50 Months of Karen Bass

On a Friday afternoon in late February 2025, six weeks after the Palisades fire took twelve people and six thousand structures, Karen Bass stood at a Los Angeles podium for sixty-one minutes and fired her own fire chief. The press conference opened at +0.76 polarity on the linguistic-stance instrument that runs over every transcript in this analysis. "Acting in the best interest of Los Angeles public safety," she said. "I just met with Chief Crowley." Twenty-five minutes in, the polarity reading on the same speaker collapsed to −0.82. "I felt absolutely terrible," she said, "not being here for my city."

Two readings — opening, admission — in the same room, on the same day, in the same voice. They are the kind of numbers this article is about. 50 Months of Karen Bass is a pre-registered linguistic analysis of forty-three first-person videos covering April 2022 through May 2026 — Bass's 2022 campaign, her first term as Mayor of Los Angeles, her response to the Palisades fire, her June 2025 confrontation with federal immigration enforcement, and the opening of her 2026 re-election. Twenty-one and a half hours of audio. Eight hundred and seven scored segments. One hundred and fifty thousand transcript words. Every hypothesis locked in writing to an OSF-style pre-registration before any of the data was opened.

What the data show is not a story about whether Bass is a good mayor or a bad one. The rubric makes no such determination, and this article will not either. What it shows is that, scored on the same instrument across fifty months, the words she chose to put into a microphone moved in measurable patterns. Her enthusiasm for her own flagship homelessness program drained from +0.331 in 2023 to +0.012 in 2025. Her language on the 2028 Olympics, by the same measure, ran +0.679 in the same year. Her language on ICE was the only language in the entire corpus that went negative.

December 2022. The Launch.

Karen Bass took office on December 11, 2022 — the forty-third person to be Mayor of Los Angeles. Ten days later, on December 21, she signed Executive Directive 1, called it Inside Safe, and held a press conference. The phase-mean linguistic polarity on Inside Safe segments across her honeymoon year was +0.331 (n=22) — the highest sustained reading on that topic in her entire term.

Inside Safe is the right anchor for this analysis because it is the policy she campaigned on and the policy she made the marker of her first term. The on-camera language at the launch was specific: executive directive, motel rooms, encampment resolution, permanent housing. The transcripts read forward-leaning. The rubric registered forward-leaning. Both things were true.

Honeymoon year metrics on Bass's other anchor topic — LAPD funding — also ran high: +0.456 (n=9). The 2023 State of the City address landed in April 2023. Per-segment polarity across that 56-minute speech averaged +0.445 over 104 scored segments. By every measure on this instrument, her first year as Mayor of Los Angeles registered at the warm end of the scale.

January 2025. The Crisis.

The Palisades fire ignited on January 7, 2025, while Bass was in Ghana. The fire would burn for more than three weeks, kill twelve people, and destroy more than six thousand structures across the western edge of the city. Bass's arrival back in Los Angeles became the moment her first-term narrative split. NBC News captured her at LAX on January 9 declining to answer fire-related questions on camera (NBC News). NBC Los Angeles published text-message-based reporting on her remote management from the return flight (NBC LA); FOX 11 separately reported that texts from the period had been deleted (FOX 11). The District has not independently reviewed the messages; we cite the reporting because it is on the public record. The characterizations belong to the outlets that did the work.

What the linguistic instrument does is more boring and more specific. The full Palisades-response cell across the crisis phase (January 7 — April 30, 2025) is +0.339 (n=35). The recovery phase that followed registered higher, +0.493 (n=31). Both readings sit firmly on the positive side of zero. When Bass speaks about Palisades — about recovery, about rebuilding, about families coming home — the rubric reads the words as positive. That is a finding worth surfacing in plain terms: this analysis does not show Bass speaking bitterly about the fire. It shows her speaking optimistically about it.

