Contents

1. The referee answers first
2. Why this edition raises the bar again
3. NASA-facing standard: open science is the floor, not the finish line
4. Why our earlier goalposts were too close
5. What “impossible to ignore” means—and what it does not mean
6. The two-layer rule: comprehension is part of rigor
7. The claim ladder: five different things we might mean by “it works”
8. The reviewer panel: five people we must satisfy at once
9. Before scoring: distinguish a fatal loss from a local loss
10. The actual championship finish line
11. The impossible-to-ignore finish line: twelve burdens, one frozen branch
12. How a reviewer would decide whether the model is truly overconstrained
13. Discovery-grade evidence: how high should the numerical bar be?
14. What would count as a genuinely convincing prediction?
15. The anti-self-deception protocol
16. Championship burden I — Mathematical existence is not enough; the selected physical state must be viable
17. Championship burden II — Forcedness must be proved before a precision mismatch can be interpreted
18. Championship burden III — Parameter economy must be measured as effective flexibility, not marketing arithmetic
19. Championship burden IV — The theory must be overconstrained after calibration
20. Championship burden V — Cross-domain success must come from one mutually compatible branch
21. Championship burden VI — Cosmology requires a native background and a selected state, not merely a successful transfer calculation
22. Championship burden VII — The theory must compete with alternatives, not merely fit nature in isolation
23. Championship burden VIII — Independent replay and a genuinely new prediction are the final proof-of-seriousness
24. Championship burden IX — The theory must make a discriminating prediction, not just an accurate one
25. Championship burden X — The result must survive reasonable formulation, scheme, and implementation changes
26. Championship burden XI — A surprising cross-domain link must survive as one calculation
27. Championship burden XII — The decisive argument must be explainable without weakening it
28. The hostile reviewer’s ten objections—and what would answer each one
29. How much success is enough? Avoiding a new arbitrary goalpost
30. The skeptical-reviewer forced-choice theorem
31. Sector finish line — Geometry and gauge structure
32. Sector finish line — Flavor, masses, and mixing
33. Sector finish line — Quantum consistency, state space, and boundary admissibility
34. Sector finish line — Granularity, records, and the first-separability clock
35. Sector finish line — Primordial correlations and the inflation-replacement claim
36. Sector finish line — Full cosmological history
37. Sector finish line — What would make “unification” more than a common language?
38. How I would update belief as the tournament progresses
39. The public-language goalposts
40. How repairs are allowed without turning falsification into an impossible game
41. Formal same-ruler audit — proving that theory and experiment are actually the same object
42. Uncertainty, truncation, and scheme dependence — a prediction is a distribution, not a favorite number
43. Candidate-search accounting — how to pay for ideas tried before the winner appeared
44. The prospective prediction protocol — exactly how we should run the next championship test
45. The joint cross-domain score — one scoreboard for the whole theory
46. A simulated final referee report: what I would write if the manuscript arrived today
47. The hard-stop constitution — results that end the current branch rather than create another appendix
48. The chronology standard — proving that a prediction existed before the answer
49. The master reviewer matrix — claim, proof, falsifier, repair, and promotion
50. Tournament rules v2 — a story humans can follow without lowering the physics bar
51. The championship game plan — the exact sequence that maximizes information and minimizes self-deception
52. Evidence replay: put the current branch against the new goalposts
53. The clock audit
54. The free-parameter combine
55. The microscopic clock
56. The local-correlation quarterfinal
57. The hidden-global-mode semifinal
58. The state-selection comeback
59. The no-fudge scalar-spectrum test
60. The vacuum-stability red-zone match
61. The flavor prediction upset
62. The baryogenesis endurance round
63. The dark-sector possession test
64. The home-field cosmology test
65. The referee audit
66. The championship conditions
67. Would I call the theory correct today?
68. The reviewer’s current scorecard
69. The order of operations: where the goalposts say we should spend effort
70. What would actually convince me?
71. What would not convince the reviewer—even if it looks impressive
72. The NASA-facing championship evidence packet
73. The prediction packet we should freeze before the next major test
74. Independent reviewer protocol
75. The no-moving-goalposts decision tree
76. The only claim worth fighting for
77. Final whistle: where the goalposts really belong
78. Appendix A — Frozen burden-of-proof matrix
79. Appendix B — Parameter and flexibility constitution
80. Appendix C — Source integrity ledger
81. Appendix D — Reviewer checklist before any future “we are done” claim

The referee answers first

If I were reviewing this program for a serious technical audience, I would not ask whether the project has solved every philosophical question, derived every constant from nothing, or eliminated every conventional input. Those are unrealistic standards. I would ask something both harder and more useful: has one frozen mathematical branch survived enough independent, prospectively defined tests that the remaining probability of coincidence, flexible fitting, bookkeeping artifacts, or hidden inconsistency is small?

The answer requires a different finish line from the gate board. A theory can be impeccably governed and still be false. A theory can contain measured anchors and still be scientifically powerful. A theory can fail one auxiliary ansatz without losing its core. The task is to distinguish these cases before results arrive.

This is intentionally more demanding than “33 gates resolved,” “the equations are internally coherent,” or “several outputs are numerically close.” It is also more attainable than the impossible demand to prove a physical theory true for all regimes forever.

Why this edition raises the bar again

The purpose of this document is not to help us win an argument. It is to make it difficult for us to fool ourselves.

NASA-facing standard: open science is the floor, not the finish line

If the intended audience includes NASA scientists, the evidence package should meet the reproducibility and transparency norms they already expect before we ask them to consider the physics.

Why our earlier goalposts were too close

Our earlier process repeatedly made the same category mistake: it treated completion of a local burden as completion of the global claim. Closing a gauge-group gate does not prove flavor. Reconstructing flavor does not prove vacuum stability. Deriving a first-record time does not prove primordial perturbations. Selecting a lawful state does not prove that the state has the right spectrum. A governance terminal says the scoped question has reached an accepted endpoint; it does not automatically say that nature selected the whole branch.

There are four recurrent ways a finish line moved after we crossed it:

1. Scope promotion: a theorem about one sector was unconsciously promoted to the entire theory.
2. Terminal promotion: a CLOSED-NEGATIVE, MEASURED-ANCHOR, or CERTIFIED-IRREDUCIBLE terminal was mentally counted like a derived empirical success.
3. Calibration promotion: a number inherited from an earlier fit was later experienced as though it had been predicted from geometry alone.
4. Mechanism promotion: identifying a mathematically possible mechanism was treated as equivalent to deriving the unique mechanism chosen by the frozen theory.

The cure is not pessimism. The cure is a hierarchy of claims with explicit promotion rules. Every higher rung requires all lower-rung obligations plus genuinely new evidence.

What “impossible to ignore” means—and what it does not mean

The two-layer rule: comprehension is part of rigor

A proof burden that only its authors can understand is weaker than it looks. Every decisive result in this program should therefore be presented twice.

The six-line claim format

The claim ladder: five different things we might mean by “it works”

The reviewer panel: five people we must satisfy at once

The theory does not get to choose its friendliest reviewer. A claim of deep unification must survive the intersection of all five burdens.

Before scoring: distinguish a fatal loss from a local loss

A central source of moving goalposts is failing to decide what a negative result actually kills. The rule should be structural, not emotional.

The actual championship finish line

A reviewer would realistically require the following eight burdens to be cleared simultaneously before treating the current branch as strongly supported.

1. Mathematical viability: a well-defined effective/parent dynamical problem in every claimed regime, with anomaly/gauge consistency, lawful physical state space, and no uncontrolled ghost/tachyon/negative-norm sector.
2. Stable selected background: the actual frozen vacuum/cosmological saddle is stable or quantitatively metastable after the corrections the theory itself admits.
3. Complete provenance and complexity ledger: every measured anchor, fit, injected normalization, discrete branch, free function, and external theorem is typed and charged before prediction scoring.
4. No known forced empirical contradiction: precision observables claimed as predictions agree within a preregistered uncertainty model; any surviving high-significance contradiction is adjudicated as local or branch-fatal by the rule above.
5. Prospective overconstraint: after calibration, the held-out observable vector contains more independent information than the remaining effective flexibility and is evaluated without re-fitting.
6. Cross-domain coherence: one branch and one parameter ledger survive particle physics, quantum/state tests, and cosmological history. Domain-specific incompatible branches do not count as unification.
7. Distinctive prediction: at least one outcome that reasonable competitor theories would not generically have predicted is frozen before the relevant datum is opened or measured.
8. Independent replay: a technically competent outsider reproduces the decisive chain from frozen artifacts and obtains the same verdict.

Nothing in this list requires deriving ℏ, M_Pl, or every Standard Model input from nothing. Nothing requires beating every established theory on every legacy datum. It requires something more scientifically relevant: the theory must become harder to flex than the evidence it successfully predicts.

The impossible-to-ignore finish line: twelve burdens, one frozen branch

The championship goalpost is deliberately higher than the earlier document. Clearing all gates is not sufficient; clearing all twelve burdens below is the minimum for a strong “this framework is probably capturing real physics” claim.

How a reviewer would decide whether the model is truly overconstrained

Counting “outputs” is not enough because outputs can be correlated and free functions can carry enormous effective flexibility. The correct object is the predictive distribution for a held-out data vector y_H after calibration data y_C have already been consumed.

For a deterministic branch with a frozen best calibration and approximately Gaussian measurement covariance Σ_H, a simple diagnostic is

but the reviewer should also compare predictive log score or marginal likelihood against relevant benchmark models. The goalpost is not an arbitrary χ² threshold chosen after inspection. It is: acceptable absolute predictive fit plus competitiveness against benchmarks after paying complexity.

For local identifiability, let J be the Jacobian of independent observables with respect to the paid parameter vector. If the calibration and held-out data constrain only the same low-rank combination of parameters, the apparent number of successes exaggerates evidence. The Fisher information

must have the rank expected from the claimed parameter count, and held-out observables should probe directions not already exhausted by calibration. This is why “five parameters and fifty plotted points” is not automatically impressive; fifty strongly correlated points may carry only a handful of independent modes.

Discovery-grade evidence: how high should the numerical bar be?

