It reads the filings. It refuses to invent the rest.

Ingested, gate-passing, and deliberately not shown

What follows is the publication ledger verbatim — the reviewer’s own note from the day each deal was held, quoted rather than tidied. They are dated records and read as such: Alaska’s counts in months how stale its filing data was on the day it was held, and that count has grown every day since. Restating these in the present tense would turn someone’s dated review into a claim about today.

• Alaska Air Group, Inc.Held 2026-08-16

  Held, and the reason is staleness rather than the peer set — which is cohesive, 5 of 5 included peers sharing SIC 4512. Alaska's 10-K for the year ended 2025-12-31 was filed on 2026-02-12, but its XBRL facts have not reached the companyfacts API, so the latest year this pipeline can build is 2023-12-31 — over 31 months old, and G5 says so. Publishing a deal whose Key Financials stop two years short of its peers' would make every cross-deal comparison in the corpus wrong in a way no gate can see. Re-ingest when the facts land.

• AT&T Inc.Held 2026-08-13

  Held. Phase 6 audit evidence, deliberately outside the shown corpus — unchanged by this review. The surface independently rates its peer set 'mixed': only Verizon (SIC 4813) shares AT&T's code, while Comcast and Warner Bros. Discovery are SIC 4841 cable operators. Worth knowing if it is ever considered for publication.

• Bloomin' Brands, Inc.Held 2026-08-19

  Held. Four of six included rows are royalty businesses - RBI twice, Yum Brands, Papa Johns - against a company-operated casual dining target, so the majority rule holds it. Cheesecake Factory and Cracker Barrel are the two genuine comparables and they are outnumbered. The margin split is stark in this table: 5.1 to 8.8 percent EBITDA for the operators against 26.5 to 33.8 percent for the franchisors. Deduplicating RBI moves the median from 11.97x to 12.73x. Round 3 asked whether a peer set can pass every gate and still be wrong. This is half the answer and the hotels are the other half. Two separate faults, both invisible to G11 because every row is exact-SIC 5812 and cohesion therefore scores 4.00 of 4. FAULT ONE, mechanical and now fixed in code: Restaurant Brands International Inc. and Restaurant Brands International Limited Partnership each file a 10-K under SIC 5812 reporting identical revenue, because they are one business in an umbrella-partnership structure, and screenPeers had no dedup. Both landed in this shortlist. Worse than a double count - the two rows carry DIFFERENT multiples (5.66x against 13.52x EV/EBITDA on identical EBITDA), because the partnership's non-affiliate public float is a fraction of the corporation's and this pipeline strikes enterprise value on float. So the median was being computed over the same company at two valuations 2.4x apart, one of which is a float artefact rather than a market view. Phase 13 added dedupeDualEntities to screenPeers, which drops the partnership, keeps the corporation, and records the collision in the ingestion report. The fix is in the code; these deals were built before it, so their shortlists still carry the duplicate. FAULT TWO, structural and not fixed: SIC 5812 does not distinguish a company that owns kitchens and leases from one that collects royalties. Yum, RBI, Papa Johns and McDonald's are royalty businesses with no restaurant-level cost of sales; the targets here own their restaurants. EBITDA margins in these tables run 5 to 11 percent for operators and 26 to 48 percent for franchisors, and a median across both is not a multiple of anything. No gate can see this and the reviewer's majority rule is what decides it.

