RBI Bakeoff: Coval results

Every simulation in the Coval org, pulled through the REST API on 2026-08-12. Completion Rate is Coval's judged verdict on whether the AI took the order to completion.

Overall completion rate
67.2%
1,633 judged conversations
Simulations pulled
1,744
With transcripts
1,635
No transcript
109
Runs
107

Why conversations fail

Every Completion=NO conversation (536 of them), classified by what actually happened. 455 failures (85%) are the audio-only Coval persona unable to see the order-confirm screen: the AI asks "is your order on the screen correct?", the persona asks for a readback, the AI refuses or escalates. Excluding those from numerator and denominator, completion is 93.1% against the 67.2% headline.

Harness artifact or protocol miss Real product failure
Persona can't see screen, AI escalates
62.3%
334/536
Persona can't see screen, AI refuses readback
22.6%
121/536
Escalated for other reasons
9.7%
52/536
Order fine, never directed forward
2.6%
14/536
Other (loops, cut off, misc)
2.1%
11/536
Wrong or incomplete order
0.7%
4/536
Table view
Failure causeCountShare
Persona can't see screen, AI escalates33462.3%
Persona can't see screen, AI refuses readback12122.6%
Escalated for other reasons529.7%
Order fine, never directed forward142.6%
Other (loops, cut off, misc)112.1%
Wrong or incomplete order40.7%

Completion rate by persona

The widest spread in the data. Background noise and customer confusion cost more than any single scenario does, which points at the audio front end rather than the ordering logic. Persona names click through to every conversation for that persona, with verdicts and transcripts.

81.4%
83/102
Table view
PersonaCompletionPassed
High Background Noise Speaker16.7%17/102
Confused Customer18.8%19/101
Impatient Customer44.1%45/102
Accent - Hispanic/Latino51.0%52/102
Behavioral - Complex-Order61.8%63/102
Behavioral - Rush66.7%68/102
Standard Customer69.2%72/104
Low Volume Speaker71.6%73/102
Interruptive Speaker71.6%73/102
Behavioral - Mind-Changer79.4%81/102
Accent - AAVE81.4%83/102
Accent - General American84.3%86/102
Super Fast Speaker87.3%89/102
Accent - Louisiana Creole88.2%90/102
Accent - Southern US89.2%91/102
Behavioral - Too-Talkative93.1%95/102

Completion rate by test case

Test case names link to the filtered runs list in Coval. Combo-offer scenarios cluster at the bottom.

51.0%
49/96
54.2%
52/96
57.3%
55/96
59.4%
57/96
64.6%
62/96
68.4%
65/95
70.8%
68/96
71.9%
69/96
74.0%
71/96
75.0%
72/96
79.2%
76/96
81.2%
78/96
Table view
Test caseCompletionPassed
UAT #951.0%49/96
Combo offer after tenders53.1%51/96
UAT #2954.2%52/96
UAT #257.3%55/96
UAT #6159.4%57/96
UAT #1064.6%62/96
Combo offer after signature chicken65.6%63/96
Alternative side offering66.7%64/96
UAT #6968.4%65/95
Combo offer after sandwich68.8%66/96
UAT #870.8%68/96
UAT #3671.9%69/96
UAT #4274.0%71/96
UAT #6475.0%72/96
UAT #1179.2%76/96
UAT #181.2%78/96
Mid-order quantity change82.3%79/96

Completion rate by batch

Four official passes, no movement. Whatever changed between them did not move the headline metric.

Simulation Run 2026-07-20 23:03 (scratch)
83.5%
71/85
RBI Bakeoff
67.1%
365/544
Simulation Run 2026-07-20 23:58 (scratch)
67.1%
57/85
RBI Bakeoff 2026-07-24
66.5%
181/272
RBI Bakeoff 2026-07-24 Pass 5
66.4%
180/271
RBI Bakeoff 2026-07-23
65.8%
179/272
Simulation Run 2026-07-20 22:35 (scratch)
62.7%
64/102
Table view
BatchCompletionPassed
Simulation Run 2026-07-20 23:03 (scratch)83.5%71/85
RBI Bakeoff67.1%365/544
Simulation Run 2026-07-20 23:58 (scratch)67.1%57/85
RBI Bakeoff 2026-07-2466.5%181/272
RBI Bakeoff 2026-07-24 Pass 566.4%180/271
RBI Bakeoff 2026-07-2365.8%179/272
Simulation Run 2026-07-20 22:35 (scratch)62.7%64/102

UAT #64: Cajun Rice

Customer asks for Cajun Rice, an item that is not on the menu. The AI is expected to say so and recommend an alternative. Judged completion is 75.0% (72/96). Below is what the AI actually said first, classified into mutually exclusive buckets.

Correct handling Stalled or evaded Wrong item risk
Denied, offered alternative
84.4%
81/96
Stalled, never answered
6.2%
6/96
False affirmation
4.2%
4/96
Silent substitution
3.1%
3/96
Off-topic / no answer
2.1%
2/96
Table view
BehaviorCountShare
Denied, offered alternative8184.4%
Stalled, never answered66.2%
False affirmation44.2%
Silent substitution33.1%
Off-topic / no answer22.1%

Every conversation that did not handle it cleanly

Simulation IDs link straight to the transcript in Coval.

BehaviorSimulationCompletionWhat the AI saidFinal cart
False affirmationhgyuPP8KDqJJEsKMvESaU3YESYes. We do. Anything else?1x Red Beans & Rice
Silent substitution4FR4jwHRfqUvqgAJrF2XPTNOI understand. You like red beans and rice instead?1x Red Beans & Rice
False affirmationVgYhUwitqqw9P598U7DTZ6YESYes. We have cinnamon apple pie. Anything else?1x Red Beans & Rice
False affirmationNrUhuZ4jqkBsHCxiSwvGA5NOYes. We still have Cajun rice. What else can I get for you?1x 2 Piece Signature Chicken Regular Combo (S
Silent substitutiondmAFv6oQ8GWM769WUYxndvNOI understand. You like red beans and rice instead?1x Red Beans & Rice
Silent substitution45XWUpe4n9PRL3RRg2gNFTYESWe have red beans and rice, regular or large.1x Red Beans & Rice
False affirmationiq3YZ5keVVpMYJ658SihtcYESYes. We do. Anything else?1x Red Beans & Rice
Stalled, never answered75NkfSQrS7ozRZzwQk4myTNOMoment while I check.1x Red Beans & Rice
Off-topic / no answerfGV8TzvyGwyeNYb2i5pYPZYESSorry. I didn't catch that. What drink would you like?1x Red Beans & Rice
Stalled, never answeredcfni4H85i9EDdNmfiEtSvnNOOf course, take your time. What can I get started for you?1x Red Beans & Rice; 1x 12 Piece Signature Ch
Stalled, never answeredJgFG4PQBVwfSgcFVUJWhqrYESMoment while I check.1x Spicy Chicken Sandwich Large Combo (Red Be
Off-topic / no answerMAYfprdnawRdU4QxacycXnNOOf course. Regular or large?1x Red Beans & Rice
Stalled, never answered5ik9KqeSZ5q6EBFuEZFE6XYESLet me check that for you.1x Cajun Fries; 1x Biscuit; 1x Cane Sweeeet T
Stalled, never answeredaBpxBKFoEnNoNUrCwcRrhLYESOf course, take your time. What can I get started for you?1x Red Beans & Rice; 1x 12 Piece Signature Ch
Stalled, never answeredfriVsaj8bnB4e4eTNVVuKoYESOne moment while I check.1x Red Beans & Rice