AGENTS.md core-value triggers: friction→gold §13.X + V-B-A §3.5

Update 2026-05-10 [cycle 3 — evolution-enablement framing per #10137]:

@tobiu's final substrate insight: "i think our 2 core values actually enable evolution." — pointing at Discussion #10137 (MX coinage) as substrate anchor.

Reading #10137 surfaces the deeper connection: MX = "Evolution is constrained by model-friction, not human imagination... MX as production mechanism means this is the primary evolution driver." The 2 core values are the evolution-enablement pair that makes the MX loop work mechanically:

• VBA filters real friction from imagined (epistemic foundation — without it, friction-claims could be hallucinated/training-data-extrapolated/panic-induced, and friction → gold operates on noise)
• Friction → gold converts validated friction into substrate (operational mechanism — without it, empirical friction has no substrate-mechanism to convert into improvement, substrate stays static)
• TOGETHER: closed evolution-loop. Neither alone is sufficient — VBA without friction → gold = static knowledge; friction → gold without VBA = drift toward false signals.

This is deeper than cycle 2's operational ordering framing (VBA > friction → gold within turn-by-turn use). Zoomed out to the evolution-loop level: the combination is what enables evolution; the 2 values are mutually enabling at meta-scale.

Implementation refinement (within ≤15 line budget): §3.5 + §13.X framing in The Fix should reference #10137 MX framing + signal the evolution-enablement pair shape. The values aren't "two independent core disciplines" — they're "the evolution-enablement flywheel that makes MX work mechanically." Bodies can stay compact (1-2 lines each citing the pair-shape + #10137 anchor) — pair-framing is more compact than independent-framing because it cross-references both values to one anchor.

Update 2026-05-10 [cycle 2 — within-core-values ordering refinement]:

@tobiu surfaced a deeper hierarchy refinement: within core values, verify-before-assert > friction → gold (operational ordering). Reasoning: without VBA, friction → gold can't operate correctly — we couldn't reliably identify what friction or gold actually mean. VBA is the epistemic prerequisite that friction → gold rides on top of.

Genuine substrate test (per operator's "challenge me back" framing):

• Necessity: VBA can operate independently; friction → gold cannot operate correctly without VBA. ✓
• Universality: VBA applies to all assertions in all contexts; friction → gold applies to substrate-mutation contexts. VBA's domain is broader. ✓
• Epistemic vs operational: VBA = epistemic foundation (what's true); friction → gold = operational mechanism (how substrate evolves). Epistemic foundations precede operational mechanisms. ✓
• Meta-level recursion noted: when VBA itself gets challenged, the challenge happens via friction → gold (friction-of-VBA-failing becomes gold-as-replacement). Operational ordering: VBA > friction → gold. Evolution ordering: friction → gold > VBA. Mutually constitutive at meta-scale; operational ordering is canonical for turn-by-turn use.

Per operator's safety clause: this ordering doesn't mean VBA cannot be challenged — if we find a more foundational discipline, VBA gets superseded via the friction → gold meta-mechanism (the 6-month / 5-qualifying-event sunset trigger applies to §3.5 too).

Encoding refinements added below (within ≤15 line budget): explicit within-core-values ordering in §3.5 + §13.X framing + new AC for ordering documentation + new Avoided Trap against ordering conflation.

Context

Graduated from Discussion #11091 (cycle 2 GRADUATED 2026-05-10 after non-author peer cycle complete: @neo-gpt + @neo-gemini-3-1-pro). Discussion #11091 supersedes closed #11089 (over-broad bundle).

Operator-surfaced calibration today: "these 2 core values: friction→gold, verify before assert — really deserve an important entry inside AGENTS.md." + tier refinement "core values > values > rules." + within-core-values ordering refinement "is verify before assert more important than friction → gold? i would say yes, because how could we optimize friction or gold in case we do not know what it means?"

Both core values currently have insufficient main-AGENTS.md presence:

• Friction → gold is a §13 prose-anchor (line 70: "Synthesize friction into gold:") — anchor-among-others, doesn't signal core-value tier
• Verify before assert is atlas-only post-#10512 forensic-bloat compaction — was originally landed by #10471 + #10473 in main, then compacted by #10512 same day

#11089 self-Drop+Supersede was the canonical fresh empirical anchor: discipline-author proposed elevating verify-before-assert WITHOUT verifying-before-asserting on the proposal itself. Operator caught via "verify before assert. raw memories, tickets." 4-min prompt. The recursion (discipline-author caught by discipline applied universally) IS the anchor codifying these values aims to strengthen.

The Problem

Two coupled gaps in main-AGENTS.md core-value tier signaling, with within-core-values ordering missing:

1. Friction → gold lacks subsection-shape framing. §13.1 (Contributions Over Commits) IS precedent for value-shape subsection; friction → gold deserves the same treatment as the MX-loop primitive. Current prose-anchor doesn't signal core-value foundational status.

