Claim Court — every commerce claim, checked against its evidence

Every commerce claim, checked against its evidence. Claim Court is the commerce surface of why-agent. The claims below are the ones a customer actually reads — a salon's hours and phone number, a listing's materials, a sale price. Each is treated the way why-agent treats a work item: sources make claims, one deterministic reconciler groups them by aspect and returns CORROBORATED, PARTIAL, CONFLICTED, or NO EVIDENCE. Model lanes advise; the gate decides. Tonight's sitting: 18 claims — 12 CONFLICTED, 5 PARTIAL, 1 CORROBORATED (computed by the reconciler from the rows on this page, not arranged).

Labels, honestly: 7 cases are RECORDED — fetched 2026-10-30 from the public pages of our own pilot business (Shenail Hair Salon & Nail Bar, Niagara Falls) and the directories that list it; their conflicts are real. 11 cases are SYNTHETIC FIXTURES, labelled in every title, modelled on real listing patterns; no fixture names a real third-party seller. The gate verdicts are deterministic and were computed tonight. Reviewer lanes have NOT run over the commerce corpus yet (no provider key on the build machine — see RUN_PACK.md), and the Nemotron Court's rebuttal round below is likewise NOT RUN. The page does not quote a lane verdict that does not exist.

1 · The claims

Shenail — published address — PARTIAL RECORDED

The claim: the salon's own site publishes this address — 8043 Lundy's Lane, Niagara Falls, ON L2H 1H3 (https://shenailhairlounge.ca/ (Location))

Aspects nobody spoke to: cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
shenail-site	state	the salon's own site publishes this address	`8043 Lundy's Lane, Niagara Falls, ON L2H 1H3`	https://shenailhairlounge.ca/ (Location)
google-listing	state	the Google listing shows the same address	`8043 Lundy's Lane, Niagara Falls, ON L2H 1H3`	Google places record 942693713797309, read 2026-10-30
yellowpages-directory	state	the Yellow Pages directory shows the same address	`8043 Lundy's Lane, Niagara Falls, ON L2H 1H3`	https://www.yellowpages.ca/bus/Ontario/Niagara-Falls/Shenail-Hair-Lounge/6656388.html
fresha-listing	outcome	the Fresha booking listing is for the same address	`8043 Lundy's Lane, Niagara Falls, ON L2H 1H3`	https://www.fresha.com/a/she-nail-hair-lounge-niagara-falls-8043-lundys-lane-dccczyh5

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

Shenail — founded 2007 — PARTIAL RECORDED

The claim: the site's hair page says 'Shenail Hair Lounge started as a home based nail and spa business in 2007' — 2007 (https://shenailhairlounge.ca/hair/)

Aspects nobody spoke to: outcome, cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
shenail-site	state	the site's hair page says 'Shenail Hair Lounge started as a home based nail and spa business in 2007'	`2007`	https://shenailhairlounge.ca/hair/

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

Shenail — published opening hours — CONFLICTED RECORDED

The claim: the salon's own site publishes these opening hours — Mon 10:00-16:00; Tue-Fri 09:30-18:00; Sat 09:30-17:00; Sun closed (https://shenailhairlounge.ca/ (Our Hours))

Conflict on state: Mon 10:00-16:00; Tue-Fri 09:30-18:00; Sat 09:30-17:00; Sun closed vs Mon 10:00-17:00; Tue-Fri 09:30-19:00; Sat 10:00-18:00 — cited: shenail-site — https://shenailhairlounge.ca/ (Our Hours); google-listing — Google places record 942693713797309, read 2026-10-30; yellowpages-directory — https://www.yellowpages.ca/bus/Ontario/Niagara-Falls/Shenail-Hair-Lounge/6656388.html.

