Sources.

Baseline recording of how the questions shifted between 2018 and 2024. The vocabulary of the human equation, of restraint ("just because we can does not mean we should"), is audibly present here and audibly missing from the 2024 deck.

linkhttps://www.youtube.com/watch?v=K37mMa-qOy4

The single source document for every public communication that followed. Slide 3 carries the ChatGPT-written definition of AI. The deck names the vendors, imports the TeachAI Seven Principles, and promises reporting cadences the district has not since published.

acquisition19-slide board study-session deck. Local archive; request via IPRA if not posted on the SFPS site.

First public framing of the rollout. The phrase 'thought partner' enters circulation here. The article carries the vendor names and the early governance scaffold.

acquisitionCanonical URL not captured in codebase; search the Santa Fe New Mexican archive for 'SFPS' and 'thought partner' in October 2024.

District-affiliated podcast episode that documents the framing and the room composition of the recurring deliberation call. Vendor CTOs and peer-district CIOs are present. Parents, teachers, and tribal-government representatives are not.

linkhttps://www.youtube.com/watch?v=45sfxHRAlpc

Marker of the SFPS leadership transition. The Lawton departure opens the procedural re-examination window the campaign is built around.

linkhttps://www.lawtonps.org/article/2630731

The independent diagnostic anchor. A reporter outside the district describes the rollout as policy-free experimentation, in the same vocabulary the campaign uses. The May 14, 2026 board meeting on the same day puts this article and the deliberation in the same news cycle.

linkhttps://www.santafenewmexican.com/news/education/pockets-of-experimentation-santa-fe-teachers-embrace-ai-without-clear-policies/article_2af23b13-2f91-4031-86f2-7921de6ddddf.html

The four-day jump from local to national press. Documents that the Salkin diagnostic was found relevant enough to syndicate at the GovTech national K-12 desk. The campaign reads this as a leverage event.

linkhttps://www.govtech.com/education/k-12/how-santa-fe-schools-are-adopting-ai-without-specific-policy

Documents the prior leadership transition the rollout was launched under, by the same reporter who would write the May 14, 2026 diagnostic. Sets the institutional context for the September 26, 2024 board deck and the rollout cadence that followed.

linkhttps://www.santafenewmexican.com/news/education/new-santa-fe-superintendent-hired-from-smaller-arizona-school-district/article_72617c32-c7b7-4ddf-884d-23ccf3b79f00.html

The deliberative-record corpus the campaign reads. The January 15, 2026 meeting carries the most direct AI roadmap presentation by Weaver, Grunwald, Romero, and Lewis. The other meetings document the agenda items, the executive sessions, the Weaver farewell, and what was deliberated, when, and in what setting.

linkhttps://www.youtube.com/watch?v=C32s-_zg6t8

acquisitionPrimary link is the Jan 15, 2026 meeting (AI roadmap presentation). Companion recordings: Dec 10 2025 (https://www.youtube.com/watch?v=04k-tNtMm7c), Jan 22 2026 executive session (https://www.youtube.com/watch?v=zuDg6wulMoc), Feb 2026 (https://www.youtube.com/watch?v=mktNFduzNNQ), Feb 19 2026 study session (https://www.youtube.com/watch?v=vTGzg9YzdL8), April 2026 covering March (https://www.youtube.com/watch?v=nyaRYGiNRo0), Apr 16 2026 finance study session (https://www.youtube.com/watch?v=HecOADndU1I), May 14 2026 (https://www.youtube.com/watch?v=RTs5B8Y4iPE), Weaver farewell and Quantum Education resolution postponement (https://www.youtube.com/watch?v=ULI1CnsN_ZU).

The district's own public self-description of the AI rollout: the timeline, the named training PD, the Chat for Schools deployment, the leadership chain. The reference point the campaign compares its findings against.

linkhttps://www.sfps.info/o/sfps/page/ai-projects-and-highlights

acquisitionCompanion: SFPS Superintendent's Office roster page (https://www.sfps.info/page/superintendents).

Two days of sworn-in-effect public record. Day 1: the contract's piggyback structure through Cooperative Educational Services, the RFP timeline toward a 2027-28 classroom start, the troubled-child alert, the voice-recording retention rationale, and the committee's own accent, bias, and privacy questions. Day 2: six public comments answering the presentation — including a Roswell district administrator naming a Secretary-level memo that requires Amira for progress monitoring. The campaign's timestamped digest, every quote verified against the audio, is the Hearing Record page.

linkhttps://sg001-harmony.sliq.net/00293/Harmony/en/PowerBrowser/PowerBrowserV3/20260715/-1/79911?startposition=20260715151000&viewMode=3

acquisitionThe deep link opens the day-1 archived player at the start of the Amira item (approximately 3:10 PM MDT; recording begins 1:36:03 PM). Day-2 afternoon session (public comment from approximately 4:20 PM; recording begins 1:05:09 PM): https://sg001-harmony.sliq.net/00293/Harmony/en/PowerBrowser/PowerBrowserV3/20260716/-1/79916?startposition=20260716162010&viewMode=3. Hearing agenda: https://www.nmlegis.gov/Agendas/ALESCageJuly15.26.pdf. Digest with per-quote timestamps: /advocacy/sfps/lesc-hearing.

