Independent reviews

Reviews of this program by parties outside the family of AI models that built it — published verbatim, criticisms included. A review we could edit would be worth nothing.

Review № 1 — Independent Reproducibility Review, Economic Assessment, and Roadmap (July 31, 2026)

Reviewer: 5.6 Sol, an out-of-family AI model (a different family from the models that produced this work)
Access given: this public website and the public replication kit only — no internal records, no vault access
Commissioned by: the project owner (disclosed; the reviewer's conclusions were not subject to approval or edit)
Kit reviewed: SHA-256 11a13f03…6fd8 (605 entries)
Independent harness run: macOS arm64 · Python 3.12.13 · NumPy 2.3.5 — 72 of 82 discovered programs reproduced within tolerance; 7 mismatches classified as runtime-only, legacy numeric, or additive-schema drift (none verdict-reversing under the reviewer's classification); 3 helper scripts with no registered output
What this is: machine replication and machine criticism. What this is not: human peer review. The human tests — independent economists, real buyers, a live identity pilot — remain open, and remain the point.
"EDEN deserves support — but support should be directed toward validation infrastructure, not toward launch theater. … EDEN is not yet a better economic system. It is a better-than-average attempt to make a radical economic proposal falsifiable."
— from the review's executive judgment and final assessment

Read the full review (PDF, unedited)

The review found real defects — stale reproduction counts, a claims register that lagged the newest simulations, unpinned software versions, a promised document missing from the kit archive, and public claims (including a retired inflation headline) that had drifted from the internal record. We regard the fact that it found problems as the evidence of its independence. The defects it identified are being corrected under the adoption record below.

Addendum — same reviewer, same day (July 31, 2026)

Scope: change memo after our public response, the site corrections, the R2 packaging changes, and a separate review of the live EdenQuest app
Kit reviewed: SHA-256 bd1c514c…d3f (617 entries) — differentially audited against the original corpus: no simulation engine or result changed; the reviewer therefore did not re-claim a fresh replication
Note: the kit is updated in place during the R2 release-cleanup window, so the live zip's hash moves with each commit; each review records the exact hash it audited, and per-release checksums ship with the DOI archive
What moved: team credibility and reproducibility upgraded on behavior; the launch question split by layer (EdenQuest public beta reasonable; monetary/identity/lending/conduct/floor layers still not ready); the identity criticism narrowed after our corrections
What did not move: the two decisive hinges — buyer willingness to pay and live identity — remain untested, and the reviewer says so

Read the full addendum (PDF, unedited)

The addendum found the next layer of the same disease it named in round one — "reconciliation debt": stale counts and provenance statements surviving in start-here documents and one homepage line beneath the corrected surfaces. Those items were corrected the same day; the ledger below tracks every finding across both documents.

Our response (July 31, 2026)

What we adopted

The reviewer's release-engineering standard (a canonical engine manifest, pinned environments, a mismatch taxonomy, dual fast/clean-room reproduction ledgers, externally verifiable preregistration timestamps), a claims register extended through v39, an incidence measurement workstream behind "machines pay," most of the recommended claim-wording corrections — including the retirement of the "≈0% inflation" headline and a narrower money-creation opener — and upgrades to the field-experiment instruments. The full adoption record, item by item, lives in the project vault and travels with the next replication-kit release.

Where we pushed back — identity

The review recommends splitting identity into a uniqueness proof plus many unlinkable personas with no cross-context reputation. Our uniqueness layer is already separate (nothing biometric on-ledger; pseudonymous identifiers; no global trust score; aggregate-only detection). What the review actually objects to is cross-context consequence — and that coupling is not an oversight but the priced alternative to surveillance: it is what makes account rental irrational, lending viable without credit bureaus, and shell-company evasion expensive. The review recommends the swap without pricing what it breaks. We are answering in our own methodology instead of by argument: a simulation cell (v40, spec registered July 31, frozen until external review completes) will measure exactly what fraud, lending, and ownership enforcement cost under compartmentalized reputation. If the loss is small, the fork reopens; if it is large, the coupling finally has its number. In parallel we adopted what the review got right: a consolidated rights charter, a sanctions-and-recovery matrix, and the removal of "permanent exile" deterrence language that misrepresented a due-process design.

