Phase 0 Findings — GBMT-9 (Media)

Date: 2026-08-03 Scope: First pass per §4's phase gate — verify the ★ Anchor Table rows (plus two bonus non-starred rows closely coupled to them), build the §2 baseline as a committed pipeline, scan the three live precedents named in the protocol (C-18/Meta ban, CPB's rescission aftermath, the NY tax credit). Go/no-go input for full execution.

Verdict: GO — no kill condition fires as a stop, but two fire as headlines and the constitutional screen resolves to "not yet," not "no"

Kill condition (1) — the news-desert count — does not fire as a strict stop the way housing's shortage number did: the figure is arithmetically stable and reproducible year over year. But it fires as a softer version of the same headline: the count rests on a single research lineage with two undocumented, opposite-direction biases that nobody has sized. Kill condition (2) — the civic-harms literature audit (§4) — was not attempted in Phase 0's time budget and is queued, not resolved; this is itself a finding, logged as a deviation. Kill condition (3) — whether the constitutional screen forecloses whole instrument families — resolves to a real answer, but not the one the protocol anticipated: the screen is directionally confirmed but not yet operative, because the two marquee statutes it depends on are still in active litigation.

One prior broke outright (the ad-duopoly claim), one was corrected upward in severity (CPB didn't merely lose funding, it dissolved), and one load-bearing claim used by bargaining-code advocates got real evidentiary teeth against it (distribution to incumbents, not local outlets) ahead of full §6/§9 execution.

1. The news-desert count is single-root with two undocumented, offsetting biases (KILL CONDITION → SOFT HEADLINE)

The protocol's kill-condition test asked whether the desert count "moves by >2× under defensible alternative inclusion criteria." It does not move by 2× — the count is stable across the 2024 and 2025 Medill reports (206→212-213 zero-outlet counties; the "over half of counties have one or no local outlet" arithmetic holds both years, ~55-56%). That much is confirmed, not refuted.

But the housing-shortage lesson applies in a different shape here, not a weaker one:

• Every circulating figure — Medill's own, and every journalistic restatement checked (Poynter, Nieman Lab, Axios, Rebuild Local News) — traces to one continuous lineage: Penny Abernathy's UNC Center for Innovation and Sustainability in Local Media (through ~2020) → Northwestern's Medill Local News Initiative (2022–present, after Abernathy moved institutions). No independent national outlet census exists. The one genuinely separate-lineage instrument found — Muck Rack/Rebuild Local News's "Local Journalist Index" (journalist-density, not outlet-presence) — measures a different quantity and is directionally consistent, but does not corroborate Medill's specific county counts.
• Medill's own methodology has two documented biases pulling in opposite directions, and no publication — Medill's or any critic's — sizes either one:
  1. Commercial TV and commercial radio are entirely excluded from the database by design. Most US counties sit inside a broadcast market with at least one local TV affiliate; this almost certainly makes some "zero-outlet" counties look worse than they are on a coverage-availability basis (though not necessarily on a civic-accountability-reporting basis, which is the harder question TV news often doesn't answer either). [Verification Phase 2, 2026-08-10: "most US counties" is an understatement that matters — per the FCC's own DMA report to Congress (DA 16-613), Nielsen assigns every US county outside parts of Alaska to one of 210 markets. That makes DMA membership useless as a sizing test, since it holds for well-served counties too; the §3 pass 2 check built on it is corrected accordingly. The parenthesis above is the right caveat and it did not survive onto the public page. See ws03-04-findings and steelman-log, target A.]
  2. "Ghost papers" — a still-publishing masthead with few or no remaining reporters — count as "outlet present," which the ghost-paper critique (present in the literature since Abernathy's own UNC-era work) argues inflates the not-a-desert column in the opposite direction.
  3. Medill's own methodology changed in 2024 (added a new "network digital-only sites" category, 742 outlets; expanded ethnic-media verification) without publishing a reconciliation of how much of the year-over-year county-count change is definitional versus real market change.

Effect on the protocol: §3 cannot simply cite the desert count as a stable input to §10/§11's scoring. It must (a) explicitly flag the count as single-root, (b) attempt to size at least one of the two offsetting biases using a sub-national case study (a state where TV-market and print-desert maps can be overlaid), and (c) treat "outlet exists" and "accountability coverage capacity exists" as the two distinct quantities the protocol's own §3 language already anticipated ("measure coverage capacity, not outlet counts, wherever the data permits") — Phase 0 confirms this distinction is not just prudent but load-bearing.