Aspects nobody spoke to: outcome — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
shenail-site	state	the salon's own site publishes these opening hours	`Mon 10:00-16:00; Tue-Fri 09:30-18:00; Sat 09:30-17:00; Sun closed`	https://shenailhairlounge.ca/ (Our Hours)
google-listing	state	the Google listing shows the same hours as the site	`Mon 10:00-16:00; Tue-Fri 09:30-18:00; Sat 09:30-17:00; Sun closed`	Google places record 942693713797309, read 2026-10-30
yellowpages-directory	state	the Yellow Pages directory shows longer hours than the site and the Google listing	`Mon 10:00-17:00; Tue-Fri 09:30-19:00; Sat 10:00-18:00`	https://www.yellowpages.ca/bus/Ontario/Niagara-Falls/Shenail-Hair-Lounge/6656388.html
birdeye-profile-note	cause	the Birdeye profile states it has not been claimed by the business owner or representative — directory data for this business is not owner-maintained	`directory profiles not owner-claimed`	https://reviews.birdeye.com/shenail-hair-lounge-170031239435949

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

Shenail — published phone number — CONFLICTED RECORDED

The claim: the salon's own site publishes this phone number — +1 905-353-8889 (https://shenailhairlounge.ca/ (Location))

Conflict on state: +1 905-341-6888 vs +1 905-353-8889 — cited: shenail-site — https://shenailhairlounge.ca/ (Location); google-listing — Google places record 942693713797309, read 2026-10-30; fresha-listing — https://www.fresha.com/a/she-nail-hair-lounge-niagara-falls-8043-lundys-lane-dccczyh5.

Aspects nobody spoke to: outcome, cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
shenail-site	state	the salon's own site publishes this phone number	`+1 905-353-8889`	https://shenailhairlounge.ca/ (Location)
google-listing	state	the Google listing shows the same phone number as the site	`+1 905-353-8889`	Google places record 942693713797309, read 2026-10-30
fresha-listing	state	the Fresha listing for the same address shows a different phone number	`+1 905-341-6888`	https://www.fresha.com/a/she-nail-hair-lounge-niagara-falls-8043-lundys-lane-dccczyh5

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

Shenail — 'premiere' superlative — PARTIAL RECORDED

The claim: the site says the salon 'has proven to be Niagara Falls premiere Nail Bar & Hair Salon' — premiere (https://shenailhairlounge.ca/ (Welcome))

Aspects nobody spoke to: cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
shenail-site	state	the site says the salon 'has proven to be Niagara Falls premiere Nail Bar & Hair Salon'	`premiere`	https://shenailhairlounge.ca/ (Welcome)
google-listing	outcome	the Google listing rates the salon 4.1 of 5; no fetched source calls it the top or premiere salon, and a BestProsInTown list shows other Niagara Falls salons rated 4.7	`4.1`	Google places record 942693713797309, read 2026-10-30

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

Shenail — the standing 20% off promo — PARTIAL RECORDED

The claim: the site serves a '20% off' promo image uploaded in 2024-02; no terms, dates, or second mention appear on the fetched pages — 20% off (https://shenailhairlounge.ca/wp-content/uploads/2024/02/Shenail-20-Off.png)

Aspects nobody spoke to: outcome, cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
shenail-site	state	the site serves a '20% off' promo image uploaded in 2024-02; no terms, dates, or second mention appear on the fetched pages	`20% off`	https://shenailhairlounge.ca/wp-content/uploads/2024/02/Shenail-20-Off.png

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

Shenail — the rating a customer finds — CONFLICTED RECORDED

The claim: the Birdeye profile displays this rating for the salon — 4.5 (251 reviews) (https://reviews.birdeye.com/shenail-hair-lounge-170031239435949)

Conflict on outcome: 4.1 (28 reviews) vs 4.5 (251 reviews) vs 4.7 (110 reviews) — cited: birdeye-profile — https://reviews.birdeye.com/shenail-hair-lounge-170031239435949; google-listing — Google places record 942693713797309, read 2026-10-30; bestprosintown-listing — https://www.bestprosintown.com/on/niagara-falls/shenail-hair-lounge-/.

