GUBMENTPlain talk · policy frontier
Filings / Mental health / Sources / Red-Team Pass (§9 scorecard) — GBMT-11
GBMT-11 · Research record · No. 11

Red-Team Pass (§9 scorecard) — GBMT-11 Mental Health

mental-health/research/ws09-red-team-log.md
This is a working research document from the mental health filing, published as written — including the parts later corrected. It is the underlying record for Whitepaper No. 11, not a summary of it.

Date: 2026-08-11. Pass 1 only — attacks the pass-1 scorecard (ws09-scorecard.md). Does not run the structurally blinded re-score (M6 next step). Precedent format: drugs/research/red-team.md — Attack / Response / Amendment; net-effect close.

Brief: attack the scorecard synthesis as hard as the evidence allows. Where an attack lands with a cite from ws01ws08 / Phase 0, amend the cell and/or honesty box — do not defend. Where it fails, say so.


Attack 1: CCBHC’s monopoly of first place rests on O1=5 / O2=4 that the record does not earn

#2 leads all three weightings only because it is the sole 5 on the board (O1) and carries O2=4. The O1=5 basis cites Mathematica/ASPE DY1→DY2 adult time-to-eval 9.0→5.4 days, client volume, and 94% open-access (ws05). Attack:

  1. Scale mismatch. Anchored O1 band 4 literally names “wait-to-evaluation”; band 5 requires “Direct multi-setting causal or strong quasi-experimental evidence.” The wait metrics are clinic-reported demonstration before/after, not a secret-shopper or claims causal design. The same evaluation family’s claims DID for ED/hospital is heterogeneous — not a uniform access win (ws05).
  2. DY4 softening omitted from the cell. DY1→DY4 RTC: adult mean days 9.1→8.4; within-10-days share ~69–73% stable — far softer than the DY1→DY2 headline the cell quotes (ws05).
  3. O2=4 is service-scope theater. Basis: “crisis is a required CCBHC service; ED/hospital DID heterogeneous.” Required scope ≠ measured improvement in boarding, acute throughput, or crisis-system outcomes. Heterogeneous DID is mixed → band 3, not 4 (ws05).

Response: lands. O1=5 and O2=4 over-read the evaluation package relative to the anchored scales and the DY4 / DID disclosures already in §5.

Amendment (cells):

Rank consequence: after recompute, #2 no longer leads all three weightings — #5 crisis continuum leads W1 (3.65 > 3.55); #2 still leads W2 and W3. See amended scorecard.


Attack 2: KC1 proxy scoring treats incommensurable proxies as one O1 ladder

Under KC1, every O1 cell is a proxy — but the board stacks different proxy classes as if they ranked one object: #2 clinic time-to-eval, #5 Lifeline answered volume, #6 NQTL network corrections, #10 Bishop/Brahmbhatt secret-shopper / directory-realized appointments (ws02, ws04, ws05, ws06). Phase 0 / §2 say the flagship realized-access methods are secret-shopper and claims — not clinic demo waits or contact-center KPIs (phase0, ws02).

Response: lands on framing; cells already corrected under Attack 1 for the worst inflation (#2 O1). Comparing remaining O1=4s across rows still equates unlike instruments.

Amendment (honesty box): O1 ranks are proxy-class ranks, not a single realized-access ladder. Whitepaper must not say “CCBHC beats parity on access” without naming the proxy class each cell used. No further cell moves beyond Attack 1 / Attack 5.


Attack 3: H7’s demotion of beds is circular — Supported H7 → demote #4 → community rows “win” → H7 looks right

#4 never leads; asylum-scale is demoted; #2/#5 lead. Is that just H7 written into the ranks?

Response: does not land as circularity. H7’s formal criteria were adjudicated in §7 from evaluations (CCBHC access package; mobile/BHCC continuum) and OECD bed dispersion before ranks were computed (ws07). #4 already carries O2=4 on contemporary Construct A/B + Australia counter-steelman — the demotion is rank position under O1/O4-heavy weights, not a low acute score. McBain waiver-state bed null and KC2 still block “rebuild 1955” arithmetic (ws03).

Amendment: none to cells. Honesty box: restate that #4’s mid/low place is a weighting + O1-evidence consequence, not an O2 failure; acute/forensic pressure remains on the board.


Attack 4: All-neutral rows (#8, #9) float mid-pack and launder ignorance as “markets / integration are fine”

#9 (workforce liberalization) is all-3s and places above do-nothing. H2 warns that headcount-only instruments miss the Medicaid wedge (ws02). Should O1/O3 be 2s?

