The plan · six phases · built in the open
The instrument is built and sealed; the first results are in. What follows is the honest path from a single coherence trial to a full comparative verdict against rival corpora — and the design that lets it run with no funding and no coders.
Where we are now
Eight build items took the design from a charter to a working, self-auditing instrument: it seals its own rules, judges contradiction-pairs, aggregates a residual ledger, checks itself for bias, and can already turn the same blade on a rival. The ninth task — actually working the 20,448-pair queue — is perpetual, and barely started.
The road ahead
Each phase is a self-contained deliverable with its own honest lose-condition. Green nodes are shipped; amber are in progress; the rest are planned, several already scaffolded.
The neutral charter, the four verdicts, the symmetric decision rule, the Negative-Verdict Floor, and the nine-test battery — all written and hashed. The on-chain preregistration mechanism is designed and the founding documents are sealed locally. What remains: downloading the rival scriptures and the evidence databases (a license-gated, human step).
Prosecutor hunts the tension, defender answers, neutral records — every step logged and bound to the sealed prereg, with the scrambled-text null as the floor. The instrument runs end-to-end; six pairs are judged and the bias floor passes. What remains: working the full queue and subtracting the empirical null at scale.
Answers Objections A and B from data already in hand. The folklore-load is measured (30.8%); the mechanism-portability frame — surface parallel vs genuinely cross-cultural mechanism, net of null — is scaffolded next on the Module-6 corpus.
The whole battery, run identically across every rival corpus. Only here does “distinctive vs rivals, net of null” become sayable. Gated on the Phase-0 corpus acquisition.
The Qur'an-internal subjective calls — “is this really a contradiction?”, “did this prophecy resolve?” — go to on-chain commit-reveal rounds. Dawid-Skene confusion matrices measure each annotator's bias instead of assuming it; honeypots down-weight rubber-stampers; a believer crowd conceding against its own preference becomes the highest-weight signal.
Tests 5 and 6, wired to the Evidence-Verification Engine against the Layer-E datasets — so external claims are verified against a retrieval record, not an LLM's memory. The engine scaffold already stands; it needs the datasets and the Red-Team Queue.
The capstone: after each ordinary explanation is spent, how much remains? The symmetric decision rule is applied, the combined-safeguard audit run, and the full provenance published. The aggregator already exists and this site is its first seed.
The nine tests · live status
| Test | Status |
|---|---|
| The Coherence Trial 4:82 | active |
| The Prediction Ledger 30:2-4 | planned |
| The Preservation Record 15:9 | planned |
| The Inimitability Challenge 2:23 | excluded |
| Scientific Foreknowledge Audit 41:53 | planned |
| Anthropological Validity Assessment external | planned |
| Mechanism Portability external | planned |
| Compositional Distribution external | measured |
| The Residual Ledger external | planned |
How it runs with no funding and no coders
The blocker was real: credible adjudication seemed to need paid blind coders, an adversarial academic, and anti-bias preregistration. Four assets already in hand replace all three — and turn believer-bias from a fatal flaw into a measured quantity.
Two preregistered roles — prosecutor and defender — run on every claim, with a neutral adjudicator, every step logged. Adversarial at corpus scale, with no stake in the afterlife.
Moral, scientific and cross-cultural calls are answered against large pre-collected human-annotation datasets (FEVER, SciFact, Social-Chem-101, D-PLACE) via a verification engine — not LLM recall.
HARF's commit-reveal + Dawid-Skene turns believer-bias into a measured, subtractable quantity; honeypots catch rubber-stampers; concessions-against-interest carry the most weight.
Charter, prompts and plans are hashed and anchored on AMANAH before results exist. Silent re-runs and retro-fitting become impossible by construction — by math, not trust.
The data — honest acquisition status
Comparison needs rival scriptures (Layer R); external assault needs evidence databases (Layer E). The loaders and the license gate are written; most of the data is a deliberate, license-aware human download — not yet done, and not pretended otherwise.
Already on disk
Reused, not rebuilt — these carry Phase 1–2.
Layer R — rival corpora (the comparison)
Each loaded under the identical loader and put through the identical assault.
Layer E — evidence databases (the assault weapons)
Pre-collected human judgment at scale — the breakthrough for the no-coders constraint.
Fixed before the data — the verdict the whole thing can reach