coresearch: "CE64 is like BF16 for thinking" — exploration map + 5 .arxiv harvests

[AdaWorldAPI]

This PR adds research notes only. No code changes, nothing adopted.

• .claude/board/entries/2026-10-08-coresearch-ce64-bf16-for-thinking.md: the exploration map from a /coresearch council (premise gate, 5 scouts, 3 co-architects, a 21-row crosswalk, verdicts, pre-registered probes).
• .arxiv/: five papers, each read in full and written up in the existing harvest format (punchline, what it establishes, mapping, kill condition). They are 1905.12322 (BF16), 2209.05433 (FP8), 2603.24161 (limited-precision stochastic rounding), 0712.1182 (Jøsang, cumulative/averaging fission) and 2310.10537 (microscaling).

Headline findings

• The premise gate split the question. CE64 is a tagged instruction word, not a number format. "BF16 for thinking" fits only the F/C truth bytes (bits 24..39), with the truth functions as their ALU. Bits 46..49, 50..52 and 53..58 mean different things in the code than in the infographic: the opcode, per-plane plasticity flags, and a corpus handle.
• Bits 40..42 are a subset lattice, not a ladder. SO = 0b101 and PO = 0b011 are incomparable (pearl.rs), so a "rung ≥ k" predicate over that field is ill-posed.
• c = 255 needs a policy decision, not a float fix. Revision has an exact integer form in the u8 codes, and at c1 = c2 = 255 that form is 0/0 even in exact arithmetic. An f32 model of revision also found two exact .5 ties, so the rounding mode matters. This was checked in numpy; a Rust probe is still owed.
• Revision is cumulative fusion. Following Jøsang, cumulative fusion is only valid for independent evidence; for dependent evidence the idempotent averaging rule applies. The missing stamp gate is why revising an edge with itself raises c.
• D-RPF-4 needs no new opcode. Its weighted sum can be a per-byte c→w lookup feeding the existing strided group-sum.

Each idea is graded ADOPT-NOW, PROBE, PARK or SKIP, and every probe has a kill condition. The operator picks which to run.

🤖 Generated with Claude Code

https://claude.ai/code/session_01R2EuNHrariL6B3G3agyASK

---

Generated by Claude Code

[coderabbitai]

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Essentials
• Run ID: 76a315ed-4fbe-4926-b9f7-860d0789ffbd

📥 Commits

Reviewing files that changed from the base of the PR and between 083b71e and 448d733.

📒 Files selected for processing (7)

• .arxiv/0712.1182_fusion_needs_to_know_whether_evidence_is_independent.md
• .arxiv/1905.12322_narrow_storage_wide_accumulate_round_once.md
• .arxiv/2209.05433_special_values_are_declared_not_computed.md
• .arxiv/2310.10537_shared_scale_lives_outside_the_element.md
• .arxiv/2603.24161_stagnation_is_a_rounding_property.md
• .claude/board/entries/2026-10-08-coresearch-ce64-bf16-for-thinking.md
• .claude/board/entries/README.md

 ____________________________
< The OG: Original Grepster. >
 ----------------------------
  \
   \   \
        \ /\
        ( )
      .( o ).

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Warning

Billing warning: we have not been able to collect payment for this subscription for more than 72 hours. Please update the payment method or pay any pending invoices in Billing to avoid service interruption.

---

_{Comment @coderabbitai help to get the list of available commands.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-09T03:46:45.663887Z	448d733	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.