Response: does not land for cell demotion. Deviation #18 / symmetric evidence floor: below-neutral requires cited failure of this instrument, not scoring guidance about a related class. No project evaluation of compacts, supervision ratios, or peer-billing take-up exists (ws02, ws05). Inventing 2s would repeat the childcare / pre-amendment elder-care failure mode. #8 CoCM likewise correctly holds O1–O4 at 3 (no CoCM outcome package landed).

Amendment (honesty box only): #8/#9 ranks are evidence-density artifacts, not findings that liberalization or CoCM “work.” Whitepaper must not cite mid-pack placement as mild endorsement.


Attack 5: Commercial parity swing under KC3 — #6’s O1=4 / O4=4 and W2 “load-bearing” claim overreach the band-only rule

KC3 fires: commercial O4 for the ERISA self-funded majority is band-only / unknown (ws06). The O4 scale itself maps “KC3 band-only/unknown” to 3. Yet #6 carries O4=4 and O1=4, and the scorecard calls #6 the commercial parent’s swing architecture (2nd under W2).

Attack: DOL/HHS CAA corrections (>7.6M participants) are examined-plan spot-checks, not a census of self-funded realized access; 2024-rule “teeth” are under May 2025 nonenforcement (ws06). NQTL network/exclusion corrections are not measured appointment offers — a weak O1 proxy under the filing’s own §2 methods (ws02).

Response: lands on O1; partially lands on O4 framing.

Amendment:

Rank consequence: #6 drops from 2nd→ tied mid under W2; from 3rd→ mid/low under W1; remains near-bottom under W3.


Attack 6: Circularity of same-session scales + scores

Anchored scales, weight vectors (deviation #17), and every cell were authored in one pass by one scorer. Rank stability across W1–W3 measures internal consistency, not external truth — same failure mode as drugs Attack 1 / childcare Attack 5.

Response: cannot be fully rebutted from inside. Mitigations (scales before cells; cite-or-neutral; this red team) reduce but do not eliminate the risk.

Amendment: honesty box — numeric ranks remain provisional until the structurally blinded re-score. State results as “evidence as synthesized on this board ranks…,” not “the answer is.” No cell moves from this attack alone.


Attack 7: Measurement infrastructure should be its own architecture row

KC1’s protocol consequence: if the national by-payer series is missing, “the whitepaper becomes a measurement-infrastructure filing with supply architectures scored on proxies” (research-inquiry.md §9 / KC1). §7 already steels VA published access standards as the measurement precedent the civilian system lacks (ws07). Architectures 1–10 are all delivery/financing instruments — none scores “publish a national by-payer wait/acceptance series / CMS secret-shopper instrument.”

Response: lands. The board is incomplete relative to the filing’s own headline shape. Adding a full #11 mid-red-team without a workstream evaluation package would invent cells; the fix is disclosure + an owed scored row (or explicit exclusion) before whitepaper lead.

Amendment (honesty box + deviation #19): name measurement infrastructure as a missing scored architecture; whitepaper / next pass must either score it as a row or record an explicit exclusion. Do not silently treat KC1 as only a caveat on O1 cells.


Net effect on §9

Attack Lands? Cell moves Framing / honesty-box
1 CCBHC O1/O2 inflation Yes #2 O1 5→4; #2 O2 4→3 CCBHC loses all-three lead
2 KC1 proxy commensurability Yes (framing) (via Attack 1) Proxy-class disclosure
3 H7 bed demotion circular No Restate: weighting, not O2 fail
4 All-neutral mid-pack No (cells) Ignorance ≠ endorsement
5 Commercial/#6 under KC3 Yes (O1); partial (O4) #6 O1 4→3; O4 kept w/ tighter basis W2 swing narrowed
6 Same-session circularity Standing Blind re-score still owed
7 Measurement as architecture Yes Missing row named (dev #19)

Survives: #10 do-nothing still last under every weighting. H7 demotion of asylum-scale #4 still holds (never leads). #2 CCBHC remains a top-tier community-capacity architecture (leads W2/W3; 2nd under W1). #5 crisis continuum is the W1 leader after O2 honesty. Symmetric floor for #8/#9 unchanged.

Does not survive:#2 CCBHC leads all three weightings” as the pass-1 headline. Post-red-team: #5 leads W1; #2 leads W2 and W3.

Blind re-score still owed before ranks are treated as anything more than pass-1-post-red-team structure.

← All Mental health research documents Sources digest Read the whitepaper