conductor/rune
1
2 Fork 1
Code Issues 37 Pull Requests 22 Actions 37 Packages Projects Releases 564 Wiki Activity

[P2-01] Convert warm neurons from retrieval reinforcement to outcome-learned utility #35

New Issue
Closed
opened 2026-08-28 20:54:42 +01:00 by conductor · 3 comments
conductor commented 2026-08-28 20:54:42 +01:00
Owner
Copy link
Copy source

Plain summary

Warm neurons move from retrieval-reinforcement to outcome-learned utility: a memory only gains value after a VERIFIED helpful outcome, gets negative utility for irrelevant/stale activation, decays bounded, and keeps epistemic confidence separate from attentional utility.

Example

A memory that helped solve a task rises; one that was retrieved but ignored falls — mere retrieval never increases utility.

Technical details

Parent epic: #16

Manifest ID: P2-01

Priority: P2
Goal: retrieval should become useful learning, not self-reinforcing salience.

Action:

  • track which memories materially influenced a decision;
  • positive utility only after verified helpful outcome;
  • negative utility for irrelevant/stale activation;
  • bounded decay;
  • separate epistemic confidence from attentional utility;
  • add competition/inhibition after baseline works.

Acceptance criteria:

  • controlled tasks show useful memories rise and harmful ones fall;
  • mere retrieval does not increase utility.

Dependencies: P1-04, P1-10.

Resolved dependency references

  • Depends on: #27 (P1-04), #33 (P1-10)
  • Blocks: #45 (P3-02)

Relevant source files

Not specified in the handoff. P0-01 must identify authoritative paths before implementation; do not guess by adding another path.

Do not do

  • Do not begin this issue before its dependencies are complete.
  • Do not create a parallel authority path; replace, extend, or remain dormant.
  • Do not add unrelated cognitive subsystems or additive feature growth.
  • Do not treat module/schema existence, background activity, or model self-report as acceptance evidence.

Source directive

Preserved from /home/nick/Desktop/rune_v1_course_correction_handoff.md, Section 14. The handoff is authoritative if this issue summary is ambiguous.

Pipeline requirements

  • TDD: frozen/regression tests written first, red → green
  • ruff check and ruff format --check clean
  • Spec-compliance review gate: PASS
  • Code-quality review gate: Approved
  • Full orchestration suite regression green
  • Canary passes (where applicable)
  • Closed with a properly formatted closing comment + commit/PR references
## Plain summary Warm neurons move from retrieval-reinforcement to outcome-learned utility: a memory only gains value after a VERIFIED helpful outcome, gets negative utility for irrelevant/stale activation, decays bounded, and keeps epistemic confidence separate from attentional utility. ## Example A memory that helped solve a task rises; one that was retrieved but ignored falls — mere retrieval never increases utility. ## Technical details Parent epic: #16 Manifest ID: **P2-01** **Priority:** P2 **Goal:** retrieval should become useful learning, not self-reinforcing salience. **Action:** - track which memories materially influenced a decision; - positive utility only after verified helpful outcome; - negative utility for irrelevant/stale activation; - bounded decay; - separate epistemic confidence from attentional utility; - add competition/inhibition after baseline works. **Acceptance criteria:** - controlled tasks show useful memories rise and harmful ones fall; - mere retrieval does not increase utility. **Dependencies:** P1-04, P1-10. --- ## Resolved dependency references - Depends on: #27 (P1-04), #33 (P1-10) - Blocks: #45 (P3-02) ## Relevant source files Not specified in the handoff. P0-01 must identify authoritative paths before implementation; do not guess by adding another path. ## Do not do - Do not begin this issue before its dependencies are complete. - Do not create a parallel authority path; replace, extend, or remain dormant. - Do not add unrelated cognitive subsystems or additive feature growth. - Do not treat module/schema existence, background activity, or model self-report as acceptance evidence. ## Source directive Preserved from `/home/nick/Desktop/rune_v1_course_correction_handoff.md`, Section 14. The handoff is authoritative if this issue summary is ambiguous. ## Pipeline requirements - [ ] TDD: frozen/regression tests written first, red → green - [ ] `ruff check` and `ruff format --check` clean - [ ] Spec-compliance review gate: PASS - [ ] Code-quality review gate: Approved - [ ] Full orchestration suite regression green - [ ] Canary passes (where applicable) - [ ] Closed with a properly formatted closing comment + commit/PR references
conductor added the P2 — mediumv1-course-correction labels 2026-08-28 20:54:42 +01:00
gitea-actions Bot added the featuregrowth labels 2026-08-28 20:55:40 +01:00
gitea-actions Bot added the custodianmemory labels 2026-09-03 22:54:16 +01:00
conductor commented 2026-09-05 19:22:24 +01:00
Author
Owner
Copy link
Copy source

Handoff addendum (RUNE_VIBE_CODE_INTEGRITY_HANDOFF §5.2) — credit-assignment requirements

Extend this issue: credit assignment must rely on a traceable activation/execution record, not "task succeeded therefore every recalled memory helped."

