# Apeiron AP1-M1 – background multiscale architecture preflight

Updated UTC: `2026-09-02T18:43:47Z`

Classification:
`M1_BACKGROUND_MULTISCALE_ARCHITECTURE_PREFLIGHT_PASS_NO_CANDIDATE`

This checkpoint is a complete PASS at architecture-preflight scope only. AP1
remains **ORANGE**. No new AP1-M1 background was run, no background candidate
was assessed, no tolerance was frozen, and the physical response kernel was
not started.

## Hash-bound input

The preflight accepts only the chronology-safe 128-shell junction seed:

- seed JSON: `cf946bfcde4c10a36d5eee99a18fe1a8745483e111304d52f4cc17b3fdf103b4`;
- seed NPZ: `644126925fcff32df6f881ee604787751dbe48829072671bf541552bd65dc729`;
- all 16 seed arrays reproduce their registered hashes;
- four PV sectors × 128 momentum shells are present;
- no later-time vacuum reset is authorized.

## Computed architecture diagnostics

The physical span from the exact seed to the shared present anchor is
`56.97244594022879` e-folds. It implies:

- scale-factor range `5.531194016276229e24`;
- moving physical-split transport across `82.19386522528264` octaves;
- initial `omega/H` range `184.71535214209112` to
  `14273.833598421304`;
- shortest initial phase period in N: `0.0004401890538975115`.

One explicitly nonphysical diagnostic envelope held the initial quantum
triplet fixed. It stored no trajectory rows and was not used as a background
candidate. Its purpose was solely to expose numerical failure modes. On this
envelope:

- the first sampled zero-mode tachyonic point occurs by relative
  `N=0.055637154238504676`;
- the unscaled diagnostic growth exponent is `3835.1675406957315`, exceeding
  the Float64 log range `709.782712893384`;
- the legacy fixed-comoving UV split rejects fail-closed because physical
  sector 0 enters `omega2 <= 0`;
- the endpoint H diagnostic remains `5.866941643680344e53` times the shared
  H0, but this fixed-source number is not a physical prediction.

## Required solver architecture

Before a long background pilot, the following are mandatory:

1. transport of the seed state without later vacuum reinitialization;
2. a moving physical momentum split with adiabatic state transport;
3. log-amplitude or equivalent rescaling through tachyonic bands;
4. signed PV cancellation under explicit precision control;
5. coupled background–mode backreaction iteration;
6. independent Ward and Wronskian audits;
7. nested time-, momentum-, tail-, and rounding-pilots before any tolerance
   freeze.

Eight new tests pass; the full junction-plus-architecture suite passes
`40/40`. The cumulative registered project total is `158/158` PASS. No model
equation, physical assumption, parameter, or existing gate threshold changed.

Next allowed step: implement an independent moving-split/log-amplitude mode
transport reference and validate it against the existing Decimal propagator on
overlapping short tachyonic windows. The long background candidate, physical
kernel, observables, fits, AP2, and AP3 remain blocked.