• Darden Restaurants IncHeld 2026-08-19

  Held. Four of six included rows are royalty businesses - RBI twice, Yum Brands, McDonald's - against a target that operates its own restaurants, so the majority of this peer set is a different business model and the majority rule holds it. Only Chipotle is a clean structural comparable. Deduplicating RBI moves the median EV/EBITDA from 14.01x to 14.51x, which understates the problem: the duplicate is the smaller of the two faults here. Round 3 asked whether a peer set can pass every gate and still be wrong. This is half the answer and the hotels are the other half. Two separate faults, both invisible to G11 because every row is exact-SIC 5812 and cohesion therefore scores 4.00 of 4. FAULT ONE, mechanical and now fixed in code: Restaurant Brands International Inc. and Restaurant Brands International Limited Partnership each file a 10-K under SIC 5812 reporting identical revenue, because they are one business in an umbrella-partnership structure, and screenPeers had no dedup. Both landed in this shortlist. Worse than a double count - the two rows carry DIFFERENT multiples (5.66x against 13.52x EV/EBITDA on identical EBITDA), because the partnership's non-affiliate public float is a fraction of the corporation's and this pipeline strikes enterprise value on float. So the median was being computed over the same company at two valuations 2.4x apart, one of which is a float artefact rather than a market view. Phase 13 added dedupeDualEntities to screenPeers, which drops the partnership, keeps the corporation, and records the collision in the ingestion report. The fix is in the code; these deals were built before it, so their shortlists still carry the duplicate. FAULT TWO, structural and not fixed: SIC 5812 does not distinguish a company that owns kitchens and leases from one that collects royalties. Yum, RBI, Papa Johns and McDonald's are royalty businesses with no restaurant-level cost of sales; the targets here own their restaurants. EBITDA margins in these tables run 5 to 11 percent for operators and 26 to 48 percent for franchisors, and a median across both is not a multiple of anything. No gate can see this and the reviewer's majority rule is what decides it.

• Heineken N.V.Held 2026-08-19

  Held, and deliberately not dressed as a refusal. Phase 12's EU pilot: an IFRS/ESEF issuer built end to end from the ESMA filings index, four years of financials, the whole path proving that one licence clause reaches 26 countries. It publishes no comparables and no valuation, which looks like Waste Management and is not the same thing at all. WM's screen ran, read what came back, and declined. Heineken's screen never ran and cannot: ESEF carries no industry classification, so there is no counterpart to a SIC code to screen on and no peer universe to screen against - and a cross-taxonomy set, an IFRS target against US-GAAP peers, would compare two differently-defined EBITDAs. That is a capability gap awaiting an industry-classification source, not an editorial position, and the third registry status added in Phase 13 exists precisely so the two cannot be confused. Publishing this as a refusal would be claiming credit for a decision nobody made. It stays held until either a classification source lands or the corpus has a second EU issuer to make the path worth showing on its own.

• Hilton Worldwide Holdings Inc.Held 2026-08-19

  Held, for the same reason as Marriott and with the sharper illustration. Four of five included peers are casino owner-operators; Marriott is the only genuine comparable. The excluded row is the finding: Choice Hotels International, a pure hotel franchisor, is shown and excluded at 0.13x revenue - two hundredths outside the screen's 0.15x floor - while Boyd Gaming is included at 0.34x. The screen dropped the best available structural peer on size and kept a casino, and no rule in the pipeline is capable of noticing that this was the wrong trade. This is the case round 3 was built to find, and it is worse than the question anticipated. SIC 7011 holds three unlike businesses under one code: asset-light franchisors (Marriott, Hilton, Choice, Wyndham), casino owner-operators (MGM, Las Vegas Sands, Wynn, Boyd, Caesars) and hotel REITs (Park, Xenia). The franchisors earn fees on other people's buildings; the casino operators own and run enormous physical estates. There is no reading of EV/EBITDA under which they belong in one median. Every row is exact-SIC 7011, so peer cohesion scores 4.00 of 4, G11 has no basis to refuse, and the review surface reports the set as cohesive. It is cohesive. It is also wrong, and the two statements are about different questions. THE SIZE BAND MAKES IT WORSE RATHER THAN BETTER, which is the finding worth keeping: Hilton's screen EXCLUDED Choice Hotels at 0.13x revenue, just under the 0.15x floor, and INCLUDED Boyd Gaming at 0.34x. Choice is a pure hotel franchisor and the single best structural peer for Hilton in the US market; Boyd is a regional casino operator. The industry code and the size band both point away from the right answer at the same time, independently, and every gate passes. A THIRD FACTOR COMPOUNDS IT, and it is a documented design choice rather than a defect: with no licensed price source, enterprise value is struck on non-affiliate public float from the 10-K cover page, which understates EV for any controlled company. SIC 7011 is full of them - Las Vegas Sands reads 2.18x EV/EBITDA on that basis. So the casino rows are not merely the wrong business, several are the wrong business at an artificially low multiple. WHAT THE CORPUS DOES ABOUT IT: holds both deals and records this, rather than inventing a business-model signal. Classifying franchisor against owner-operator cannot be done from any structured field either filer tags - it needs a judgement about what the business is, which is the thing a human reviewer is for and the thing a gate is not. The honest position is that the refusal mechanism has a stated blind spot: it can see that peers are the wrong INDUSTRY (Waste Management, Republic Services, both refused and both now published as refusals) and cannot see that they are the wrong BUSINESS inside the right industry. Recording that is the deliverable; pretending a gate covers it would be worse than the gap.

