How the TIME Score Works

The TIME Score measures one thing: how much human time a system liberates or destroys. It is a number from 0 to 100, calculated from four components, grounded in a signal-noise framework adapted from information theory. Every score on this site traces back to the formulas below.

The Core Idea

Time is value. Not a metaphor — a measurement. When a tool saves you three hours of work, those three hours are the value. When a platform holds your attention for two hours without producing anything you can point to afterward, those two hours are the cost.

The TIME Standard makes this visible. For every system we score, we ask: how many hours of human time does it return? How much of that time produces real change — output created, knowledge gained, conditions improved? How much is wasted on redundancy, friction, drift, and dissipation? What's the net direction?

The Four Components

QATU — Quality-Adjusted Time Units (35% of score)

How much human time the system liberates, weighted by utility (is the saved time high-value or low-value?) and distribution (does everyone benefit or just a few?). One QATU equals one hour of human time returned at maximum utility and perfect distribution. A medical AI saving emergency triage time earns more QATUs per hour than an entertainment app saving content browsing time.

QATU = Σ (Hours saved × Utility coefficient × Distribution coefficient)

TSNR — Temporal Signal-to-Noise Ratio (25% of score)

Of all the time this system touches, how much produces real state change versus how much is wasted? Adapted from Claude Shannon's signal-to-noise ratio in information theory — the same mathematical structure used in electrical engineering, MRI imaging, and quantitative finance, applied to time.

TSNR = (ΔO + ΔK + ΔS) / (R + F + D + Dp)

TSNR greater than 1 means the system produces more signal than noise. Less than 1 means more noise than signal. TikTok's TSNR is 0.18 — for every hour of signal, it produces 5.5 hours of noise. Wikipedia's TSNR is 32 — for every hour of noise, it produces 32 hours of signal.

CPI-T — Consumption-Productivity Index (25% of score)

The net direction. After everything is accounted for, is this system giving you time or taking it? Ranges from −1 (pure time destroyer) to +1 (pure time creator). Zoom scores exactly 0.000 — it saves exactly as much time as it consumes.

CPI-T = (Signal hours − Noise hours) / Total hours

TLR — Time Leverage Ratio (15% of score)

How much human time is returned per unit of computational time consumed. A system that uses enormous GPU resources to save a small amount of human time scores lower than one that uses minimal compute to save significant human time. This component naturally decays in weight as compute becomes cheaper.

TLR = Human hours liberated / Compute hours consumed

Signal and Noise

The framework classifies every unit of time expenditure as either signal or noise.

Signal: The State Change Test

Time is signal if and only if it produces a measurable state change:

ΔO (Delta Output) — a tangible artefact was produced or a task reached completion. Code was written. A document was finished. A transaction was processed.

ΔK (Delta Knowledge) — information was acquired, synthesised, or transferred that did not previously exist in the recipient. Someone learned something. A decision was informed by new data.

ΔS (Delta State) — the condition of a system, person, or process moved from one state to a different, intended state. A patient was treated. A relationship was maintained. A mood was regulated.

Noise: The Four Components

Time is noise if nothing changed and resources were consumed:

R (Redundancy) — repeated processing without incremental value. Regenerating the same AI output. Re-reading the same Slack thread. Retrying an identical search.

F (Friction) — unnecessary intermediation, latency, or poor design. Loading screens. Authentication loops. Fee calculations. Format conversions.

D (Drift) — movement away from an intended objective. Content rabbit holes. Algorithmic recommendations that pull you off-task. Scope creep without value.

Dp (Dissipation) — energy consumed with zero output. GPU cycles on discarded inference. Human attention on content that leaves no memory, skill, or changed state. Browsing without choosing.

Grading Scale

90–100  A+   |   80–89  A   |   70–79  B+   |   60–69  B
50–59  C   |   40–49  D   |   25–39  F   |   0–24  F-

Data and Sources

All scores on this site are currently marked ESTIMATED, meaning they are calculated from publicly available data: published monthly active user counts, published session duration statistics, published benchmark results, pricing pages, app store analytics, and industry research reports. Specific sources for key inputs include Sensor Tower, SimilarWeb, company earnings reports, and official platform transparency disclosures.

We distinguish three data confidence levels:

OBSERVED — calculated from direct telemetry, API logs, or first-party operational data. The gold standard. Available when companies share data directly or when we instrument systems ourselves.

ESTIMATED — calculated from public data, published benchmarks, and industry norms. The current basis for all scores. Assumptions are stated and defensible but not verified against internal data.

PROJECTED — calculated from stated capabilities, roadmaps, or theoretical performance. Used only for pre-release or early-stage systems.

We welcome corrections. If you have better data for any scored entity, contact us. We will update scores when provided with evidence that materially changes any input assumption.

Frequently Asked Questions

Who decides the scores?

The formulas decide the scores. The human role is estimating the inputs — user counts, session durations, completion rates, noise fractions. The mathematical framework produces the score from those inputs deterministically. Two people with the same inputs will get the same score.

Why does TikTok score so low?

TikTok has 1.9 billion monthly active users spending an average of 95 minutes per day. That's approximately 5.7 billion hours consumed monthly. The signal — educational content, creative output, genuine knowledge transfer — is real but accounts for a small fraction of total consumption. The noise — engagement loops, content drift, dissipation — accounts for the majority. The TSNR of 0.18 means for every hour of signal, the platform produces 5.5 hours of noise. This is a measurement of time impact, not a judgment of content quality.

Why do all frontier AI models score so close together?

Because at the composite level, they are close. The differentiation happens in the components: Claude leads on TSNR (signal quality), GPT-5.2 leads on raw QATU volume (largest user base), Gemini leads on distribution (Workspace integration reaches the broadest population), and Llama leads on TLR (open-source efficiency). The composite score compresses these differences. The component breakdown is where the real story is.

Can a system that consumes time score well?

Yes. Headspace scores 58.5 despite consuming time (meditation takes time) because its CPI-T is positive — the consumed time is almost entirely signal (genuine ΔS from mental state improvement). The TIME Score measures whether time is well-spent, not only whether time is returned. Some systems create value by consuming time effectively rather than eliminating it.

Why isn't [company X] scored?

We're expanding the index continuously. If you'd like a specific entity scored, contact us with the entity name and we'll add it to the queue.

How often are scores updated?

Scores are recalculated when material input data changes — new user count reports, significant product updates, published usage statistics. The TIME Standard includes a rolling baseline mechanism where the reference point updates annually to reflect the current state of the art.

Can I use these scores?

You can cite, reference, and share any score with attribution to timescore.co. The methodology is described in the TIME Standard paper, linked in the footer. Commercial use of the scoring engine requires permission.

The TIME Standard

The full mathematical framework, intellectual lineage, and extended methodology are documented in the TIME Standard paper by Ash Koosha (2025). The paper covers dynamic baselines, temporal deflation, time-energy entanglement, currency implications, and the complete derivation of every formula used on this site.

Read the TIME Standard ↗

Contact

Corrections, data submissions, scoring requests, and press inquiries:

ashkoosha@gmail.com