On February 19, 2025, Bass told ABC7 the Ghana trip had been "a mistake" (ABC7). Two days later, on February 21, she fired LAFD Chief Kristin Crowley in the press conference that opens this article. The full Crowley-firing transcript runs sixty-one minutes; thirty-five scored segments. Two of those segments are the readings the lede quoted: the +0.76 opening and the −0.82 admission. The mid-firing polarity averaged to +0.008 — almost exactly neutral. The instrument registered the act as procedural even as individual phrases moved up and down inside it.

2025. The Fade.

By the summer of 2025, Bass was still discussing Inside Safe in press conferences and recovery briefings. The transcripts continued to register the program. The phase-mean polarity on those segments, though, had dropped to +0.012 (n=8) across the recovery-governing window — statistical near-zero. The same instrument that scored her 2023 Inside Safe segments at +0.331 scored her 2025 Inside Safe segments at one-twenty-seventh that value.

This is the curve in the hero. The Los Angeles Homeless Services Authority publishes its own data dashboard on the program's outcomes; that data is what it is. What the linguistic instrument measures is the language — how Bass chose to talk about her own program as the years went by. The instrument shows a slow drain of the launch-era register: enthusiasm and forward language replaced by procedure and recovery-frame language. The thing on the same instrument that her language did not drain on was LAPD: +0.456 (n=9) in 2023 → +0.440 (n=30) in 2025. One topic stayed inside a tenth of itself. The other lost a third of an absolute polarity unit.

February 2026. The Pivot.

Bass delivered the 2026 State of the City address on February 3, 2026, ten weeks earlier than the traditional April slot. The 75-minute speech scored 112 segments. Inside it, two cells diverged sharply on the instrument. The Olympics-readiness segments registered at mean polarity +0.679 (n=10). The Palisades-response segments scored within the same window registered at +0.493 (n=31). The pre-registered effect-size statistic for the comparison — Cliff's δ — came out at 0.397, exceeding the locked 0.30 threshold. Pre-registered Hypothesis 4 passes.

What does this finding say? It says that within the same recovery window, the same person, scored on the same instrument, used measurably warmer language to describe a thing she is building (the 2028 Games) than a thing she is managing (the Palisades recovery). The hypothesis is not a claim about her private feelings about either topic. It is a measurement of the words she put into the air. The pre-registration committed in advance to publishing whatever Cliff's δ the comparison returned, in either direction. The comparison returned +0.397, with Olympics on top.

June – August 2025. The Confrontation.

On June 8, 2025, federal immigration enforcement operations escalated visibly in Los Angeles. By June 11, Bass was holding press conferences alongside Governor Gavin Newsom. By June 21, she was on camera responding to Vice President JD Vance's framing of sanctuary-city policy. By August 14, she was outside a Newsom press conference reacting on camera to a federal operation that had arrived while the Governor was speaking nearby. The full P5 ice_sanctuary cell registered at mean polarity −0.424 (n=11) — the only negatively-valenced topic in the entire 807-segment corpus. Every other topic Bass discussed across fifty months — Inside Safe, LAPD, Palisades, housing, Olympics, the budget — registered positive on average. ICE did not.

The single most negatively-valenced Bass segment in the corpus comes from the August 14 reaction outside the Newsom press conference. The instrument scored her opening sixty seconds at −0.85. "I heard Bovino say that he was here to make Los Angeles safe," she said. "This is not making Los Angeles safe. And to disrespect one of our iconic …" Two minutes later, on the same camera, the instrument scored her at −0.92: "There is no way in the world that I believe that, and I'm frankly disappointed in Bovino." (Gregory Bovino is a Border Patrol commander assigned to Los Angeles operations.)

The Unmoved Message.

One topic in the corpus did almost nothing across fifty months. LAPD-funding segments registered at +0.456 (n=9) in the honeymoon-governing window and +0.440 (n=30) in the recovery-governing window — a difference of 1.6 percent of a polarity unit across nearly three years. The pre-registered Hypothesis 3 was a Mamdani-precedent null hypothesis: that LAPD funding language would survive campaign-to-governing without statistically significant adjacent-phase shifts. In the Mamdani piece, the equivalent test on the police_funding topic returned d = −0.066, p = 0.88 — consistency confirmed. The Bass corpus did not have cell density across every adjacent phase to run the same five-pair test cleanly, but the two cells that did clear the gate registered within a tenth of each other.

Across a fire that took twelve lives, a federal immigration confrontation, the firing of a fire chief, a three-year drain on the language of her own flagship homelessness policy, and the opening of a re-election campaign, the one thing that stayed steady in the transcripts was how she talked about LAPD funding. The data don't tell us why. They tell us it did.

May 2026. The Debate.

On May 6, 2026, at the Skirball Cultural Center, Bass debated Spencer Pratt and Councilmember Nithya Raman in the NBC4 LA Mayoral Debate. The full 58-minute broadcast scored 41 segments. Bass's in-debate polarity averaged near zero. The most negative Bass segment in the debate came on the public-safety question, scoring −0.66. Pratt, at one point in the broadcast, called Bass "an incredible liar" on the fire response. He is a candidate and the line is campaign-attack speech; we record it once, with the date and venue, and move on. The rubric makes no deception determination. Bass and Raman both withdrew from the May 11 follow-up debate before this analysis was finalized.

The P6 phase — May 6 through May 12, 2026 — is a seven-day window by design. It is also the thinnest cell in the corpus: only the debate itself sits inside it, plus a handful of clip releases from the campaign. The data are not dense enough to support most cross-phase tests against P6. The pre-registered Hypothesis 1 — a test for tonal inversion on Palisades-response language between the P4 crisis window and P6 — was gated out for insufficient n on the P6 side. The May 6 debate registered 41 segments, but none of them matched the Palisades-response keyword set; Bass's framing in the debate moved past the fire itself into broader fire-readiness language, which the linguistic regex did not catch. The pre- registration commits to publishing that result honestly. We do.

What This Is and Is Not.

The numbers in this article are linguistic-polarity readings on transcript text, produced by a pinned RoBERTa sentiment model (cardiffnlp/twitter-roberta-base-sentiment-latest) running over AssemblyAI Best transcripts. They are not vocal-prosody readings. They are not facial-action readings. They are not a description of what Karen Bass believes about Inside Safe, the Palisades fire, ICE, LAPD funding, or the 2028 Olympics. The pre-registration is explicit about this; the analyst-prior disclosure attached to it is more explicit still: the byline holder, Sunil Rajaraman, holds an adverse view of the subject. The pre-registered methodology — the locked phase boundaries, the locked topic keywords, the locked n ≥ 8 cell-count gate, the commitment to publish whatever the data return — is the operational protection against that prior bending the analysis. The findings ordering above is part of the same protection. Inside Safe decay, LAPD consistency, Olympics divergence, ICE confrontation: each beat got the same treatment regardless of whether it ran with or against the prior.

What the data do show, repeated across forty-three videos: a governing register that runs warm on programs Bass is launching or building, cools on programs she has been managing for several years, holds steady on the public-safety budget across a period of real political pressure, and turns negative only on federal immigration enforcement. Five hypotheses were pre-registered. The one that tested cleanly (Hypothesis 4 — Olympics > Palisades within recovery, Cliff's δ = 0.397) tested positive. Three more (H1 Palisades tonal inversion, H3 LAPD consistency, H5 budget-gap externalization flip) ran into thin cells. One (H2 cross-modal disagreement on Inside Safe) is structurally untestable until the full multimodal Tier-1 build runs the vocal and joint lanes that this Tier-2 v2 deferred. The Tier-1 follow-up is on the queue.

Three weeks from this writing, Angelenos vote in the June 2 primary for the next Mayor of Los Angeles. Karen Bass is on the ballot. What this article will not do is tell you how to vote. What it tries to do is make legible — measurably, reproducibly — the patterns in fifty months of public language by the incumbent.