# UD frozen new-seed phase campaign v0.1

Four fresh bounded constructor records certify unique matching to the pinned
fitted reference. A fifth, deliberately oversized impulse yields a retained
model/order HOLD. Physical promotion 0; development — no registry entry;
independent review pending.

| New record | Learned temporal order (floating discovery) | Phase certificate, radians/sample |
| --- | --- | --- |
| dense_216104 | 105 | <0.000500262 |
| dense_216105 | 105 | <0.000499343 |
| dense_216106 | 105 | <0.000502489 |
| block_512_216107 | 105 | <0.000496848 |
| oversized_impulse_216108 | 106 | Proposer model/order HOLD |

Bounds are a posteriori errors relative to the fixed learned reference, not
observed parameter errors or physical-frequency measurements. The block-sign
case holds each error sign for512 consecutive samples. Three new dense seeds
and this dependence pattern all meet both declared additive gates. Every bounded
62-vector uses exactly31/32 of the squared vector-error budget. The oversized
impulse intentionally violates both gates. Its even learned order106 is outside
the proposer's stationary-plus-conjugate-pairs model. That HOLD is not a claim
that the record's true full-Hankel signature fails.

## Frozen execution and checks

CAMPAIGN_FREEZE.json binds all recipe seeds, code and input hashes before noise
generation. DATA_FREEZE.json binds all five complete raw records and exact error
checks before fitting. No seed, threshold, reference, estimator or mathematical
proof was adjusted after these outcomes. Four successful original proposals are
byte-identical to the instrumented ones; the fifth pair has equal HOLD receipts.
Ten fresh original/instrumented executions were performed.

All93,950,770 actual additive scalar differences were checked against the
losslessly restored baseline. Four saved-array phase residual replays agree;
13 tests pass, including scalar/vector boundary distinctions and large-integer
squared-norm overflow protection. All five raw records were separately
regenerated with identical hashes and gate results. This is same-constructor
deterministic reproduction, not independent review or blinded execution.

The PR189 proof helpers and original/instrumented proposer code retain their
published bytes (SOURCE_IDENTITY.json). Only the adapter's record/report/witness
lineage changes; patches are included. Exact residual formulas, fixed phases,
frame-constant recomputation, angle rounding and strict matching gates are
unchanged. THEOREMS.md reproduces the direct-span argument; EXACT_GATES.md proves
the encoded additive gates and states the scope.

## Portable phase replay

Extract the core ZIP and all four WITNESS archives listed in WITNESS_INDEX.json
into one directory. Each witness archive contains its actual floating arrays
and the corresponding original proposal needed for byte comparison. Then run:

```
OPENBLAS_NUM_THREADS=1 python replay_package.py
```

Python3.12, NumPy2.3.5, SciPy1.17.0, python-flint0.9.0 are the recorded runtime.
This verifies package hashes and recomputes all four available actual-array
phase certificates in a temporary copy. It checks the oversized HOLD/audit
receipts but does not rerun its proposer. Every float array, auxiliary reduction,
actual proposal and original proposal required by this replay is retained.

## Optional full raw-record regeneration

Raw records and the parent lossless codec are omitted to avoid duplicating
large source evidence. Obtain the UD repository with the canonical codec path
and exact hashes specified by CAMPAIGN_FREEZE.json. From the extracted package:

```
OPENBLAS_NUM_THREADS=1 python reproduce_records.py --repo /path/to/UD --destination /new/empty/path
```

This restores the baseline from the hash-checked parent codec, regenerates all
five raw records, rechecks every additive coordinate, and compares all resulting
record hashes/gate receipts with DATA_FREEZE.json. The destination must not
exist. The normal portable phase replay does not silently claim this source
reconstruction. RAW_REGENERATION_RECEIPT.json records the separate execution
performed for this campaign.

## Remaining scope

This is five synthetic fixtures on one known baseline; four certified phase
runs do not establish stochastic coverage, uniform whole2^-20-ball phase
matching, arbitrary excitation loss or global fitter convergence. Floating
component discovery is distinct from exact algebraic rank and the full1893-value
Hankel criterion. No new full-Hankel verification or dense Hankel allocation is
claimed. No timing/preparation/SI calibration, acquisition/metrology admission,
independent source custody, independent review, hidden-state recovery or physical
identification is inferred. The nominal747-dimensional observer kernel and the
separation of62 history channels from189 diagnostic outputs remain unchanged.
Task178 and review HOLD177 remain open. O03 tolerances remain protocol controls.