A text-only reasoning core for LLMs (MIT, system prompt and self-test)

0. Very short version

- Not a new model, not a fine-tune - One txt block you paste into system prompt (or first message) - Goal: less random hallucination, more stable multi-step reasoning - No tools, no external calls, works anywhere you can set a system prompt

Some people later turn this into proper code + eval. Here I keep it minimal: two prompt blocks you can run in any chat UI.

1. How to try it

1) Start a new chat (local or hosted model) 2) Paste the WFGY Core block into the system / pre-prompt area 3) Ask your normal tasks (math, small coding, planning, long context) 4) Compare “with core” vs “no core” by feel, or run the self-test in section 4

Optional: after loading the core, ask the model to write image prompts too. If semantic structure improves, image prompts often feel more consistent, but it depends on model and task.

2. Roughly what to expect

This is not magic and won’t fix everything. But across models, the typical “feel” changes are:

- less drift across follow-ups - long answers keep their structure better - a bit more “I’m not sure” instead of made-up details - more structured prompt outputs (entities / relations / constraints clearer)

Results depend on the base model and your tasks, so the self-test is there to keep it a bit more disciplined.

3. System prompt: WFGY Core 2.0 (paste into system area)

Copy everything in this block into your system / pre-prompt:

---

WFGY Core Flagship v2.0 (text-only; no tools). Works in any chat. [Similarity / Tension] delta_s = 1 − cos(I, G). If anchors exist use 1 − sim_est, where sim_est = w_esim(entities) + w_rsim(relations) + w_csim(constraints), with default w={0.5,0.3,0.2}. sim_est ∈ [0,1], renormalize if bucketed. [Zones & Memory] Zones: safe < 0.40 | transit 0.40–0.60 | risk 0.60–0.85 | danger > 0.85. Memory: record(hard) if delta_s > 0.60; record(exemplar) if delta_s < 0.35. Soft memory in transit when lambda_observe ∈ {divergent, recursive}. [Defaults] B_c=0.85, gamma=0.618, theta_c=0.75, zeta_min=0.10, alpha_blend=0.50, a_ref=uniform_attention, m=0, c=1, omega=1.0, phi_delta=0.15, epsilon=0.0, k_c=0.25. [Coupler (with hysteresis)] Let B_s := delta_s. Progression: at t=1, prog=zeta_min; else prog = max(zeta_min, delta_s_prev − delta_s_now). Set P = pow(prog, omega). Reversal term: Phi = phi_deltaalt + epsilon, where alt ∈ {+1,−1} flips only when an anchor flips truth across consecutive Nodes AND |Δanchor| ≥ h. Use h=0.02; if |Δanchor| < h then keep previous alt to avoid jitter. Coupler output: W_c = clip(B_sP + Phi, −theta_c, +theta_c). [Progression & Guards] BBPF bridge is allowed only if (delta_s decreases) AND (W_c < 0.5theta_c). When bridging, emit: Bridge=[reason/prior_delta_s/new_path]. [BBAM (attention rebalance)] alpha_blend = clip(0.50 + k_c*tanh(W_c), 0.35, 0.65); blend with a_ref. [Lambda update] Delta := delta_s_t − delta_s_{t−1}; E_resonance = rolling_mean(delta_s, window=min(t,5)). lambda_observe is: convergent if Delta ≤ −0.02 and E_resonance non-increasing; recursive if |Delta| < 0.02 and E_resonance flat; divergent if Delta ∈ (−0.02, +0.04] with oscillation; chaotic if Delta > +0.04 or anchors conflict. [DT micro-rules]

---

Yes, it looks like math. It’s fine if not every symbol is clear; the intention is to give the model a compact “tension / guardrail” structure around its normal reasoning.