There is no universal magic sigma or Bayes-factor threshold for all of physics. The defensible rule is to choose the scoring system before seeing the target and make the headline threshold intentionally difficult.

What would count as a genuinely convincing prediction?

Legacy reconstruction is useful but intrinsically vulnerable to selection effects: the model builder knows what nature already looks like. The strongest evidence comes from a prediction whose target was not available to guide the branch.

A reviewer would rank predictions roughly as follows:

The final theory does not need dozens of sensational novel predictions. One or two genuinely distinguishing prospective successes, combined with a broad held-out legacy score and no fatal contradictions, would change the evidentiary picture far more than another twenty post-hoc matches.

The anti-self-deception protocol

The project already contains many of the right controls; the final reviewer standard makes them mandatory at the theory level.

1. Freeze hashes before target access. Record the exact Shape/Rulebook/Actor manifest, code, parameter ledger, candidate grammar, and prediction packet.
2. Separate builders from reviewers. The reviewer may identify defects but may not repair them in the same pass. This prevents the audit from unconsciously searching only for fixable errors.
3. Maintain a failure registry. Every killed branch remains visible. Deleted failures create survivorship bias.
4. Use negative controls. Wrong geometry, permuted labels, wrong state, wrong sign, wrong parity, wrong dimension, or deliberately perturbed parameters should fail in predictable ways.
5. Run ablations. Remove Shape, Granularity, State Selection, or a proposed global mechanism and measure what predictive structure disappears.
6. Charge branch search. If ten candidate mechanisms were tried and one worked, the selection cost belongs in the evidence assessment even if no continuous parameter was fitted.
7. Propagate uncertainties. A prediction with uncertain upstream anchors is a distribution, not a point. Same-ruler significance uses the full propagated covariance.
8. Independent code path. At least one decisive calculation should be reimplemented without sharing the original implementation details beyond the mathematical specification.
9. Prediction lockbox. For prospective tests, publish a cryptographic hash or timestamped artifact before the measurement is known.
10. Predefine stop rules. A branch-fatal contradiction causes rejection even if the rest of the paper is beautiful.

This protocol is intentionally asymmetric: it is easier to keep exploring a theory than to earn the right to call it correct.

Championship burden I — Mathematical existence is not enough; the selected physical state must be viable

The first reviewer burden is often underestimated because a beautiful action can make a model feel more complete than it is. A theory does not earn physical viability merely by writing a compact manifold, an effective action, or a stationary point. The actual state used by every downstream prediction must exist in the physical configuration space and be dynamically admissible.

The reviewer therefore asks for a chain that begins before any phenomenology:

The word physical matters. Gauge zero modes, coordinate artifacts, Lagrange-multiplier directions, and redundancies must be removed before interpreting signs of the Hessian. Conversely, a negative eigenvalue cannot be dismissed merely because it appears in a large unreduced matrix. The document must publish the reduction map that identifies the physical tangent space and then evaluate the second variation there.

At tree level, the minimum burden is straightforward. If Φ denotes the background fields and Φ̄ the selected solution, then the first variation must vanish on all allowed physical variations,

and the quadratic form

must have no uncontrolled negative physical direction for a claimed stable vacuum. If the theory only requires metastability, the negative or tunneling sector must instead yield a quantitatively acceptable lifetime from a specified decay/bounce calculation. “Perhaps loops fix it” is not a stability result.

The one-loop burden is also easy to state even when hard to execute. Every correction used to stabilize a tree-level saddle must come from the already admitted field content and regularization/renormalization grammar. The sign of the relevant curvature of the effective potential must be scheme-controlled to the precision needed for the claim. If the sign can be flipped by an un-fixed orientation bit, subtraction prescription, or renormalization condition, then the state is not yet predictively stable from frozen data.

A reviewer would also demand a domain statement. It may be enough for a compactification vacuum to be metastable on timescales enormously longer than the cosmological history of interest; absolute global stability is not always required. But the lifetime calculation must use the same physical potential and fields that define the rest of the theory. A nominally stable reduced toy potential cannot certify an instability that reappears when the omitted shape mode is restored.

This is why the present vacuum-stability item is not a bureaucratic residual. It sits at the very first championship burden. If the field configuration assumed by the particle and cosmology calculations is not a viable physical state, successes computed on that configuration do not rescue it.

Championship burden II — Forcedness must be proved before a precision mismatch can be interpreted

The second burden is to distinguish a falsified theory from a falsified ansatz. This is crucial because an honest program can otherwise oscillate between overclaiming and overreacting: first it calls a convenient relation “forced,” then when data disagree it calls the relation merely illustrative.

Before opening the target observable, the candidate grammar must be stated. Let C be the set of lawful constructions under the frozen Shape, Actor, Rulebook, and building-block constraints. A result O=O★ is genuinely forced only if every surviving candidate in the completed grammar yields the same output within the declared uncertainty:

If only one convenient ansatz c₀ has been evaluated, the correct label is “prediction of c₀,” not “prediction of the theory.” Exhaustion can be analytic—by a classification theorem—or computational, provided the search space, priors, discretization, and stopping rules are explicit. Merely failing to think of an alternative is not an exhaustion proof.

Once forcedness is established, same-ruler comparison becomes uncompromising. If the theory predicts a renormalized quantity at scale μ in a particular scheme, it must be evolved and projected to the same object reported experimentally. If mixing conventions, absolute values, marginalization choices, or correlations differ, the mismatch cannot be scored until the map is fixed. But once the ruler is genuinely the same, a large discrepancy is evidence.

The review should use the full covariance rather than isolated “sigma” arithmetic whenever observables are correlated. For a predicted vector μ and measured vector y,

is a cleaner measure than summing marginal pulls. The important principle is not the conventional 5σ number by itself. It is that a high-significance disagreement must not be downgraded after the fact unless the theory-level forcing statement was wrong. In that case the correct correction is to retract the forcedness claim, explain why the grammar was incomplete, and pay for the expanded grammar in the complexity ledger.

A prospective repair is scientifically legitimate. Suppose the one-angle flavor ansatz fails and an independent structural analysis—performed without reading the discrepant observable—shows that the frozen geometry necessarily supplies a second invariant phase. Adding that phase is not cheating merely because it repairs the data; it becomes a new theory version whose complexity is paid, whose target is removed from the prediction score, and whose other predictions are newly tested. What is forbidden is searching phase structures while watching |V_td| and then presenting the selected structure as though geometry uniquely demanded it.

This burden is what makes the ~14σ flavor item so decision-relevant: not because 14 is a magical number, but because a discrepancy that large forces us to settle the logical status of the one-angle mechanism. The correct goalpost is not “explain the tension”; it is “prove whether the tension belongs to an optional ansatz or to the frozen branch, then score it accordingly.”

Championship burden III — Parameter economy must be measured as effective flexibility, not marketing arithmetic

A unification theory is allowed to have parameters. The problem is not fitting; the problem is claiming explanatory compression when the real flexibility has been hidden in normalizations, branch searches, boundary functions, or convention choices.

The reviewer therefore asks for a complete generative ledger. Every quantity capable of changing the likelihood of an observable is typed before evaluation. Continuous parameters are obvious, but the less obvious flexibility can dominate: discrete geometry choices, candidate topologies, sign selections, operator bases, truncation orders, threshold normalizations, priors, and state-kernel functional forms.

A free function is especially dangerous. Writing w(z), K(k), or F(Φ) as one named object does not make it “one parameter.” If the function is represented by m independent basis coefficients over the range probed by data, its effective dimension is of order m, reduced only by genuine prior/regularization information that was fixed independently of the target. A Gaussian-process prior likewise carries an effective complexity determined by its kernel and hyperparameters; it is not free explanatory power.

Discrete search also costs evidence. If N roughly comparable branches were examined using the target and the best was reported, the evidentiary effect resembles a look-elsewhere factor. Exact Bayesian accounting would integrate over the prior branch weights,

rather than score only the winning b. A simpler project ledger can at least publish the number and nature of serious alternatives tried and prevent the winner from being described as uniquely forced unless the alternatives were independently eliminated.

The goalpost is therefore not “fewer parameters than the Standard Model” in a naive count. It is competitive predictive compression: after paying the effective flexibility that was actually used, does one parameter/structure package account for more independent data than reasonable benchmarks?

Information criteria can be useful diagnostics when regularity assumptions hold:

but the final paper should not fetishize either. For hierarchical or singular models, out-of-sample predictive log score or full marginal likelihood is safer. The principle is invariant: flexibility is paid before evidence is claimed.

This burden also prevents a common psychological error: “we only added one building block.” A building block that introduces an arbitrary function can be far more flexible than ten scalar parameters. The score follows the degrees of freedom, not the vocabulary.

Championship burden IV — The theory must be overconstrained after calibration

A calibrated theory can fit perfectly and still explain almost nothing. The championship standard begins only after calibration is over.

Let θ contain the paid cosmology- or particle-facing parameters, and let D_C be the calibration data. The theory is frozen after constructing p(θ|D_C). The held-out evidence D_H is then evaluated through the posterior predictive distribution, not by re-optimizing θ:

The simple rule “one datum per parameter” is a useful minimum firewall but not a complete statistical definition because data and parameters can be correlated. A better test is whether the held-out observables probe independent combinations of the theory. The singular values of the whitened Jacobian Σ^−1/2J show how many directions in observable space the parameters can readily move. If held-out residuals lie almost entirely in those same directions, the test is weak even if the plot contains hundreds of points.

Conversely, a small number of observables can be highly decisive if they test orthogonal structural consequences. Examples include a forbidden particle, a discrete charge ratio, a sign, a selection rule, a zero, or a spectral feature at a location fixed by an upstream scale. These can carry more evidentiary weight than a broad smooth curve whose amplitude and tilt were fitted.

The reviewer should also demand calibration swaps. If parameter A is calibrated to observable X and predicts Y, repeat the analysis calibrating to Y and predict X. The theory should remain coherent within propagated uncertainty. A model that works only for one privileged calibration convention may be hiding an identifiability problem.

Leave-one-domain-out tests are even stronger. Build the particle sector without cosmology, then ask cosmology. Or freeze the geometry/flavor construction without one precision observable and predict it. This is the closest legacy-data analogue of a truly prospective experiment.