Aspects nobody spoke to: state, cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
birdeye-profile	outcome	the Birdeye profile displays this rating for the salon	`4.5 (251 reviews)`	https://reviews.birdeye.com/shenail-hair-lounge-170031239435949
google-listing	outcome	the Google listing displays a different rating and a much smaller review count	`4.1 (28 reviews)`	Google places record 942693713797309, read 2026-10-30
bestprosintown-listing	outcome	the BestProsInTown listing displays a third rating and count	`4.7 (110 reviews)`	https://www.bestprosintown.com/on/niagara-falls/shenail-hair-lounge-/

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

SYNTHETIC FIXTURE — 'As seen on TV' — PARTIAL SYNTHETIC FIXTURE

The claim: the listing carries an 'As seen on TV' badge; nothing else in the corpus speaks about the item — As seen on TV (fixture:listing/tv)

Aspects nobody spoke to: outcome, cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
listing (synthetic fixture)	state	the listing carries an 'As seen on TV' badge; nothing else in the corpus speaks about the item	`As seen on TV`	fixture:listing/tv

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

SYNTHETIC FIXTURE — award banner, no award found — CONFLICTED SYNTHETIC FIXTURE

The claim: the site's hero banner claims a 2025 'Best of the City' award — Best of the City 2025 — award-winning (fixture:site/award)

Conflict on state: vs Best of the City 2025 — award-winning — cited: site hero banner (synthetic fixture) — fixture:site/award; award winners list (synthetic fixture) — fixture:web/award-winners-2025.

Aspects nobody spoke to: outcome, cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
site hero banner (synthetic fixture)	state	the site's hero banner claims a 2025 'Best of the City' award	`Best of the City 2025 — award-winning`	fixture:site/award
award winners list (synthetic fixture)	state	the claimed award does not appear on the award programme's published 2025 winners list — a recorded non-match, kept as a claim with an empty value		fixture:web/award-winners-2025

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

SYNTHETIC FIXTURE — the clean listing — CORROBORATED SYNTHETIC FIXTURE

The claim: the listing describes a solid walnut serving board with a food-safe oil finish — solid walnut, food-safe oil finish (fixture:listing/board)

Every expected aspect is spoken to and no two sources disagree — the only shape that earns CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
listing (synthetic fixture)	state	the listing describes a solid walnut serving board with a food-safe oil finish	`solid walnut, food-safe oil finish`	fixture:listing/board
nemotron-vl photo read (synthetic fixture)	state	the vision lane's read of the product photo reports solid walnut grain and an oil finish, no veneer seam	`solid walnut, food-safe oil finish`	fixture:photo/board
supplier invoice (synthetic fixture)	outcome	the supplier invoice for the board lists solid walnut and a food-safe oil finish	`solid walnut, food-safe oil finish`	fixture:invoice/board
audit note (synthetic fixture)	cause	the audit note records that the supplier invoice matches the listing's stated materials	`supplier invoice matches the listing's stated materials`	fixture:audit/board

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

SYNTHETIC FIXTURE — 'genuine leather' bag — CONFLICTED SYNTHETIC FIXTURE

The claim: the listing claims the bag is genuine leather — genuine leather (fixture:listing/bag)

Conflict on state: PU synthetic leather vs genuine leather — cited: listing (synthetic fixture) — fixture:listing/bag; manufacturer spec (synthetic fixture) — fixture:spec/bag.

Aspects nobody spoke to: outcome, cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
listing (synthetic fixture)	state	the listing claims the bag is genuine leather	`genuine leather`	fixture:listing/bag
manufacturer spec (synthetic fixture)	state	the manufacturer's spec for the same SKU lists PU synthetic leather	`PU synthetic leather`	fixture:spec/bag

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

SYNTHETIC FIXTURE — '100% handmade' mug — CONFLICTED SYNTHETIC FIXTURE

The claim: the listing claims the mug is 100% handmade ceramic — 100% handmade ceramic (fixture:listing/mug)

Conflict on state: 100% handmade ceramic vs machine-printed decal on a moulded blank — cited: listing (synthetic fixture) — fixture:listing/mug; nemotron-vl photo read (synthetic fixture) — fixture:photo/mug.

Aspects nobody spoke to: cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
listing (synthetic fixture)	state	the listing claims the mug is 100% handmade ceramic	`100% handmade ceramic`	fixture:listing/mug
nemotron-vl photo read (synthetic fixture)	state	the vision lane's read of the product photo reports a printed decal and uniform mould lines	`machine-printed decal on a moulded blank`	fixture:photo/mug
supplier spec (synthetic fixture)	outcome	the supplier spec for the blank describes machine printing on a moulded blank	`machine-printed decal on a moulded blank`	fixture:spec/mug

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

SYNTHETIC FIXTURE — 'open 7 days' vs its own hours table — CONFLICTED SYNTHETIC FIXTURE

The claim: the site's hero banner claims the business is open 7 days a week — open 7 days a week (fixture:site/hero-hours)

Conflict on state: Sunday: closed vs open 7 days a week — cited: site hero (synthetic fixture) — fixture:site/hero-hours; same site's hours table (synthetic fixture) — fixture:site/hours-table.

