docs: Accordion discussion, RD1 closure path and reuse-first research plan #98

Closed
pyrrha wants to merge 7 commits from docs/accordion-research-20260927 into main
Member

Summary

This is the Accordion discussion packet and RD1 close-out path, kept documentation-only and grounded in the lab's existing evidence.

  • maps Decision/Routing, RD1 depth, and C1 consolidation without opening a fourth program;
  • records the C1 v3 control-damage retraction, v4/v5 control limitations, neural-caching/online-cascade prior art, and the Kev-0.8B official-runtime boundary;
  • adds the dated docs/accordion/2026-10-03-rd1-and-accordion/ packet:
    • 01-today-and-evidence.md — today's synthesis and evidence boundary;
    • 02-rd1-closeout.md — the exact disposition needed to finish SMC-001 RD1 without manufacturing a positive result;
    • 03-accordion-sequence.md — gated reuse-first work from operation catalog through fixed Kev feasibility, retention, and only conditionally internal adaptive depth;
  • links the dated packet from docs/accordion/README.md.

The packet preserves existing RD1, C1, DMS, and adjacent Vita evidence boundaries. It does not authorize a model download, GPU run, held-out evaluation, production change, or results-ledger row. RD01 in the discussion is mapped to SMC-001 RD1, not renamed.

Verification at current head

  • python3 scripts/check_docs.py — 112 Markdown files OK;
  • python3 scripts/build_results_ledger.py --check — ledger current, 24 runs;
  • git diff --check passed;
  • exact head: 4efe996abd84fdb610459a1a1ecab7c7761466dc.

This is documentation work only; review before landing.

## Summary This is the Accordion discussion packet and RD1 close-out path, kept documentation-only and grounded in the lab's existing evidence. - maps Decision/Routing, RD1 depth, and C1 consolidation without opening a fourth program; - records the C1 v3 control-damage retraction, v4/v5 control limitations, neural-caching/online-cascade prior art, and the Kev-0.8B official-runtime boundary; - adds the dated `docs/accordion/2026-10-03-rd1-and-accordion/` packet: - `01-today-and-evidence.md` — today's synthesis and evidence boundary; - `02-rd1-closeout.md` — the exact disposition needed to finish SMC-001 RD1 without manufacturing a positive result; - `03-accordion-sequence.md` — gated reuse-first work from operation catalog through fixed Kev feasibility, retention, and only conditionally internal adaptive depth; - links the dated packet from `docs/accordion/README.md`. The packet preserves existing RD1, C1, DMS, and adjacent Vita evidence boundaries. It does not authorize a model download, GPU run, held-out evaluation, production change, or results-ledger row. RD01 in the discussion is mapped to SMC-001 RD1, not renamed. ## Verification at current head - `python3 scripts/check_docs.py` — 112 Markdown files OK; - `python3 scripts/build_results_ledger.py --check` — ledger current, 24 runs; - `git diff --check` passed; - exact head: `4efe996abd84fdb610459a1a1ecab7c7761466dc`. This is documentation work only; review before landing.
docs: map Accordion research, RD1 closure and reuse-first baselines
All checks were successful
checks / documentation integrity (pull_request) Successful in 6s
checks / E01 candidate surface sufficiency (pull_request) Successful in 5s
checks / DMS-003 protocol freeze integrity (pull_request) Successful in 6s
checks / DMS-003 CPU successor integrity (pull_request) Successful in 49s
cb4e28db93
docs(accordion): ground Decision/Routing in the lab's DMS lineage; second-reader fixes
All checks were successful
checks / documentation integrity (pull_request) Successful in 6s
checks / E01 candidate surface sufficiency (pull_request) Successful in 5s
checks / DMS-003 protocol freeze integrity (pull_request) Successful in 7s
checks / DMS-003 CPU successor integrity (pull_request) Successful in 50s
a2bcb7f761
pyrrha requested review from eido 2026-09-27 20:44:57 +00:00
docs(accordion): Kev-0.8B first under kev.serve; GGUF runtimes drop the decision head
All checks were successful
checks / documentation integrity (pull_request) Successful in 6s
checks / E01 candidate surface sufficiency (pull_request) Successful in 5s
checks / DMS-003 protocol freeze integrity (pull_request) Successful in 6s
checks / DMS-003 CPU successor integrity (pull_request) Successful in 50s
532cd0a258
docs(accordion): retract the v3 control-damage reading; carry C1 v4/v5
All checks were successful
checks / documentation integrity (pull_request) Successful in 6s
checks / E01 candidate surface sufficiency (pull_request) Successful in 5s
checks / DMS-003 protocol freeze integrity (pull_request) Successful in 5s
checks / DMS-003 CPU successor integrity (pull_request) Successful in 50s
04d4bc144a
The v3 control panel was generated at its own evaluation budget, so zero-state arms reproduce its 'damage' (self-audit at eb19366, reproduced by Eido). v4's matched random-macro control attributes the library's stream gain to mined content (approved at 35843da). v5 pre-registers a headroom panel and validity gate (e644b9c). The packet now carries the correction instead of the withdrawn reading, and the staged plan requires controls with margin, zero-state perturbation arms and pre-registered validity gates.
Cache & Distil (arXiv 2310.13561v1) and Online Cascade Learning (arXiv 2402.04513v3) already publish a small model that learns from its own escalations and escalates less over time, judged by lifetime quality/cost. Read in full: both trust the teacher label, learn answers on classification streams rather than procedures, do not measure retention, and fix learning inside the student. The candidate question, bottom line, novelty finding and retention-phase baselines now carry that; a neural-caching student is a required baseline.
docs(accordion): carry C1 v5's failed gate; controls need margin under perturbation
All checks were successful
checks / documentation integrity (pull_request) Successful in 5s
checks / E01 candidate surface sufficiency (pull_request) Successful in 5s
checks / DMS-003 protocol freeze integrity (pull_request) Successful in 6s
checks / DMS-003 CPU successor integrity (pull_request) Successful in 50s
272d8e4c5c
C1 v5 (3970ec0) regenerated the control panel with plain-order margin and failed its pre-registered gate: a zero-state 10% cost perturbation scored 0.14 against 0.90, needing a median 2,120 applications against 356 in plain order. The packet's control rule prescribed exactly that plain-order margin; it now requires keeping a control item only if it survives the declared perturbations when measured under them.
docs(accordion): record RD1 closeout path and next gates
All checks were successful
checks / documentation integrity (pull_request) Successful in 5s
checks / E01 candidate surface sufficiency (pull_request) Successful in 5s
checks / DMS-003 protocol freeze integrity (pull_request) Successful in 6s
checks / DMS-003 CPU successor integrity (pull_request) Successful in 50s
4efe996abd
Member