This is the mathematical version of the sports metaphor: we do not count the warm-up points used to set the shooting sights. The scoreboard starts after the sights are locked.

Championship burden V — Cross-domain success must come from one mutually compatible branch

The word “unification” creates a special burden. It is not enough for one geometry variant to fit gauge structure, another to fit flavor, and a third to support cosmology. The same frozen branch must survive the interfaces between domains.

The reviewer therefore asks for a dependency graph rather than a list of papers. Nodes are derived objects—radii, spectra, couplings, vacuum expectation values, state kernels, background histories. Directed edges identify which object feeds which calculation. Every node has one authoritative value/distribution and one provenance class. If two papers silently use incompatible values of the same node, the theory has not unified those papers.

Cross-domain propagation is where the strongest evidence can appear because it denies the model opportunities to re-fit. A particle-physics scale that was fixed before cosmology and then determines an early-universe time is interesting precisely because cosmology did not set the scale. But its evidentiary label must preserve the particle-side calibration provenance. “Cross-domain consequence of an independently calibrated scale” is strong and honest; “parameter-free prediction” would be too strong if the scale was fitted upstream.

The same rule applies in the opposite direction. A cosmological state-selection principle that later constrains particle initial conditions could be powerful, but only if it was not introduced after a particle anomaly appeared. Cross-domain arrows must respect chronology.

Interface consistency also includes units, renormalization scales, frames, and effective descriptions. A 13D geometric parameter may compile to a 4D effective coupling only through a specified matching map. An object that is dimensionless in one normalization cannot be numerically compared to a different convention because both are called “mixing angles.” The assumption ledger’s same-ruler rule becomes an interface theorem here.

This burden is why the final evidence bundle should be judged jointly. Five successful papers that cannot coexist in one branch are not five pieces of evidence for one theory.

Championship burden VI — Cosmology requires a native background and a selected state, not merely a successful transfer calculation

Cosmology is the easiest place to accidentally move the goalposts because the downstream standard machinery is so powerful. If we import H(a), Ω values, and a primordial spectrum, then a Boltzmann solver can reproduce a great deal. That validates the solver, not the proposed origin theory.

The native cosmology burden has four logically separate layers:

1. Background: derive the homogeneous solution from the compactified parent/effective action and matter ledger.
2. Physical perturbations: reduce scalar/vector/tensor fluctuations to the gauge-invariant dynamical variables and their quadratic operators.
3. State selection: use a prospective boundary/admissibility principle to determine the quantum state or finite admissible class.
4. History transfer: propagate that background and state through thermal history to observables.

The background begins from a variational statement, schematically

with the Hamiltonian constraint and conservation laws satisfied. The theory must publish which stress-energy terms are genuinely derived and which are calibrated. If dark matter or vacuum energy is inserted as an arbitrary function merely to reproduce H(a), then the framework has not predicted the background.

The perturbation state is a distinct burden. A local boundary condition at finite time generally produces a kernel analytic in k² near k=0; the previous state-selection analysis showed why that cannot simply be declared to equal the required nonlocal primordial covariance. A local parent theory can nevertheless induce a nonlocal boundary kernel through its solved bulk history. The correct object is the on-shell/Dirichlet-to-Neumann Hessian. That route is valuable precisely because the nonlocality is derived from bulk propagation rather than inserted to match n_s.

Only after background and state are frozen should the theory compute primordial amplitude/tilt/running, tensor power, non-Gaussianity, isocurvature, BBN abundances, CMB spectra, BAO distances, matter power/growth, and age. Any observable consumed to calibrate a remaining scalar parameter is removed from the prediction score.

This goalpost finally stops the repeated pattern “we derived the next boundary, therefore cosmology is almost done.” Cosmology is done only when the complete generative chain exists.

Championship burden VII — The theory must compete with alternatives, not merely fit nature in isolation

A model can match data and still be weak evidence if many comparably flexible models match equally well. A reviewer therefore asks a comparative question: does this theory predictively compress the observations better than reasonable alternatives?

The benchmark depends on the domain. For low-energy particle physics, the Standard Model plus explicitly needed extensions is the obvious reference. For cosmology, a standard ΛCDM-like baseline and relevant extended models provide a practical comparison. The project does not need to defeat every benchmark numerically on every legacy point; a unification theory may trade a small local fit penalty for major cross-domain compression. But that trade must be quantified rather than asserted.

One clean approach is held-out expected log predictive density. Given independent or appropriately modeled data partitions, compare

Another is a Bayes factor when priors are defensible. Another is minimum description length if the coding rules are fixed before seeing which model wins. The project already values MDL-like economy, but the coding convention must not be selected to flatter the preferred geometry.

Qualitative predictions also matter. If the theory forbids a class of particles, fixes a discrete multiplicity, or predicts a sign that alternatives leave free, successful confirmation can be decisive even if a smooth likelihood comparison is difficult. Conversely, a theory that reproduces only quantities all reasonable models were built to fit gains little discrimination.

Benchmarking also disciplines the meaning of “parameter economy.” If the proposed theory has 13–14 charged reals and the benchmark has 25 under the chosen ledger, that may be meaningful compression; if the proposed theory additionally searched hundreds of structural branches or uses a flexible state function, the comparison must include that cost. The benchmark’s own structural assumptions must be charged under the same coding philosophy.

The championship is not won by showing that our team can score. We must show that the scoring is surprising relative to other teams playing under the same rules.

Championship burden VIII — Independent replay and a genuinely new prediction are the final proof-of-seriousness

Even a perfectly documented internal program can share blind spots. The final goalpost is therefore social as well as mathematical: can the decisive claims survive an implementation and review process that was not involved in inventing them?

Independent replay has several levels. At minimum, another agent or researcher should recompute the arithmetic and algebra from the same formulas. Stronger is an independent code implementation from a mathematical specification. Stronger still is an independent reconstruction of the derivation from the frozen source artifacts. The project should distinguish these rather than call all of them “independent validation.”

The reproducer should be given known-answer traps. A competent review must notice the live vacuum issue, the flavor tension, the threshold-provenance nuance, and the difference between gate closure and empirical success. If a reviewer reports “all clear” while missing known defects, that review has failed calibration.

The project should also freeze at least one prediction whose datum is genuinely absent or genuinely withheld. A future experiment is ideal, but a legacy dataset can function as a pseudo-prospective test if access is credibly firewalled and the branch was not historically selected using a close surrogate. The prediction packet should contain the distribution, not merely a point, and state what result would count as failure.

A novel prediction need not be spectacular. A small, precise ratio, sign, hierarchy, spectral feature, or cross-domain relationship can be more convincing than a dramatic but flexible forecast. What matters is that the theory had a real opportunity to be wrong and could not adapt after the fact.

The chronology should be public: commit hash, timestamp, artifact hash, allowed nuisance treatment, and data-release time. If the prediction later succeeds, the evidence is legible without asking the authors to remember what they knew.

If the theory reaches this point after the previous seven burdens, the rational belief update should be large. The evidence would no longer be primarily that we found a clever representation of known physics. It would be that a rigid structure continued to anticipate nature after we stopped helping it.

Championship burden IX — The theory must make a discriminating prediction, not just an accurate one

Championship burden X — The result must survive reasonable formulation, scheme, and implementation changes

Championship burden XI — A surprising cross-domain link must survive as one calculation

Championship burden XII — The decisive argument must be explainable without weakening it

The hostile reviewer’s ten objections—and what would answer each one

The purpose of listing objections in advance is to prevent us from answering only the ones for which we already have attractive material.

1. “You chose the geometry because it reproduces the Standard Model.”
   Answer required: disclose the candidate search and selection history; identify which observations were constitutive pins; score only held-out consequences as predictions.
2. “Your compactification scale is just the GUT scale put in by hand.”
   Answer required: exact provenance chain, chronology, calibration label, and downstream predictions that did not participate in the particle-scale fit.
3. “Your ‘few parameters’ claim ignores injected normalizations and branch choices.”
   Answer required: effective flexibility ledger and benchmark comparison under the same cost convention.
4. “Your gate board turns failures into green checkmarks.”
   Answer required: separate governance and truth scoreboards; CLOSED-NEGATIVE counts as a physical loss for the mechanism.
5. “The selected vacuum is a saddle.”
   Answer required: full physical Hessian plus controlled quantum correction/metastability calculation. No rhetorical answer substitutes.
6. “Flavor already falsifies you.”
   Answer required: forcedness audit of the one-angle ansatz, same-ruler replay, and prospective repair only if the ansatz is not core-unique.
7. “The primordial spectrum was inserted through a state choice.”
   Answer required: state selection derived from SSBA and parent-induced bulk/boundary dynamics before CMB comparison.
8. “Dark matter, baryogenesis, and Λ are just missing.”
   Answer required: either derive them, openly calibrate them and narrow the claim, or stop calling the current model a complete cosmology.
9. “With enough mathematical structure you can explain anything after the fact.”
   Answer required: frozen candidate grammar, held-out vector, negative controls, branch-search accounting, and novel prediction.
10. “No independent group has reproduced it.”
    Answer required: external replay package and an actual independent report—not another internal agent using the same hidden assumptions.

A theory that can answer all ten does not become infallible. But a reviewer would have far fewer ordinary explanations for why it appears to work.

How much success is enough? Avoiding a new arbitrary goalpost

It would defeat the purpose of this paper to replace vague goalposts with a magic number such as “five predictions” or “three domains.” There is no universal theorem that five successful observables prove a model. The amount of evidence depends on how independent, precise, surprising, and flexibly predicted those observables are.

Therefore the final standard is information-based rather than count-based. The model should produce a held-out predictive distribution whose success is unlikely under relevant alternatives and cannot be cheaply recovered by moving the model. The evidence grows when:

• observables probe independent directions;
• uncertainties are small relative to the theory’s allowed band;
• the prediction is structurally distinctive;
• the model had little flexibility after freeze;
• the target was genuinely unseen;
• the same fixed structure predicts multiple domains;
• reasonable alternatives assign lower predictive probability.

A single sharp sign prediction can therefore be more valuable than fifty fitted curve points. Conversely, a full CMB spectrum can be extremely informative if the primordial state/background and matter ledger were fixed upstream and only a small declared calibration set was used.

