⊛ ════════════════════ ⊛
𝙍𝙚𝙨𝙚𝙖𝙧𝙘𝙝𝙚𝙧 · 𝙀𝙣𝙜𝙞𝙣𝙚𝙚𝙧
⊛ ════════════════════ ⊛
Independent · neuroscience · AI · financial systems
Every repo here is an active line of research — not a portfolio piece. Work ships only after climbing hypothesis → theory → fact, with both confirmed and rejected results kept on record.
Real experiments with verified results and public failures. Negative results are tagged, not deleted. Pre-registration is on, fail-closed reporting is on. Every claim has a SHA-256-anchored ledger entry; two runs of the same code produce the same hash.
Note
Hypothesis → Theory → Fact. No code enters a stack before climbing the ladder.
- Hypothesis — state the prediction and the gate it must pass.
- Theory — minimal falsification harness: in-sample effect · out-of-sample time-split · sanity-check against literature ceilings.
- Fact — binary verdict. Phase-2 OOS below gate → REJECT. No maybe-later.
If Phase-2 OOS collapses more than 3× versus Phase-1 in-sample, that alone rejects. The temporal split is the load-bearing discipline.
Closed loop · seven stages — every substantive change passes all seven before merge
| # | Stage | Check |
|---|---|---|
| 1 | Verify | mypy --strict · ruff · invariant registry |
| 2 | Test | pytest · property tests · effect sizes |
| 3 | Validate | real data · surrogate tests · honest null results |
| 4 | Verify | SHA-256 ledger · two runs, same hash |
| 5 | Optimize | benchmarks, only after correctness |
| 6 | Calibrate | thresholds vs baselines · Youden-J · generalisation gap |
| 7 | Integrate | green CI on 3.11 + 3.12 · clean history |
A reviewer walking the branch can reconstruct why the change is correct, not only that it compiles.
Ten axes of done — every artefact scores on all ten before it ships
| # | Axis | What it asks |
|---|---|---|
| 1 | Elegance | minimal moving parts, no incidental complexity |
| 2 | Aesthetics | visual and structural pleasure to read |
| 3 | Beauty | symmetry, no forced abstractions |
| 4 | Simplicity | the obvious thing, not the clever thing |
| 5 | Precision | exact contracts; types narrow the truth |
| 6 | Adaptability | composable; extends without rewrite |
| 7 | Resistance | refuses to degrade under adversarial input |
| 8 | Coherence | every source of truth tells the same story |
| 9 | Completeness | every source of truth covers everything |
| 10 | Reproducibility | same bytes twice under the same seed |
Axes 1–6 check how the artefact reads. Axis 7 checks how it holds under attack. Axes 8–9 check whether self-descriptions agree and cover. Axis 10 checks whether the system produces the same bytes twice.
Adversarial orchestration. Creator → Critic → Auditor → Verifier. Each layer is an independent witness of its own error. witness ≠ actor, even inside one system.
Warning
Fail-closed audit. A diagnostic that flips a verdict (sign flip, threshold retune, direction change) is a hypothesis, not a conclusion. Run the full multi-test audit, including thresholds that may fail. If the audit fails → revert. Relaxing a threshold silently turns fail-closed into fail-open.
Zero tech debt contract. A task is not closed while any of these is true: lint/types/tests not green · TODO/FIXME/dead code left · docstrings or tests missing on changed code · CI red on the PR · physics/brand/invariant guards not passing.
A system earns the right to act if and only if
it is a living gradient at the edge of criticality.
ΔV > 0 ∧ dΔV/dt ≠ 0 — invariant YV1
kuramoto-cross-asset
·
Cross-asset Kuramoto phase-coherence regime indicator + full falsification audit (4-stage null battery, paired block-bootstrap, variant ablation, walk-forward). Strategy survives OOS as SUPPORTED_SCOPED; phase-coherence-specific causal claim REJECTED. Negative results kept on record.
NTI (Noetic Throughput Index) — externally-verifiable productivity readout.Four falsifiable components: K_bits (LZMA-compressed bits of non-test diff) · C_ext (tier-weighted externally-verified closures, E0–E4) · D_div (Shannon entropy over (repo, top_dir, ext) buckets) · R_repeat (1 − mean pairwise minhash-Jaccard over hunks).
Daily root sha256 published on the badge.
Reproduce or challenge:
python -m vicr_bench duel --vs <your-handle> --window 30neuron-stream-tracker