The printed record handed to committee members: PED's 15-slide assessment-system deck and Amira's 17-slide vendor deck. Carries figures the audio could not — the vendor's own “Mandated Screener” classification for New Mexico, the ~574,000 progress-monitor count and 55.5 million tutoring minutes, grade 4-6 usage at scale, printed 2026 proficiency splits, the Louisiana study's prepared-for-Amira-by-Instructure cover, and PED's 113-response teacher survey with 8% reporting no major concerns.

link/docs/lesc-2026-07-15-amira-handout-packet.pdf

acquisitionPhotographed copy of the printed packet distributed at the hearing, archived on this site as a compressed PDF; page citations on the Hearing Record page refer to this file.

The countable public reaction: 578 comments in five days, 10 self-identified educators and 36 self-identified parents or family members, 88 comments raising privacy or consent, zero comments defending the deployment. Cited for aggregate volume and distribution only — anonymous comments are not treated as verified fact.

linkhttps://www.instagram.com/reel/Da5RMffxDGS/

acquisitionCounts computed from a July 22, 2026 export of the comment thread (ExportComments.com).

The Legislature hears the vendor: Amira's founder-CEO and PED's senior assessment manager, scheduled testimony. Coverage of the hearing carries the CEO's data-storage assurances without specifics and PED's confirmation that the state assessment contract opens for competitive bids in school year 2027-28.

linkhttps://www.nmlegis.gov/Agendas/ALESCageJuly15.26.pdf

The op-ed that made the parent campaign public: consent gap, accuracy concerns for students with accents and speech differences, and the opt-out instrument. Cited as a fact in the record; this campaign neither directs nor speaks for that effort.

linkhttps://www.santafenewmexican.com/opinion/my_view/get-unvetted-ai-out-of-new-mexico-schools/article_ddbb9ced-013f-40e5-bb19-c557eedd95c9.html

Independent parent resource: opt-out forms, Spanish-language materials, and the request-it-in-writing discipline. The existence of organized parent demand is the data point.

linkhttps://www.amiraoptoutsfps.com/

The strongest available evidence that hints-first, Socratic, teacher-prompted AI tutors can help learning where a general chatbot harms it. Validates the design pattern SFPS chose (Khanmigo's Socratic posture). The campaign rests on indefensible process, not on a claim that AI harms learning.

acquisitionPublished in the Proceedings of the National Academy of Sciences, 2025. Search PNAS for Bastani and 'GPT tutor.'

Reports the largest cleanly attributed effect from a major K-12 AI tutor trial: about four percentage points in the better-served arm, less than half a letter grade, and the effect attenuates where home setup is thin.

acquisitionMicrosoft Research publications. Search Microsoft Research and 'Tutor CoPilot.'

Finds zero statistically significant difference between an AI tutor and a general web search on the common end-of-unit assessment outcome. The strongest available counterpoint to vendor pilot claims.

acquisitionAggregated across districts; consortium publication. Operator to confirm canonical citation.

Documents the speed of the adoption curve: roughly ten percent of teachers reporting AI use in 2024, roughly seventy-six percent by 2026. Sets the pace the institutional clock has to keep up with.

acquisitionAggregated from multiple US K-12 teacher surveys; definitions of 'use' vary across waves.

Parent support for AI in classrooms dropped thirteen points between 2023 and 2025. The ecological clock signal the campaign uses to show that the public consent the rollout is assuming is not there.

acquisitionPDK International longitudinal polling on K-12 education. Operator to fill specific data release URL.

The district next door, in its own words: the merger-forced switch from Istation to Amira means APS needs "at least one full school year" to re-establish baselines for its five-year literacy goal. A vendor transaction reset the institution's measuring stick.

linkhttps://www.aps.edu/news/news-from-2025-2026/new-amira-assessment-complicates-aps-literacy-goal-tracking

APS staff report inconsistent scoring and the tool's difficulty deciphering the speech of children with impediments or accents — the accuracy-equity concern, on the record, from practitioners in the state's largest district.

linkhttps://www.abqjournal.com/news/students-read-aloud-ai-scores-them/3029121

Source of the four-measure evaluation scaffold the campaign uses on the proposal surface (Listening / Deciding / Doing / Adjusting). The international reference frame against which the SFPS rollout is being read.

linkhttps://doi.org/10.54675/KECK1261

acquisitionISBN 978-92-3-100784-2. Pages 108–127 carry the evaluation framework cited.