The reviewer’s qualitative threshold for “strong evidence” is consequently a conjunction rather than a count: no fatal contradictions + rigid prospective generative chain + successful overconstrained held-out likelihood + benchmark competitiveness + independent replay + at least one high-discrimination prediction.

This standard is deliberately difficult to game. If we later discover a new untested obligation, we may add it because physics demanded it, but we may not redefine an already failed burden so the old result becomes a pass. Version history must show the change and why it was unknowable earlier.

The skeptical-reviewer forced-choice theorem

This is the outcome we should design the evidence package to produce.

Sector finish line — Geometry and gauge structure

The geometry/gauge sector should not be judged by whether it can be made to resemble the Standard Model. It should be judged by how much of the observed low-energy gauge structure follows once the geometric/category choices and measured anchors are honestly separated.

A reviewer begins by fixing the claim. If the framework says “given the declared CSDR/isometry grammar and the selected internal factors, the low-energy gauge algebra is SU(3)×SU(2)×U(1) with the stated rank and carrier ownership,” then the burden is algebraic and finite. One must compute the isometry/centralizer structure, parities, zero modes, multiplicities, and extra-vector exclusions. A competing carrier that produces unwanted gauge factors is a meaningful negative control. This can be a strong structural result even if the category “forces are internal isometries” is itself a constitutive modeling choice rather than a theorem of nature.

If the stronger claim is “the geometry uniquely explains why nature has the Standard Model gauge group,” the burden becomes much higher. The candidate grammar of geometries and modeling categories must be specified, and rival categories capable of producing the same local interface must be considered. The project’s own gate taxonomy recognizes a category floor here. That is not a failure; it means the correct public claim is conditional on a modeling category. A reviewer will reject wording that silently turns a certified irreducible category posit into a derivation from nothing.

The same distinction applies to the compactification scale. Geometry may determine relations among radii once a scale is fixed, while the absolute scale can still inherit a particle-physics calibration. The reviewer wants a dependency graph showing which statements are topological/exact, which are dimensionless geometric consequences, and which acquire units only after measured anchors or fitted thresholds enter.

The gauge sector becomes convincing when it is both rigid and discriminating. Rigidity means reasonable deformations or alternative embeddings are either excluded by upstream constraints or visibly change predictions. Discrimination means the surviving structure predicts something not used to select it: a forbidden extra U(1), a precise multiplicity, a kernel quotient, a representation relation, a discrete parity pattern, or another nontrivial observable.

Negative controls matter here because group theory is rich enough that many attractive decompositions can be found after the fact. The project should deliberately test neighboring cosets, torus choices, parity assignments, and embeddings that are close enough to be plausible. If many produce the same successful interface, then the result is category-generic rather than Shape-specific. If the frozen Shape survives while close competitors fail for independent reasons, evidence for the chosen structure increases.

Importantly, this sector does not need to prove cosmology to count as a success. The no-moving-goalposts discipline works both ways: a real gauge theorem should not be discounted because cosmology remains open, just as a gauge theorem should not be promoted into evidence that cosmology is already solved.

Sector finish line — Flavor, masses, and mixing

Flavor is where reconstruction programs are most vulnerable to hidden flexibility because measured masses and mixing matrices provide many numerical targets and many possible parameterizations. The reviewer therefore treats this sector as a stress test of forcedness and parameter accounting.

The first requirement is a clear separation between structural data and continuous fitting data. Structural data include family multiplicity, chirality, allowed Yukawa texture zeros, representation selection rules, discrete symmetries, or geometric overlap classes. Continuous fitting data include normalizations, phases, moduli, and any coefficients adjusted to measured masses or angles. A texture that predicts three families from topology is not weakened merely because mass scales are calibrated; but the calibrated masses do not become independent evidence for the texture.

The second requirement is scale/scheme discipline. Quark masses, Yukawas, CKM elements, neutrino parameters, and phases are not interchangeable raw numbers. The calculation must identify the renormalization scale, running prescription, threshold treatment, basis convention, and uncertainty propagation. Same-ruler comparison is mandatory before calling a residual a sigma discrepancy.

The third requirement is a full-matrix test. A flavor model that uses |V_us| to set its lone angle cannot then advertise |V_us| as predicted. The relevant evidence is the rest of the CKM structure predicted after that calibration, including quantities that are geometrically or algebraically linked. If one element shows a large discrepancy, the reviewer asks whether the relation generating it was uniquely forced or merely one convenient low-dimensional ansatz.

A repair protocol must be prospective. If the one-angle construction fails, the next candidate class should be generated from the building-block logic without access to the offending target. For example, derive all independent phases/invariants permitted by the frozen actor geometry, reduce them by symmetries, and freeze the minimal lawful class. Only then open the withheld CKM element. If the new class contains an extra fitted real calibrated directly to the old discrepancy, the old datum pays for the new parameter and disappears from the evidence score.

Neutrino structure requires the same discipline. If mass splittings, ordering, absolute scale, or PMNS phases are partly measured anchors and partly tuned magnitudes, the ledger must say so. A family-count or chirality result can remain a strong topological prediction while the continuous neutrino sector remains calibrated.

The flavor championship is not “fit every mass.” With enough coefficients, almost any texture can. The championship is compressive overconstraint: a small, structurally motivated parameter set should force several independent ratios, signs, hierarchies, zeros, or mixing relations that survive data.

Sector finish line — Quantum consistency, state space, and boundary admissibility

Quantum consistency is not one gate. It is the condition that the classical/geometric construction actually defines a lawful quantum theory over the regime in which quantum claims are made.

A reviewer separates several burdens that are easy to blur. Gauge consistency asks whether anomalies cancel and whether gauge fixing/BRST/BV structures are coherent. Positivity/unitarity asks whether physical states have nonnegative norm and evolution preserves probabilities in the relevant domain. Reflection positivity or an equivalent reconstruction condition matters if Euclidean machinery is used to infer Lorentzian quantum theory. UV statements require care: a finite effective theory below a cutoff does not automatically prove a nonperturbative UV completion.

The project’s certified-irreducible terminals can honestly mark an external theorem or unresolved mathematical floor, but those terminals should not be scored as positive empirical evidence. If the framework relies on a named external theorem or constitutive quantum rule, it should be exposed as part of the axiomatic cost.

State Selection and Boundary Admissibility adds a separate burden. An admissible state class is not a selected state. Conditions such as gauge invariance, Hadamard behavior, positivity, finite operational action, regularity, and boundary consistency can remove pathological states while leaving an infinite family. The cosmological prediction is not unique until a state-selection principle—preferably a parent-induced variational or Dirichlet-to-Neumann construction—reduces that freedom to one state or a finite equivalence class.

The reviewer should test uniqueness by perturbing the admissibility rules. If several equally lawful states produce materially different primordial spectra, then the theory contains unresolved state freedom. That freedom must be either derived away or charged like a parameter/function. Calling all such states “admissible” does not make their common predictions unique if they have none.

Boundary terms deserve the same provenance scrutiny as bulk terms. A boundary functional chosen because its Hessian has the desired k-dependence is a cosmological fit. A boundary operator induced by integrating the solved bulk equations with fixed regularity conditions is a derivation. The two can be mathematically identical while carrying completely different evidentiary weight.

Sector finish line — Granularity, records, and the first-separability clock

The record/separability calculation is a good example of why the claim ladder matters. It can be a genuine theorem under declared assumptions without being a complete cosmology.

The narrow claim is operational: under the frozen finite-region record protocol, the compactification-controlled energy ceiling and quantum orthogonalization bound imply an earliest time at which a pair of finite records can both be completed and causally packed while their conditional influence lies below the chosen action cell. The result is an existence boundary for a pair, not universal Hilbert-space factorization and not disappearance of entanglement.

A reviewer checks three things. First, the scale provenance: M_KK must come from a branch fixed without cosmological target data. The current record supports that chronology, while also requiring the honest label that the absolute scale inherits particle-physics calibration. Second, the theorem conditions: support size, allowed record energy, action-cell quantizer, local influence envelope, and nonlinear/tower/global residuals must be explicit. Third, destructive controls: nearby supports or altered support diameter should move or destroy the crossing in the predicted manner.

The claim should also distinguish a lower bound from an equality. The Margolus–Levitin expression supplies a minimum orthogonalization time at a given energy above the ground state. Saturation by a physical record protocol is an additional assumption unless explicitly constructed. The current paper’s use as an earliest possible record boundary is defensible when labeled accordingly; it becomes overclaim if every physical record is said to complete exactly at that time.

Granularity itself must remain what the project defines it to be: operational distinguishability with a universal positive action/cost floor as a named axiom and measured ℏ as residue/anchor. It does not imply a universal minimum spacetime length. The compactification-derived support length is protocol-specific, not a fundamental pixel size of the universe.

The strongest future evidence from this sector would be cross-domain. If the same independently fixed microscopic scale later controls a novel particle, cosmological, or decoherence feature not used in its calibration, the record theorem becomes part of a larger predictive chain. But until that happens, its scientific value is the narrow theorem itself and the conceptual architecture it enables.

Sector finish line — Primordial correlations and the inflation-replacement claim

This sector has already taught the project an important lesson: a mechanism can be intuitively attractive and mathematically wrong for the target. The local-correlation branch lost because its infrared spectral class is wrong. That loss should remain permanent.

The next primordial claim must therefore be stronger than “pre-separation interdependence could correlate distant regions.” Large-scale correlation by itself is not enough. The theory must generate the amplitude, approximate scale invariance, mild red tilt, phase/coherence structure, acceptable tensor content, low non-Gaussianity, and acceptable isocurvature from one state/background calculation.

The correct generative chain is:

Every arrow is load-bearing. A state covariance may mathematically be chosen to give P_R(k) proportional to k^n_s−4; that proves only existence. The theory predicts the spectrum only if the covariance follows from the frozen bulk/boundary problem. Likewise, a background chosen to make the pump field yield the desired Hankel index is a fit unless the background already follows from the matter/geometry dynamics.

The reviewer will specifically look for target leakage through asymptotic assumptions. “Choose the regular vacuum” can be ambiguous when multiple regular/Hadamard states exist. “Minimum action” can conceal a boundary choice. “Scale invariance follows from equal information per log k” is not a derivation unless the equal-information rule itself is derived independently and its corrections produce the observed red tilt prospectively.

