Three Voices,
One Mayor’s Race.

Spencer Pratt’s audiences keep cheering for him. Joe Rogan’s two-hour podcast, Megyn Kelly’s broadcast, the Skirball debate, the Adam Carolla show — the response on the night has been some version of warmth in rooms that otherwise have very little in common. What he says back to them isn’t warm. His transcripts skew negative on the words alone. Point a multimodal model at the same footage and it agrees: across the 194 clips where Pratt is doing the talking, it reads him as arguing against something in 72% of them and arguing for something in only 21%.

The bar chart above is one way to read what each candidate’s words look like on average, plus the range of where the true average likely sits. Bass’s range and Pratt’s range don’t overlap at all — there is a clear gap between the bottom of Bass’s and the top of Pratt’s, wider than either range is. That is the version of a real difference you would show a friend. Raman sits in the middle, with a wider range because we have fewer clips of her own voice; her range just barely includes zero at its upper edge. The next four sections are about what each of those positions means.

The yardstick is the same for all three. The words get scored for how positive or negative they read. The audio gets scored for how heated and how commanding the voice sounds. A model that takes in the words, audio, and video together calls whether the candidate is arguing for or against something. None of this is psychic — it can’t tell whether a candidate privately means what they say, whether a program is working, or whether one of them would be a better mayor than the others. It tells you what each of them sounds like under the same instrument.

“Negative” on a word-level scale doesn’t separate a grieving mother from a budget hawk from a defense attorney on cross-examination. Reading the words alone isn’t enough to tell those apart, so a multimodal model ran on the same footage — one that watches the picture and listens to the audio while it reads the transcript — and scored each clip on three things: how warm or cold the speech sounds, how worked up the speaker is, and which way they are arguing.

The model places Bass at the warm end of its valence scale, which runs from 0 to 1 with 0.5 as neutral. Across her 430 voice-gated clips she reads 0.65 for warmth and 0.58 for emotional intensity, and 70% of her clips read as arguing for something rather than against. That is what a mayor delivering a State of the City sounds like on this instrument.

Raman points in the same direction at a cooler temperature. Across her 121 clips she comes in at 0.58 on warmth — a few hundredths below Bass — and 71% of her clips read as arguing for something, statistically the same rate as Bass’s 70%. Where they differ is set-piece access, not stance: Bass speaks from a podium, Raman from a couch across from a host.

Pratt is the outlier. His warmth reads 0.31 — the lowest in the field by a clear margin — and his arousal reads 0.70, the highest. Of his 194 clips, 72% land as arguing against something. He’s the only candidate of the three whose own voice runs against the room he is running to lead, in almost every room he enters.

Two different reads — one looking only at the words, one watching and listening to the video — agree on the direction of each candidate’s register on Bass 80.7% of the time across 430 paired clips, with a correlation of r = +0.74 — strong, by the standards of how this kind of measurement usually behaves. When two unrelated tools point at the same three speakers and rank them the same way, the ranking is probably picking up something real about the speakers, not something quirky about either tool.

Spencer Pratt is a professional. Not a political professional — a media one. He and his wife Heidi Montag spent four years on the MTV docuseries The Hills, where the producers needed a heel and Pratt obliged. After the show ended in 2010 he kept the camera. He hosts a podcast called The Fame Game, has 2.3 million TikTok followers attached to it, and has been a working entertainment-industry name for fifteen consecutive years. When he announced his run for mayor on January 7th, 2026 — one year to the day after losing his Palisades home in the fire — he announced it from the burned lot, on Snapchat, in vertical video. He is not new to a camera.

His party registration is Republican (publicly reported in coverage of the campaign and stated on his own Ballotpedia entry). When CBS News asked him in May whether he leans Democrat or Republican, he declined to put himself in either box. In a May 8th sit-down with NBC4 LA, after the anchor referenced a prior Pratt interview comparing his community-service background to Barack Obama’s at the same career stage, Pratt told the anchor: “President Obama actually didn’t even have awards when he was a community organizer. He was able to become a senator and then a president for eight years. So I feel like him and I have the same experience.” By April his campaign had raised more than half a million dollars, with at least fifty donors maxing out their contributions per Newsweek’s reporting, which names entertainment-industry contributors including the Lakers owner Jeanie Buss, the singer Katharine McPhee, the producer Jeremy Latcham, and Heidi Montag’s sister Holly Wilson among the maxed-out donors. The Obama comparison, the entertainment-industry roster, and the Republican party registration are what a first-time candidate gets when twenty years of camera training meet a city primary. The analysis doesn’t resolve the contradictions. It records the way it sounds when he tries to fit them all into one campaign.

