Changelog

A running record of research findings, documentation updates, and Evidence Vault changes. For the narrative version of each entry, see the blog.

---

September 2026

2026-10-01 - GENESIS, EXODUS and AIGARTH, a new test, and a refined design

• New chapter: GENESIS, EXODUS and AIGARTH: what is on chain. It covers:
  • the three community assets and how they are linked by one source of funds;
  • where the riddle texts are and what they say;
  • where the "March 3" date comes from;
  • which popular readings were not confirmed.
• New result in where the Anna Matrix comes from: letting Aigarth units grow before training does not reproduce the unusual linearity of the Anna Matrix (140 runs, protocol fixed in advance).
• Rewritten: Anna 0.22 and Bitcoin price data. It now covers:
  • how the program works;
  • the same network design in Qubic's public mining algorithm BPP9000;
  • a first test of a saved state on unseen hours (48.9 %);
  • the corrected table size of 9,456 hours (earlier 9,458).
• Extended: CFB and Satoshi gained a section on what holds from our earlier work on the 1CFB address.
• Corrected:
  • the origin of the "March 3" date, which comes from the GENESIS token text, on the re-check page;
  • the on-chain paragraph of the April 2026 Genesis post;
  • the wording about the Anna bot, which belongs to Qubic's Aigarth project.
• Design: one consistent visual system across the site. It covers type scale, spacing, contrast, tables, callouts, the documentation layout, the header and footer, and the network monitor.

---

2026-09-30 - What the Anna Matrix is: new chapters, a rebuilt Research Lab, and a leaner site

We went through the whole site against the raw data. The narrative version is in the blog post, and the full list of changes with reasons is on the re-check page.

• New chapters:

  • Where the Anna Matrix comes from: the table as bot replies, entry errors, a corrected candidate table, the matching public Aigarth unit, and the comparison with trained units, including the open question of its unusual linearity.
  • Anna 0.22 and Bitcoin price data.
  • How the earlier claims arose.

• Research Lab: the evidence page was rebuilt around what holds:

  • a matrix explorer comparing the table as transcribed with the corrected candidate;
  • the fingerprint comparison with trained Aigarth units;
  • the key findings with links to their methods.

• Rewritten:

  • the start page;
  • research motivation;
  • CFB and Satoshi: what the primary sources say;
  • verified facts;
  • the discussion chapters.

• Discord sources: every quote now carries a message ID. Quotes that could not be matched, and quotes by other community members that had been attributed to CFB, were set aside. Appendix D gained CFB's statements on the GENESIS token, on Satoshi, on the 13 April 2027 date and on Anna's design.

• Set aside:

  • 88 archived pages;
  • 15 chapters and methods pages whose main claims were not confirmed: Bitcoin bridge, time lock, Genesis puzzle, seven-position signature, Grok path, JINN, statistical "validation" and identity percentages;
  • unused template pages;
  • research data files with test keys generated during the matrix experiments (none belong to anyone's wallet).

  Old addresses redirect to the re-check page.

• Web pages: the timeline, the labs, the companion and the books page no longer present open, re-checked findings as verified. The books remain available as narrative works.

• New data files: anna-matrix-clean.json, a corrected candidate, and itu-fingerprint.json, the comparison data. The original matrix file is unchanged.

• Security: the site's server routes were hardened.

---

2026-09-29 - Re-check of our own research

We re-checked our own findings and published the results on the re-check page. The short version:

• The Anna Matrix is a table our team compiled by hand from replies of the Anna bot on X. It is not a file from Aigarth, JINN or Qubic source code.
• About half of the 68 symmetry exceptions come from one run in column 30 that is very likely an entry error in the typed table. The seven-position signature and the "1161 = 27 x 43" sum depend on it.
• No bridge, node system, time lock or activation has been demonstrated.
• Appendix D was rewritten. Statistics we could not reproduce, unsourced quotes and a quote attributed to a relative of CFB were set aside.
• Affected chapters now carry a short re-check notice.