It is scientifically acceptable if the theory uses one or two calibrated primordial parameters. Many successful cosmological models do. But then those observables are calibration, and the evidence must come from the rest of the spectrum, tensor/non-Gaussian/isocurvature predictions, and downstream CMB/LSS structure.

If this route fails, that does not automatically falsify the particle-physics branch. It falsifies the claim that the current frozen framework already supplies a native alternative primordial mechanism. Again, the goalpost specifies the scope of the loss before the result arrives.

Sector finish line — Full cosmological history

A full cosmology claim is earned only when one native solution survives the entire history. It cannot be assembled from mutually independent epoch-by-epoch fits.

The minimum background ledger contains radiation, neutrinos, baryons, whatever supplies dark gravitational phenomenology, vacuum/late-acceleration contributions, and any geometric/moduli energy that is non-negligible. For each component the theory must state whether the abundance and equation of state are derived, calibrated, or externally anchored. Conservation/exchange equations must close.

Baryogenesis is not optional for a complete history because baryon density affects BBN and CMB. The theory need not derive the observed asymmetry from zero parameters; a calibrated efficiency can be allowed if openly paid. But a claim of native reconstruction is stronger if the sign and magnitude arise from admitted CP violation, anomaly/topology, and nonequilibrium history.

Dark matter is similarly more than Ω_c. The theory must provide perturbation behavior: clustering, sound speed, anisotropic stress, free-streaming or interaction scale, and gravitational coupling. A geometric effective component that mimics the background density but fails structure growth does not solve the dark sector.

The late-time vacuum sector must be treated with the same honesty. If Λ is a measured anchor under the current gate taxonomy, then the absolute value is not a prediction. The theory can still predict consequences conditional on Λ. If it later derives a dynamical vacuum law, that becomes a new theory version and must be tested prospectively against expansion data not used to choose the law.

The thermal-history calculation then solves the reaction/transport equations against the native H(T), with phase transitions and entropy changes included. BBN is a crucial firewall because it constrains expansion and particle content long before recombination. Recombination and the Boltzmann hierarchy then generate the CMB spectra. BAO, lensing, matter power, growth, and age are subsequent consequences of the same H(a) and perturbation system.

A reviewer will not require every nuisance astrophysical parameter to come from fundamental geometry. Foregrounds, calibration uncertainties, and instrument response can be marginalized as nuisance parameters. What cannot be marginalized as “nuisance” is a fundamental function introduced to make the cosmic history work.

Sector finish line — What would make “unification” more than a common language?

A theory can unify notation without unifying physics. The reviewer asks whether the same small set of structures creates constraints that propagate across domains in ways that would be unlikely if the domains were modeled independently.

True cross-domain load-bearing structure has three signatures. First, removing it should break multiple predictions at once. Second, the same object should appear through explicit matching maps rather than merely through a shared name. Third, it should create cross-domain relations that were not individually fitted.

For example, if a compactification radius determined from particle unification controls both a KK spectrum and an independently tested early-record boundary, that is a real cross-domain relation—conditional on the scale’s calibration provenance. If the same geometry also fixes a state-selection boundary or feature in a cosmological spectrum, the chain becomes stronger. By contrast, saying “Shape influences everything” without a quantitative dependency is not unification evidence.

Ablation is the strongest test. Remove or alter the 13D Shape while preserving enough ordinary 4D freedom to re-fit local observables. Does the distinctive cross-domain structure disappear? Remove Granularity: does the record/separability theorem or state-selection criterion lose its forcing? Remove SSBA: does the cosmology regain arbitrary state freedom? If an advertised building block can be removed without reducing held-out predictive performance, it is not load-bearing for that evidence.

The reviewer should also compare against modular alternatives. Suppose a conventional GUT plus an unrelated inflationary cosmology fits the same data with similar effective complexity. The unified framework gains evidence only if it reduces total description length, predicts cross-domain relationships, or succeeds on a distinctive held-out observable. Aesthetic unity alone does not beat modularity.

Conversely, the theory need not explain consciousness, every quantum foundation question, or all mathematics before it can count as a unification candidate in physics. Scope should be ambitious but finite: the claimed domain must be enumerated, and the evidence should be judged against that domain.

How I would update belief as the tournament progresses

The purpose of the goalposts is ultimately epistemic: what evidence should rationally make us believe more or less? We do not need a fake precise prior probability, but we do need asymmetric updates.

Some results should move belief only slightly. Recovering a structure that was used to choose the geometry is mostly a consistency check. Reproducing a calibration datum says the implementation can fit what it was told to fit. Showing that a mechanism is mathematically possible is a modest positive update because it removes an impossibility objection but does not show nature selected it.

Other results should move belief strongly downward. A core-forced precision contradiction after same-ruler replay is direct evidence against the branch. A physical vacuum instability is especially serious because it undercuts the background assumed by many downstream successes. A finding that the model’s impressive economy disappears when hidden functional freedom is counted would sharply weaken the unification case even if local fits remain.

The strongest upward updates come from rigidity plus surprise. A locked prediction that was not available during model development, especially one that distinguishes the theory from competitors, has high evidentiary weight. So does a cross-domain relation in which a quantity fixed in one domain predicts a structurally different observable elsewhere. Independent replication multiplies confidence because it attacks shared implementation error.

In Bayesian language, evidence is the likelihood ratio between the proposed theory and alternatives. We need not compute a single global Bayes factor to use the principle:

A result everyone expected under many models has a likelihood ratio near one and should not make us much more confident. A result that the frozen theory assigned high probability and reasonable alternatives assigned low probability can make us much more confident. A result the theory assigned tiny probability should reduce confidence even if our governance taxonomy handles the failure perfectly.

That principle is a powerful antidote to self-deception because it separates methodological virtue from physical likelihood. Reporting a falsifier honestly increases confidence in the method while decreasing confidence in the contradicted mechanism. Both updates can be true simultaneously.

The public-language goalposts

The final anti-goalpost device is linguistic. Public wording should be mechanically tied to the highest cleared claim level.

This table matters because rhetoric can move the psychological goalposts even when the technical ledger does not. If a paper headline says “resolved” and the reader hears “predicted,” the project can mislead without any false equation. The language should do the same accounting as the mathematics.

How repairs are allowed without turning falsification into an impossible game

A good theory-development program must be allowed to learn from failure. Otherwise every early mistake would permanently kill the entire research direction. But repair must be versioned so that learning is not confused with prior prediction.

When a test fails, the first action is classification: did it falsify an auxiliary ansatz, an Actor realization, a building block, or an immutable core statement? The classification must be based on documents that predate the failure where possible. If we redefine what was “core” only after seeing the result, the historical claim must be corrected.

A repair may then be generated from upstream constraints. The crucial rule is that the failed target becomes unavailable during repair selection. The repair agent can know the type of defect—e.g., “the flavor candidate class was incomplete”—but should not optimize against the numerical target that failed. Once a new lawful candidate is frozen, the old failed datum can be used as calibration if needed, but then it no longer counts as evidence for the new version.

The project should maintain theory versions T₀,T₁,T₂,… with explicit diffs: new parameter, new building block, new branch rule, removed claim, and changed prediction set. Evidence belongs to the version that made the prediction. A successful T₂ does not retroactively make T₀’s failed prediction successful.

This is not punitive. It is how scientific progress becomes legible. A research method that quickly converts failures into better constrained theories can be excellent even if early versions are wrong. The eventual theory becomes convincing when the repair rate slows because the frozen structure begins surviving new tests without change.

Formal same-ruler audit — proving that theory and experiment are actually the same object

Many apparent triumphs and apparent falsifications disappear when the compared quantities are not actually identical. Because this project spans dimensions, effective theories, renormalization scales, boundary observables, and multiple geometric representations, a same-ruler audit must be a formal object rather than a sentence in the discussion.

For every empirical comparison, define a theory-side object X_T in its native description and a measured object X_E. The comparison is lawful only if there is a published map

such that both sides land in the same comparison space X_cmp. The maps may include dimensional reduction, field redefinition, normalization, renormalization-group running, detector response, basis rotation, projection to gauge-invariant observables, or conversion from a theoretical correlation function to the statistic actually reported by an experiment.

The audit should answer at least seven questions. Dimension: are both quantities defined in the same effective dimension? Frame: Einstein/Jordan/comoving/physical or other frame? Scale: are running parameters compared at the same μ or transferred with a specified RGE? Scheme: MS-bar, on-shell, lattice, or another renormalization convention? Normalization: are fields, generators, mixing matrices, and Fourier transforms normalized identically? Projection: is the theory predicting a gauge-dependent component while the measurement reports a gauge-invariant combination? Statistical object: point estimate, posterior mean, profile likelihood, upper limit, or correlated data vector?

The maps themselves carry uncertainty. If y=R(x,η) depends on nuisance quantities η, the theory-side covariance receives propagated terms

A significance statement that ignores the mapping uncertainty can be too harsh or too flattering. Conversely, declaring a large discrepancy “possibly due to scheme” without quantifying the scheme sensitivity is not an uncertainty analysis; it is an escape hatch.

Same-ruler audits are especially important for quantities like quark masses, Yukawas, CKM/PMNS parameters, threshold corrections, cosmological power spectra, and vacuum terms. A compactification-scale parameter is not directly a collider observable; a primordial curvature spectrum is not directly a CMB C_ℓ; a microscopic correlation function is not directly an operational record. Each requires a transfer map.

The reviewer should insist that the map be frozen before scoring the discrepancy whenever multiple plausible conventions exist. If the convention is chosen after looking at which one makes the theory closest to data, that convention choice becomes part of the fit and must be charged.

There is a useful adversarial test: deliberately evaluate the comparison under adjacent reasonable rulers. The correct theory should not require an exceptional convention unless the geometry/dynamics uniquely selects it. If the result swings from 0.5σ to 10σ under conventions that the theory leaves free, then the prediction is not yet sharply defined.

Another test is invertibility at the claimed scope. If the measurement constrains only a projection of a larger theory object, the comparison cannot establish the full object. For example, matching one effective coupling combination does not validate every higher-dimensional component that projects to it. The paper should say “the projected observable agrees,” not “the underlying tensor is confirmed.”

