← Complete research archive
Architecture researchClosed / no-go85 lines

Shohin ETTR Experiment History

This file is the index for experimental Git histories that have been folded into the private repository's canonical main history. It exists so rejected, superseded, incomplete, and custody-only work remains discoverable without reactivating obsolete source files in the live archi…

EXPERIMENT_HISTORY.mdOpen original Markdown ↗

Shohin ETTR Experiment History

This file is the index for experimental Git histories that have been folded into the private repository's canonical main history. It exists so rejected, superseded, incomplete, and custody-only work remains discoverable without reactivating obsolete source files in the live architecture.

Archive policy

  • Private main is the only canonical implementation line.
  • A historical branch is incorporated with an ancestry-preserving merge when its working tree is obsolete. Its commits remain inspectable with git show <tip> and git log <tip>, but its old tree does not overwrite the current architecture.
  • NO-GO, rejected, and superseded mean "do not resume without a new hypothesis and a new preregistered falsifier." They do not mean that the evidence should be deleted.
  • Preservation and repository-topology branches are retained as provenance, not counted as scientific results.
  • The sanitized public repository is a separate publication surface. Private ETTR history must never be pushed there.

Canonical line

HistoryTip before consolidationDisposition
codex/ssqac-law-compiler960333fCurrent private ETTR architecture and evidence line. This becomes private main.
codex/acw-g27433062Already contained in the canonical ancestry. Historical ACW custody baseline.
codex/carry-recovery-v9a0c258eAlready contained in the canonical ancestry. Historical carry-recovery candidate.
codex/public-ssqac-pre-sanitization-2026072778b7653Already contained in private ancestry; records the point before public sanitization.

Folded experimental histories

Historical branchPreserved tipClassificationWhy it is not the live tree
codex/cgrfc-mechanicsdcfe501Superseded mechanics / neural-fit NO-GOPreserves the source-sealed joint compiler, counterfactual-repair, equilibrium, and custody mechanics. The branch's own runbook records that mechanics existed while neural fitting and reasoning promotion remained NO-GO.
codex/er-tt-ordinal-route7d1b936Rejected route with useful diagnosticsPreserves ordinal witness routing, fresh-board receipts, opcode coupling, and the near-gate marginal-route evidence. Fresh ordinal-route v1 was formally rejected; later work replaced this compiler family.
codex/private-wip-acw-202607274215256Incomplete preservation snapshotACW G2 development work was preserved before the public scrub. Its runbook states commit/install/release remained NO-GO pending independent review; it is evidence and unfinished machinery, not a promoted result.
codex/private-wip-carry-recovery-20260727539e01dIncomplete preservation snapshotPreserves the causal-carry motor-recovery preregistration, implementation, job, and test that were not promoted into the current architecture.
codex/private-wip-er-ordinal-202607271528b7fRejected-route preservation snapshotExtends the ordinal-route history with the final pre-scrub handoff. It is retained for reconsideration but remains downstream of the rejected ordinal-route result.

Folded repository-provenance histories

Historical branchPreserved tipClassificationPurpose
codex/main-integration182738cRepository integration provenanceRecords the earlier ETTR/public integration sequence. It introduces no unique canonical tree relative to its parents.
codex/public-pre-sanitization-20260727062cc0cPublic-scrub provenanceRecords the public repository state before sanitization. Its ancestry is retained privately, but its tree is not allowed to replace private ETTR main.

Preserved non-Git evidence

The obsolete ER worktree contained two unique local result directories. They were moved intact into the canonical private evidence store before that worktree was removed:

  • artifacts/r12/er_dual_stream_fresh_score_5499768532556522119 (compiler.pt SHA-256 01cbaa8c9de2c59ce75ff0eb95b6414e1cbb4c2636ca59fca03ee59d07ef3106; assessment SHA-256 ddd607f326bf1b8cab6a5a67989ba5d5908a3323f4efba2b6c22d1184cf2e483).
  • artifacts/r12/er_dual_stream_ordinal_1790361034717866861 (compiler.pt SHA-256 99be7b89e0b7dfe35f745abf1320c6640ad61f2fb62624b288fb8f9502cd97e7; report SHA-256 6cc31e67734afa2320bc825d8bc368c780125fb7e6bb98409841e08950a2dece).

Large tensors remain intentionally ignored by Git; the paths and hashes above make their local custody explicit rather than leaving them hidden in an old branch worktree.

How to revisit an archived experiment

  1. Read the relevant result/preregistration and its contemporaneous section in AGENT_RUNBOOK.md from the preserved tip.
  2. Inspect the exact historical tree without changing the live workspace: git show <tip>:<path> or git diff <tip>^ <tip>.
  3. State what new evidence invalidates the old rejection or what defect the new design repairs.
  4. Create a fresh branch from private main; do not resurrect the old branch as the canonical line.
  5. Preserve the old evaluator and add a preregistered comparison before using new compute.

This structure keeps every failed idea recoverable while preventing abandoned implementations from cluttering or silently changing the current architecture.