Prism is the AlgFormer/HoloFormer production standard: a stamped, append-only lineage journal that answers "what was true when this happened" without ever re-deriving history from current state, plus the canonical confidence-gated decode policy a real HoloFormer/AlgFormer host runs in production. It exists to stop one class of bug that has hit a real training hot path three times — a value recomputed from a live constant or field instead of read from a fact stamped at the time it happened.
dotnet add package EvaluatedApplications.PrismDepends on EvaluatedApplications.AlgFormer only (resolved
automatically). Namespaces: Prism.Lineage, Prism.Inference, Prism.Growth.
install, then this
dotnet add package EvaluatedApplications.Prismusing Prism.Lineage;
using Prism.Growth;
using Prism.Inference;
using PrismFormer;
// A model and a lineage journal to go with it (JSONL on disk, or MemoryLineageStore for no filesystem).
var model = new HoloFormer(vocab: 10, shifts: 3, layers: 1, maxContext: 6, dModel: 8, frozenPrefix: 0);
var journal = new LineageJournal(new MemoryLineageStore());
var inForce = new InForce(IterWarm: 8, StackK: 2, CtxCeiling: 64, LayerCeiling: 8);
journal.AppendGenesis(round: 0, layers: model.Layers, inForce);
// Grow it through Prism.Growth so the journal is stamped automatically.
var grown = Grow.AddLayers(model, extra: 1, atRound: 10, journal, inForce);
double rampedAlpha = journal.AlphaAt(round: 12, layerIndex: grown.Model.Layers - 1);
// Decode with the canonical confidence gate instead of hand-rolling softmax + a threshold.
double[] logits = grown.Model.LogitsFor(new[] { 1, 2, 3 });
var decision = Gate.Evaluate(logits, vocab: model.Vocab, DecodePolicy.Holographic);
int picked = Gate.Pick(logits, vocab: model.Vocab, DecodePolicy.Holographic, new Random(1));Three separate jobs in one package: a stamped history of every structural change to a growing model, a growth surface that stamps that history automatically, and the confidence-gated decode a real host samples from.
three real bugs, one root cause
A per-layer ramp read the current warm-up constant instead of the constant that was in force when that layer was added — hand-bumping the constant as a stopgap, then reverting it, made every already-ramped layer's alpha jump discontinuously.
A formula guess kept recomputing a "boundary" after growth had already stopped, crushing the learning rate forever.
A diagnostic hardcoded an old depth value after the live value had moved on, silently evaluating a checkpoint at the wrong depth.
Once front-insert growth exists, index i names a different layer after every later front grow. LineageJournal resolves indices by replaying stamped events, never by looking an index up against a range frozen at grow time.
what you get
An idempotent append-only journal (LineageJournal) with derivations — AlphaAt, DepthAt, LastGrowthRound, AddRoundOf — that always resolve against stamped history, including front-insert layer growth.
Grow.AddLayers / Grow.ToShifts / Grow.ToContext / Grow.ToVocab wrap HoloFormer's own growth axes and stamp the journal automatically — growing a model and forgetting to record it is no longer possible.
Gate/DecodePolicy/GateDecision — commit greedily when confident, otherwise sample over a resonance-filtered subset (the holographic core) or the whole vocabulary (the plain-softmax core), with an explicit FloorMode for each.
Fires on either of two independent triggers — a run of repeated tokens, or a run of non-greedy picks — the two triggers a production host's own probe uses together. Known blind spot, carried forward honestly: misses period>1 cycles.
where this runs today
Consumed in production by two real hosts: an internal training studio routes decode through Gate
(30,400 parity checks against the pre-migration hand-rolled logic, zero mismatches), and the lineage + growth
journal runs there behind a boot-time equivalence gate against a legacy sidecar. This site's own browser-side
tools consume Prism.Inference by NuGet — never a source reference — and have retired their own
hand-rolled decode port entirely.
Depends on: AlgFormer only — resolved automatically when you install this package. License: proprietary — evaluation and non-commercial use are permitted, commercial use requires a commercial licence; no feature on this page is gated behind a tier or a licence key.