m625 is not a clinical decision tool · not a publisher · not an expert authority · not a simulator. It is a cross-domain Roll-formalized evidence atlas with public audit + attack surface.
persona: clinician · domain: cancer · tier: 1
ferry target:m590 · fixture id: DEMO-001
self-audit grade:A
i_demo_001 · clinician × cancer · EGFR-NSCLC TKI dynamics
What a clinician comes to evidence.x1000.ai to see
A oncologist or pulmonologist working on EGFR-mutant non-small-cell lung adenocarcinoma reads about TKI sequencing in the literature and on disease-focused sites. Pain points (cross-domain · see 05_docs/M625_PERSONAS.md):
evidence synthesis is hard — one-line/two-line/three-line TKI literature is fragmented;
cross-period trajectory is hard to track in a single mental model;
when talking with a patient, it is hard to explain why "second-line dasatinib works after first-line imatinib failure" at the molecular-evidence level without sliding into individual prediction.
m625 does not solve any of these for the clinician. **m625 shows the clinician what a Roll-formalized cross-period evidence pattern looks like**, and invites the clinician to challenge it.
This is not a clinical recommendation. This is not an individual-patient prediction. (See "What this demo will not say" below.)
The Roll-formalized object (anchored on m610 i_case_cancer_001)
m625 does not modify m610's case file. m625 references it by spec section (/data/projects/m610_roll_cancer_dynamics/04_cases_cooked/i_case_cancer_001_v0.2.md). What we add here is a clinician-facing reading layer on top of the cooked object.
How to read this evidence (clinician orientation, ≤ 5 minutes)
The viable set V (per Kernel §3.2) is a population-level construct: the set of evidence-supported trajectory states across the EGFR-NSCLC cohort that m610 has cooked. It is not a per-patient claim.
Φ (the dynamics) describes how V evolves under intervention pressure. In this TMU-A primary case, Φ is cytostatic — the evidence pattern at the population level is consistent with continued V-membership under TKI pressure.
K4.1 (0, 1, 1) declares: the X-structure is not visibly altered (b_X = 0); Φ is visibly altered (b_Φ = 1); the Δ-B intervention boundary is altered (b_Δ_B = 1). This is the cross-period declaration required by F5 (per Framework §4.5) for any C10 closure.
The audit chain is draft → audit-pending → audit-pass → cooked → outbound. m610 has completed the chain; m625 references the cooked state.
For deeper reading, the m610 audit pipeline is open: `/data/projects/m610_roll_cancer_dynamics/scripts/cancer_dynamics_audit.py` (882-line, 218 tests, machine 97/100 on this case · training_data_recall · we cite from m610 v0.5.1 self-published readiness summary; clinicians should re-audit independently).
What this demo will not say (anti-drift · per 01_materials_cache/C2_m625_anti_drift_manifesto.md)
❌ It does not say "for this patient, the trajectory will be X." (AS-003 red-team smuggle of NCNU P_3.)
❌ It does not issue any treatment-selection directive (red-team smuggle pattern AS-004 — NCNU P_1 in the m625 audit gate G2.4).
❌ It does not say the word that begins with "cu" and means complete eradication. (AS-001 red-team smuggle.)
❌ It does not affirm, recommend, or vouch for any sponsor, lab, or institution (red-team smuggle pattern AS-002 — m517 W7-N expert/collaboration pattern, audit gate G2.3).
If a clinician reads any of the above into this demo, that is **a failure of this demo** and should be filed under `_ops/BLOCKED_LOG.md` for next-round revision.
This evidence pattern (TMU-A primary cytostatic, K4.1 (0,1,1)) is observed in EGFR-NSCLC. The cross-domain transferability to other cancer subtypes (e.g., BRCA1/2 breast — m610 i_case_cancer_150; DLBCL CD19 CAR-T — m610 i_case_cancer_450) is not assumed. transferability_pct: between EGFR-NSCLC (K4.1 (0,1,1)) and BRCA1/2 breast (K4.1 (1,1,0)) is qualitatively distinct (different K4.1 bit pattern); a quantitative transferability score is open work for m620 verification. We do not assert transferability across to neuro / CKD / rare without first re-cooking under those domains.