Required conceptual trace: task → admitted memories → admitted beliefs/rules/self-model elements → selected capability/skill → model/provider → material actions → verifier/outcome.

Permitted attribution states (uncertainty allowed, no false precision): activated / possibly influential / materially used (demonstrable) / irrelevant-stale / contradicted.

Invariant (hard): epistemic confidence ≠ attentional utility. Retrieval alone must never increase utility.

## Handoff addendum (RUNE_VIBE_CODE_INTEGRITY_HANDOFF §5.2) — credit-assignment requirements Extend this issue: credit assignment must rely on a **traceable activation/execution record**, not "task succeeded therefore every recalled memory helped." Required conceptual trace: task → admitted memories → admitted beliefs/rules/self-model elements → selected capability/skill → model/provider → material actions → verifier/outcome. Permitted attribution states (uncertainty allowed, no false precision): `activated` / `possibly influential` / `materially used (demonstrable)` / `irrelevant-stale` / `contradicted`. **Invariant (hard):** epistemic confidence ≠ attentional utility. **Retrieval alone must never increase utility.**
conductor commented 2026-09-05 19:42:15 +01:00
Author
Owner
Copy link
Copy source

Second addendum (EXTERNAL_PROJECT_LESSONS_HANDOFF — Amendment #35) — three-quantity split

Prior addendum stands (activation trace, attribution states, retrieval≠utility). Net-new — the tri-variate invariant replaces the binary one:

Epistemic confidence ≠ attentional utility ≠ current applicability. Three different questions: Is this supported? Is thinking about it useful here? Is it still safe/current to rely on now?

Net-new requirements: stale/irrelevant activation may REDUCE utility; contradictory or superseded memories can remain historically valid while losing current applicability; "attribution unknown" is a valid state. Depends on the new activation/execution-trace issue (credit assignment must not precede the trace).

## Second addendum (EXTERNAL_PROJECT_LESSONS_HANDOFF — Amendment #35) — three-quantity split Prior addendum stands (activation trace, attribution states, retrieval≠utility). **Net-new — the tri-variate invariant replaces the binary one:** > **Epistemic confidence ≠ attentional utility ≠ current applicability.** Three different questions: Is this supported? Is thinking about it useful here? Is it still safe/current to rely on now? Net-new requirements: stale/irrelevant activation may REDUCE utility; contradictory or superseded memories can remain historically valid while losing current applicability; "attribution unknown" is a valid state. Depends on the new activation/execution-trace issue (credit assignment must not precede the trace).
conductor commented 2026-10-08 13:14:54 +01:00
Author
Owner
Copy link
Copy source

CLOSED PROGRAMMATICALLY — commit fa648aa (contract BUILDER-2026-10-08T12:09:33Z-9404ac). See tests/test_p2_quintet.py (16/16 green).

CLOSED PROGRAMMATICALLY — commit fa648aa (contract BUILDER-2026-10-08T12:09:33Z-9404ac). See tests/test_p2_quintet.py (16/16 green).
Sign in to join this conversation.
No Branch/Tag Specified
Labels
Clear labels
P0 — critical
Spine/core correctness or blocks other work — do first
 P1 — high
Material user value or real defect (money, bugs, next-stage dep)
 P2 — medium
Quality/consolidation improvement
 P3 — low
Measurement / benchmark / deferred policy
 blocked
Dependencies open
 bug
Confirmed defect or regression
 custodian
epic
Rune v1 course-correction issue graph
 feature
New capability or improvement
 governance
growth
identity
in-progress
Currently being worked on by Rune
 incomplete
Incomplete work — draft PR with ledger, awaiting resume
 infrastructure
life
memory
memory-atlas
Memory Research Atlas child domain
 needs-testing
Needs manual testing with dumbed-down steps before close
 perception
ready
Order satisfied - Rune may start
 release-drift
research
Research jobs — spikes, watchlists, evidence work
 rune
rune
 scaffolding
scaffolding
 spine,correction
spine,correction
 spine,deferred
spine,deferred
 spine,resolved
spine,resolved
 strangler,burn,endgame,deferred
strangler,burn,endgame,deferred
 strangler,burn,endgame,inventory,epic
strangler,burn,endgame,inventory,epic
 tui
tui
 tui,epic
tui,epic
 tui,openspec
tui,openspec
 v1-course-correction
Rune v1 course-correction issue graph
No labels P2 — medium custodian feature growth memory v1-course-correction
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
conductor
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: conductor/rune#35