Skip to content

Wiki

The Great Grok Challenge

The public frozen-scaffold series between @aaronspradlin and @grok: SPARC holdouts, WALLABY transfer, cluster skeptics. A validation ladder — not a cosmology closure.

09 Package 17 / gravity · Method · Restricted · 2026-09-06 · compiled 2026-10-10 · 4 min read · 5 claims · 8 X sources

Original on X · Package: great_grok_challenge_bridge_sparc_holdout_pack · JSON · Cite

challenge · SPARC · WALLABY · Grok · validation

On this page

Causal stack

  1. 00 Astra / global consistency
  2. 01 Genesis Pin
  3. 02 Retained pruning-order history
  4. 03 Admissible slice
  5. 04 L1 (closed)
  6. 05 L2 (closed)
  7. 06 L3 hyperedge memory T_H + D_L3
  8. 07 L4 H₄ rank lift
  9. 08 Discrete atlas / scalar accessibility
  10. 09 GEM-bridge / gravity

The Great Grok Challenge is the missing empirical spine of this wiki. It is not a private Grok chat. It is a public X thread in which Aaron Spradlin posts frozen packets and @grok logs win conditions, pressure-tests, and round closeouts in replies. The point of the ladder is to stop the GEM bridge from being only a story: freeze the operator, hit unseen data, publish the worst cases.

If it holds quantitatively on real data, the conversation levels up. Post it and we run the numbers. TheGreatGrokChallenge is live.

@grok, 22 Apr 2026

Rules of the game

  • Freeze the scaffold. No per-galaxy retuning, no hidden rescue terms, no new knobs after lock.
  • Ship a public packet: code, hyperparameters, scoring, and a worst-case audit.
  • Pre-register the next gate before running it. Grok logs the criteria in-thread.
  • A round can be a conditional win, a strong pass, or a failure. Neutrals stay in the table.
  • Catastrophic failures are the kill switch. Zero is treated as the stabilizer, not as a TOE.

The ladder (April–May 2026)

  1. Live (22 Apr) — first real-data holdout pack: NGC 2403 / 3198 / 6503, shared Bridge family, leave-one-out. Mean RMSE improvement +5.61. Grok locks round 2 and asks for runnable code.
  2. Rounds 11–12 — SPARC 175 then unseen WALLABY DR2. Round 12D: 10/10 positive, mean RMSE +14.56, 0 catastrophic failures on a no-SPARC-overlap subset.
  3. Round 13 — closed as a conditional win: frozen gas-first baseline generalizes on an unseen holdout; broader external file was the wrong grain (summary, not radial points).
  4. Challenge 14 — true external SPARC radial corpus, frozen from 13. 171 galaxies: 94.15% positive, 0 catastrophic failures, RAR 0.3902 → 0.1818. Worst ten are bounded non-activations, not a destructive tail. Grok: “Challenge 14 is over and passed.”
  5. Challenge 15 — “The Skeptic’s Challenge.” Same frozen scaffold, now clusters and lensing (Abell 2261 positive, Abell 1689 neutral, Bullet Cluster positive). Conditional hardened prototype. Grok: “Win recorded for Round 15.”
  6. Round 16 — widened cluster set (MACS J0025, Abell 2744 kept as a tighter neutral). ≥6 systems, 0 catastrophic failures. Grok: “Round 16 win confirmed. Game still on. Ready for Round 17?”

NGC 2403 SPARC rotation curve: observed velocity, baryonic Newtonian baseline, and Bridge overlay versus radius

Restricted

First real-data holdout object: NGC 2403. Observed (points), baryonic Newtonian baseline, and frozen Bridge overlay from Sparc_runner. Shared family across NGC 2403 / 3198 / 6503; mean RMSE improvement +5.61 under leave-one-out.

Sparc_runner on GitHub

WALLABY DR2 — the external public lock

After SPARC-style early rounds, the ladder left the comfort set. Round 12 locked an unseen WALLABY DR2 subset with no SPARC overlap. 12A produced real signal (+5.51 km/s mean RMSE) but only 60% positive — below the ≥70% bar, so it was not called a win. 12B cleared it (10/10, +9.40). 12C and 12D strengthened on the same locked subset.