• Marriott International IncHeld 2026-08-19

  Held. Four of five included peers are casino owner-operators - MGM, Las Vegas Sands, Wynn, Boyd - against an asset-light hotel franchisor. Hilton is the only real comparable in the set and the median discards it: the included multiples are 2.18x, 5.06x, 5.70x, 10.84x and 21.24x, so the median is Boyd Gaming's 5.70x while Hilton's 21.24x sits at the top as an apparent outlier. A franchisor valued at a regional casino's multiple, with a full provenance ledger and eleven passing gates behind it. Park Hotels & Resorts, a hotel REIT, was excluded on size at 0.1x - so the set escaped a third business model by accident rather than by design. This is the case round 3 was built to find, and it is worse than the question anticipated. SIC 7011 holds three unlike businesses under one code: asset-light franchisors (Marriott, Hilton, Choice, Wyndham), casino owner-operators (MGM, Las Vegas Sands, Wynn, Boyd, Caesars) and hotel REITs (Park, Xenia). The franchisors earn fees on other people's buildings; the casino operators own and run enormous physical estates. There is no reading of EV/EBITDA under which they belong in one median. Every row is exact-SIC 7011, so peer cohesion scores 4.00 of 4, G11 has no basis to refuse, and the review surface reports the set as cohesive. It is cohesive. It is also wrong, and the two statements are about different questions. THE SIZE BAND MAKES IT WORSE RATHER THAN BETTER, which is the finding worth keeping: Hilton's screen EXCLUDED Choice Hotels at 0.13x revenue, just under the 0.15x floor, and INCLUDED Boyd Gaming at 0.34x. Choice is a pure hotel franchisor and the single best structural peer for Hilton in the US market; Boyd is a regional casino operator. The industry code and the size band both point away from the right answer at the same time, independently, and every gate passes. A THIRD FACTOR COMPOUNDS IT, and it is a documented design choice rather than a defect: with no licensed price source, enterprise value is struck on non-affiliate public float from the 10-K cover page, which understates EV for any controlled company. SIC 7011 is full of them - Las Vegas Sands reads 2.18x EV/EBITDA on that basis. So the casino rows are not merely the wrong business, several are the wrong business at an artificially low multiple. WHAT THE CORPUS DOES ABOUT IT: holds both deals and records this, rather than inventing a business-model signal. Classifying franchisor against owner-operator cannot be done from any structured field either filer tags - it needs a judgement about what the business is, which is the thing a human reviewer is for and the thing a gate is not. The honest position is that the refusal mechanism has a stated blind spot: it can see that peers are the wrong INDUSTRY (Waste Management, Republic Services, both refused and both now published as refusals) and cannot see that they are the wrong BUSINESS inside the right industry. Recording that is the deliverable; pretending a gate covers it would be worse than the gap.