The words-only read scores him at −0.28 on average across the eight first-person videos in the corpus: Rogan’s two-hour podcast, Megyn Kelly’s broadcast, a sit-down with Meghan Daum on his burned lot, the NBC4 LA Obama-comparison interview, the Fox News appearance from three weeks into the campaign, the full CBS News interview, and the May 6th Skirball debate. Two-thirds of his scored segments come back negatively framed. Changing the room barely moves the number. His warmest room is Rogan’s, where he still scores −0.37. The only room where his score tips positive is the one where Megyn Kelly’s guest Michael Knowles is praising him on air — even there, the praise belongs to the host, not to anything Pratt is saying back.

The four clips above are inside that floor. The most negative segment in the entire Pratt corpus is on Rogan, talking about what happens “to the people whose houses just” burned. He returns to the same thread later in the same episode and describes Palisades homes as having been “stolen” from their owners. On NBC4 LA, the anchor opens with “Donald Trump is not popular in this city” and asks how Pratt answers opponents who have labeled him a MAGA candidate. Pratt replies: “Thankfully, it’s a nonpartisan race ... I will never do national politics. I will never talk about other states. I’m going to talk about how we fix Los Angeles.” The model scores the segment slightly negative. The only segment in the entire Pratt corpus to score above +0.9 is on Daum’s show, where he tells her about strangers on the street: “They hug me and they go, thank God for you.” That is the one room that let him sell the rescuer, and he took it.

Across the corpus, the clips where Pratt is doing the talking at his loudest cluster around a small, recurring set of topics. On Joe Rogan’s podcast he describes a Los Angeles where parents “have to have their kids in the back seat staring at an iPad and not to look out the window because meth addicts will just be having sex on the side of the street,” and where “naked zombies” roam the streets (Joe Rogan Experience #2483, April 15 2026, ~15–28 min). On Meghan Daum’s show he describes people on the street as “drug zombies” and the abuse of shelter animals as “Nazi-level experiments on animals at a rate you can’t even comprehend” (May 2026, Unspeakeasy with Meghan Daum, 27–28 min). In the same conversation, on what he will do to repeat offenders, he says: “I’m going to destroy their lives” (18 min). Each of these quotes is verbatim from the diarized transcript with Pratt as the dominant speaker for the segment. None of the other candidates in the field use this imagery on the record. The article doesn’t adjudicate whether his characterization of LA disorder is correct; it records that he is the only candidate on the June 2 ballot whose own voice operates in this register this often.

The policy record around the rhetoric is consistent. Pratt has called the city’s leadership “criminally negligent” on the fire response (Fox News, January 28th, 2026). His homelessness plan, as reported by VINnews on May 10th, gives an open encampment a 2-to-3-week window to come down before what he calls mass arrests of street homeless and street criminals begin. His public-safety pitch, as NBC4 LA reported, is to expand the LAPD to twelve thousand officers, paid for by reallocating the city’s homelessness-services budget. He has called for audits of the nonprofits that deliver those services. The platform is consistent with the register: a candidate running as a sharp break from the way LA has been governed for the past decade, spoken in the cadence that break sounds like. Voters who think LA has been governed well by the current establishment will hear that cadence as alarming; voters who think it hasn’t will hear it as overdue. The analysis scores the cadence; it doesn’t adjudicate which voter is correct.

What the analysis also doesn’t tell us is whether Pratt’s twenty years of camera training, his 2.3-million TikTok reach, or his half-million-dollar donor roster translate into running the seventh-largest city budget in the country. He has never held elected office, managed a public agency, written a budget, or supervised unionized employees. Neither has any other candidate the first time they ran. The argument for Pratt the candidate is largely an argument that the city needs an outsider; the argument against him is largely that an outsider is precisely what a fifteen-billion-dollar municipal government doesn’t need. The analysis scores how he sounds. June 2 is when LA decides what that sound is worth.