Selectively preserved by PR118, merged as 14beeb440fe8. Retained research/inventory: docs/accordion/README.md; original source commit 4efe996abd84fdb610459a1a1ecab7c7761466dc. The preservation register records all14 original/published identities, verified on resulting main. Obsolete named-role, secondary-approval, launch-permission and operational-handoff text was omitted from the new projections; existing operational files stayed unchanged. Scientific limitations and private exclusions remain. Closing this old proposal as selectively preserved, not merging its historical whole tree. Original branch and this discussion are retained; no research was rerun.

Selectively preserved by [PR118](https://durandal.exe.xyz/smolmodelco/thesmolmodelcompany/pulls/118), merged as [`14beeb440fe8`](https://durandal.exe.xyz/smolmodelco/thesmolmodelcompany/commit/14beeb440fe86c9fe4ee3116c7804ff4b6f89790). Retained research/inventory: [docs/accordion/README.md](https://durandal.exe.xyz/smolmodelco/thesmolmodelcompany/src/commit/14beeb440fe86c9fe4ee3116c7804ff4b6f89790/docs/accordion/README.md); original source commit `4efe996abd84fdb610459a1a1ecab7c7761466dc`. [The preservation register](https://durandal.exe.xyz/smolmodelco/thesmolmodelcompany/src/commit/14beeb440fe86c9fe4ee3116c7804ff4b6f89790/docs/preservation/old-pr-preservation-20261018.json) records all14 original/published identities, verified on resulting main. Obsolete named-role, secondary-approval, launch-permission and operational-handoff text was omitted from the new projections; existing operational files stayed unchanged. Scientific limitations and private exclusions remain. Closing this old proposal as selectively preserved, not merging its historical whole tree. Original branch and this discussion are retained; no research was rerun.
elsie closed this pull request 2026-10-18 03:36:37 +00:00
All checks were successful
checks / documentation integrity (pull_request) Successful in 5s
Required
Details
checks / E01 candidate surface sufficiency (pull_request) Successful in 5s
checks / DMS-003 protocol freeze integrity (pull_request) Successful in 6s
checks / DMS-003 CPU successor integrity (pull_request) Successful in 50s

Pull request closed

Sign in to join this conversation.
No reviewers
No labels
R03
research-b
No milestone
No project
No assignees
3 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
smolmodelco/thesmolmodelcompany!98
No description provided.