2. Prior broken — the ad-duopoly claim is now false, not just aging

Anchor 3's second half hedged that "Google+Meta together take a majority of US digital ad revenue... may be stale as Amazon rises." It is stale, and the direction is sharp enough to call it broken rather than merely qualified:

• Google + Meta's combined share of US digital ad spend fell below 50% in 2024 for the first time since roughly 2014, driven by Amazon's rise. Independent eMarketer-derived figures (cross-checked directly, not just via the verifying subagent) put the combined share at ~47.1% in 2025, forecast to fall to ~44.8% in 2026, with Amazon's share reaching ~17.3% by 2026 — at which point the Google+Meta+Amazon "triopoly" together holds ~62%. [Verification Phase 1, 2026-08-10: every figure in this paragraph traces to a single estimator (eMarketer), whose successive vintages date the below-50% crossing to 2022 in one report and 2024 in another. IAB/PwC confirms only the $258.6B market total. Single-root, by this filing's own standard.]
• The total US digital ad market itself is large and still growing ($258.6B in 2024, IAB/PwC), so this is genuinely about concentration declining, not the market shrinking.

Effect: §6's framing of platforms as an ad-market duopoly needs updating to a widening triopoly before any antitrust-flavored architecture (§11) is scored against current market structure. This does not change the bargaining-power analysis (Google and Meta remain the two counterparties in every bargaining-code precedent examined), but it changes any argument that rests on "two firms control the ad market."

3. Prior sharpened, not broken — CPB did not merely lose funding, it dissolved

The protocol anchor stated CPB's appropriation "was rescinded... and CPB is winding down." Phase 0 confirms the funding facts precisely and finds the institutional outcome is more severe than "winding down" implies:

• Rescissions Act of 2025 (Public Law 119-28), signed July 24, 2025: rescinds CPB's already-enacted FY2026 and FY2027 advance appropriations, 535M/yeareach, 1.1B total — confirmed directly against Congress.gov/GovInfo records, independent of the verifying subagent.
• CPB's board voted to dissolve the organization at its December 10, 2025 meeting (publicly announced January 5, 2026) — confirmed directly. This is not a funding pause or a downsizing; CPB filed Articles of Dissolution after 58 years of operation, ending a body created by the Public Broadcasting Act of 1967.
• Functions did not fully transfer to a single successor: music-royalty negotiation (ASCAP/BMI/SESAC) was locked in through 2027 before closure and handed to PBS/NPR directly; the Ready To Learn children's-education grant program was cancelled outright with no identified successor; interconnection funding was distributed as a final lump sum rather than continued as an ongoing program.
• Station-level damage concentrating in small/rural markets is empirically documented from the incidence side, not merely predicted: rural public-media grantees averaged 17% of revenue from CPB versus 9% for non-rural stations (CRS/CPB-sourced data); 120 rural grantees relied on CPB for ≥25% of revenue, and 33 (many tribal) for ≥50%. Concrete station-level outcomes by mid-2026 include closures (NJ PBS, KWSU-TV), a rescue-via-acquisition (WPSU/Penn State absorbed by WHYY), and tribal stations losing the majority of their budgets (KLND −50%, KSHI −96%). No single comprehensive post-hoc academic study of the closure distribution was found — the incidence claim is well-evidenced on the pre-rescission dependency baseline, thinner on a systematic post-rescission outcome count.
• The Neff & Pickard comparative-spending figures the protocol cites are real but are two different point estimates from the same research team, not a range: $1.40/capita (federal-funding-only) versus $3.16/capita (broader total-public-funding calculation, the figure used in their actual peer-reviewed paper, ranking the US 25th of 33 countries studied). Both predate the 2025 rescission, so the true post-rescission US figure is now lower than either.

Effect: §8 and §9 (precedents) should treat CPB's dissolution as a completed institutional-failure case, not a live one — the durability question this domain most needs answered ("can an insulated, formula-funded institution survive a determined defunding push") now has a documented negative data point after half a century of survival, which is itself the single most important precedent Phase 0 surfaced for §11's "automatic, formula-based, insulated instrument" design logic. An insulating design was tried, for decades, and still lost to a simple-majority rescission vote — that is a harder finding than "public media is underfunded relative to peers," and it should be foregrounded over the comparative-spending figures in the eventual whitepaper.