---

2026-09-16 - Second book: Come-from-Beyond

A new book joins the library: Come-from-Beyond, a 47-page biography of the man behind Qubic. Where the first book, Qubic, The Long Version, mapped the mechanism, this one follows the person: a twenty-five-year line from a Minsk computer newspaper to the anna.exe deploy of summer 2026. It is free, translated from Russian, and reads in the browser or downloads as a PDF.

Open it in the Library.

---

June 2026

2026-06-15 - Anna.exe & Emergence: reverse engineering + a 15-phase emergence study

Re-checked on 30 September 2026; the readings below that were not confirmed stay open. See September 2026 re-check.

We reverse-engineered CFB's Anna.exe (an Aigarth evolution engine, a 100-cell ternary lookup-table cellular automaton trained by a (1+1) evolution strategy on 7-bit addition) and built a controlled, exactly-verified study of what can and cannot emerge by mutation and selection.

• Decomposition theorem: every component of a generalizing algorithm can emerge by selection on tiny examples, and exactly one primitive (the bare iterate-and-check loop) cannot, for a finite-description reason. Fifteen phases, each verified by exact match far beyond training size (e.g. 26/26 seeds converge to the identical full adder, exact to width 128; an invented mod-3 state machine is exact on 80,000 random strings).
• The substrate can, the search cannot: a hand-built adder reaches 0 error using under a third of the cells, yet the monolithic (1+1) search stalls (~41% on the full 7-bit task, matching CFB's own screenshot). Composition reaches 0 where monolithic search cannot.
• Honest novelty: methodological, not phenomenal, every phenomenon has a named predecessor (Kashtan-Alon 2005, Clune-Mouret-Lipson 2013, Mitchell version-space, …). Not "AI found", not consciousness.

New chapters: 32 - Anna.exe & Emergence (overview), 33 - Emergent Generalization, 34 - Reverse-Engineering Anna.exe. New interactive Anna.exe & Emergence lab tab in the Evidence section (train a module on 2-bit addition in your browser, then watch it add 128-bit numbers it never saw), with runnable Python at /data/emergence-artifacts/.

---

May 2026

2026-05-13 - Q4: free-iteration cycle maps to Anna's top 2 antipode pairs

Re-checked on 30 September 2026; the readings below that were not confirmed stay open. See September 2026 re-check.

Follow-up to the endogenous-activity finding. The 4 states in Anna's free-iteration period-4 cycle have an output-half that lands on exactly 4 of the 19 Phase-N concept attractors, and those 4 are the top 2 antipode pairs by cluster mass:

Cycle order: concept_5 → concept_1 → concept_4 → concept_0 → concept_5 → ... (alternating between the two antipode pair families). Seeds enter at different phases; the cycle itself is unique.

The dominant attractors under input pressure and under free iteration are the same set. This is a measurable architectural bridge between the two regimes, not a consciousness claim. Chapter 30 gets §14 documenting this.

Audit: __, __.

2026-05-13 - Endogenous activity test (free-iteration extension to Chapter 30)

Pre-registered test of whether Anna's matrix architecture sustains intrinsic temporal dynamics under free iteration sign(M · v) with NO clamped input.

• Anna result: 25/25 random initial states (T=200×20 + T=10,000×5) converge to a period-4 cycle. Transient ~7-9 ticks before entering cycle. Verdict locked: ANNA SUSTAINS ENDOGENOUS ACTIVITY in bounded periodic form.
• Random antipodal control: 0/5 cycles within 10,000 ticks; orbits, if cyclic, are > 10,000 long.
• EngineeredDCOnly (Walsh DC-only): period-2 in 10/10.
• EngineeredWalsh (Anna-frequency mimic, shuffled rows): variable cycle lengths 1-4 across 10 seeds.

Two-layer reading: sustained short-cycle dynamics is a Walsh-family property; the specific period-4 is an Anna-specific property (depends on row arrangement, not just frequency spectrum). Lerchner-compatible framing, measurable architectural property, not a consciousness claim.

Integration: Chapter 30 (Period-4 Anti-Attractor) gets a new §13 documenting the free-iteration extension.

