The loop that should have run
A curriculum is a bet placed years before it pays out or fails. A programme director sets a syllabus in year one; the first graduates carrying it into the labour market appear three or four years later; the market's verdict on those graduates arrives later still, mediated by employer surveys, alumni salaries and the noisy gossip of LinkedIn endorsements. Between the bet and the verdict sits a gap wide enough for the world to have moved twice. The question this page asks is not whether curricula go stale — everyone who has sat on a curriculum committee knows they do — but what kind of intake would have let someone notice while there was still time to act, rather than two cohorts after the fact.
What arrives
Four streams run continuously through any functioning programme, whether or not anyone is watching all of them.
Assessment streams: coursework marks, rubric scores, exam pass rates, broken down by module and cohort. These arrive on a fixed academic clock — end of term, end of year — and they measure whether students can do what the syllabus says they should be able to do. They say nothing about whether that thing is still worth doing.
Engagement telemetry: attendance, time-on-task in the learning platform, which optional modules fill up and which run at half capacity, which recorded lectures get rewatched near assessment deadlines. This stream is granular and near-continuous, logged in minutes, and it is the one most programmes actually have in bulk, because the platform vendor built the logging in for reasons unrelated to curriculum design.
Curriculum changes: the committee minutes, the module specification diffs, the date a topic was added or dropped, who proposed it and what the stated justification was. This stream is sparse — a handful of events a year — and is usually held as static documents rather than as time-series data, which matters later.
Labour-market signals: job postings mentioning the programme's named skills, employer survey responses, graduate destination data, salary bands by sector. This is the slowest stream, arrives with the longest lag, and is the one that actually adjudicates whether the bet paid off.
Each stream alone is legible. Together, unaligned in time, they are four separate stories that happen to share a subject.
What is held
The difference between a system that can explain a lock-in and one that can only describe it is what gets kept from these streams and how it is timestamped. Held with provenance means: this module specification changed on this date, following this employer complaint, referencing this labour-market report, superseding this earlier version which itself referenced a different report. Held without provenance means: the current syllabus, full stop, with the reasoning behind past versions compressed into an institutional memory that lives in the heads of committee members who eventually leave.
Most programmes hold the second thing. The learning platform retains engagement telemetry at fine granularity because storage is cheap and nobody has to decide to delete it. The curriculum's history is retained badly, because curriculum documents get overwritten rather than versioned, and because the labour-market signal that justified a given change is rarely filed against the change itself — it lives in a slide deck from a meeting three years ago, if it survives at all.
This is the education-sector version of a general pattern. A corpus frozen at some cutoff — an accreditation review, an external audit — captures the current module list and the current pass rates in detail, and captures almost nothing about which employer conversation, which retiring lecturer's pet topic, or which vendor's free training license caused a given module to enter the syllabus in the first place. The event that mattered was small, local, and unrecorded at the time, exactly as the gauge of a colliery tramway was unrecorded when it later became a national rail standard.
What triggers revision
In a well-run programme, revision is meant to be triggered by discrepancy: labour-market signal diverges from assessment output, and someone notices. In practice the trigger is almost always calendrical rather than evidential — the five-year accreditation cycle, the annual programme review — because that is the only clock anyone has built infrastructure around. A five-year gap against a labour market that turns over named tools and named skills inside eighteen months to two years is not a design flaw exactly; it is what happens when the only available intake is periodic sampling of a continuous process.
The failure mode this produces has a specific shape. A data-analytics diploma built its statistics module around one dominant proprietary package because that was the departmental licence in place when the module was written. Assessment scores stay healthy for years — students can do the exercises, the exercises test the package syntax, the marks look fine. Engagement telemetry looks fine too, or looks like normal attrition. The labour-market signal — job postings shifting toward a different, open toolchain — starts moving two cohorts before anyone with the authority to change the module sees a number attached to it, because that signal sits in a different stream, on a different clock, owned by a different office (careers services, not academic affairs), and nothing routes it to the committee automatically. By the time the discrepancy surfaces at the five-year review, the programme has certified two full cohorts in skills the market stopped paying for, and the honest question — why did we keep teaching this — has no answer better than "nobody was watching that stream against this one, continuously, at the point where they diverged."
What the operator sees
The programme director's actual dashboard, where one exists, shows the fixed-clock outputs: this term's pass rates, this year's satisfaction scores, last year's graduate outcomes survey. It rarely shows drift — the first-derivative signal of a labour-market keyword losing frequency, or an employer survey free-text field mentioning a tool that isn't taught, three appearances this term against none last term. Drift is exactly the kind of thing multiple weak streams, aligned in time and kept with provenance, would surface early and any single periodic snapshot will not. The director who eventually explains the failure to an accreditation panel is not being asked to justify a decision made with bad judgement; they are being asked to justify a decision made with the information that happened to have been kept, which was mostly assessment data, because that is what the institution's systems were built to retain.
The objections
Path dependence is overclaimed in curriculum design too. Programmes don't get locked into obsolete skills through some mysterious increasing-returns mechanism; they get locked in because changing a module is administratively expensive — new specification, new exam board sign-off, staff retraining — and that cost is visible and ordinary, not some deep contingency worth a formal theory.
This lands partly. Much of curricular inertia is straightforward bureaucratic friction, not the amplifying dynamic Brian Arthur described for standards competition. But the friction argument does not remove the sequencing problem; it explains why switching is expensive without explaining why this particular module, rather than some equally defensible alternative, became the one worth defending. That answer is historical — which vendor gave a free licence in 2011, which lecturer championed the topic, which employer sat on the advisory board that year — and the historical answer is exactly what does not survive in assessment or engagement data. The weaker claim, that outcome depends on sequence and sequence is knowable only from a contemporaneous record, holds even where the mechanism is mundane administrative cost rather than exotic increasing returns.
Much of what actually decided a module's survival is a private conversation — a department head's preference, an informal deal at a conference, a committee member's loyalty to a topic they built their career on. No engagement log or labour-market feed captures that. Calling for continuous intake overstates what any instrumented system can reach.
Also correct, and worth conceding without qualification. Continuous intake means every stream that is running through the institution's actual channels — assessment systems, the platform's telemetry, minuted committee decisions, job-market feeds — not every event, including the ones that happen over coffee. That instrumented fraction is smaller than "everything." It is not fixed, though: minuted rationale, timestamped module-change logs and structured labour-market feeds are all things an institution can choose to build, and most currently do not. A programme director working from four aligned, provenance-tagged streams still misses the private conversation. They miss strictly less than a director working from a five-year review built on assessment data alone.
The terminal rung
Nothing above proposes a system that watches every meeting and every hallway aside. It proposes that the four streams already running — assessment, engagement, curriculum change, labour market — be held continuously, timestamped and cross-referenced rather than sampled periodically and reconciled retroactively at accreditation time. That is not a new kind of evidence beyond what a well-instrumented institution could in principle already collect. It is the difference between an audit trail and an excuse, produced after the fact, for why the syllabus certified something the market had already stopped buying.