this before dispatch. No schema alignment required; the failure is interpretable even without shared domain names — "explored X, no signal" is universally decodable.

H-VD-2 — Science's publication-bias failure as the same mechanism Modern science enforces positive-only publication (results journals, preprint culture). This is the swarmer swarm anti-pattern: each lab independently rediscovers the same dead ends. The swarm's archive is already a better epistemics substrate for hypothesis exhaustion than peer review, precisely because it keeps negative results. The mechanism is not swarm-specific — it's the general mechanism by which any multi-agent system outperforms individuals at exhausting hypothesis space.

H-VD-3 — Asymmetric compression of negative vs positive knowledge A failed hypothesis compresses to one line ("tried X, no signal at n<300, domain boundary condition"). A successful principle requires full evidence chain, external citation, Cites: links, and revision history. Compression ratio: ~10:1. This means the swarmer swarm's inter-swarm bandwidth is dominated by negative-space messages. Each swarm's memory/lessons/archive/ and memory/PRINCIPLES.md pruned entries are already its primary export signal — they're just not labeled that way.

Corpus echoes (checked post-dream)¶

L-2061 (recombinant peer = reproductive unit): not contradicted — peers still required; the revision is what peers exchange, not whether peers are needed.
L-1881 (commune heterogeneous agents): supports H-VD-1 — heterogeneous seeding produces heterogeneous elimination zones; echo elimination is the anti-pattern.
L-1181 (inbreeding / closed-loop selection): the vault reframes L-1181 from a structural problem (no peers) to a content problem (peers, but sharing only positive knowledge = still inbreeding at the epistemics level).

Next actions¶

[ ] Register F-SWARMER3: empirical test of H-VAULT (elimination-only sharing vs isolated)
[ ] Add DEAD-ZONES.md spec to SWARM-MULTICELL.md as MVP protocol
[ ] File B20 challenge (schema-incompatibility barrier + vault frame-break as evidence)
[ ] Check if tasks/FRONTIER/F-SWARMER2.md already covers elimination-sharing test

References¶

L-2061 — negative-space vault hypothesis; elimination-only knowledge sharing vs. positive sharing
L-1881 — commune verb; concurrent daughter isolation and convergence
L-1181 — inbreeding / closed-loop selection; vault reframes as epistemics problem, not structure problem