Provides the four-category risk frame (pedagogical drift, equity gaps, data exposure, governance capture) used on the Four Clocks page. The campaign treats it as a useful structural reference, not a final word.

acquisitionBrookings task force release. Operator to fill canonical URL.

Names the gaps in the Brookings framework: overreliance on industry data, the missing labor analysis, the missing disability analysis. The discipline check the campaign applies before importing the Brookings categories whole.

acquisitionNational Education Policy Center peer-review release. Operator to fill canonical URL.

Five minimum-standard principles for AI in classrooms, written by the largest US educator union. The comparison baseline used on the proposal sidebar against the SFPS Sep 26 2024 deck.

acquisitionNEA published policy statement. Operator to fill canonical URL.

The policy scaffold the SFPS Sep 26 2024 deck adopted. The campaign reads it skeptically because the authoring coalition is vendor-funded; reading it disclosed is part of the deliberation that did not happen.

acquisitionPublished by TeachAI. Operator to fill canonical URL.

Published assessment-scale framework that SFPS guidance materials cite by name. The campaign reads this carefully because the framework was adopted as the academic anchor for an AI rollout that did not also adopt the framework's deliberation requirements.

linkhttps://doi.org/10.53761/q3azde36

acquisitionPreprint at arXiv:2312.07086 (https://doi.org/10.48550/arXiv.2312.07086). Published in the Journal of University Teaching and Learning Practice, vol. 21, issue 06.

The state framework SFPS imported. The campaign reads its adoption as procedural offloading: deliberation that did not happen at the local level was claimed to have happened at the state level.

linkhttps://go.ncdpi.gov/AI_Guidelines

First major US K-12 collective-bargaining agreement with explicit AI provisions. Ratified by ninety-two percent of thirty-five thousand educators. Six named provisions, modeled on the proposal surface's sixth memo (SFAI-2026-06).

acquisitionUTLA / AFT contract ratification reporting, April 2026.

The state's own analyst, verbatim: New Mexico "does not currently have statewide procurement standards for AI tools, student data privacy requirements specific to AI, and a formal evaluation process for AI education tools" — and it is unclear what vetting PED used before mandating Amira. Also confirms monthly assessment for below-benchmark students and recommends formal tribal consultation before statewide AI tools.

linkhttps://www.nmlegis.gov/handouts/ALESC%20052726%20Item%205%20.1%20-%20Artificial%20Intelligence%20in%20K-12%20Education-State%20Trends%20and%20Implications%20for%20New%20Mexico.pdf

The Legislature's working-group report, one month before the statewide Amira mandate took effect: recommends minimum data privacy and security standards, a vetting process for AI tools, prohibitions on training from sensitive student data, tribal data sovereignty protections, and a clear opt-out procedure. SFPS's Chief Information and Strategy Officer presented Amira to this working group.

linkhttps://www.nmlegis.gov/handouts/ALESC%20072325%20Item%203%20.1%20%20-%20Report;%20LESC%20Artificial%20Intelligence%20Working%20Group.pdf

The mandate document: Amira ISIP required K-2 statewide (grade 3 optional), continued under the Istation contract after the merger. Carries the local-control sentence SFAI-2026-07 stands on, verbatim: "If a district or state charter does not wish for students' audio to be retained, local administrators can elect to remove the retention of the student's voice."

linkhttps://web.ped.nm.gov/wp-content/uploads/2025/09/AMIRA-ISIP-GUIDANCE-8.2025.pdf

Establishes the national baseline: more than fifteen million US K-12 students without reliable home broadband at the current FCC threshold. The number that sits underneath every claim that AI tools 'extend learning beyond the classroom.'

acquisitionFCC Form 477 filings and US Census American Community Survey datasets.

County floors that anchor the New Mexico-specific argument: Hidalgo 2.1%, Socorro 5.7%, Colfax 10.2%, Catron 14.6%. Without this data the broadband gap is abstract.

acquisitionNM Office of Broadband Access and Expansion release.

Documents the broadband gap on the largest tribal land base in New Mexico's region: between forty and one hundred percent of Navajo Nation households below ten megabit service depending on chapter geography.

acquisitionFCC tribal broadband releases plus Navajo Nation TRC reporting.