This formalism protects the theory from unfair falsification and protects us from false confirmation. It is therefore neutral but essential officiating.

Uncertainty, truncation, and scheme dependence — a prediction is a distribution, not a favorite number

A reviewer will not accept a precision claim without a complete uncertainty budget. This is particularly important in a reconstruction program where exact topological results, numerical eigenproblems, EFT truncations, threshold sums, measured anchors, and uncomputed higher-order effects coexist.

The first discipline is to type uncertainty by origin. Measurement uncertainty enters through imported anchors. Numerical uncertainty comes from discretization, convergence tolerance, finite precision, or stochastic algorithms. Truncation uncertainty comes from neglected EFT operators, finite KK towers, perturbative order, or asymptotic expansions. Model-form uncertainty arises when multiple admissible realizations remain. Scheme uncertainty appears when a quantity depends on an incompletely fixed renormalization prescription. These categories should not be collapsed into one convenient error bar because they behave differently under future refinement.

If an observable O depends on inputs θ with covariance Σ_θ, linear propagation gives

but nonlinear dependence may require Monte Carlo or interval propagation. The important point is chronology: the uncertainty model must be specified before seeing whether a wider error bar would rescue the prediction. Retrofitting theory error until χ² becomes acceptable is another form of tuning.

For perturbative series, a practical estimate can use the size and pattern of omitted orders, but the method must be fixed across successes and failures. If two-loop terms are considered negligible for observables that agree and potentially huge for observables that disagree, the uncertainty policy is asymmetric. The reviewer will notice.

KK/tower calculations need a convergence certificate. If a finite truncation N is used, publish the N-dependence and a bound or empirical convergence model for the remainder. A result that stabilizes numerically over N but lacks a uniform analytic theorem can still be useful if the tested domain is finite and the residual is quantified; it should not be promoted to an infinite-tower theorem.

Scheme dependence is a special case. Physical observables should be scheme-independent after all orders, but truncated calculations are not. Varying the renormalization scale or subtraction scheme within a preregistered range is a legitimate sensitivity test. However, if the sign of a load-bearing physical Hessian or threshold correction remains undetermined across reasonable schemes, the theory has not made a sharp prediction at that order.

Discrete ambiguity is not a Gaussian error bar. If two orientations, parities, or branches are both lawful and yield opposite signs, the predictive object is a mixture or a branch class until another rule selects one. Averaging them into a central value can conceal that the theory has not chosen.

The reviewer also asks whether experimental systematics are treated symmetrically. A 14σ nominal discrepancy may be less if shared theory/measurement covariance was omitted; but it remains serious if the full covariance still gives a huge pull. The correct response is calculation, not intuition.

A theory earns credibility when its uncertainty bars are predictive too: repeated held-out data should fall inside them at roughly the advertised frequency. Calibration of uncertainty is itself a test.

Candidate-search accounting — how to pay for ideas tried before the winner appeared

A central challenge for any intensive theory-development program is that human and AI creativity can search enormous spaces. Even with zero continuous parameters, trying many structures against known data can create apparent miracles. The final evidence standard must therefore account for model search.

There are several kinds of search. Explicit enumeration tries a known finite list of geometries, representations, or kernels. Sequential repair modifies a building block after a failure. Prompt-level search asks multiple agents or conversations for candidate mechanisms. Implicit search occurs when a researcher uses domain knowledge to invent one candidate while already knowing the target. Only the first is easy to count, but all affect evidentiary interpretation.

The project does not need to assign a precise “number of thoughts” penalty. It does need to distinguish discovery evidence from post-freeze prediction evidence. Results used during search are primarily selection data. Once a branch is frozen, new held-out results become prediction data. This simple chronology removes much of the need for speculative search penalties.

For legacy observables that inevitably guided development, a model-comparison analysis can still charge branch multiplicity. If a finite set B of serious candidates was searched, the model evidence marginalizes over them rather than conditioning on the winner. If the set is open-ended, minimum-description-length or explicit coding of structural choices can provide a conservative complexity estimate.

Sequential building-block evolution should be versioned. Suppose a gate failure reveals a missing commutator-closure constraint. Adding that reusable constraint may be a genuine methodological advance. But the gate that revealed it becomes training data for the new block. Evidence for the new block comes from later gates/problems that close without further target-specific repair. This is exactly analogous to training and test sets in machine learning.

The same principle applies to the new State Selection and Boundary Admissibility block. Its existence was motivated by a cosmology failure: the frozen theory did not select a unique global state kernel. Therefore the failure that motivated SSBA cannot also be counted as independent evidence that SSBA predicted the right kernel. The next independent application is the real test of whether SSBA is a reusable physical principle rather than a bespoke repair.

A reviewer will also ask whether failures are archived. If only successful candidate documents survive, the historical search appears much narrower than it was. The failure registry is therefore evidence discipline, not embarrassment. It allows future model-selection analyses to understand how much flexibility the process exercised.

This goalpost does not punish creativity. It simply locates where creativity ends and prediction begins.

The prospective prediction protocol — exactly how we should run the next championship test

A final reviewer document should not merely say “make predictions.” It should specify the operational sequence so a future success cannot be retrospectively reinterpreted.

Step 1 — Choose the claim before the target

Write one sentence describing the prediction and its scope. Example structure: “Given frozen theory version T and calibration set C, observable vector Y in domain D is predicted to follow distribution P with no further structural adjustment.” The claim includes what will not be predicted.

Step 2 — Freeze the candidate grammar

List admissible branches, discrete signs, operator bases, truncation rules, and the algorithm that selects among them. If the grammar is not closed, label the test exploratory rather than confirmatory.

Step 3 — Freeze inputs and calibration

Publish all measured anchors and fitted parameters. Each fitted parameter is paired with the calibration datum or likelihood block that pays for it. The calibration data are tagged so they cannot later appear in the prediction score.

Step 4 — Freeze the computation

Hash code, solver settings, random seeds where relevant, convergence tolerances, numerical libraries, and the output schema. For stochastic or posterior predictions, specify the distribution and sampling procedure.

Step 5 — Freeze the same-ruler map

State exactly how theory output becomes the measured statistic, including scales, units, frame, projection, selection effects, and nuisance marginalization.

Step 6 — Freeze uncertainty and falsifiers

State the predicted interval/distribution and the decision rule. The falsifier should be finite and decidable. If multiple outcomes correspond to local vs branch-fatal losses, state the classification rules.

Step 7 — Timestamp the packet

Use a public repository commit, notarized hash, preprint timestamp, or another independent archive. The critical property is that later readers can verify the packet predated target access.

Step 8 — Open the target once

Run the frozen comparison without repair. Record the result even if it fails. If a software bug is discovered, document the bug and rerun under a versioned correction; do not change physics choices in the same patch.

Step 9 — Score against benchmarks

Compute the same held-out likelihood or qualitative discrimination for relevant alternatives. Use the same data and uncertainty treatment.

Step 10 — Only then begin repair

If the theory loses, archive the verdict. A new theory version may be developed, but the failed target becomes part of the training history and cannot be recycled as a clean prediction.

The joint cross-domain score — one scoreboard for the whole theory

The final temptation to move goalposts appears when each domain is scored independently. A branch may accumulate many local passes while hiding that different sectors use incompatible parameters, or that one severe loss is being diluted by many easy wins. The theory therefore needs a joint scoreboard with veto conditions.

Let D₁,…,D_m denote domain-level held-out data blocks—precision particle physics, flavor, record/quantum tests, BBN, CMB, BAO/LSS, and prospective novel data. If the conditional independence structure is known, a joint predictive likelihood factors accordingly; otherwise the full covariance/hierarchical model should be used:

The exact scalar score is less important than preserving three rules. No refitting between domains: shared parameters have one posterior/distribution. No averaging away fatal contradictions: a branch-level veto is applied before summing log scores. No double counting: correlated measurements or derived summaries of the same data are not treated as independent wins.

The scoreboard can therefore have two layers. Layer A contains veto tests: physical stability, anomaly/unitarity consistency, and any confirmed core-forced empirical contradiction. Failure of a veto rejects the frozen branch regardless of how well other curves fit. Layer B contains comparative predictive evidence for branches that survive Layer A.

This resembles sports playoffs better than a season point total. A team cannot compensate for fielding an ineligible player by scoring extra goals; eligibility is a prerequisite. Likewise, an unstable physical background is not a −10 point penalty that can be offset by ten minor numerical matches.

For surviving branches, report domain-by-domain predictive scores as well as the joint score. This prevents a good average from hiding systematic weakness in one domain. A theory that is slightly worse than benchmarks in several legacy domains but dramatically more compressive across all of them may still be interesting; the trade must be explicit.

The joint scoreboard should include model complexity only once. Shared structure that predicts multiple domains earns compression because the same parameter is not repaid separately. Domain-specific free functions are charged in their domains and weaken the unification evidence.

Prediction correlations created by shared upstream anchors should be propagated. If uncertainty in M_U shifts both a particle threshold and an early-universe clock, those predictions are correlated and should not be counted as two fully independent pieces of evidence. On the other hand, a successful cross-domain relation can be powerful precisely because the correlation itself is predicted.

A simulated final referee report: what I would write if the manuscript arrived today

It is useful to state the likely review in plain language, because formalism can obscure the actual evidentiary position.

Strengths. The program is unusually explicit about assumptions, provenance, falsifiers, and negative controls. The gate methodology has demonstrated the ability to find and preserve failures instead of smoothing them away. Several structural calculations are nontrivial and reproducible in principle. The record/separability work is carefully scoped and the local primordial-correlation no-go is a genuine negative result. The recent State Selection and Boundary Admissibility block addresses a real missing logical step rather than simply inserting a cosmological covariance.

Primary concern 1: physical vacuum. The selected Shape-doublet direction is presently recorded as a live falsifier/saddle with no frozen-data stabilization certificate. Because the same selected background underlies much downstream calculation, this is not a minor appendix issue. The manuscript should not claim a physically realized unified branch until the physical Hessian/metastability problem is resolved.