• Texas Roadhouse, Inc.Held 2026-08-19

  Held, and it is the one that would have published on the majority rule alone - five of eight included rows (Brinker, Bloomin' Brands, Cracker Barrel, Cheesecake Factory, Dave & Buster's) operate their own restaurants, so the majority is right and only RBI and Yum are not. It is held on the duplicate instead, and the arithmetic is why: removing the RBI partnership row moves the median EV/EBITDA from 11.55x to 14.97x, a 30 percent move in the headline multiple of the whole screen, caused by one company appearing twice at two different float-derived valuations. That is the largest single distortion Phase 13 found anywhere in the corpus, and publishing a table carrying it would be publishing a figure known to be wrong. It becomes publishable on re-ingest, when the dedup fix applies to the screen - which is the concrete value of that fix, recorded here rather than asserted. Round 3 asked whether a peer set can pass every gate and still be wrong. This is half the answer and the hotels are the other half. Two separate faults, both invisible to G11 because every row is exact-SIC 5812 and cohesion therefore scores 4.00 of 4. FAULT ONE, mechanical and now fixed in code: Restaurant Brands International Inc. and Restaurant Brands International Limited Partnership each file a 10-K under SIC 5812 reporting identical revenue, because they are one business in an umbrella-partnership structure, and screenPeers had no dedup. Both landed in this shortlist. Worse than a double count - the two rows carry DIFFERENT multiples (5.66x against 13.52x EV/EBITDA on identical EBITDA), because the partnership's non-affiliate public float is a fraction of the corporation's and this pipeline strikes enterprise value on float. So the median was being computed over the same company at two valuations 2.4x apart, one of which is a float artefact rather than a market view. Phase 13 added dedupeDualEntities to screenPeers, which drops the partnership, keeps the corporation, and records the collision in the ingestion report. The fix is in the code; these deals were built before it, so their shortlists still carry the duplicate. FAULT TWO, structural and not fixed: SIC 5812 does not distinguish a company that owns kitchens and leases from one that collects royalties. Yum, RBI, Papa Johns and McDonald's are royalty businesses with no restaurant-level cost of sales; the targets here own their restaurants. EBITDA margins in these tables run 5 to 11 percent for operators and 26 to 48 percent for franchisors, and a median across both is not a multiple of anything. No gate can see this and the reviewer's majority rule is what decides it.

None of these has a link, because a held deal is not in the generated index and has no route to open. That is the mechanism working, not a gap in this page.

Published because they refused

The same ledger records the opposite decision. These 2 ship with no comparables table and no peer-derived valuation, and lead with the reason instead. They do open, because a refusal is a published deal.

• Republic Services, Inc.Published 2026-08-19

  Published as a refusal, alongside Waste Management and for the same reason. This one is the more valuable of the two, because it is the second: SIC 4953 widened to major group 49 independently, on a different company, and came back with Dominion, Targa, ConEd, Vistra, Kinder Morgan, FirstEnergy and Edison International - a different peer list, the same mean shared SIC digits of 2.00 of 4, the same refusal. Two unrelated companies failing the same way is what makes this a mechanism rather than a bad deal, and it is why the pair is published rather than just the first. Also carries an absent revenue split, on the separate G-check that a segment breakdown must reconcile to consolidated revenue: srt:ProductOrServiceAxis sums to 6,182m against 16,591m consolidated, 62.7 percent apart, so the chart is absent with its arithmetic stated. Three declared absences on one deal, each with a different cause, each written down.

• Waste Management IncPublished 2026-08-19

  Published as a refusal, reversing the Phase 9 hold. Nothing about the deal changed; the reason for holding it did. SIC 4953 (Refuse Systems) is thin at four and three digits, so screenPeers widened to major group 49 - 'Electric, Gas & Sanitary Services' - and got back NextEra, Exelon, Constellation, Southern, AEP and NRG, all SIC 4911/4931. Mean shared SIC digits 2.00 of 4. The screen declined to return a set at all and wrote the reason into absent.comparables; the deal therefore ships no comps table and no peer-derived valuation. G11 audits that decision and passes, because a refused screen leaves it nothing to check - the refusal is the screen's, not the gate's. Phase 9 held it on the grounds that a deal whose Comparables and Valuation screens are both a stated absence is a thin company document. That was a defensible judgement about a company profile and the wrong one about the product: the argument this thing makes is about the comps screen, and this is the corpus's clearest instance of it. Held-but-correct made the differentiator unviewable. The third registry status (published-as-refusal, Phase 13) exists so that leading with an absence is a decision on the record rather than something a surface infers from an empty array.