Where we pushed back — the shadow ledger

We accept the constraint entirely (dollar settlement, no transferable unit before evidence, no token, nothing to buy). But a shadow ledger nobody can use produces no behavior to measure. Our registered alternative keeps the constraint and fixes the measurement: a non-transferable internal credit redeemable for exactly one thing — governed AI queries over the contributed corpus — tested in two arms (dollar-denominated vs. unit-denominated at identical rates), so the only difference is the unit itself. Internal redemption will never be cited as external demand evidence; the willingness-to-pay experiment remains the hinge.

Corrections to the review

Two calibration notes, for the record: the $600/contributor-year figure the review criticizes as "presented as a market price" is labeled on this site as an exogenous assumption with a pre-registered failure bar — it is the hinge, and marked as such. And several "do not launch yet" cautions land on doors already locked: there is no token, the floor is already published as a separately funded module rather than a promise, and the program's own gates already sequence review before pilots before any settlement question.

Adoption ledger

One row per finding, across both review documents. "Resolved" links are checkable in the public kit and site; nothing here is graded by the reviewer — this is the team's own accounting, offered for audit.

FindingStatusWhere
Canonical engine manifestRESOLVEDkit: manifest.yaml (82 programs across 63 folder rows; roles + per-engine statuses)
Current reproduction ledgersRESOLVEDkit: dated standard + clean-room ledgers, LATEST.json
Mismatch taxonomyRESOLVEDkit: classify_report.py; classes enumerated on /replication and this page
Claims register through v39RESOLVEDkit: register + claims.json + drift_check.py; prose reconciled Jul 31
Economist Brief missing / staleRESOLVEDpackaged; rewritten v2 (Jul 31) around the 82-engine record
Environment specificationPARTIALfloors + 3 attested matrices ship; hash-lock + CI owed
Clean-room recomputationPARTIAL69/82 from scratch; 4 engines lack a from-scratch path (tracked S13)
Preregistration evidencePARTIALv40 frozen spec ships in kit; external timestamps pending DOI
Immutable release + DOIIN PROGRESSconsole repo committed; Zenodo release = owner steps
"≈0% inflation" headlineRESOLVEDretired on all six surfaces; index-governance limits stated
Dynasty metric conflationRESOLVEDstatistic defined on C-30; /whitepaper C.8 reconciled
Stale reproduction counts (49/54)RESOLVEDreplaced on every claim surface; dated sim-report pages carry retirement notes
Bank-money openerRESOLVED/paper §01, /start, /blueprint, homepage (Jul 31)
"Permanent exile" languageRESOLVED/evidence rewritten to the due-process story
Whistle-market card missing limitsRESOLVEDC-54 carries its two named limits
Provenance precision ("no external review")RESOLVEDsite-wide: one out-of-family AI review exists; no human review yet
$600 "presented as market price"DECLINEDH-1 already labels it an exogenous assumption with a FAIL bar
Unlinkable-persona identity architectureTESTED INSTEADv40 frozen preregistration; rights charter + appeals published (/rights)
Shadow EVE ledgerMODIFIED-ADOPTEDtwo-arm non-transferable credit design + liability/outage/label guardrails
Full stack unbundlingPARTIALadopted as validation strategy; compositions stay first-class claims
Incidence workstreamADOPTEDWTP field-kit addendum (who ultimately pays)
Reviews discoverabilityRESOLVEDthis page: indexed, in footers and sitemap (Jul 31)
EdenQuest on the economist paperRESOLVED/paper §12 Stage-0 deployment & identification note
EdenQuest privacy items (operator ID, child-data, E2E default, crash-report disclosure, AI safety)OPENapp-team punch list, tracked S14
Conventional PDF working paper / TODO markersOPENclears at submission (Unjournal path); TODOs are unmined register keys

Standing invitation: we commission and host hostile reviews with unrestricted publication rights. If you are an economist, security researcher, identity or civil-rights specialist and want the kit plus a fixed fee to attack this work in public, see the replication page for the current kit and contact route. · Back to the paper