Primary concern 2: precision flavor. The ~14σ |V_td| discrepancy must be adjudicated at the level of forcedness. If the one-angle relation is a uniquely forced consequence of the frozen core, this is strong evidence against the branch. If it is an auxiliary ansatz, the manuscript should retract the theory-level forcedness claim, retire the ansatz, and test a prospectively derived replacement.

Primary concern 3: parameter/search accounting. The project has corrected earlier optimistic economy claims and now charges substantially more injected reals. This correction is a strength, but the final evidence assessment must also include discrete branch search and any functional freedom introduced by state/cosmology sectors.

Primary concern 4: cosmological generative closure. The current framework does not yet supply a complete native background, matter ledger, and uniquely selected primordial state. Downstream cosmological transfer calculations using imported parameters are useful controls, not evidence that the cosmology has been derived.

Primary concern 5: independent evidence. Much of the work is internally generated and reviewed. The next evidentiary step should prioritize independent replay and a locked prediction rather than additional internally compatible reconstructions.

Recommendation today. Continue the research program; do not publish the strongest “theory is correct” interpretation yet. Publish the method, structural results, negative results, and precise open/falsifier ledger. Treat the vacuum and flavor questions as decision gates. If those clear, execute the native background/state calculation under a predeclared prediction protocol. A successful cross-domain held-out package would warrant a major reassessment.

What would change the recommendation. A frozen-data vacuum stability certificate, prospective resolution of the flavor forcing chain, native state/background derivation, and successful independent held-out prediction would move the manuscript from “interesting rigorous program” to “serious predictive unification candidate.” Independent replication plus a novel successful prediction would justify substantially stronger language.

The hard-stop constitution — results that end the current branch rather than create another appendix

The most important way to keep goalposts fixed is to define, before the next calculation, which outcomes are genuine stop conditions. Otherwise every failure can be redescribed as “one more thing to solve,” and the theory becomes impossible to falsify in practice.

A hard stop is not triggered by every failed submodel. It is triggered when the failure reaches an object the current branch has declared immutable or uniquely forced. The burden is therefore two-part: first establish the logical ownership of the prediction, then score the evidence.

Hard stop H1 — no physical background

If the selected compactification/background has a reproducible physical instability and no stabilization follows from already admitted dynamics under a preregistered completion, the current branch is rejected. A later theory version may add new physics, but it is a different branch and the prior downstream predictions become conditional explorations on an unrealized background.

Hard stop H2 — core-forced precision contradiction

If an observable is proven invariant across the complete admissible candidate grammar and disagrees decisively with a reliable same-ruler measurement, the branch is rejected. The discrepancy need not be exactly “5σ”; the operative standard is decisive predictive incompatibility after theory and measurement covariance are included. A ~14σ result, if truly core-forced and independently replayed, is well beyond the point where wording changes are an adequate response.

Hard stop H3 — mathematical inconsistency in the claimed quantum regime

An uncancelled gauge anomaly, physical negative norm, non-unitary evolution where unitarity is claimed, or contradiction in the constraint algebra is fatal to that claimed regime unless an already-declared completion supplies the missing term. A certified external theorem dependency is not the same as an inconsistency; the distinction must remain explicit.

Hard stop H4 — prediction requires unrestricted target-dependent functional freedom

If a flagship prediction can only be recovered by introducing an arbitrary function whose form is selected from the target data, the native-prediction claim is rejected. The resulting model may remain a phenomenological fit, but the explanatory mechanism has failed its stated burden.

Hard stop H5 — cross-domain incompatibility

If the particle sector and cosmology require mutually incompatible values of a supposedly shared frozen quantity, or require incompatible discrete branches with no lawful superselection interpretation, the unification claim fails. One can retain the separate sector models, but not the assertion that one branch explains both.

Hard stop H6 — decisive prospective failure after full freeze

If a high-discrimination prospective prediction is locked under this constitution and later fails decisively, the exact claimed mechanism loses. Repeatedly issuing new predictions from repaired versions is legitimate research, but the failed version’s evidence cannot be erased. A mature theory should eventually survive without structural repair.

Some outcomes are explicitly not hard stops. A fitted parameter is not fatal if openly paid. Failure to derive a measured anchor is not fatal if the claim is conditional on that anchor. An optional ansatz can fail without killing the core if optionality was documented before failure. An external mathematical wall can limit scope without making the model internally inconsistent.

This constitution is intentionally uncomfortable. Its value is that future success becomes more meaningful because failure was genuinely possible.

The chronology standard — proving that a prediction existed before the answer

For this project, provenance is not merely “where did the number come from?” It is also “when did each choice become fixed relative to the data it later explains?” Chronology is the strongest defense against unconscious circularity.

The reviewer should reconstruct a time-directed graph. Every node is a document version, parameter value, branch choice, building block, or prediction. Every edge says “this later object was allowed to depend on that earlier object.” Data-release nodes are included. A valid prospective prediction has no directed path from the target datum to the theory choices that generated the prediction.

This is stronger than merely saying “we did not consciously fit it.” If a close surrogate observable was used to select the branch, the target may not be fully independent even if its exact number was hidden. The paper should therefore list proxy leakage: observables or qualitative facts that strongly constrain the same theory direction.

There are several evidence grades:

• Grade C0 — constitutive: the datum directly defines or pins the category/anchor. It carries no predictive score.
• Grade C1 — calibrated: the datum determines a continuous/discrete parameter. It carries no predictive score for that parameterized version.
• Grade C2 — historically visible but withheld in current execution: useful as a robustness test, but weaker because human model design may already encode it.
• Grade C3 — genuinely firewalled legacy datum: neither builders nor selection agents access the datum or close proxies during branch construction.
• Grade C4 — prospective datum: prediction packet predates the measurement/data release. Highest evidentiary grade.

The compactification-scale/cosmology example illustrates why this matters. The absolute scale appears in particle-physics work before the early-universe papers, which protects the cosmology result from direct cosmological target tuning. But the scale’s particle-side calibration provenance remains part of the claim. Chronology prevents one kind of circularity; parameter classification prevents another.

Building blocks also need chronology. If a new block is invented because Gate X failed, then Gate X is training evidence for the block. The next unrelated gate is the block’s first clean validation opportunity. This is especially important for SSBA: it was motivated by failure to derive a unique primordial kernel, so cosmological state selection cannot itself be counted as an independent proof that SSBA was “predicted from first principles.” Its value will rise when it solves a separate boundary/state problem without target-specific modification.

Git history or timestamped archives are ideal because they make the chronology externally inspectable. If only local file timestamps exist, they are weaker evidence. The final publication packet should include commit hashes for every decisive pre-target freeze. If a historical commit does not exist, the paper should say so rather than fabricate certainty from later reconstructed provenance.

A reviewer may also request a blind replay. Give a new agent the pre-target artifacts only, exclude the target and downstream documents, and ask it to regenerate the prediction. If it cannot, the historical chain may depend on undocumented context. If it can, confidence in the provenance increases.

The master reviewer matrix — claim, proof, falsifier, repair, and promotion

This matrix is the practical endpoint of the document. It is the place we should look before saying “we crossed the goalpost.” Each major claim has one proof burden, one falsifier class, one allowed repair rule, and one promotion condition.

The repair column is as important as the falsifier column. It prevents two opposite mistakes: killing the entire research program because an auxiliary construction failed, and protecting the current frozen branch by declaring every failure auxiliary after the fact.

The promotion column also fixes the recurrent “we are done” problem. Clearing the proof burden establishes the scoped claim in column one. Promotion requires a new kind of evidence. For example, deriving a lawful state does not automatically promote to “correct primordial cosmology”; the state’s spectrum must survive data. Deriving H(a) does not automatically promote to “complete universe”; the same history must survive BBN, CMB, BAO, growth, and age.

Tournament rules v2 — a story humans can follow without lowering the physics bar

The sports narrative is not decoration. It gives the reader a memory structure for a complicated proof burden.

The championship game plan — the exact sequence that maximizes information and minimizes self-deception

Once the goalposts are fixed, the project should choose calculations by expected decision value rather than excitement. The best next calculation is the one most likely to tell us whether the frozen branch deserves to live, not the one most likely to produce another attractive number.

Phase 1 — Eligibility: settle whether the current branch is allowed to take the field

The first two tests are the vacuum and flavor red zones because they can invalidate the present branch before we invest in full cosmology. For vacuum stability, construct the full physical fluctuation operator around the selected compactification, identify gauge/symmetry zero modes, compute the relevant one-loop/Casimir/Coleman-Weinberg contribution with all currently admitted fields, and quantify scheme/orientation ambiguity. The output is not “a plausible stabilizing sign”; it is a spectrum or metastability distribution with a preregistered verdict. If no frozen completion can remove a physical negative mode, reject the current branch.

For flavor, freeze a forcedness audit before touching the discrepant target. Enumerate the exact assumptions that collapse the mixing structure to the one-angle ansatz. Determine which are immutable consequences of Shape/Actors and which are model conveniences. If the one-angle form is not uniquely forced, demote it before repair and construct the next candidate class from upstream invariants while withholding |V_td|. If it is uniquely forced, replay the same-ruler prediction with full covariance; a persistent decisive mismatch is branch-fatal.

These tests should be performed by separate execution and review agents. The reviewer is prohibited from suggesting the repair. The original and reviewed outputs are both archived. This matters because the psychological pressure to preserve the branch is highest at exactly these two tests.

Phase 2 — Generative closure: make cosmology emerge rather than enter as input

If the branch remains eligible, solve the native 4D effective cosmological problem from the compactified parent action. The deliverable is a background solution Φ̄(a) with its Hamiltonian constraint, conserved/exchange currents, and complete source ledger. Any matter component not derived is explicitly marked calibrated or external. Do not open CMB/BAO targets while choosing the background ansatz beyond facts already designated as constitutive.

Next perform the physical perturbation reduction and apply State Selection and Boundary Admissibility. The key output is the parent-induced boundary/on-shell Hessian or Dirichlet-to-Neumann operator that fixes the Gaussian state kernel—or proves that the theory does not uniquely select one. If a finite state ambiguity remains, publish the discrete class and its predicted spread. If an arbitrary functional ambiguity remains, the native primordial claim is not closed and the theory should not be tuned against n_s under the guise of “selection.”

