# Apeiron AP1-M1 – chronology-safe 128-shell junction seed checkpoint

Updated UTC: `2026-09-02T18:19:58Z`

Classification: `PASS_JUNCTION_SEED_ONLY`

AP1 remains **ORANGE**. This checkpoint releases only an exact early M1
junction seed for a bounded, non-production background pilot. It does not
release a late-time background, the physical response kernel, observables,
fits, or significance claims.

## Chronology repair

- The first 96-shell result existed before the strict 96-shell tolerance
  artifact. Its seed eligibility is therefore revoked.
- The 96-shell result remains pilot evidence only and was never used to start
  a new background.
- The unchanged strict caps were frozen for the previously unseen 128-shell
  evaluation in the `20260902T181030Z` artifact (internal UTC
  `2026-09-02T18:10:41Z`).
- The 128-shell heldout was evaluated once afterwards in the
  `20260902T181054Z` artifact (internal UTC `2026-09-02T18:11:09Z`) and passed
  every frozen gate.

Authoritative hashes:

- 128 tolerance freeze: `2b1205481c62d7d87b3e5265cbfa5dd01f77d020595740b975dce3e2cb0e5323`
- 128 heldout: `b4bb291379e2556e2b673fb680c6d5034672728111295f89372281dd58c7470b`
- 128 seed NPZ: `644126925fcff32df6f881ee604787751dbe48829072671bf541552bd65dc729`
- 128 seed JSON: `cf946bfcde4c10a36d5eee99a18fe1a8745483e111304d52f4cc17b3fdf103b4`

## Exact released seed

- Stored frozen index: `0`; no interpolation and no later-time vacuum
  reinitialization.
- `N_physical = -56.97244594022879`.
- `H = 4.296911897956572e-7 M_pl`.
- Required standard density: `9.017902368676568e-26 M_pl^4`.
- Friedmann normalized residual: `-8.91158373533706e-28`.
- Independent Raychaudhuri normalized residual:
  `-3.625504617408686e-8`, below the frozen `4e-8` cap.
- Positive frequency margin: `6.299255678190488e-9 M_pl^2`.
- Heldout Decimal Wronskian error: `4.4e-78`.
- Binary canonical Wronskian error: `4.440892098500626e-16`, below the
  unchanged `1e-11` limit.
- Four Pauli–Villars sectors with 128 momentum shells; all 512 frequencies
  are positive and all 16 stored array groups round-trip exactly.

The 96→128 changes in physical N, required density, `rho_q/H^2`, Friedmann,
and all other registered diagnostics pass the pre-frozen difference caps; the
largest nonzero registered relative change is
`6.132085916940108e-16`.

## Verification and boundary

The chronology-safe junction chain passes `32/32` canonical tests; the
cumulative registered project total is `150/150` PASS. No NONPASS was stored
or used as a seed. The old solver and Physical Map were not called. No model
equation, parameter, physical assumption, or existing gate threshold changed.
The released report binds all 19 direct and transitive code/data authorities
actually used for construction, including the utility-only functions imported
from the superseded 96-shell seed module.

Next allowed step: implement a bounded new `N=ln(a)`-parameterized AP1-M1
background **pilot** from this seed on nested resolutions, then freeze its
time-, momentum-, curvature-tail-, and rounding-error tolerances before any
background candidate is assessed. The physical two-time response kernel
remains blocked until a new M1 background and pre-frozen pilot-based physical
tolerances both exist.