Aspects nobody spoke to: outcome, cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
site hero (synthetic fixture)	state	the site's hero banner claims the business is open 7 days a week	`open 7 days a week`	fixture:site/hero-hours
same site's hours table (synthetic fixture)	state	the same site's hours table lists Sunday as closed	`Sunday: closed`	fixture:site/hours-table

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

SYNTHETIC FIXTURE — 'organic cotton' tee — CONFLICTED SYNTHETIC FIXTURE

The claim: the listing claims the shirt is 100% organic cotton — 100% organic cotton (fixture:listing/tee)

Conflict on state: 100% organic cotton vs 60% cotton / 40% polyester — cited: listing (synthetic fixture) — fixture:listing/tee; nemotron-vl label read (synthetic fixture) — fixture:photo/tee-label.

Aspects nobody spoke to: outcome, cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
listing (synthetic fixture)	state	the listing claims the shirt is 100% organic cotton	`100% organic cotton`	fixture:listing/tee
nemotron-vl label read (synthetic fixture)	state	the vision lane's read of the sewn-in label reports a cotton-polyester blend	`60% cotton / 40% polyester`	fixture:photo/tee-label

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

SYNTHETIC FIXTURE — review-count badge — CONFLICTED SYNTHETIC FIXTURE

The claim: the site displays a badge claiming 500+ five-star reviews — 500+ five-star reviews (fixture:site/reviews-badge)

Conflict on state: 28 reviews (4.1 stars) vs 500+ five-star reviews — cited: site badge (synthetic fixture) — fixture:site/reviews-badge; google-style listing (synthetic fixture) — fixture:web/listing-reviews.

Aspects nobody spoke to: outcome, cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
site badge (synthetic fixture)	state	the site displays a badge claiming 500+ five-star reviews	`500+ five-star reviews`	fixture:site/reviews-badge
google-style listing (synthetic fixture)	state	the public listing for the same business shows 28 reviews at 4.1 stars	`28 reviews (4.1 stars)`	fixture:web/listing-reviews

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

SYNTHETIC FIXTURE — price list vs booking price — CONFLICTED SYNTHETIC FIXTURE

The claim: the salon's site price list shows a shellac manicure at $35 — Shellac manicure $35 (fixture:site/price-list)

Conflict on state: Shellac manicure $35 vs Shellac manicure $45 — cited: site price list (synthetic fixture) — fixture:site/price-list; booking page (synthetic fixture) — fixture:web/booking.

Aspects nobody spoke to: outcome, cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
site price list (synthetic fixture)	state	the salon's site price list shows a shellac manicure at $35	`Shellac manicure $35`	fixture:site/price-list
booking page (synthetic fixture)	state	the booking page charges $45 for the same shellac manicure	`Shellac manicure $45`	fixture:web/booking

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

SYNTHETIC FIXTURE — shipping promise vs policy — CONFLICTED SYNTHETIC FIXTURE

The claim: the listing promises dispatch in 1-2 business days — ships in 1-2 business days (fixture:listing/shipping)

Conflict on state: made to order: ships in 7-10 business days vs ships in 1-2 business days — cited: listing (synthetic fixture) — fixture:listing/shipping; shop policy page (synthetic fixture) — fixture:policy/shipping.

Aspects nobody spoke to: outcome, cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
listing (synthetic fixture)	state	the listing promises dispatch in 1-2 business days	`ships in 1-2 business days`	fixture:listing/shipping
shop policy page (synthetic fixture)	state	the same shop's policy page says items are made to order and ship in 7-10 business days	`made to order: ships in 7-10 business days`	fixture:policy/shipping

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

SYNTHETIC FIXTURE — 'was $80' sale price — CONFLICTED SYNTHETIC FIXTURE

The claim: the listing advertises a was/now discount: was $80, now $40 — was $80, now $40 (fixture:listing/was-price)

Conflict on state: price recorded as $40 for the last 90 days; no $80 price recorded vs was $80, now $40 — cited: listing (synthetic fixture) — fixture:listing/was-price; price history (synthetic fixture) — fixture:history/price.