Audit trail: __, TASK_ENDOGENOUS_ACTIVITY.md, endogenous_activity_results.json, endogenous_extended_results.json.

---

2026-05-12 - Cognitive Lab + Anna Concept Lab (17 interactive modules)

Re-checked on 30 September 2026; the readings below that were not confirmed stay open. See September 2026 re-check.

Two full Evidence-Vault labs built since the last public push:

Cognitive Lab (Evidence Vault → Cognitive Lab tab), live HyperIdentity-pressure evolution on a pure-logic engine (bit-packed sign-only AIT, Web Worker thread).

• HyperIdentity Evolution, live 3D Anna Brain + particle-life evolution plane; population converges toward the HyperIdentity Config0 reference structure.
• Concept Emergence, discovers stable basins in a small population; cross-linked to Cluster Discovery for the honest emergence demo.
• Multi-Task Pareto, fitness scatter across multiple tasks; shows substrate ceiling and multi-task interference onset.
• Structural Convergence, radar chart of Anna's 7-axis structural fingerprint over generations (5 functional axes that emerge under selection + 2 identity axes that stay at 0%); Anna's fingerprint as dashed reference.
• Cognitive Trajectory (tech), full per-generation state trace, exposed for researchers.
• Visual Metaphor marked deprecated, earlier "Anna is a brain" framing was misleading; new default: "we measure how matrices change under selection."

Anna Concept Lab (Evidence Vault → Anna Concept Lab tab), built on the production AIT TypeScript port (matches reference algorithm across 16,384 / 16,384 known inputs).

• Findings, 12 pre-registered predictions vs live measurement. 11/12 passed; 1 falsified and removed.
• Try Your Input (easy), interactive 64-bit input panel; runs the production AIT path and shows the closest of 19 known concepts.
• Concept Universe (easy), all 19 attractors in 2D PCA; antipodal pairs connected; PC1 axis carries 81.7% of variance.
• Similarity Matrix (easy), 19×19 Hamming-distance heatmap over concept centroids.
• Dendrogram, single-linkage Hamming dendrogram; cophenet 0.68, 45σ above random. The "19 concepts" finding lives here.
• Cluster Discovery (tech), silhouette sweep k=2..32. Anna peaks at k=2 (silhouette ≈ 0.89); random peaks at k≈3 (≈ 0.13). 7× tightness ratio is the legitimate emergence signal.
• Trajectory Cloud, premium 3D scene showing AIT iteration as brain-tissue-like flow; validated tick-by-tick against the reference algorithm.
• Architecture, overview of AIT structure: clamped-input iteration, ternary state, convergence criterion.
• Bit-flip Game (easy), interactive challenge demonstrating Anna's avalanche bimodality.
• Live Inference (tech), every tick of AIT rendered live with state vector and convergence diagnostic.
• Anna vs Random, side-by-side comparator: Anna converges to a stable concept; random typically does not.

Explainability layer. Every module supports four reading levels via the ExplainPanel component: kid (5-year-old metaphor), simple (curious non-specialist), researcher (methods + params), math (KaTeX, lazy-loaded). Glossary tooltip layer covers technical terms (bit, matrix, AIT, attractor, basin, antipodal symmetry, etc.) without dense prose.

Brutal-review verification pass (2026-05-12 evening) on top of the labs:

• Parity 60% claim re-verified at n=20 seeds: mean best-train 60.94% ± 11.59%, mean held-out 52.66% (4.6× random). The 60% threshold is crossed at the mean, not as a lucky single seed.
• Walsh-Hadamard mechanism finding generalizes from popcount to parity: a hand-designed DC-only Walsh matrix beats Anna on parity by 0.60σ. Anna is one valid member of the sparse Walsh-Hadamard family, not the optimum within it.
• Aigarth+Qubic Vision doc (/docs/04-discussion/05-aigarth-qubic-vision): two disclaimer callouts added. (1) Quote provenance, no public source located for the four attributed CfB Discord quotes despite targeted web search. (2) Reading scope, each section explicitly classified fact / interpretation / speculation.
• Aigarth Lab Section 14 wording softened ("Mathematical breakthrough" → "Spectral characterization"); new parity-engineering table rendered alongside popcount table.
• Aigarth Lab Section 13 re-sourced to n=20 numbers; explicit mean ± SD; per-seed-variance caveat preserved.
• Cluster Discovery silhouette chart, labels moved from in-chart positions to header badge; eliminates overlap with Y-axis and with each other at narrow widths.
• Evidence tabs mobile overflow, tab strip at 375px viewport no longer scrolls the page horizontally; internal scroll with gradient-fade affordance.