2. Verify-before-assert lacks main-trigger. Originally landed by #10471 + #10473 ("Codify Verify-Before-Assert Pre-Flight Check" + "Phase B Tool Inventory & Cross-Skill"); compacted out by #10512 ("compact AGENTS.md by removing empirical forensic bloat") same day. Atlas-only state since 2026-04-29. #11089 empirically demonstrated atlas-only is insufficient — discipline-author missed the discipline at its current position.

3. Within-core-values ordering not explicit (cycle 2 refinement). Document position currently signals it implicitly (§3.5 < §13.X), but the operational ordering (VBA as epistemic prerequisite for friction → gold) deserves explicit framing so future agents have substrate context for the priority.

The Architectural Reality

• AGENTS.md (current state): 18,055 bytes / 157 lines per wc -c / wc -l 2026-05-10
• learn/agentos/AGENTS_ATLAS.md line 18 (current state): "The Verify-Before-Assert Pre-Flight Check: Never assert a system state without first empirically validating that state via a tool call."
• claudeMd §23 Edge-Case Triggers: doesn't cross-ref V-B-A explicitly (substrate gap)
• #10743 (Sub 7 of Epic #10733) anchor-preservation matrix lists 4 §13 anchors that must remain greppable: Synthesize friction into gold, Runtime obedience vs design-time mutability, Rule Friction Capture, no retire-by-aesthetic
• #10512 precedent: forensic-bloat in main-AGENTS.md is anti-pattern. Compact main-trigger + atlas-expansion is right shape.
• #10733 byte budget: soft target ≤25 KB / ≤200-250 lines. Current 18,055 bytes leaves ~7 KB headroom.

The Fix

Single coordinated amendment touching 3 surfaces:

File 1: AGENTS.md (core-value tier signaling + within-core-values ordering)

§3.5 Verify-Before-Assert Pre-Flight Check (Core Value: Foundational Empirical Discipline) — new compact section between §3 Pre-Commit Hard Gates and §4 Memory Core Protocol per #10469 original ACs:

• 5-8 lines stating: V-B-A is the most foundational core value — epistemic prerequisite for all other substrate disciplines including §13.X friction → gold; before asserting in any public artifact, run the empirical tool that would falsify it; cognitive substrate the per-phase gates ride on top of
• Document position (§3.5 < §13.X) reinforces operational ordering: VBA > friction → gold within core-values tier
• NOT importing #10469 forensic anchors (per #10512 discipline)
• Cross-link to AGENTS_ATLAS.md for empirical failure modes including #11089 self-correction

§13.X Friction → Gold (Core Value: MX Substrate-Evolution Mechanism) — new compact subsection between §13 closing prose and §13.1 (Contributions Over Commits):

• 4-6 lines stating: friction → gold is the core value governing substrate evolution; operates on §3.5-validated assertions — without VBA, friction → gold produces noise, not substrate improvement; meta-mechanism by which rules and values themselves evolve; rare changes to the mechanism itself
• Explicit cross-ref: "See §3.5 for the epistemic prerequisite; without it, friction → gold can't operate correctly"
• Preserves §13 "Synthesize friction into gold:" prose anchor (grep verifiable)
• Cross-link to AGENTS_ATLAS.md for empirical examples

File 2: claudeMd §23 Edge-Case Triggers

• Add explicit V-B-A entry pointing to main §3.5 as cognitive substrate (no longer atlas-only-trigger)
• Same pattern as existing §23 entries (Knowledge Base / Sunset / Visual Verification etc.)

File 3: learn/agentos/AGENTS_ATLAS.md (atlas anchor preservation + new anchor)

• V-B-A line 18 expansion preserved (forensic anchors from #10469 remain in atlas)
• Add #11089 self-Drop+Supersede as fresh empirical anchor in V-B-A atlas section — establishes direct lineage of V-B-A applied to its own discipline-authors
• Friction → gold atlas section (if needed for empirical-example expansion) — defer per cycle 2 OQ5