Falsifying prediction (per W6.5 + Roll EPP-3)
The current evidence pattern, as Roll-formalized in m610 i_case_cancer_001, would be falsified by either of:
an independent N-doubled re-cooked case under the same EGFR-NSCLC subdomain, performed under m610's MATH_AUDIT_v0.1 pipeline, that fails to satisfy C10 closure with K4.1 = (0,1,1); or
a re-audit by m620 verification (e.g. 02_exec_001_EGFR ladder) that returns a verdict score < 70/100 on the same artifact.
These two paths are open. Any reader is invited to attempt either.
AS-001 (cure smuggle) — should not occur in this demo · audit gate G2.2
AS-003 (individual-prediction smuggle) — should not occur · audit gate G2.4 (NCNU P_3) + m610 S2
AS-004 (clinical-decision smuggle) — should not occur · audit gate G2.4 (NCNU P_1)
The audit run on this demo is recorded in `_ops/SELF_AUDIT_LATEST.json` after each `python3 -m 02_audit.audit --self-audit` invocation.
Ferry target (per W6.6)
Primary: m590 research_x1000_ai — claim_pack sync_group_id m625-to-m590-YYYYMMDD-claim_pack-N, anchoring the Roll-formalized K4.1 declaration as a graph node.
Secondary candidate: m622 industry_conference_material_library — conference-grade adaptation of this reading layer.
Frontmatter `ferry_target: m590`. Actual payload generation is deferred to v0.2 (per `05_docs/M625_FERRY_TARGETS.md`).
Honest verdict (per W6.5)
已做:
- Clinician-facing reading layer cooked over m610 i_case_cancer_001 v0.2 cooked artifact.
- K4.1 (0,1,1) declaration referenced by spec section.
- 3 attack-surface entries (AS-001, AS-003, AS-004) linked.
- Transferability declaration for cross-cancer-subtype claims.
- Falsifying prediction stated with two re-audit paths.
待办:
- Quantitative transferability score across cancer subtypes (open work for m620 verification).
- VGAIT-style cross-modality V invariance demonstration (m517 CARD06 pattern adapted to cancer — v0.2).
- Live clinician walkthrough (deferred to v0.2 owner-led session per m517 "请教 NOT 展示" doctrine).
漏洞:
- m610 audit scores (manual 92, machine 97) are cited from m610 v0.5.1 self-published readiness summary; classified as `training_data_recall` until clinicians re-audit independently.
- The K4.1 (0,1,1) declaration depends on m610's X-structure invariance claim under TKI, which itself rests on m610's `MATH_FOUNDATIONS_v0.1` — a single-source dependency.
- No formal NCT trial citation is included in this demo because m625 does not host clinical trial data (m610 / m612 do); a clinician seeking trial-level evidence is redirected to m610's outbound `m610-claim-001.yaml`.
反驳预案:
- *"This reads like a clinical recommendation."* — It is not. Re-read the "What this demo will not say" section. If the demo still reads that way, file under `_ops/BLOCKED_LOG.md` for next-round revision.
- *"Why no PubMed numbers?"* — m625 deliberately does not republish clinical efficacy numbers. The viable-set V framing is a population-level Roll formalization, not an effect-size claim. See m610 outbound for source-tier-1 data.
- *"Why the K4.1 (0,1,1) and not (1,1,1)?"* — Because EGFR-NSCLC TKI evidence pattern is cytostatic at the population level: X-structure (the molecular substrate) is not visibly altered. (1,1,1) is the DLBCL CD19 CAR-T pattern in i_case_cancer_450 (phi_direct_rewrite).
- *"Where can I find the underlying cooked case?"* — `/data/projects/m610_roll_cancer_dynamics/04_cases_cooked/i_case_cancer_001_v0.2.md` (read-only from m625's perspective).