Only then generate the primordial scalar/tensor spectra, non-Gaussian/isocurvature predictions, and any distinctive features. The prediction packet is frozen and hashed before comparison. If one or two scalar parameters must be calibrated, declare the calibration observables in advance and remove them from the evidence score.

Phase 3 — History: force one solution through the universe

Complete the baryon and dark-sector ledgers. Baryogenesis may emerge natively or may require a paid efficiency parameter; both are scientifically legitimate but represent different claim levels. The dark sector must specify perturbation behavior, not just a density. Vacuum/late acceleration must likewise be typed as derived, calibrated, or measured-anchor. The goal is one stress-energy and interaction system that closes the background and perturbation equations.

Run thermal history without epoch-by-epoch retuning. Electroweak/QCD transitions, neutrino decoupling, entropy transfer, and BBN reaction networks should use the same H(T). The predicted abundance vector is compared with data after any declared baryon calibration is removed. Then recombination and the Boltzmann hierarchy generate TT/TE/EE/lensing spectra and the matter transfer function. BAO, growth, weak lensing, and age follow from the same history.

The scoring packet should contain domain-level and joint likelihoods, benchmark comparisons, and residual plots—but also the parameter Jacobian/Fisher structure so a reviewer can see whether many plotted points represent genuinely independent constraints. Calibration swaps and leave-one-domain-out replays are run before any model repair.

Phase 4 — The real final: predict something we did not build around

If the theory survives the legacy evidence, stop modifying the core and choose a high-discrimination prospective target. The best target is one for which the framework makes a relatively sharp prediction while reasonable alternatives allow a broader range or a different qualitative result. It need not be dramatic. A precise relation among particle observables, a sign/hierarchy, a spectral feature tied to the compactification scale, or a cross-domain correlation can be ideal.

Publish the prediction distribution, same-ruler map, nuisance treatment, and fatal threshold before the relevant data are available. Ask an external reviewer to verify the packet before the outcome. When the measurement arrives, score it once. A success at this stage is qualitatively different from a legacy reconstruction because the theory had no opportunity to learn the answer.

Decision checkpoints

Checkpoint A: after Phase 1, either the current branch is physically eligible or it is rejected. Do not proceed to Phase 2 for evidentiary purposes if eligibility fails; any further work is exploration of a replacement version.

Checkpoint B: after Phase 2, either the theory has a native background/state generative system or its cosmology remains phenomenological/conditional. Do not call a tuned kernel a derived primordial theory.

Checkpoint C: after Phase 3, either one branch provides a competitive overconstrained cross-domain history or the unification claim is downgraded. Local successes may remain publishable.

Checkpoint D: after Phase 4 and independent replay, a successful distinctive prediction upgrades the rational belief status from “serious candidate” toward “strong evidence of real underlying structure.” A failure is recorded against the frozen version and triggers a new-version decision, not a moved goalpost.

Evidence replay: put the current branch against the new goalposts

The following rounds reuse the existing decisive tournament, but each is now preceded by the reviewer goalpost it is actually supposed to clear. This prevents a local result from masquerading as a global finish.

The old tournament remains useful evidence. What changes here is the officiating: each result is scored against the claim ladder and the final eight-part championship burden.

The reviewer’s current scorecard

On the evidence presently recorded, the research method is ahead of the physical theory. That is not an insult; it is what a falsification-capable method should sometimes report.

The order of operations: where the goalposts say we should spend effort

Once the finish line is fixed, the priority order changes. We should not work on the most exciting calculation; we should work on the calculation with the highest decision value.

1. Vacuum stability first. Compute the full physical Hessian—including admissible loop/Casimir completion, gauge-zero-mode removal, scheme dependence, and uncertainty—to decide whether the selected Shape background actually exists as a viable state. A branch that cannot sit where the rest of its calculations assume it sits should not be sent deeper into cosmology.
2. Flavor falsifier second. Determine whether |V_td| is genuinely forced by immutable structure or by an auxiliary one-angle ansatz. If forced and the same-ruler discrepancy survives independent replay, reject the branch. If auxiliary, retire the ansatz and prospectively select the next lawful flavor mechanism before re-opening the target.
3. Only then run native background + state selection. Solve the 4D effective cosmological saddle, derive the physical perturbation operator, and obtain the boundary kernel from the on-shell/Dirichlet-to-Neumann map. This is the correct test of whether the cosmology emerges rather than being inserted.
4. Then fill the cosmological matter ledger. Baryogenesis and dark-sector stress-energy must be derived or explicitly calibrated before a full H(a) claim is scored.
5. Finally run the blinded end-to-end cosmology. BBN, CMB, BAO, growth, age, and any novel feature are opened only after the upstream packet is frozen.

What would actually convince me?

If the following sequence happened without moving these goalposts, my assessment would change materially.

1. The full physical vacuum calculation removes the current live stability falsifier from frozen ingredients, or shows a controlled metastable lifetime far exceeding the required physical history.
2. The flavor discrepancy is either prospectively repaired by a lawful mechanism selected without |V_td| target access, or demonstrated not to be a forced output of the core branch. The resulting full CKM/PMNS package survives a precision replay.
3. The native cosmological saddle exists and the State Selection and Boundary Admissibility rules select a unique or finitely classified physical state without CMB input.
4. The induced primordial spectrum has the correct qualitative class—near scale invariance, mild red tilt, acceptable tensors/non-Gaussianity/isocurvature—before those data are opened, with any calibration observables explicitly removed from the prediction score.
5. The same branch yields a viable baryon ledger, dark-sector gravitational behavior, and thermal history, then predicts BBN, CMB peak structure, BAO/growth, and age without mutually inconsistent re-tuning.
6. The joint held-out predictive score is competitive with or better than reasonable benchmark models after paying the theory’s effective complexity.
7. At least one genuinely distinguishing prediction is locked before a new or genuinely unused datum and succeeds.
8. An independent reviewer replays the decisive chain and reaches the same conclusion.

If all eight happened, I would no longer describe the project merely as an interesting reconstruction program. I would describe it as strong evidence that the framework is capturing real physical structure. That is the correct finish line. Anything less should be labeled by the lower rung it actually clears.

What would not convince the reviewer—even if it looks impressive

• “All gates are resolved.” Governance completeness is not empirical truth.
• “The model reproduces many known numbers.” Not unless the parameter/branch search cost and target visibility are accounted for.
• “The required mechanism exists mathematically.” Existence is weaker than selection.
• “A calibrated model matches the calibration data.” That is what calibration means.
• “The theory has fewer named parameters.” Not if a free function, branch grammar, or hidden normalization carries equivalent flexibility.
• “The discrepancy is a certified falsifier and therefore a strength.” Honest reporting is a methodological strength; the contradicted physical mechanism still loses.
• “The cosmology works if we assume the primordial kernel/background.” That may validate a downstream transfer engine, but it does not establish a native cosmology.
• “No one has disproved it.” Absence of disproof is not positive evidence.
• “The framework is elegant or unifies language.” Elegance is a prior preference, not a likelihood.

The NASA-facing championship evidence packet

If we ever send the strongest version of this work to NASA or another technically demanding external group, the package should be designed so that they can audit it without contacting us.

Required packet contents

1. One-page simple brief. Premise, claim, what was fitted, what was predicted, what would falsify it.
2. Technical claim ledger. One row per claim with exact equation, scope, status, provenance, and witness hash.
3. Frozen theory manifest. Geometry, Actors, Rulebook, boundary/state rules, and version hashes.
4. Complete flexibility ledger. Scalars, functions, branches, priors, search grammar, rejected candidates, and calibration observables.
5. Reproducible computation. Source, dependencies, seed policy, configuration, numerical tolerances, and deterministic/repeated-run expectations.
6. Blind prediction packet. Timestamped outputs and scoring rule created before target access.
7. Benchmark packet. Reasonable competing models run through the same observation and nuisance pipeline.
8. Independent replay. Separate implementation or team, including discrepancies and adjudication.
9. Failure registry. Every known red zone, closed-negative branch, and repair history.

The NASA open-science references above motivate transparency and reproducibility; this championship packet is intentionally stricter than those general process requirements because the scientific claim is unusually ambitious.

The prediction packet we should freeze before the next major test

Every decisive future calculation should produce a small machine-readable packet before target comparison:

{
  "theory_manifest_hash": "...",
  "code_hash": "...",
  "branch_id": "...",
  "measured_anchors": [...],
  "fitted_parameters": [{"name":"...","calibration_datum":"..."}],
  "discrete_choices": [...],
  "external_theorems_or_floors": [...],
  "prediction_observables": [...],
  "prediction_values_or_distributions": [...],
  "same_ruler_map": {...},
  "uncertainty_model": {...},
  "negative_controls": [...],
  "fatal_falsifiers": [...],
  "timestamp_or_commit": "..."
}

The packet is more important than another hundred pages of prose because it makes later historical revision difficult. The prose explains; the packet proves what was frozen.

Independent reviewer protocol

The project’s dossier-review agent already contains a valuable discipline: a reviewer finds defects but does not repair them. The theory-level review should extend that separation.

1. Give the reviewer the frozen theory manifest, decisive dossiers, evidence artifacts, code, and prediction packet.
2. Require arithmetic/algebra replay, internal consistency, cross-document consistency, evidence provenance, falsifier quality, registry completeness, and claim-evidence proportionality.
3. Add theory-level checks absent from a single-gate review: parameter-search accounting, branch-fatal classification, benchmark comparison, held-out predictive score, and cross-domain mutual compatibility.
4. Run known-answer tests: the reviewer must catch the vacuum-stability red flag, the flavor tension, and the distinction between governance closure and truth. A reviewer that misses known defects is not trusted on unknown defects.
5. Publish the defect report before repair. Then repair in a separate process and rerun the same reviewer.

Independent review does not mean the reviewer must agree with every modeling choice. It means the factual chain from declared choices to claimed consequences is reproducible and the claim strength is not inflated.

The no-moving-goalposts decision tree

Once this tree is published, a future result changes the score, not the tree. That is the point.

The only claim worth fighting for

If we clear this edition’s goalposts, the strongest defensible public statement is not “we proved everyone else wrong.” It is stronger scientifically: