# Global Geometry II — U2 evaluation checkpoint report

**Checkpoint date:** 2026-08-30

**Overall status:** `UNRESOLVED`

**Decision authority:** `NONE`
**Physical validation:** `NOT_RUN`

## Executive verdict

The current work is shareable as a content-addressed **partial evaluation
checkpoint**, not as a completed confirmatory evaluation of U2.

Two independent hosts completed the bounded screening-v3 census and a strict
partial campaign.  Those executions substantially improve the evidence for
the implementation, validation, failure semantics, and portability of the
research system.  They do not establish the preregistered universality
candidate.  The strict partial ledger derives Gate 1 as `PASS` and Gates 2--8
as `UNRESOLVED`; the full confirmatory campaign is `NOT_RUN`.  U2 is therefore
neither accepted nor rejected.

This report keeps three states separate:

1. the original finite preview remains `NOT_EVALUATED` and has no decision
   authority;
2. the evaluation-v2 checkpoint is `UNRESOLVED` and has no decision authority;
3. the full confirmatory U2 evaluation remains `NOT_RUN`.

The machine-readable source of truth is the
[evaluation-v2 index](artifacts/global-geometry-ii/u2/evaluation-v2/evaluation-index-v2.json).
Digest values are intentionally not copied into this prose: the addressed
index and comparison records bind the exact files without creating a second,
manually synchronized hash ledger.

## 1. Evidence ladder

| Layer | Executed state | What it establishes | What it does not establish |
|---|---|---|---|
| Finite preview | `NOT_EVALUATED` | Deterministic bounded examples and measurement plumbing | Any U2 gate, scaling law, or universality decision |
| Screening v3 | `COMPLETED_SCREENING`; 3,840/3,840 records on each host | Complete execution of the bounded screen, fresh semantic validation, and explicit failure records | Preregistered decision estimators, robustness ensembles, Gate 8, acceptance, or rejection |
| Strict partial | `COMPLETED_PARTIAL`; U2 `UNRESOLVED` | Exact Gate 1 fixtures and partial finite diagnostic profiles under the strict source boundary | Gates 2--8, a completed perturbation layer, or a confirmatory U2 conclusion |
| Full confirmatory | `NOT_RUN` | Nothing outcome-bearing | A preregistered U2 decision |
| Passive-sheet dry preflight | declared digital dry checks `PASS` | Analytic and rendered A4 fit inside the sanitized imaging envelope | Printer output, scale on paper, cutting, assembly, material response, or camera reconstruction |
| Physical and camera validation | `NOT_RUN` | Nothing empirical | Any physical-shape, material, sensing, actuation, or safety claim |

Screening and convenience archives have scientific decision authority
`NONE`.  A successful replay of them is evidence that the declared bytes and
validation procedures were reproduced, not evidence that U2 is true.

## 2. Registration and source boundary

The evaluation is interpreted under the frozen U2 registration and its
source-bound implementation supplements.  In particular:

- the registered margins, estimator precedence, failure semantics, negative
  controls, and global decision order are not changed after observing a run;
- screening remains screening even when every record validates;
- a missing decision-bearing estimator is represented as `UNRESOLVED`, not
  imputed from a descriptive alternate;
- a rounded diagnostic comparison cannot replace exact cross-platform
  replication; and
- neither a content address nor a passing test suite is a proof of a continuum
  statement.

The earlier screening development surfaces are retained only as historical
engineering records.  Screening v3 is the evaluation-v2 screening surface;
it still has no confirmatory decision authority.

## 3. Screening-v3 execution

macOS and Windows each completed all 3,840 registered screening records.  The
per-host records, summary envelope, runtime record, checkpoint manifest, and
writer-capability record were independently validated.  The
[cross-platform screening comparison](artifacts/global-geometry-ii/u2/evaluation-v2/screening-v3-cross-comparison-v1.json)
reports that the raw record streams are not exact across hosts: exact raw
replication is `FAIL`.  The comparison does not hide that mismatch and does
not convert the screen into Gate 8 evidence.

The shared failure ledger contains 112 estimator-unavailable records: 9 for
the comb discriminant and 103 for the small-world discriminant.  Those 103 are
genuine unavailable-estimator records in screening v3; they are not the false
control-shortfall labels found and repaired during the superseded screening-v2
engineering run.

Every confirmatory decision gate remains outside screening authority.  The
scientifically correct reading is therefore:

```text
screen completed != confirmatory evaluation completed
screen validation != U2 acceptance or rejection
```

The compressed raw streams are convenience transports.  Their decompressed
bytes, rather than the gzip containers themselves, are bound back to the
authoritative source records:

- [macOS screening raw archive](artifacts/global-geometry-ii/u2/evaluation-v2/screening-v3-macos-raw.jsonl.gz)
- [Windows screening raw archive](artifacts/global-geometry-ii/u2/evaluation-v2/screening-v3-windows-raw.jsonl.gz)

## 4. Strict partial campaign

The strict partial campaign is a finite diagnostic campaign under the frozen
source boundary.  It is not the complete perturbation and inference programme
specified for confirmatory U2.

Its normalization artifact is validated and content-bound, but the current
partial campaign builder does not consume the fitted normalization values in
the numerical measurements.  The strict partial channels must therefore not
be described as calibration-normalized; the binding proves provenance and
readiness, not numerical use.

| Gate | Partial-campaign state | Interpretation |
|---|---|---|
| 1. Exact carrier/topology/spectral checks | `PASS` | The declared exact and numerical-oracle fixtures passed in their finite scope. |
| 2. Exponent atoms | `UNRESOLVED` | Required decision-grade estimates and hierarchical uncertainty are incomplete. |
| 3. Profile atoms | `UNRESOLVED` | Required primary normalized profile comparisons are incomplete. |
| 4. Einstein atoms | `UNRESOLVED` | The registered conjunction is incomplete. |
| 5. Transport atoms | `UNRESOLVED` | Decision-grade converged transport comparisons are incomplete. |
| 6. Robustness atoms | `UNRESOLVED` | The registered perturbation layer is not complete. |
| 7. Control separation | `UNRESOLVED` | The registered control-count conjunction is not satisfied by the partial evidence. |
| 8. Independent-host replication | `UNRESOLVED` in the partial ledger | The separate exact cross-host comparison does not satisfy this gate. |

The host-specific compact projections and their exact comparison are:

- [macOS strict-partial compact record](artifacts/global-geometry-ii/u2/evaluation-v2/strict-partial-macos-compact-v1.json)
- [Windows strict-partial compact record](artifacts/global-geometry-ii/u2/evaluation-v2/strict-partial-windows-compact-v1.json)
- [strict cross-platform comparison](artifacts/global-geometry-ii/u2/evaluation-v2/strict-cross-comparison-v1.json)

The strict comparison exposes two different questions:

- exact cross-platform campaign replication is `FAIL`; and
- the explicitly nondecisional projection that rounds exposed finite
  diagnostics to `1e-12` is `PASS`, with decision authority `NONE`.

The second statement is useful portability evidence, but it cannot repair the
first or satisfy Gate 8.  No path-difference count or causal explanation for
the exact mismatch is asserted in this prose; the comparison artifact is the
authoritative record of what was compared.

The full campaign transports are available for audit as deterministic gzip
archives:

- [macOS strict-partial campaign archive](artifacts/global-geometry-ii/u2/evaluation-v2/strict-partial-macos-campaign-v2.json.gz)
- [Windows strict-partial campaign archive](artifacts/global-geometry-ii/u2/evaluation-v2/strict-partial-windows-campaign-v2.json.gz)

These archives have scientific authority `NONE`.  They support exact
decompression and validation of the bound campaign bytes.

## 5. Physical-readiness evaluation

The
[physical preflight record](artifacts/global-geometry-ii/fabrication/physical-preflight-v1.json)
passes only the declared **digital dry-preflight** surface: the q=5,6,7 vector
foreground envelopes fit the frozen sanitized A4 imaging box with positive
analytic clearance, the dry-filter observations use unscaled placement, and
the renderer-specific foreground stays inside the dry printable raster.

No physical output was produced.  Printer connectivity and paper feed, scale
on paper, cutting dimensions, hinge incidence, assembly, material response,
camera permission and calibration, fiducial decoding, 3-D reconstruction,
repeatability, and physical safety remain untested.  Physical validation and
camera validation are both `NOT_RUN`.

## 6. Why this checkpoint is share-worthy

The checkpoint is suitable for public technical review because it exposes,
rather than suppresses, the limits of the evaluation:

- both completed host executions are content addressed;
- validation and comparison are mechanically derived from bound artifacts;
- failed exact cross-platform comparisons are published as `FAIL`;
- the nondecisional `1e-12` diagnostic is labeled with authority `NONE`;
- unfinished gates remain `UNRESOLVED`;
- the preview, partial evaluation, full confirmation, and physical experiment
  are separately named; and
- clean-clone reproduction includes quick and deep modes.

“Share-worthy” here means reproducible enough to inspect and criticize.  It
does not mean peer reviewed, accepted, physically validated, or ready for a
universality claim.

## 7. Reproduction and release map

The current release-v2 surface is:

- [evaluation-v2 index](artifacts/global-geometry-ii/u2/evaluation-v2/evaluation-index-v2.json)
- [release-v2 reproducibility manifest](global_geometry_ii_reproducibility_manifest_v2.json)
- [release-v2 evidence index](artifacts/global-geometry-ii/release-index-v2.json)
- [release-v2 macOS/Windows comparison](artifacts/global-geometry-ii/certificates/macos-windows-comparison-v2.json)

Run the exact deep verification command from the repository root:

```bash
node website/global-geometry-lab/reproduce-global-geometry-ii-v2.js --verify --mode deep
```

Deep verification validates the registered compact release surface and
replays the platform-native evaluation validators.  It does not rerun the
full confirmatory campaign and cannot establish physical validation.

The v1 release index and v1 platform comparison remain available as
historical prior-release records only.  They do not certify the enlarged v2
source, artifact, evaluation, or test surface.

## 8. Remaining work before a U2 decision

A decision-bearing continuation would need the complete registered primary
estimators and hierarchical uncertainty, the full perturbation layer,
adequate control separation, and exact independent-host replication under the
confirmatory source boundary.  Those requirements must be completed without
reinterpreting thresholds after outcomes are known.

Until then, the only valid top-level statement is:

\[
\boxed{\text{U2 evaluation checkpoint: UNRESOLVED; no acceptance or rejection claim.}}
\]