4. The constitutional screen is directionally right but not yet operative — both marquee statutes remain unresolved

The protocol's §7 workstream exists to determine which instrument families the First Amendment forecloses before economics or politics are scored. Phase 0's scan of Moody v. NetChoice refines this in a way that matters for how confidently §11 can lean on the screen:

• The Court's holding that platform feed curation is protected editorial speech (drawing on Miami Herald v. Tornillo) commanded a real 6-justice majority on the legal principle (Kagan's opinion, joined in full by Roberts/Sotomayor/Kavanaugh/Barrett, and in relevant part by Jackson) — stronger backing than "guidance dicta," which is a more confident basis for the screen than the protocol's cautious anchor phrasing implied. [Verification Phase 1, 2026-08-10: corrected. The platform-curation reasoning is in Part III–B, which Jackson did not join — she joined Parts I, II and III–A only, and wrote that she "would not go on to treat either like an as-applied challenge and preview our potential ruling on the merits." Five justices, not six. Alito, Thomas and Gorsuch wrote that "everything else in the opinion of the Court is nonbinding dicta." Verified against the slip opinion, supremecourt.gov 22-277.]
• But neither Texas's HB 20 nor Florida's SB 7072 has been finally resolved on remand as of mid-2026. Texas litigation was still in scoping/amended-complaint stage as of the last confirmed filings (Aug 2025); Florida's case has a summary-judgment hearing set for June 2026 with no outcome yet located. Both laws' core anti-deplatforming provisions remain enjoined or contested, not settled either way.
• Downstream, the screen's practical bite is visible in one confirmed sibling ruling: the Ninth Circuit's September 2024 decision in X Corp. v. Bonta held California's algorithmic-transparency mandate (AB 587, requiring platforms to report their content-moderation definitions to the state) is likely unconstitutional as content-based compelled speech — a live, concrete instance of the Tornillo/NetChoice logic actually striking something down, separate from the still-pending TX/FL cases themselves.
• On the money side: Maryland's digital-advertising tax is the closest real-world analogue to a "platform tax funding journalism" architecture, and its litigation history (Fourth Circuit struck the pass-through-ban provision on First Amendment grounds in Aug 2025; the core gross-receipts tax itself survives) suggests structural, viewpoint-neutral platform taxes are more durable than content-based mandates, consistent with the screen's predicted shape — but Maryland's revenue isn't earmarked for journalism, so it's an imperfect test case, not a direct precedent.

Effect: §11's constitutional screen should be applied as "this instrument family is trending toward foreclosed / trending toward safe based on the doctrinal direction of travel," not "this instrument family is foreclosed / safe," since the two statutes that would make either determination final are both still live. This is a real, useful finding — it just resolves to a probabilistic screen rather than a binary one at this stage of the litigation.

5. Bonus finding (beyond the ★ rows) — bargaining-code money goes disproportionately to incumbents, and the effect on small outlets is asymmetric

Rows 6 and 7 weren't both starred, but scanning them together surfaced the sharpest early evidence for the protocol's H11.1 rhyme (demand-side money without distribution reform inflates incumbents rather than local coverage):

• Australia: the widely cited AU$200M+/yr bargaining-code figure traces almost entirely to one person's estimate (Rod Sims, former ACCC chair, in a Judith Neilson Institute report) — not an independently audited government total, since individual deal values are confidential. The one peer-reviewed distributional estimate found (Brevini & Ward, Media International Australia 2023) puts ~90% of the money flowing to three large incumbents (News Corp, Nine, Seven West), with ~90-95% of commercial radio and other outlets excluded from deals entirely. [Verification Phase 1, 2026-08-10: no such paper exists — Brevini's 2023 bargaining-code paper is sole-authored, in Javnost — The Public 30(2), and is paywalled and unread. Verification Phase 2, 2026-08-10: the Australian Treasury's own statutory review of the Code was never consulted by this filing, and it (a) states the review "has not been provided with the details of these agreements by either digital platforms or the relevant news businesses," so no distributional split is auditable from any source; (b) concludes "it is reasonable to conclude that the Code has been a success to date," recording over 30 agreements "with a broad range of news businesses, both large and small, and in metropolitan and regional areas," Country Press Australia's collective deal for small regional publishers, and Minderoo's for 24 more; and (c) records the ABC appointing "57 regional positions, including reporters in 19 locations, 10 of which did not previously have them." See steelman-log, target B.]
• Canada: Meta's response to C-18 (a full news ban, still in effect as of mid-2026, per the Media Ecosystem Observatory's tracking) hit local outlets far harder than national ones — engagement down 85% for local vs. 64% for national at the one-year mark [Verification Phase 1, 2026-08-10: withdrawn. The cited MEO report contains no local-versus-national engagement split and the string "64%" does not appear in it; its 85% is an all-outlets Facebook+Instagram figure. The asymmetry finding survives, re-based on the report's own local-specific numbers: 30% of the 713 local outlets previously active on social went dormant, and 212 of the 217 outlets that went dark (98%) were local. See media/research/verification-log.md, finding 1.] — while Meta's own Canadian usage stayed essentially flat (corroborated by both a Reuters/analytics-vendor report and MEO's own survey, satisfying the two-different-types rule). Roughly 30% of previously active local outlets went dormant on the platform; local outlets accounted for 96% of all such dormancies.
• The asymmetry that gives platforms leverage in these fights is confirmed directionally by the Canada case: Meta could walk away with negligible usage cost to itself, while the outlets that depended on referral/engagement paid the price — this is the leverage-direction finding the protocol's §6 seed hypothesis needed and did not yet have evidence for.