Aspects nobody spoke to: outcome, cause — an unknown is not a pass, so this cannot read CORROBORATED.

Evidence rows (each carries its receipt)

Source	Aspect	Statement	Value	Receipt
listing (synthetic fixture)	state	the listing advertises a was/now discount: was $80, now $40	`was $80, now $40`	fixture:listing/was-price
price history (synthetic fixture)	state	the price record shows the item listed at $40 for the last 90 days, with no $80 price in the record	`price recorded as $40 for the last 90 days; no $80 price recorded`	fixture:history/price

Reviewer lanes over this corpus: NOT RUN — no provider key on the build machine; the gate verdict above is deterministic and computed from the rows shown. Live lanes: RUN_PACK.md step 3.

2 · Nemotron Court — the lanes, cross-examined

The reviewer tiers' Round 1 verdicts below are RECORDED from the 2026-10-28 live Token Factory run. On 8 of 14 items a lane disagreed with the deterministic gate; every disagreement is preserved with its citations, never averaged. The citation audit is computed, not narrated: for each lane verdict it lists the evidence that verdict fails to account for. The rebuttal round is NOT RUN — it is a live model round and no provider key is present tonight (RUN_PACK.md, step 4).

bm-card-lies — gate: CONFLICTED · planted truth CONFLICTED

Round 1 (recorded): fast (nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B): CORROBORATED ($0.0002) — disagrees with the gate; deep (nvidia/Nemotron-3-Ultra-550b-a55b): CONFLICTED ($0.0018) — agrees with the gate.

Audit — fast: verdict not supported as stated. Unanswered: conflict on state (card — bench/cards/bm-card-lies.md:1; ci — bench/ci/2); missing on cause.

Audit — deep: verdict supported; it answers every conflict and missing aspect the gate found.

Rebuttal round: NOT RUN — cross-examination is a live model round and no provider key is present on the build machine; run `NEBIUS_API_KEY=... python3 court.py --live` to hold it. The citation audit above is the deterministic layer and is complete.

bm-clean-ship — gate: CORROBORATED · planted truth CORROBORATED

Round 1 (recorded): fast (nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B): CONFLICTED ($0.0001) — disagrees with the gate; deep (nvidia/Nemotron-3-Ultra-550b-a55b): CORROBORATED ($0.0014) — agrees with the gate.

Audit — fast: verdict not supported as stated. Unanswered: agreement on cause (git — bench/git/1c); agreement on outcome (deploy — bench/deploy/1); agreement on state (card — bench/cards/bm-clean-ship.md:1; git — bench/git/1; ci — bench/ci/1).

Audit — deep: verdict supported; it answers every conflict and missing aspect the gate found.

bm-forged-aspect — gate: PARTIAL · planted truth PARTIAL

Round 1 (recorded): fast (nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B): CONFLICTED ($0.0003) — disagrees with the gate; deep (nvidia/Nemotron-3-Ultra-550b-a55b): UNPARSEABLE ($0.0125) — disagrees with the gate.

Audit — fast: verdict not supported as stated. Unanswered: agreement on outcome (ops — bench/ops/12); agreement on state (card — bench/cards/bm-forged-aspect.md:1; git — bench/git/12).

Audit — deep: verdict not supported as stated. Unanswered: the gate’s verdict itself — a preserved disagreement.

bm-green-check-wrong-scope — gate: PARTIAL · planted truth PARTIAL

Round 1 (recorded): fast (nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B): PARTIAL ($0.0002) — agrees with the gate; deep (nvidia/Nemotron-3-Ultra-550b-a55b): CONFLICTED ($0.0011) — disagrees with the gate.

Audit — fast: verdict supported; it answers every conflict and missing aspect the gate found.

Audit — deep: verdict not supported as stated. Unanswered: agreement on outcome (deploy — bench/deploy/3); agreement on state (card — bench/cards/bm-green-check-wrong-scope.md:1; ci — bench/ci/3).

bm-missing-cause — gate: PARTIAL · planted truth PARTIAL

Round 1 (recorded): fast (nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B): CORROBORATED ($0.0001) — disagrees with the gate; deep (nvidia/Nemotron-3-Ultra-550b-a55b): CORROBORATED ($0.0012) — disagrees with the gate.