Cross-source aggregation that puts a high-end seventy percent undercount estimate on the count of Native K-12 students in district-level reporting. Names the equity gap as a measurement gap first.

acquisitionAggregated from NMPED releases, US Census ACS 5-year tribal estimates, and tribal-government enrollment data.

Revenue trajectory from fifty-three million dollars (FY2018) to one hundred twenty-eight million dollars (FY2023), with roughly seventy-seven percent of revenue from contributions. Establishes the foundation-dependent operating model that is the relevant context for any 'free' deployment.

acquisitionPublic IRS Form 990 filings. Search ProPublica Nonprofit Explorer or IRS Tax Exempt Organization Search for Khan Academy.

Schedule B contributor identity is statutorily redacted on the nonprofit side. Naming the named figures is done from the donor side: cross-referenced Gates, Stand Together (Koch network), Doerr-led Schmidt, and Yass-aligned Susquehanna disclosures.

acquisitionDonor-side public foundation databases.

Documents the trajectory from roughly two hundred million dollars (2023) to twenty billion dollars annualized (2026). K-12 segment is not separately disclosed in any public filing, which is itself a finding.

acquisitionAggregated from The Information, Reuters, and Bloomberg reporting.

Permits retention of de-identified learning-pattern telemetry for product improvement. Multi-year retention of interaction logs is typical. The standard DPA against which other vendors are compared.

acquisitionStandard Khanmigo DPA, public version. Available on the Khan Academy site or via IPRA request.

Permits use of de-identified content for 'maintenance, development, support, operation, or improvement' of the service. The 'improvement' word is the long-tail commitment the campaign keeps returning to.

acquisitionOpenAI Student DPA, public version. Available on the OpenAI site or via IPRA request.

Combined post-merger footprint of roughly eighteen hundred districts. The principal data asset of the combined company is the voice-recording corpus of K-2 children reading aloud.

acquisitionJune 2024 transaction announcement, Macquarie Capital as facilitator.

Strongest of the three vendor DPAs on substance: no external LLM inference on student voice, no model training on identified student data. Sets the comparison baseline. The post-merger caveat this entry carried from launch matured in July 2026: see the Portland contract dispute and the privacy-policy version comparison at /advocacy/sfps/amira-policy-diff.

acquisitionStandard Amira Learning DPA, public version.

Reports that two of the three vendors deployed at SFPS (Khan Academy and OpenAI) sit inside the funding network of the organization reviewing them. The independence-of-reviewers problem named explicitly in mainstream reporting.

acquisitionPOLITICO reporting, April 1, 2026. Operator to fill canonical URL.

The moving target, captured: the April 12 Wayback capture and the July 20 version (archived PDF hosted on this site), compared section by section with every change marked at /advocacy/sfps/amira-policy-diff.

linkhttps://web.archive.org/web/20260412172021/https://amiralearning.com/amira-privacy

acquisitionJuly 20, 2026 version archived at /docs/amira-privacy-policy-2026-07-20.pdf on this site; live document at amiralearning.com/amira-privacy.

Records-request-based reporting: Amira's $5,160 Portland pilot contract carried checked boxes for student disability status, migrant status, and living situation. The company states no such data was transmitted; why the boxes were checked in the contract went unanswered.

linkhttps://www.wweek.com/news/schools/2026/07/05/ai-company-tells-portland-school-board-it-does-not-collect-student-data/

Federal statute governing student-record privacy. The statute carries the per-incident civil penalty estimates and the loss-of-federal-funding remedy that make the OpenAI 'free teacher tier' question more than an editorial concern.

linkhttps://www.law.cornell.edu/uscode/text/20/1232g

State statute governing public-records access. § 14-2-8 sets the response window (three business days for acknowledgment; fifteen business days for production or written explanation of delay). § 14-2-12 carries the Attorney General complaint pathway.

acquisitionNew Mexico Compilation Commission code; see the New Mexico Foundation for Open Government IPRA guide for a plain-language walkthrough.

New Mexico becomes the first state to prevail at trial against a major technology company for harming young people: a $375 million jury verdict for violations of state consumer-protection law, with an injunctive-relief phase following. One of the two facts in the Scorecard's state-context panel.

linkhttps://nmdoj.gov/press-release/new-mexico-department-of-justice-wins-landmark-verdict-against-meta/

acquisitionVerdict coverage: CNBC, March 24, 2026.

National record on classroom freemium mechanics: the FTC complaint documents membership advertising outnumbering math problems in home play and premium benefits visible in class. A complaint, not an adjudication — cited here as the reason the IPRA asks whether Prodigy runs in SFPS, not as a finding about SFPS.

linkhttps://fairplayforkids.org/feb-19-2021-advocates-to-ftc-prodigy-math-game-preys-on-kids-and-families/