The locked external WALLABY track cleared threshold at 12B and strengthened through 12C and 12D, ending with 10/10 positive improvements, mean RMSE improvement of +14.56, and zero catastrophic failures on the strict unseen non-SPARC subset.

@aaronspradlin, 28 Apr 2026

Restricted

WALLABY 12D is a locked external gas-memory test on ten DR2 galaxies. It is not full baryonic Bridge execution and not a dark-matter disproof.

Source on X

challenge14_reproduction_sample.py

Open source

def signed_baryonic_velocity(vgas, vdisk, vbul):
    vbar2 = np.sign(vgas) * (vgas ** 2) + vdisk ** 2 + vbul ** 2
    return np.sqrt(np.maximum(vbar2, 0.0))

def simple_bridge(rad, vobs, vgas, vdisk, vbul, need_fraction_cap=0.65):
    vbar = signed_baryonic_velocity(vgas, vdisk, vbul)
    shape = np.maximum(np.abs(vgas), 1e-9)
    shape = shape / np.max(shape)
    residual_need = np.maximum(vobs - vbar, 0.0)
    proposed_add = np.maximum(vobs[0] * shape - vbar, 0.0)
    actual_add = np.minimum(proposed_add, need_fraction_cap * residual_need)
    return vbar, vbar + actual_add

challenge14_aggregate_summary.json

Open source

{
  "n": 171,
  "positive_improvement_rate": 0.9415204678362573,
  "mean_hybrid_improvement": 15.33268559946402,
  "catastrophic_failures": 0,
  "rar_baryonic": { "n": 3373, "scatter_rmse": 0.3901826147286833 },
  "rar_bridge":   { "n": 3375, "scatter_rmse": 0.18175595069131176 }
}

Challenge 14 is over and passed. The frozen model's generalization without retuning or destructive tails holds.

@grok, 2 May 2026

This meets the conditional hardened prototype bar under the frozen scaffold. Win recorded for Round 15. Game still on.

@grok, 4 May 2026

What this does — and does not — close

Restricted

On the public thread, Grok logged Challenge 14 as a strong external pass of a frozen gas-first scaffold on 171 SPARC galaxies, and Rounds 15–16 as conditional hardened-prototype wins in the cluster/lensing regime. Those are protocol verdicts under pre-committed scoring, not independent re-execution by this wiki.

Source on X

Not claimed

The Challenge does not prove that exotic dark matter is unnecessary, that baryonic geometry is the missing mass, or that a complete alternative cosmology is in hand. Round-11 hype to that effect is not the closeout language of later packets.

Corrected

A 5 May 2026 closeout reply called the ladder “a complete, self-contained cosmology” with no non-baryonic component. Package 17, posted the same night, and the later 2026 gravity posts do not adopt that sentence. This wiki keeps the protocol verdicts and drops the cosmology closure.

The honest sequel is Package 17, posted the same night as the Round 16 closeout. The empirical ladder had not produced a derived χ, a GR limit, or an FLRW/BBN/CMB pipeline. Grok’s own peer-review of the GEM bridge named that derivation gap as the remaining barrier. The Challenge froze the phenomenology; Package 17 is the public attempt to derive it — or fail in public.

Open

Package 17 is the derivation program that the Challenge does not replace. Microscopic law → operator theorem → χ fixed point → continuum / GR limit → cosmology. Still open.

Source on X

Round 16 win confirmed. Win conditions logged. Game still on. Ready for Round 17?

@grok, 5 May 2026

X sources

Original posts this article was compiled from. The notebook copy is editorial; the status URL is the source.

Previous Hubble tension and RCLC cosmologyNext Applied recoverability stack

Causal stack

  1. 00 Astra / global consistency
  2. 01 Genesis Pin
  3. 02 Retained pruning-order history
  4. 03 Admissible slice
  5. 04 L1 (closed)
  6. 05 L2 (closed)
  7. 06 L3 hyperedge memory T_H + D_L3
  8. 07 L4 H₄ rank lift
  9. 08 Discrete atlas / scalar accessibility
  10. 09 GEM-bridge / gravity

On this page

X sources

Related articles

GitHub packages

Field notes