Audit — fast: verdict not supported as stated. Unanswered: missing on cause.

Audit — deep: verdict not supported as stated. Unanswered: missing on cause.

bm-no-evidence — gate: NO EVIDENCE · planted truth NO EVIDENCE

Round 1 (recorded): fast: not run — the item has no claims; the harness does not run lanes on an item with nothing to judge (recorded 2026-10-28 behaviour); deep: not run — the item has no claims; the harness does not run lanes on an item with nothing to judge (recorded 2026-10-28 behaviour).

bm-receipt-mismatch — gate: CONFLICTED · planted truth CONFLICTED

Round 1 (recorded): fast (nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B): CONFLICTED ($0.0001) — agrees with the gate; deep (nvidia/Nemotron-3-Ultra-550b-a55b): CONFLICTED ($0.0013) — agrees with the gate.

Audit — fast: verdict supported; it answers every conflict and missing aspect the gate found.

Audit — deep: verdict supported; it answers every conflict and missing aspect the gate found.

bm-reviewer-split — gate: CONFLICTED · planted truth CONFLICTED

Audit — fast: verdict supported; it answers every conflict and missing aspect the gate found.

Audit — deep: verdict supported; it answers every conflict and missing aspect the gate found.

bm-stale-card — gate: CONFLICTED · planted truth CONFLICTED

Round 1 (recorded): fast (nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B): PARTIAL ($0.0001) — disagrees with the gate; deep (nvidia/Nemotron-3-Ultra-550b-a55b): CONFLICTED ($0.0011) — agrees with the gate.

Audit — fast: verdict not supported as stated. Unanswered: the gate’s verdict itself — a preserved disagreement.

Audit — deep: verdict supported; it answers every conflict and missing aspect the gate found.

bm-unexpected-aspect — gate: CORROBORATED · planted truth CORROBORATED

Round 1 (recorded): fast (nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B): PARTIAL ($0.0002) — disagrees with the gate; deep (nvidia/Nemotron-3-Ultra-550b-a55b): PARTIAL ($0.0017) — disagrees with the gate.

Audit — fast: verdict not supported as stated. Unanswered: agreement on cause (reviewer-x — bench/rev/11); agreement on outcome (qa — bench/qa/11); agreement on state (card — bench/cards/bm-unexpected-aspect.md:1; git — bench/git/11).

Audit — deep: verdict not supported as stated. Unanswered: agreement on cause (reviewer-x — bench/rev/11); agreement on outcome (qa — bench/qa/11); agreement on state (card — bench/cards/bm-unexpected-aspect.md:1; git — bench/git/11).

bm-vl-receipt — gate: CORROBORATED · planted truth CORROBORATED

Round 1 (recorded): fast (nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B): PARTIAL ($0.0003) — disagrees with the gate; deep (nvidia/Nemotron-3-Ultra-550b-a55b): CORROBORATED ($0.0028) — agrees with the gate.

Audit — fast: verdict not supported as stated. Unanswered: agreement on cause (audit — bench/audit/13); agreement on outcome (nemotron-vl — receipt.png via nvidia/Nemotron-3.5-Super-VL; bank — bench/bank/13); agreement on state (card — bench/cards/bm-vl-receipt.md:1).

Audit — deep: verdict supported; it answers every conflict and missing aspect the gate found.

Exhibits: vision read receipt.png via nvidia/Nemotron-3.5-Super-VL → paid-100.

bm-web-confirms — gate: CORROBORATED · planted truth CORROBORATED

Round 1 (recorded): fast (nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B): CORROBORATED ($0.0001) — agrees with the gate; deep (nvidia/Nemotron-3-Ultra-550b-a55b): CORROBORATED ($0.0016) — agrees with the gate.

Audit — fast: verdict supported; it answers every conflict and missing aspect the gate found.

Audit — deep: verdict supported; it answers every conflict and missing aspect the gate found.

Exhibits: live-evidence read https://status.example.com/ok → live.

bm-web-contradicts — gate: CONFLICTED · planted truth CONFLICTED

Audit — fast: verdict supported; it answers every conflict and missing aspect the gate found.