Karen Bass started in South Los Angeles in the 1980s as a community organizer responding to the crack epidemic. She founded the Community Coalition. She served six years in the California Assembly, where she became the first Black woman in U.S. history to lead a state legislative house. She spent twelve years in Congress. When she was sworn in as the 43rd mayor of Los Angeles on December 11th, 2022, she was the first Black woman to hold the office. When the analysis we’ve pointed at her hears Karen Bass on a podium, it hears what someone with that résumé sounds like on a podium: warm, in-charge, in full control of the room. The two State of the City addresses she has delivered in office contain the warmest single rooms in the entire scored corpus. The 2026 State of the City, on February 2nd, opens with her on the dais telling Angelenos that “welcoming the world for the World Cup, Los Angeles is already showing what it does best”, scored at +0.97. The 2023 SOTC, ten months earlier in her first term, contains a +0.85 segment on “Los Angeles is open for business.” When Karen Bass gets to the podium, the model hears a mayor in full possession of her job.

The trouble for Karen Bass is that three and a half years in office have included rooms that are not the podium. The signature program of her first term is Inside Safe, signed ten days into her mayoralty under Executive Directive 2. By the end of 2025, per Governing’s reporting, Inside Safe had moved about 5,800 people into interim housing — mostly hotels and motels — at a total cost of more than $300 million. About 40 percent of the people moved — roughly 2,300 of the 5,800 — were back on the street by the end of year three. About one in four had reached permanent housing. Across the small subset of segments where Bass names the program out loud, her 2023 readings — when the model heard a mayor pitching a new directive — average +0.72; her 2024 readings average +0.17. Small samples in both years, but the direction isn’t ambiguous. The mayor at the 2026 State of the City podium telling the city it’s open for the World Cup is the same speaker who has cooled on her own homelessness program by half a polarity unit over thirty months.

The fire makes the cooling specific. Per a CBS News timeline, Mayor Bass was in Ghana on January 7th, 2025, attending the presidential inauguration of John Mahama as part of the U.S. delegation; the inauguration ceremony ended around mid-morning local time and the Palisades fire broke out approximately ninety minutes later. She traveled back to Los Angeles overnight and arrived around 11 a.m. Pacific on January 8th, by which time about 1,000 structures had burned. Five weeks before the fire, on December 4th, 2024, LAFD Chief Kristin Crowley had sent the Fire Commissioners a memo stating that a 7-million-dollar overtime cut had “severely limited” the department’s ability to respond to large-scale emergencies. On February 21st, 2025, Bass fired her. The press conference where she did it — the third clip in the reel above — is the one room in her entire corpus where the polarity range spans from −0.82 to +0.85; the mean over the full 62-minute press conference is +0.015 — net neutral, the room she had to walk into and answer questions in for an hour. The most direct version of her contrition during this stretch came three days earlier, on FOX 11’s The Issue Is on February 18, when she told host Elex Michaelson she “felt absolutely terrible not being here for my city,” a phrase widely re-reported in the days around the firing. The full LAFD after-action review did not come out until October 8th of that year, nine months after the firing, limited in scope to the first 36 hours of the response. The analysis can’t answer the question voters are weighing on June 2nd. It can tell you what a mayor sounds like when she has to discuss it.

A piece of context this section doesn’t score. NBC Los Angeles and FOX 11 have reported on text-message exchanges from Bass’s flight back from Ghana and her first-week management of the fire. NBC published characterizations of those messages; FOX reported that some had been deleted. The District has not independently reviewed them. We cite both stories because both are on the public record, and we mention them at all because they shape the political weather around the candidate. The analysis scores what she said on a microphone. It does not score what she texted on a phone.

The other reason Bass’s polarity goes durably hostile in the ICE-response clip above — and stays there across most of her summer-2025 corpus — is the summer the city had. Federal immigration operations swept through MacArthur Park, Boyle Heights, Pico-Union, and Vermont Square in June, July, and August of 2025. Trump deployed the National Guard and Marines to back the operations; Bass and Governor Gavin Newsom filed parallel federal lawsuits against the operations and gave overlapping pressers. Bass imposed a downtown curfew. Whatever else 2025 was for her, it was the year a Democratic mayor of the second-most-Latino major city in the country spent her summer in open conflict with the federal executive — and the analysis hears her against that specific opponent in a register that doesn’t show up anywhere else in forty-four months of recordings. The Aug 14 segment in the reel above is what that register sounds like.

