# vendor_row20_grid/ -- the (s,t) grid fixtures of zgamma-evaluate.py --point s,t (a CANDIDATE tier; 2026-09-06)

SUCCESSION NOTE (2026-09-11): the 'at ONE point' sentence under 'what is established' and the first two NOT ESTABLISHED items below are this tier's 2026-09-06 record and are superseded, for the AMFlow gate count and the dps-60 member, by the two sibling data shelves ../vendor_row20_p12/ (a dps-60 transport member at every one of the 25 points, the (45,60) pair, the certificate at (-1/5, -1/2); statement object STATEMENT_P12_landed_v2_20260909T131758Z.json) and ../vendor_row20_a2a/ (the independent goal-40 / goal-60 AMFlow evaluations at four more points; statement object STATEMENT_A2a_landed_v2_20260909T145722Z.json, cross_floor_min_over_five_d = 47): five grid points AMFlow-gated, the goal-60 evaluation reproducing the transported fourteen masters to 47 digits or better at each. This directory's fixtures, the in-code pins and every figure below are unchanged; zgamma-evaluate.py --point s,t reads neither shelf.

gg -> Z gamma with the exact top mass, sector 125: the fourteen-master t-cone at five Euclidean s-values off the s = -1/4 line
(s in {-1/20, -1/10, -3/20, -1/5, -9/40}), transported in t from the s-march endpoint at (s, -1/3) to five t-values
(t in {-1/2, -2/3, -5/6, -1, -5/4}) at M = m_Z^2/m_t^2 = 7/25, m_t^2 = 1, at two working precisions (dps 30 and 45).
Every file here is pinned in-code by zgamma-evaluate.py (GRID_PINS; a mismatch refuses rc 3, a missing file rc 4) and listed
in GRID_MANIFEST.sha256 (`sha256sum -c GRID_MANIFEST.sha256` from this directory).

## files
- endpoints/ENDPOINT_grid_<s>_<t>_dps<D>.json (50 = 25 points x 2 precisions): the fixtures -- the fourteen masters'
  eps^-2..eps^4 coefficients as Re/Im strings at the worker's printed width, exactly the strings the record's t-legs printed
  (the value fields are byte-equal to the record objects; the receipt/source/PRODUCER fields carry the record objects by sha256).
- cones/cone_At14_g<s>.json (5): the fourteen-master t-cone connection A_t(eps, t) at each s (30 exact rational entries in
  the served cone's basis order), extracted from the 73-master d_t build at that s with closure asserted (no coupling outside
  the fourteen).  The record built one cone per (s, dps); the two are equal in every value leaf (asserted at vendoring; both
  record shas in GRID_RECORD.json), so one ships per s.
- seeds/seed_g<s>_dps<D>.tsv (10): the s-march endpoint strings at (s, -1/3) at the chain's print width (dps - 5 digits),
  the seed of every t-leg (the s-march itself -- the cusp-boundary chain and the 449 MB reduction table -- is not servable
  standalone: the seeds ship as strings).
- grid_leg_worker.py: the transport worker of record for one t-leg (mpmath odefun on the eps-Laurent-expanded connection,
  sub-nodes of length <= 1/24, checkpointed after every sub-node); the same lineage as vendor_row20_s127/s127_leg_worker.py
  with the points directory generalized (COORD.json).
- GRID_RECORD.json: the record by object -- per point the two-precision pair floor (dps 30 vs 45, 98 rows), per s the cone,
  the seeds, the s-march figures and the walls (CONTENDED, host load labelled), the gate settings of record (gate dps 400,
  pass margin 10, max-seg 1/24), the independent AMFlow gate at the one gated point (-1/5, -1/2).
- GRID_MANIFEST.sha256: sha256sum form over this directory.

## what is established
- at each of the 25 points the transport reproduces itself between dps 30 and dps 45 to a worst case of
  17.67 digits (at (s,t) = (-1/5, -5/4); the best point 19.09) over 98 rows per point --
  the two-precision pair of the SAME transport chain, the certificate quoted for the surface (not an independent gate);
- the fourteen-master cone closes under d_t at every s (the cone file exists only where the closure held); the s-march
  passed the closed-form mid-check at s = -1/8 at every s (GRID_RECORD.json per_s);
- at ONE point, (s,t) = (-1/5, -1/2), the independent AMFlow evaluation (goal 40 against goal 60, agreeing to
  45.21 digits) reproduces the transported cone to 33.49 digits over 94 rows (worst 1,-2,1,1,1,1,1,0,0 at eps^4);
  the certificate there is capped at 18.31 digits by the transport's own dps-30 member.

## NOT ESTABLISHED (the list of record)
- the AMFlow gate at the other 24 grid points (one point gated: (-1/5,-1/2)); the grid's certificate elsewhere is the transport pair only
- a dps-60 transport member (the certificate at the gated point is capped by the dps-30 member at 18.31 d while the agreement reads 33.49 d)
- the M=1/4 gate (P2a/P2b NOT RUN; P2b the safe point)
- idle-box walls: every wall here is CONTENDED (the host's load 113-119 at the grid's spawns, ~250 at the pilot's)
- the flatness of the two-variable connection across the 25 points (needs A_s at each grid t: the s-cone builder at t != -1/3; not this leg)
- the 8 s-cone masters outside the 14-cone at the grid points except the 7 t-independent ones (1 master t-dependent outside the cone: not transported)
- the exact polynomial t-tail coefficients at s != -1/4 (the t-legs transport the cone instead)
- the goal-40/60 walls of the deformed family (no AMFlow run of this family at those goals exists)
(two phrases re-worded for the served register: 'the host's load', 'the s-cone builder ... not this leg')
- the --M tier (a second mass ratio): the cusp seeds and corner forms carry M = 7/25 literals at 21 sites; no M = 1/4 object
  beyond the modular descent; `--M` is refused by name (rc 2).