Audit — deep: verdict supported; it answers every conflict and missing aspect the gate found.

Exhibits: live-evidence read https://status.example.com/incident → down.

bm-web-nonmatch — gate: CONFLICTED · planted truth CONFLICTED

Audit — fast: verdict supported; it answers every conflict and missing aspect the gate found.

Audit — deep: verdict supported; it answers every conflict and missing aspect the gate found.

Exhibits: live-evidence read https://example.com/other → (no literal match).

3 · Rigor — the adversarial benchmark underneath

Before it judged a single commerce claim, the gate was scored on a planted-truth adversarial corpus (agent-work claims, recorded live run 2026-10-28, reviewer-lane cost $0.033). The lanes' false approvals are published, not hidden — they are the evidence for keeping lanes advisory. Scores below are computed from the committed files by the same counting rule as benchmark/score.py.

Scorer	Model	Correct	False approvals
Deterministic gate (the reconciler)	`evidence.py — no model`	14/14	0
Deep lane (advisory)	`nvidia/Nemotron-3-Ultra-550b-a55b`	9/14	1
Fast lane (advisory)	`nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B`	6/14	2

The original agent-work cases, with lane receipts and narration checks, are in the first Case Lab: the Case Lab — bm-card-lies (CONFLICTED); bm-clean-ship (CORROBORATED); bm-no-evidence (NO EVIDENCE); bm-reviewer-split (CONFLICTED); bm-web-nonmatch (CONFLICTED); why-alpha (CONFLICTED); why-beta (PARTIAL).

4 · Live-evidence check — Tavily (measured 2026-10-30)

Tavily is our evidence-expansion layer, not an answer layer. RECORDED On 4 claims / 19 pages it widened the evidence surface the reconciler could test — and the open web disagreed with itself on all four: PARTIAL → CONFLICTED is the system becoming safer, because a conflict that exists in the wild was invisible in the static corpus. The 5/19 verbatim rate is a property of the web, and it is exactly why raw search output must pass a verbatim gate before it can corroborate anything. We use Search → Extract, never the generated answer as evidence.

One claim, end to end. Our own planning claim: the Nebius deadline, Oct 30. Query: “Nebius NVIDIA Global AI Hackathon submissions close deadline”. Tavily returned 4 pages — https://nebiusglobalaihackathon.devpost.com/rules; https://www.startupgrantsindia.com/nebius-nvidia-global-ai-hackathon; https://nebiusglobalaihackathon.devpost.com; https://nebius.com/events/nebius-nvidia-ai-builders-hackathon. Pages containing the claimed value verbatim: 0 of 4. Without web rows the reconciler returned PARTIAL; with them, CONFLICTED.

Claimed value	Without web	With web — verbatim	Pages credited	With web — normalization v1	Pages credited
`Oct 30`	PARTIAL	CONFLICTED	0/4	CONFLICTED	3/4
`November 3`	PARTIAL	CONFLICTED	0/5	CONFLICTED	0/5
`1,000 credits`	PARTIAL	CONFLICTED	1/5	PARTIAL	5/5
`Nemotron`	PARTIAL	CONFLICTED	4/5	CONFLICTED	4/5

Normalization v1 is a second, labelled matching mode (dates, numbers and phones canonicalised at ingest; every hit is labelled normalized:<rule> on its row). Over the same receipts it credits 12 of 19 pages, and the “1,000 credits” claim improves to PARTIAL — all five of its pages credited by the number rule. The other three claims stay CONFLICTED. “Oct 30” stays CONFLICTED even with 3 of its 4 pages matched by the date rule: the fourth page states no date, and a recorded miss counts as disagreement — an open design question, deliberately not changed here. Reconciliation itself stayed exact.

Receipts — the raw Tavily responses and both ablation result files: docs/tavily-live-2026-10-30/ in the repo. Harness: ablation.py; lane: tavily_lane.py; matching rules: normalize.py.

Generated by claimcourt_lab/build.py from claimcourt/cases.jsonl, court/debate-replay.json, benchmark/ and the recorded Tavily ablation receipts in docs/tavily-live-2026-10-30/. The builder makes no network request; the only external call on this page is the optional ECharts CDN chart, and the table above carries the same numbers without it.