The data also complicates the framing of Bass as a low-energy incumbent. Across her 430 clips where she is the one doing the talking, her average emotional intensity reads 0.58 on a 0-to-1 scale — the same range Raman registers in interviews, and well below Pratt’s 0.70. But her *peak* moments — the clips that score 0.85 and above — aren’t at debates or in interviews. They’re at the DNC podium praising Vice President Harris in August 2024, at the closing of the 2026 State of the City address (“I have never believed in us more”), and at the August 14th 2025 press response to the ICE operation outside the Newsom presser. Bass has heat. She spends it on civic-pride set-pieces and on federal-opposition moments. She does not appear to spend it on her primary opponents. Whether that is a strategic choice or a temperamental one is a question the analysis can’t answer.

Rick Caruso, who lost to Bass by nine points in 2022, has sat out 2026 publicly — but his post-fire nonprofit, Steadfast LA, has been a de-facto opposition government in the Palisades rebuild (permitting fights, debris-removal coordination, donor lobbying of Sacramento). Pratt’s candidacy is, in some real way, the shape that money and frustration have taken in the absence of another Caruso run. The analysis scores three voices on the June 2 ballot; the fourth voice driving the politics of the Palisades isn’t one of them.

Nithya Raman became a Los Angeles City councilmember in 2020 by beating an incumbent. She was thirty-nine at the time, an urban planner raised in Louisiana who had worked as an adult in Chennai on slum-housing reform — and the first South Asian ever elected to the council. The district she represents is Council District 4. After the 2021–2022 redistricting cycle it looks very different than the one she won in 2020 — still Hollywood, Los Feliz, and Silver Lake on the Eastside, but now also most of the East Valley: Sherman Oaks, Studio City, Encino, North Hollywood east of the 405. The constituency Raman built her 2020 insurgency on (Eastside renters) is now sharing a district with Valley homeowners. The moderating pattern her left flank keeps flagging is downstream of that change. She is a Democratic Socialists of America member. National political reporters started calling her the next Zohran Mamdani within weeks of her February announcement; she has spent the closing weeks of the primary drawing a sharper contrast with Pratt, framing him as a MAGA-aligned outsider in campaign appearances and on progressive podcasts.

What complicates the Mamdani comparison is that Raman has been governing for five years, and the record of those years contains a pattern her opponents have spent the campaign trying to exploit. In August 2023 the council approved an LAPD officer pay raise twelve to three; Raman was one of the three no votes, along with Eunisses Hernandez and Hugo Soto-Martinez, and her statement at the time said the contract would constrain the city’s ability to stand up unarmed public-safety alternatives. That vote is the left-flank credential the Mamdani comparison rests on. The January 2026 vote complicates it. That month she introduced a motion for a ballot measure that would carve out a fifteen-year exemption from Measure ULA — the documentary transfer tax on luxury real estate that her base spent the 2022 campaign passing — for newly built and substantially rehabilitated multifamily, commercial, and mixed-use development (The Real Deal). The motion was referred to committee and has not advanced. A DSA councilmember proposing a fifteen-year tax exemption for new construction is how her mayoral campaign began. Her base reacted poorly and the proposal stayed in committee.

Her own homelessness platform, articulated in April 2026 in an LAist interview, is essentially a cost-arithmetic argument against Inside Safe. The Time-Limited Subsidy Program costs about $24,000 per household per year per LAist’s reporting; Inside Safe’s motel rooms cost about $85,000. Her plan redirects spending toward short-term rental vouchers, shared-housing modular units, and street-medicine outreach, and calls for the city to be prepared to “make significant changes including taking on direct contracting for some funds and potentially contracting directly with the county” rather than continuing to route homelessness spending through LAHSA. On May 11th she criticized a Bass budget proposal to allow second-home short-term rentals during the 2028 Olympics, telling a college audience — per USC Annenberg’s coverage (which headlined the moment “up for sale to big tech”) — “In other cities, they are taxing people who own second homes. In LA we are cutting deals from the corporations that profit from them.” The campaign is a candidate running to the incumbent’s left and to the outsider’s left at the same time.

A bigger sample of Raman’s own speech also changed what the model heard. An earlier version of this analysis, drawn from just 26 of her clips, scored her as 85% neutral and 0% for-something. On the bigger sample of 121 clips, those numbers are 12% and 71%. That is what a larger sample does to a noisy estimate, not a change in the underlying candidate. On words alone, her average reads +0.02 — essentially at zero, with a range that touches both sides. On the model that watches and listens, her average warmth reads 0.58 — positive, a few hundredths below Bass at 0.65. Her for-something rate of 71% sits inside the range of Bass’s 70%.