Verification surface. 20 audit gates green (10 unit + 4 audit + 6 multi-seed). Production TypeScript build clean. Every numeric claim in the UI sourced from a JSON results file.

Path 2, Scale-up verification (2026-05-12 night, pre-registered before any compute):

• Task 2 (mod_11 protocol audit, LOCKED): Anna-evolved Phase 3 mod_11 winners decode to integers at 5.3% (first-tile) / 6.0% (majority), BELOW the 9.1% chance for 11-class. The +4.08σ Phase 3 signal is a substrate-bias detected at the bit-correlation level, NOT end-to-end integer capability. The two protocols (Phase 3 sign-matrix evolution vs ITU framework) measure different things; no contradiction.
• Task 3 (Walsh on gridworld, LOCKED): EngineeredDCOnly within 0.46σ of Anna on 4×4 best, pre-reg verdict "GENERALIZES". BUT: SIGN_RANDOM init beats both Anna and DC_ONLY by 0.93σ. Goal-directed tasks at this scale are dominated by initialization noise, not substrate structure. Walsh substrate generalizes weakly to non-count-aligned tasks; random init does best.
• Task 4 (Anna-vs-Random, LOCKED): Random-init Phase 4 addition mean = 0.144 in pre-reg [0.10, 0.25] band → verdict "ANNA HELPS, FRAMEWORK CAN". Framework alone reaches ~56% of Anna's level on adaptive task; popcount specifically remains Anna-necessary (random 2.5% held-out).
• Task 1 (parity n=100): ABORTED at ~47% to free CPU; n=20 result stands. Pre-reg verdict on n=20: APPROACHES (mean 60.94% > 0.60, 95% CI [0.559, 0.660] includes both 0.58 and 0.60). The 60% claim is supported at the mean but not locked at 95% confidence, would need n ≥ 40-100 to tighten.

Refined publishable claim line (post-Path-2): task-conditional priors. Anna's substrate matters when the task family aligns with its spectral signature; otherwise initialization noise dominates.

• Count-aligned tasks (popcount, parity): Walsh substrate IS the engine; Anna is one member, EngineeredWalsh/DC-Only beat her.
• Adaptive tasks (Phase 4 addition): Anna seed provides ~80% boost (0.256 vs 0.144) but framework alone reaches half the level.
• Goal-directed tasks (gridworld): Neither Anna nor Walsh substrate helps; random init wins.

Path 2 pre-reg: __. Per-task audits at __.

UI polish (in same release):

• ParetoScatter (Multi-Task Pareto in Cognitive Lab), fixed axis-tick 0.4/0.4 corner collision; ticks now start at 0.5, X-axis label has more bottom clearance.

• StructuralRadar (Structural Convergence in Cognitive Lab), hid radial-tick labels (the "1.0" tick was overlapping with the "antipodal antisymmetry" axis label at the top of the radar).

• Blog post

---

April 2026

2026-04-30 - First Book Published

• New: Books, Qubic, The Long Version by Qubic Church (18 chapters, ~200 pages, distributed freely).
• Three downloads: English PDF, original Russian PDF, and the Anna Matrix Companion, a 41-page forensic appendix designed to be loaded into an LLM alongside Chapter 13 to verify every claim with copy-paste Python snippets against the published Anna Matrix data.
• Companion canonical home at /companion; the verification snippets in the PDF now resolve correctly via redirects.

