Four clocks, one rollout.
The Four Clocks diagnostic is a framework primitive (one of the basic instruments this project applies everywhere). The short form is that every decision in a complex system runs on four clocks at once, and the fastest clock in the room sets the pace for the others unless somebody puts a brake on it. The full background lives at /research/four-clocks. This page applies the diagnostic to the SFPS AI rollout.
The diagnosis: the machine clock is forcing the institutional clock, the institutional clock is overrunning the human clock, and the ecological clock, the children, is the one taking the damage. Re- examination is the clutch the rollout never had.
§01The four clocks at SFPS
Each card names the clock, what it runs on at SFPS, what its natural tempo is, where it is being overrun, and the damage signature you should expect to see if the overrun continues. Tap a card to expand.
Machine clockseconds · minutesmodel inference, vendor release
Inside the tool, a model answers in seconds. Around the tool, OpenAI, Khan Academy, and Amira Learning ship new models, new features, and new pricing tiers on quarterly or faster cadences. A model that was the right choice in fall 2025 may be deprecated, retrained, or sold to a new owner by spring 2026.
Compute time is bounded by latency. Vendor release time is bounded by competitive pressure. Both are optimized for the company's revenue clock, not for the cognitive development of a particular cohort of children.
Vendor release cycles are forcing district procurement timelines that are not built for that speed. Contracts written this year may govern features that did not exist when the contract was signed.
Lock-in. Contracts that point at a moving target. Teacher training that becomes obsolete before the school year is over. The procurement function quietly cedes its review window.
Human clockweeks · monthshabit, skill, behavior change
Habits form over weeks. A teaching practice takes a semester to bed in. A student's relationship to a tool, what they reach for first, what they trust without checking, what they no longer try to do unaided, settles over months. Parent perception of the rollout adjusts on the same scale.
Human behavior change is slower than software change and faster than institutional change. It is the layer where the rollout actually becomes lived. Skipping the months it takes for a practice to bed in produces brittleness that does not show up in the first quarter.
The rollout treats teacher attention and student attention as elastic. The number of AI tools, dashboards, and verification responsibilities grows on the machine clock; the human capacity to integrate them does not.
Teacher burnout. Compliance behavior in place of considered review. Students using the tool without the framing that would make the tool actually instructional.
Institutional clockmonths · yearsboards, contracts, curriculum
Board meetings convene monthly. A records request returns in fifteen business days. Curriculum review cycles run a semester or a year. Vendor procurement under a normal cadence takes one to three months. A change in district direction, fully through the system, takes one to three years. The Sep 26 2024 study-session deck named a reporting cadence (Quarterly BOE AI Updates in Q2, Q3, and Q4 of SY2024-25) published to the board itself. The Q2, Q3, and Q4 updates do not appear on the board agenda record. Published promise versus published delivery is a pattern visible on the institutional clock when somebody bothers to check. And the clock mismatch cuts the measuring stick itself: a vendor merger forced the statewide switch from Istation to Amira mid-goal-cycle, and Albuquerque Public Schools' own accountability office says it needs at least one full school year to re-establish the baselines for its five-year literacy goal. A corporate transaction on the machine clock reset the institution's instrument.
Deliberation has a tempo. The community-input function of public schooling exists because slow deliberation is what makes decisions reversible by the community that has to live with them.
Machine-clock cadence is overrunning institutional review. Procurement is being run on a peer-network deliberation surface (the recurring Friday call) that is faster than the board cycle and not subject to the public-meeting record.
The community's review function quietly stops being load-bearing. Decisions that should be reversible accumulate as fait-accompli. The board hears about decisions after they are operationally locked in.
Ecological clockdecades · generationsthe cohort, the next cohort
A child in 7th grade today graduates in 2031 and is 28 in 2041. By then she has lived two-thirds of her life with an AI in the loop of her own thinking. The cognitive scaffolding she is building right now compounds for forty years. The cohort she raises inherits the practice she normalized.
Cognitive development runs on cohort time. Cultural transmission runs on generation time. The damage signature shows up years after the choice that produced it, and the people who made the choice are often no longer in the building.
The clock that should set the pace for the others is being treated as if it were the longest waiting room. The choices being made on the machine clock are landing on a clock that cannot send a complaint for ten years. The early complaint is already on the record: between 2023 and 2025, parent support for AI use in lesson planning dropped thirteen points in PDK's longitudinal poll. Familiarity is moving in one direction.
A generation that has outsourced parts of its own thinking before it built the muscle to know what was outsourced. The damage is not legible until the cohort is voting, working, and raising children of its own. Parent-side responses are no longer petitions at the margin — they are on the public record at scale. In New York City, a majority of the City Council backed a two-year moratorium on generative AI in the nation's largest school system, and the education department paused its AI guidance after nearly 6,500 public comments. In New Mexico, Santa Fe parents organized the independent Amira Opt-Out campaign. The campaign neither endorses nor opposes those vehicles; the existence of organized parent demand is the data point.
§02Where Brookings 2026 lands on the four clocks
The Brookings 2026 Task Force on AI in K-12 named four risk categories. Each lands on a different clock, which is why a single-clock review cannot see all four at once.
- Pedagogical drift (Human clock). Teaching practice and student habit reshape over months; the drift is visible in the semester after it starts.
- Equity gaps (Institutional clock). Procurement, broadband floor, and assessment cycles are where gaps either close or compound.
- Data exposure (Machine clock). Vendor model retraining and DPA "improvement" clauses run on the vendor's release cycle, not the district's.
- Governance capture (Institutional clock). Imported frameworks, vendor-funded reviewers, and peer-network deliberation surfaces all do their work here.
NEPC's peer review of the Brookings task force flagged overreliance on industry data, light treatment of labor, and weak treatment of disability. The four-category map is useful even with those caveats noted. The New Mexico Legislature's own May 2026 policy brief cites the same Brookings work — the state's analyst and this page are reading the same evidence.
§03Why the ecological clock matters most
A vendor whose product fails recovers in a quarter. A district whose procurement was rushed recovers in a board cycle or two. A teacher whose workload was misjudged recovers over a summer. A cohort whose cognitive scaffolding was reshaped by tools the community did not have the chance to evaluate carries that reshaping into how it votes, works, parents, and reasons for the rest of its life.
The ecological clock is the only one whose damage compounds. That is the clock that has to set the pace for the others, not the other way around.
§04The clutch the rollout never had
A clutch is what lets a fast engine drive a slower wheel without tearing the drivetrain apart. The institutional clock is the community's clutch. When the institutional clock keeps up with the machine clock by skipping deliberation, the community has not built a faster clutch. It has removed the clutch.
Six concrete clutch mechanisms are named on the Proposal page: a procurement pause during the leadership transition, an independent equity audit, real tribal consultation, a parent advisory council with real say (not advisory), public posting of all vendor contracts, and support for negotiating AI protections into the next educator collective-bargaining agreement. Each is a way to put the brake back on without stopping the engine.