Raman and Bass aren’t arguing in different directions — they’re arguing in the same direction at different temperatures. Bass speaks from the podium with the warm cadence of an incumbent at the State of the City; Raman speaks across a table from a host with the cooler register of an urban planner explaining how she would do the job. What separates them is warmth, set-piece access, and subject matter — not which side of an argument they’re on. The campaign rhetoric of Raman’s last twelve weeks — the ULA-carveout proposal, the MAGA-aligned framing of Pratt on progressive podcasts, the May 11th critique of the short-term-rental proposal — sits on top of that baseline as oppositional content, not as a replacement for her resting voice. Whether the layer reads as authentic or constructed to an LA voter is the political question. The analysis doesn’t adjudicate.

The analysis scores expressed signal, not intent. A negatively-framed sentence is the same input to the model whether the speaker is grieving, indicting, performing, or simply tired. The model resolves register, not motive. It doesn’t tell us whether a candidate is sincere. It tells us what they sound like to a measurement that treats them all alike. The article is deliberately quiet about sincerity for that reason.

What it also doesn’t score is whether a policy is correct, a program is working, or a candidate would be a good mayor. Where this article cites a number from LAHSA, a memo from Crowley, a donor list from Newsweek, or a vote count from the council, the number is documentary evidence — independent of the linguistic reading and not part of it. We put them side by side because both are on the public record.

The analysis has limits we owe the reader. Sample asymmetry: Bass has 846 segments of eight seconds or longer in the scored corpus; Pratt has 541 and Raman has 487. After voice-gating, those counts drop to 473, 369, and 177 respectively, because each candidate’s interview corpus includes large stretches of anchor-dominated audio. Confidence intervals widen with smaller voice-gated samples; Raman’s interval is the only one that contains zero. Diarization: the transcription system tags every transcript by speaker label, and each segment used for inference passes the voice-gating audit (dominant_subject_fraction ≥ 0.60). The full per-segment audit file is available at _audit_la_corpus.json in the replication repo. Multimodal scoring: the joint multimodal model finished scoring all three candidates as of the v6 run on 2026-05-15 — 430 voice-gated segments for Bass, 194 for Pratt, 121 for Raman. About 5% of model outputs are rejected by an automatic citation validator that checks whether quoted evidence is actually in the transcript. Polling: Tavern Research's May 1–4 survey is one private-firm poll and the mid-March Berkeley IGS poll is one institutional survey; the NBC4 viewer poll on debate night is a self-selected audience. Neither poll, nor the viewer poll, is the basis for a forecast in this piece.

The methodology and pre-registered analysis plan are locked in bass.yaml (available on request via district@myhamlet.com) and the pre-registration was completed before any scored data was opened. All meetings and transcripts behind this analysis are searchable on Hamlet.

What the analysis shows, more than anything else, is that these three candidates are not competing within the same register. They are arguing different things about what kind of city Los Angeles is right now. Karen Bass at the podium argues a city that is working — World Cup coming, Olympics next, open for business, the existing programs running. Spencer Pratt inside every room he can find argues a city that has been broken by the people currently running it, and that needs a fixer who breaks it harder. Nithya Raman, when she is not fighting a Mamdani comparison or a left-flank backlash on a tax carveout, argues a city that needs to be reformed by someone who reads the budget line items.

The election on June 2nd is whether a plurality of Angelenos believes one of those three accounts more than the other two. Public polling in mid-May has the field within a single-digit band. The Skirball audience cheered loudest for the one selling a broken city. The analysis heard that same voice as the highest-arousal, lowest-valence on the stage. Whether that matches what Angelenos actually want from a mayor is what a quarter-million primary voters will decide.

Los Angeles has been, this year, a city that lost a neighborhood to fire, watched federal immigration enforcement move through its streets, filed parallel federal lawsuits with Governor Gavin Newsom against those operations, and is still being asked to host two of the biggest sporting events on earth in two and three years respectively. Three weeks from a primary, the candidates the city is choosing between have produced fifty-eight hours of public speech for the record. The analysis has placed each of them. The choice from here is the city’s.