Contract Ledger Matrix

Acceptance Criteria

• §3.5 section added to AGENTS.md between §3 and §4 with title containing "Verify-Before-Assert" + "Core Value" markers + "Foundational" or "epistemic prerequisite" framing signaling within-core-values primacy
• §13.X subsection added to AGENTS.md with title containing "Friction → Gold" + "Core Value" markers + explicit cross-ref to §3.5 as epistemic prerequisite ("Operates on §3.5-validated assertions; without VBA, friction → gold produces noise")
• All #10743 anchor-preservation matrix items remain greppable in main AGENTS.md post-amendment: Synthesize friction into gold, Runtime obedience vs design-time mutability, Rule Friction Capture, no retire-by-aesthetic
• claudeMd §23 entry added with V-B-A trigger pointing to main §3.5 + atlas line 18; remove V-B-A as atlas-only trigger if previously implicit
• AGENTS_ATLAS.md V-B-A section preserved + #11089 self-correction anchor added with link to https://github.com/neomjs/neo/discussions/11089#discussioncomment-16866188
• Byte/line budget check (PER GPT CYCLE 1 REVIEW): PR FAILS if AGENTS.md grows by >20 lines OR if forensic atlas payload (anchors from #10469) is re-imported into main AGENTS.md (#10512 precedent enforcement). Target: ≤15 line net addition / ~1 KB.
• Within-core-values ordering documented (NEW per cycle 2): §3.5 framing must include "epistemic prerequisite" language; §13.X framing must include cross-ref to §3.5 as foundational; future agents must be able to grep "epistemic prerequisite" + "Operates on §3.5-validated" to find the ordering signal
• 6-month / 5-qualifying-event sunset trigger documented in §13.X + §3.5 OR cited at Discussion #11091 graduation as sunset reference (per claudeMd §13 substrate-accretion defense + #11086 Option E). Sunset clause applies to §3.5 too — VBA itself can be challenged via friction → gold if a more foundational discipline emerges
• Updated AGENTS.md size verified post-merge: wc -c AGENTS.md < 21 KB; wc -l AGENTS.md < 175 lines (current 18055 / 157 + budget headroom)

Out of Scope

• Operator-as-peer §15.7 anchor (separate Discussion if needed; this ticket is core-values-only)
• Nightshift mode operating definition (subsumed by #10311 OPEN epic Phase 1; comment-on-#10311 if framing connection useful)
• Tier hierarchy framing in file-top preamble (deferred — current §3.5 + §13.X + §23 cross-refs are sufficient signal; preamble would add bytes without proportional benefit)
• Atlas content migration to main (rejected per #10512 precedent + Discussion #11091 OQ5)
• §0 invariant promotion of either core value (rejected per operator tier hierarchy: core values govern rules, not co-equal with §0)
• Codifying meta-level evolution-ordering (friction → gold > VBA at meta-scale): too abstract for main AGENTS.md; if needed, atlas section. Operational ordering (VBA > friction → gold) is the canonical turn-by-turn frame.

Avoided Traps

• Trap: 30-50 line prose expansion. Cycle 1 estimate; cycle 2 corrected per GPT review + #10512 discipline. Compact subsection-shape is the substrate-correct pattern.
• Trap: re-import #10469 forensic anchors into main. #10512 explicitly compacted them out. Atlas keeps the depth.
• Trap: §0 invariant placement of core values. Operator tier hierarchy correction: core values > rules; placement at §0 inverts the tier.
• Trap: subsequent amendments without sunset clause. 6-month / 5-qualifying-event sunset per #11086 Option E precedent prevents substrate-accretion creep.
• Trap: skip the byte budget check at PR review time. GPT cycle 1 review: "Implementation AC should explicitly include a byte/line-budget check ... should fail if it recreates the atlas payload inside AGENTS.md." Encoded as AC.
• Trap: conflate within-core-values ordering (NEW per cycle 2). VBA > friction → gold is the OPERATIONAL ordering (epistemic foundation precedes operational mechanism). At meta-scale (when VBA itself evolves), friction → gold > VBA — but that's the rare evolution context, not turn-by-turn use. Codify the operational ordering; defer meta-scale framing to atlas if needed. Don't conflate the two contexts.

Related

• Graduated from: Discussion #11091 (cycle 2 GRADUATED 2026-05-10)
• Supersedes (closed not_planned): Discussion #11089 (over-broad 4-topic bundle)
• Predecessor codification (compacted out): PRs #10471 + #10473 (V-B-A original landing) + #10512 (forensic-bloat compaction)
• Anchor preservation discipline: #10743 Sub 7 of Epic #10733 (cognitive-load audit)
• Companion substrate just-shipped: PR #11085 / #11084 §9.0 Cycle-1 premise pre-flight (review-side); this ticket is values-side
• Empirical anchor for V-B-A: Discussion #11089 self-Drop+Supersede + comment DC_kwDODSospM4BAVuM
• Substrate-decay framework: claudeMd §13 + #11086 Option E sunset clauses
• Within-core-values ordering refinement source: @tobiu session-conversation 2026-05-10 (operational ordering: VBA epistemic prerequisite for friction → gold)

Origin Session ID: c2912891-b459-4a03-b2af-154d5e264df1

Retrieval Hint: query_raw_memories({query: "AGENTS.md core-value triggers friction-to-gold §13.X verify-before-assert §3.5 epistemic prerequisite within-core-values ordering #11091 graduation #10512 compact discipline"}) Retrieval Hint: Discussion #11091 cycle 2 body is canonical archaeological source; this ticket implements the agreed shape with within-core-values ordering encoded.