← back

i_demo_011 · researcher × autism · cross-condition Roll-formalized object

Why a researcher in autism would come to evidence.x1000.ai

Researchers in autism / neurodevelopmental conditions face a particular evidence challenge: condition heterogeneity is the norm, and any cross-condition pattern claim must be defended against the "you're conflating subtypes" objection. Pain points (see 05_docs/M625_PERSONAS.md):

• formalization language varies across labs and across DSM-vs-research-criteria conventions;
• intervention boundary Δ_B is contested (behavioral / pharmacological / mixed);
• cross-sex / cross-age trajectories are difficult to compare in a single Roll frame;
• the m620 verification ladder has two autism exec dirs (005_AUTISM and 007_AUTISM_MABOSS) which differ in modeling approach — researchers want to see how this affects the verdict.

m625 does not advise on autism research priorities. **m625 shows that m620 has run two independent autism verification exec passes and that researchers can request a third via the `verify_req` ferry**. The framework is what's offered — not specific findings.

This is not a research recommendation. This is not a clinical-decision support claim. (See "What this demo will not say".)

---

The Roll-formalized object (anchored on m620 verification ladder)

Field	Value
Upstream verification	m620 02_exec_005_AUTISM + m620 02_exec_007_AUTISM_MABOSS (L1 disk verified · W0 pre-flight 2026-05-22)
Domain	autism / neurodevelopmental (cross-condition · cross-age)
Roll TMU type	mixed (per-exec-dir varying · TMU-B candidate based on m517 CARD08 cross-condition pattern)
K4.1 bit pattern proposed	K4.1 = (0, 0, 1) for cross-condition same-X claims · awaiting m620 verdict on actual exec audit scores
Verification status	two exec dirs cooked on m620 side (researcher persona) · m625 does not assert specific score because m610/m620 v0.5.1 calibration means ±1 noise band must be tolerated
m620 follow-on	02_exec_007_AUTISM_MABOSS is the second-engine cross-check (m620 v0.6 candidate per the original v0.4 cache · disk shows the dir present)

m625 does not modify m620 exec dirs or scores. Researchers requesting a third independent verify pass do so via 06_registry/outbound_to_m620/ per 07_bridges/CONTRACT_m625_TO_m620.md.

---

How a researcher should read this evidence (≤ 5 minutes)

• Two independent autism exec dirs is a feature, not a bug. m620 keeps 005_AUTISM and 007_AUTISM_MABOSS as separate verification surfaces because the second engine cross-checks the first. A pitch deck or paper that cites a single-engine verdict in autism is structurally weaker than one that cites both.
• Researcher attack path. If a researcher believes the bit-pattern interpretation K4.1 = (0, 0, 1) mis-reads m620's exec scores, the researcher should file a verify_req to m620 stating the alternative interpretation and the falsifying prediction. m625 routes the request; m620 returns the verdict.
• Cross-condition caveat. Autism is heterogeneous · the cross-condition Roll-formalized object (from m517 CARD08 Wu 2025) is a necessary structural condition, not a sufficient substantive condition. Different m620 exec passes may reveal different K4.1 bit patterns by sub-cohort — this is expected.
• m610 inheritance. The five hard-veto families from m610 apply to autism Roll formalization just as they apply to cancer · in particular, m610 F4 (axis derivable · per m604 V2.6 Framework §4.4) and F5 (cross-period K4.1 declaration · per m604 V2.6 Framework §4.5) are not optional in autism cross-period claims.

---

Cross-domain transferability declaration (per W6.3)

The K4.1 = (0, 0, 1) cross-condition X-invariant bit pattern was originally formalized in m517 CARD08 (Wu 2025 · neuropsychiatric cross-condition). transferability_pct to autism: structural-analog only · pending m620 exec_005/007 verdict on autism-specific cross-period closure. m625 does not assert any quantitative cross-domain score in v0.3.

---

What this demo will not say (anti-drift)

• ❌ It does not promise an autism therapy or intervention works.
• ❌ It does not assign per-person autism subtype classification (NCNU P_3).
• ❌ It does not affirm or vouch for any autism researcher, clinic, or sponsor (m517 W7-N inheritance).
• ❌ It does not assert m620 verdict polarity; the verdict is m620-sovereign.
• ❌ It does not use the cure / reverse / recovered / miracle lexicon.

---

Falsifying prediction (per W6.5 + Roll EPP-3)

The cross-condition X-invariant framing for autism would be falsified by:

• m620 02_exec_005_AUTISM verdict that fails to satisfy the proposed K4.1 = (0, 0, 1) bit pattern · i.e., shows X-structure altered across autism subtypes; OR
• m620 02_exec_007_AUTISM_MABOSS verdict that disagrees with 02_exec_005_AUTISM by a meaningful margin · i.e., the cross-engine cross-check itself fails.

Both paths are open via `06_registry/outbound_to_m620/` verify_req ferry.

---

Attack surface entries

• AS-002 (sponsor / collaborator endorsed by smuggle) — audit gate G2.3
• AS-008 (cross-domain conflation across autism subtypes) — audit gate G4.2 + W6.3 veto
• AS-009 (moral attribution across non-responders) — m610 S3 veto

---

Ferry target (per W6.6)

• Primary: m590 — claim_pack carrying the autism cross-condition seed.
• Secondary: m620 — verify_req for m620 to return verdict on the K4.1 (0,0,1) interpretation.

Frontmatter `ferry_target: m590`. Audit reproducibility: `python3 -m 02_audit.audit --artifact 03_demos/i_demo_011_researcher_autism/README.md`.

---

Honest verdict (per W6.5)

• 已做: researcher × autism reading layer anchored on m620 two-engine verification surface · K4.1 (0,0,1) interpretation marked as proposal · 3 attack-surface refs · falsifying prediction with two re-audit paths.
• 待办: actual m620 verdict reception + claim_pack ferry to m590 (v0.4+ once verify_req returns).
• 漏洞: the proposed bit pattern is m625's interpretation of m620's exec dirs · m620 may disagree. m517 CARD08 Wu 2025 anchor is training_data_recall · TODO_unverified per m517 readiness band. Two autism exec dirs is a good feature but does not by itself prove cross-engine convergence — the cross-check verdict is open work.
• 反驳预案:

- *"Why anchor on m620 instead of an autism-research clinical paper?"* — m625 is a cross-domain ferry · clinical-paper anchoring is m590 sovereign · m625 references the cross-domain verification engine. - *"Two exec dirs are not validation."* — Correct; they are *cross-check candidates*. Convergence is the open verdict.