The blind prediction method
An agreement only counts if neither road peeked. Forty-four times, this framework computed its own number for a cosmic milestone — from the granularity budget, before looking — and only then laid it beside standard cosmology and the measurement. The result stands at 44 tests / 30 agree / 14 indeterminate / 0 disagree, with zero numbers tuned after the fact. Three rules make those agreements evidence rather than curve-fitting: freeze before compare, two routes with no shared machinery, and no shared tuning. This page states them, and the provenance that lets anyone check.
BLOCKED_PENDING_NATIVE_LAMBDA, waiting on a from-scratch derivation
of the cosmological constant. The gate board has since settled that question the honest way: Λ is
measured, never derived — it is the fifth measured anchor
(Gap-05 value: MEASURED-ANCHOR, RESOLVED +0), the from-scratch
derivation demand dissolved under audit, and the one route by which the geometry’s own fine-grainedness might
have pinned the number misses by ~113 orders of magnitude — a negative control this program reports in the open.
The blind discipline did not dissolve with it: it produced the scoreboard.
1. Two routes, no shared machinery
Every checkpoint is computed twice. Standard cosmology solves the hot-Big-Bang equations. The granularity route asks one question of its own: when does this distinction first become an affordable, stable record? The two pipelines share no equations, no calibration, and no free parameters — so when both land on the same number at quark confinement, at neutrino freeze-out, at the forging of the first nuclei, at the release of the CMB, the agreement cannot be one method quietly feeding the other. That independence is the entire value of the exercise, and it is exactly what a wrong construction would break: one milestone landing on two different numbers is a direct “does not agree,” published as such. Across 44 checkpoints there are zero.
2. Freeze before compare — the sequencing rule
The rule that makes this blind rather than curve-fit is a strict ordering:
- Compute the framework’s number first, from the granularity budget and previously computed native inputs — never from the observed value under test.
- Write it down before the comparison value is consulted, in a record that exists independently of the comparison.
- Only then load the measured value, and record the verdict — agrees, does not agree, or indeterminate — exactly as it comes out.
Running the steps out of order — loading an observed number early, or adjusting an input after seeing a mismatch — is a rule violation, not a shortcut. And each of the 44 rows carries its own verdict, per calculation: there is no combined “the model works” score to hide a miss inside. A milestone that missed would be reported as a miss — the scoreboard has a column for it, and that column reads zero.
3. No shared tuning
Blindness would mean little if the model’s structure were quietly chosen to fit. So structural choices are selected by the framework’s own machinery, never by which branch matches data better. Big Bang nucleosynthesis runs on a single baryon count consistent with the CMB — not one count per element. The inflation-era rows carry their normalization branches openly: the tensor band is published as the union of four branches, with the operative branch shown beside it. And where a row necessarily consumes the same measured input on both roads — the standard nuclear network, the measured density ratio at matter–radiation equality — the row is marked consistency check — shared inputs, on its face. The caveat is part of the record, not a footnote.
4. Provenance — every row opens
Every card on the scoreboard carries the same six lines: the number under test, the standard-cosmology route, the granularity route, the measured value with its source, the honest agreement metric, and the link to the full derivation. Nothing is graded on trust; every line opens. That format is the provenance rule made visible — a claim you cannot open does not appear on the board.
5. What the discipline delivered
44 checkpoints, from the first payable record to today’s 13.8-billion-year age. 30 agree — confirmed by measurement, including the fractional Neff = 3.044 against Planck’s 2.99 ± 0.17, deuterium and helium-4 at 0.2–0.7σ, and recombination at the identical z* = 1089.92. 0 disagree. The 14 indeterminate are shown with the whole denominator, and none is a private failure of this method: 6 are standing bets where both routes already agree and the experiment hasn’t voted — LiteBIRD’s tensor band, r ∈ [3.5, 36]×10⁻³, chief among them — and 8 are open for all of cosmology: baryogenesis, dark-matter identity, lithium-7, the S8 tension, the primordial magnetic field, the Planck wall itself. On those, both methods are shown reaching the same honest limit, never a manufactured match.
6. Where the measured universe enters: declared anchors
The framework runs on a small, published set of measured anchors, and the cosmology work adds exactly one: Λ, the dark-energy density — the fifth measured anchor, at (2.3 meV)⁴ ≈ 1×10⁻¹²² in Planck-mass units, measured from supernovae, the CMB, and galaxy surveys. It is consumed as a declared input and compared to the Planck scale as a plain ratio — never claimed as a derivation. The gate ledger grades it MEASURED-ANCHOR, RESOLVED +0, and adds the negative control that keeps the claim honest: the one route by which the geometry’s fine-grainedness might have predicted the value misses by ~113 orders of magnitude — proof the number was not quietly reverse-engineered, because the machinery demonstrably cannot produce it.
7. Why the rules are published
Because a discipline you cannot inspect is a claim, not a discipline. The rules above are stated in public, the declared anchors are named before any comparison, and every row’s derivation opens — so anyone, expert or expert’s AI, can verify that no observed number entered earlier than the ordering allows. That is what lets the agreements carry weight: the protocol is part of the falsifiability record, not decoration around it.
See the scoreboard: 44 tests, every row open → Next: the standing bet — the inflation gate →