2026-04-19 - Grok Path Verification (Chapter 28)

Re-checked on 30 September 2026; the readings below that were not confirmed stay open. See September 2026 re-check.

• New chapter: Grok Path Verification in a new sidebar section "CFB Signature Analysis".
• Full re-verification of the 40-number sequence CFB posted on 2026-01-13 in a Grok AI exchange: all 20 matrix lookups match claimed values (100%).
• Four new Tier 1 findings: start/end rows form matrix mirror pair (45 + 82 = 127), step 1 coord sum = 137 (1/alpha), 1CFB byte 17 = 72 = PUSHBYTES_72, Satoshi Troll hash-sum arithmetic verified (2432 = 2^7 x 19, 2299 = 11^2 x 19, diff = 7 x 19).
• Satoshi Troll source traced to bitcointalk.org thread 506692 (2014-03-07); correcting earlier local-note attribution to ning.com. Full post context added to Section 4.5 of Chapter 28.
• Speculative claims (Master Key, seed IDs, probability bound, Satoshi Troll) downgraded to Tier 3 with explicit inconsistencies documented.
• Blog post

2026-04-19 - Seven-position signature (Chapter 27)

Re-checked on 30 September 2026; the readings below that were not confirmed stay open. See September 2026 re-check.

• New chapter: Seven-Position Signature & Period-4 Attractor.
• Independent verification of an external researcher's finding: every trajectory through the Anna Matrix converges to a period-4 cycle with identical asymmetry at positions [3, 18, 32, 61, 65, 101, 109].
• Null test: 0 of 1,000 shuffled-matrix trials produced any period-4 cycle.
• New observation: the entire ternary state space admits exactly two period-4 cycles.
• Blog post

2026-04-19 - Contact Cube redesign

Re-checked on 30 September 2026; the readings below that were not confirmed stay open. See September 2026 re-check.

• Evidence Vault: Contact Cube tab comprehensively redesigned.
• Bloom + Vignette post-processing on desktop, free-camera orbit, auto-tour, smooth preset transitions, color-coded anomaly columns, Genesis [72, 72] marker.
• Math correction: sum = 0 -> sum = -1 across all visualization code (Two's Complement identity).
• Address Database tab marked as coming-soon pending a properly-scoped dataset.
• Blog post

2026-04-14 - Genesis Block Puzzle (Chapter 26)

Re-checked on 30 September 2026; the readings below that were not confirmed stay open. See September 2026 re-check.

• New chapter: Genesis Block Cryptographic Puzzle.
• Thirteen verified findings covering the dual-purpose encoding of the 65-byte Genesis public key, the matrix[72][72] = -27 resonance, the period-4 cycle at step 11, the on-chain Qubic trinity (POCC / HASV / PXMARAS), and the incomplete ECDSA signature structure.
• Evidence Vault: Spectral tab marked New.
• Blog post

2026-04-13 - Spectral analysis tab

• New Evidence Vault tab: Spectral with interactive eigenvalue circle, 90.456 degree dominant eigenvalue visualization, and fullscreen scroll-reveal explainer.
• Documents the structural mechanism that produces the period-4 cooperation-rhythm observed in the ALife simulations.

---

Prior releases

Emergence Lab (March 2026)

• 10M-tick long-run validation of the Anna Matrix ALife dynamics.
• Multi-seed ensemble testing (5 seeds x 1M ticks): cooperation stabilizes at 33.0% +/- 0.1%.
• Anna-vs-Random ablation: Anna runs are 4.3x more stable with 2.8x more food-sharing and 46% less aggression.
• Hamming-3 bijection test: 341K inputs, 16,384 attractors; null-hypothesis p-value below 1 in 500M.

Matrix foundation (February 2026)

• Chapters 1-22 published covering matrix symmetry, anomaly structure, Bitcoin archaeology, Qubic-Bitcoin bridge, and source archaeology.
• Evidence Vault launched with 3D Analysis, Emergence Lab, and Address Database tabs.

---

For the full academic documentation, see Research home. For chronological updates, see Blog.