Effect: This is exactly the evidence base §6/§9's full execution should build from — it does not need to be rediscovered, only extended (deal-level data, more country cases, more outlet-size breakdowns).

6. Anchor verifications

Full agent reports (QA'd, with three headline claims independently re-verified by direct search rather than taken on the subagent's word) inform the rows below. Rows completed in §3 of the protocol.

7. Pipeline: proven, no connector required

The BLS QCEW open-data CSV-slice API works end to end for the national newspaper-publisher industry series and is committed in media/baseline/:

Notes for reuse: the NAICS code changes from 511110 to 513110 in the 2022 vintage (the NAICS 2022 recode); pull_qcew_newspapers.py detects the splice year from the data and checks the year-over-year discontinuity at the splice (observed: +1.6%, well under the 15% threshold that would flag the two codes as non-comparable) before accepting the spliced series. Two real gaps were caught rather than silently worked around, both in the incompatibility log: the modern Census CBP national-file URL pattern 404s for pre-2014 vintages, and the QCEW open API itself has no pre-2014 window for this slice despite QCEW's underlying data reaching back to 1990. Neither was resolved in Phase 0's time budget; the mid-2000s peak figures used above therefore rest on Pew/NAA secondary literature, not a direct admin-data pull, and that is stated rather than presented as pipeline-verified.

8. Re-ranking for full execution

§4's prior was that §2, §3, §6, §7, and §10 carry the most information per hour. Phase 0 partly confirms and partly reorders:

1. §7 (the constitutional screen) moves up, but its output changes shape. Rather than a one-time binary screen, it needs to track two live cases (TX HB 20, FL SB 7072) to their eventual resolution and update the instrument-family foreclosure map accordingly — the screen is a moving target through at least mid-to-late 2026, not a settled input §11 can treat as fixed.
2. §8 (public/nonprofit media) rises sharply. CPB's dissolution is a completed institutional-failure case, not a live risk — this is likely the single most important precedent for §11's design logic (durability of insulated funding structures) and deserves more depth than a single workstream paragraph.
3. §3 stays central but the deliverable changes. Not "here is the desert count" but "here is the desert count, its single-root status, and a sized estimate of at least one of its two known biases" — the sizing exercise (a state-level TV-market overlay, or a ghost-paper audit in a sampled set of "served" counties) is the highest-value original work available to this inquiry, directly analogous to GBMT-2's retained-local-discretion measurement.
4. §6/§9 (platform economics and precedents) rise. The Canada/Australia evidence base is stronger and more specific than the protocol anticipated — local-outlet-versus-national asymmetry, incumbent-capture of bargaining-code money — and full execution should extend rather than rediscover it.
5. §4 (civic-harms literature audit) falls in urgency but not in importance. It was not attempted in Phase 0 and remains completely open; per M3's monoculture check, this is also where the steelman-for-subsidy case (the Nordic direct-press-subsidy precedent, §10) should get equal-effort treatment, since Phase 0's evidence so far leans toward subsidy/instrument skepticism (ghost papers, incumbent capture, dissolved insulation) and the protocol's own tilt-check calls for a real counterweight.

One time-sensitivity to carry: both the NetChoice remand cases and California's Google-deal budget line are moving targets with decisions/appropriations expected within the execution window (FL summary judgment ~June 2026; CA's 2026-27 budget cycle). Full execution should re-check both rather than freezing Phase 0's snapshot.