Remove architecture mechanism disclosure; link Z-Next results page + live demo

Two changes:

1. IP: the card described the mechanism (state formulation, gating, and a per-component parameter breakdown naming each functional part). Replaced with grouped, non-identifying wording that preserves every measurement and the SFT conclusion.

2. Added links to the current generation: Z-Next 1B results page and the live public demo.

README.md

@@ -13,6 +13,12 @@ extra_gated_prompt: "Weights for this model are not published. This repository c
 
 # Tetracta-zkas-1B — results card (no weights)
 
+> **Newer work:** this card documents our 2025 architecture. The current generation —
+> **Z-Next 1B**, no attention and no KV cache, with a session state that does not grow
+> with context — has its own results page and a **live public demo you can talk to**:
+> · [Z-Next 1B results & benchmarks](https://www.tetracta.ai/znext.html)
+> · [**Live demo →**](https://www.tetracta.ai/zchat)
+
 **This repository publishes measurements, not weights.**
 
 `tetracta_zkas_1B` is the **twin** of [Tetracta VAN-1B](https://huggingface.co/tetracta/tetracta-van-1b):
@@ -20,7 +26,7 @@ it saw **bit-identical pre-training data** (same corpus, same order, verified) a
 instruction-tuning run** (same 11,419 examples, same recipe, same seed). The only difference is the architecture.
 
 That makes it a controlled single-variable experiment, and it is the reason our central claim about SFT is not
-a one-model anecdote. **All of those results are below.** The architecture's equations, pole parameterisation,
+a one-model anecdote. **All of those results are below.** The architecture's equations, parameterisation,
 initialisation recipe and stability rules are **Tetracta's closed research work and are not published.**
 
 - 🌐 **[tetracta.ai](https://tetracta.ai)** · 📄 **Paper:** *"What does SFT do?"* → [tetracta.ai](https://tetracta.ai)
@@ -35,14 +41,14 @@ initialisation recipe and stability rules are **Tetracta's closed research work
 | | |
 |---|---|
 | Type | **Not a transformer.** No attention. |
-| Mechanism | γ-pole **recurrent state** with a learned decay spectrum + a **multiplicative saturation gate** |
+| Mechanism | **Recurrent, constant-size state.** The mechanism itself is not disclosed. |
 | Inference | State size is **independent of sequence length** (O(1) per step) |
 | Parameters | **953,466,904** |
 | Pre-training | **315,000 steps · 10,321,920,000 tokens · one epoch**, bit-identical to VAN-1B |
 | Instruction tuning | 11,419 examples · 3 epochs · lr 1e-5 · batch 16 · 2,142 steps · 1,247 s (1×H200) |
 | Stability | **0 skipped steps out of 315,000** during pre-training; 0 during SFT |
 
-Not disclosed: equations, pole parameterisation, initialisation recipe, MFU/parallelisation
+Not disclosed: equations, parameterisation, initialisation recipe, MFU/parallelisation
 recipe, stability laws, energy-distribution findings.
 
 ---
@@ -53,7 +59,7 @@ Our claim from the VAN paper: **SFT does not add knowledge; it tunes the output
 If that is true, two very different architectures given the same data and the same SFT should end up at the
 same *interface* state even from different starting points. They do:
 
-| Measurement | VAN-1B (transformer) | **zkas-1B** (γ-pole) |
+| Measurement | VAN-1B (transformer) | **zkas-1B** |
 |---|---|---|
 | Base output entropy | 5.951 | 7.882 — *very different* |
 | **Post-SFT output entropy** | 1.795 | **1.624** — same order |
@@ -74,19 +80,19 @@ This is the cleanest evidence we have, and it is only visible here. In a transfo
 "memory" from "processing" — attention weights carry both. In this architecture components separate by function,
 so we can read directly where SFT went:
 
-| Functional component | δ |
+| Group of parameters | relative change (δ) |
 |---|---|
-| **position/phase channel — output projection** | **18.72 %** ← largest move, 3.5× the embedding |
+| the part that shapes **how the model speaks** | **18.72 %** ← largest move, 3.5× the embedding |
 | embedding / output head (tied) | 5.327 % |
-| saturation gain | 2.560 % |
-| mixing projections (in / out) | 1.299 % / 1.119 % |
-| position/phase channel — input | 0.551 % |
-| pole coefficients | 0.210 % / 0.070 % |
-| **★ decay spectrum (the time memory)** | **0.00009 %** ← **frozen** |
+| everything else in the processing path | 0.07 % – 2.56 % |
+| **★ the part that holds the model's memory over time** | **0.00009 %** ← **frozen** |
+
+*(Component names and the full per-part breakdown are not disclosed; the grouping above
+is what the measurement supports without describing the architecture.)*
 
 ![Where SFT went](fig2-where-sft-went.png)
 
-**SFT moved the memory by roughly one part in a million.** It rewired the position/phase output instead.
+**SFT moved the memory by roughly one part in a million.** It rewired the output-shaping path instead.
 *"SFT changes how the model speaks, not what it knows"* is, in this architecture, a **structural** statement
 rather than an inference.
 
@@ -98,8 +104,8 @@ rather than an inference.
 | **Alignment tax** | **+11.5 %** | **+35.9 %** |
 
 Same data, same recipe, **3× the cost**. Our pre-registered prediction was "~11 % again" — that was a **miss**,
-and it is recorded as one. Likely mechanism: the position/phase channel participates in processing *every* token,
-so moving its output by 18.7 % perturbs the plain-text distribution more. *(Untested hypothesis. Testable
+and it is recorded as one. Likely reason: the part that moved most participates in processing *every* token,
+so perturbing it by 18.7 % disturbs the plain-text distribution more. *(Untested hypothesis. Testable
 prediction: giving that projection its own lower learning rate should reduce the tax.)*
 
 ## Result 4 — blind judge panel
@@ -173,7 +179,7 @@ There are no weights here, so there is nothing to license as a model. What *is*
 quote it, replot it, argue with it — just cite it. Attribution is the whole point; these numbers exist so that
 someone else can check whether our claim about SFT holds up.
 
-**Not covered by this licence, because it is not published:** the architecture's equations, pole
+**Not covered by this licence, because it is not published:** the architecture's equations,
 parameterisation, initialisation recipe, parallelisation recipe and stability rules.
 
 ## Citation