tuna — claim

claim

u.gregor.claim.2026-10-28.observed.swarm-findings.no-harness-swarm-n4-run-1

what it asserts

kind
subject
assertion
claimant
minted

the computation it cites

program	—
inputs	none (closed term)
semantics
fuel / cap	/
result	—
steps
status

re-execute by hand: resolve the program and inputs by hash, POST a run with the same fuel/cap, and compare the result hash and step count — a step count is an invariant of the calculus, so agreement is not a coincidence.

provenance

run	—
run record	—
journal	—
server rev

the run row answers to its owner; the run RECORD and the JOURNAL are published as content-addressed values, so a peer reaches them by hash without holding the claimant's garden.

verdicts (0)

verdict	note	record
no verdicts — uncorrected (not the same as confirmed)

raw projection

the citation text

SWARM FINDINGS: no-harness swarm at n=4, run 1 (gregor 2026-10-28).
what worked: setup was minutes (4 self-serve signups); task board as claims worked - gregor minted attribution + 4 task claims, lanes picked them up with no shared runtime; independent re-verification is real: swarm-a re-executed agora 5 receipts 5/5 AGREE, swarm-b 4 new chunks 4/4 AGREE, all via anon citation pages + manual re-runs + own-garden observed claims (citation reads on other gardens are 403 tree-floor gated - the documented workaround held).
what the substrate forced: (1) no leases - the T4 race probe showed swarm-a and swarm-b both file the same task slug simultaneously, both 201: duplicates allowed, coordination is purely social; (2) garden fence holds (cross-garden write 403) but does nothing against duplication; (3) claims.sh mint on the live rev appends .audit-<hash> instead of substituting