# Tier list — the impact ladder

Normative. This file operationalizes `charter.md`: the metric there says impact is displacement in
joy, and this is that metric turned into a ranking every fact page carries. Read `charter.md` first —
it is short, and this document is meaningless without it.

Every fact page (`venture`, `person`, `helper`, `resource`, `work`) gets a **`**Tier:**`** — `S`, `A`,
`B`, `C`, `D`, `F`, or `unranked`, optionally suffixed `*` (ruling 9). `source` and `guide` pages do
not: evidence artifacts and wiki apparatus displace nothing.

## It is a Smash tier list

The analogy is load-bearing rather than decorative. Six borrowed rules decide most cases before the
rubric gets consulted:

1. **One list, everyone on it.** A defense prime, a Nobel-winning lab, a cookie franchise, and a
   county chamber of commerce are graded against the same question. There is no "best resource"
   consolation bracket, and `Type` is not a handicap.
2. **Tier is the ceiling, not the current result.** A character is top tier for what it does in the
   best hands. Rank the *maximum estimated bounds* of displacement (charter metric 3) — not this
   quarter's revenue, headcount, or press.
3. **Popularity is not tier.** Beloved does not promote. Neither does effort, nobility, hardship,
   longevity, local pride, or being the only one who showed up. A tier list of Utah's favorite
   institutions would be a different document and a less useful one.
4. **No curve.** Tiers are absolute against the world, not graded against Utah. A state is allowed to
   have empty space at the top of a sector, and most of a corpus is allowed to sit low. Grading on a
   local curve is how a tier list becomes a chamber-of-commerce brochure.
5. **Tiers are per-patch, and they move.** A tier is an estimate on a date, not a property. A bet that
   fails drops; a lab result that replicates climbs. Re-run the ladder, don't defend it.
6. **Within a tier there is no order.** Tiers are the resolution the judgment supports. Arguing
   whether Fervo outranks Recursion is a worse use of a session than finding the page that is in the
   wrong tier by two steps.

## The ladder

| Tier | Gloss | The test |
|---|---|---|
| **S** | world-historic | The world is measurably different, permanently, and the difference traces to this. A person on another continent is touched by it without knowing the name. |
| **A** | field-defining | Sets the frontier of a global industry or discipline — or is a live program whose plausible success is S and whose path there is legible. |
| **B** | structurally significant | Durably changes how a real sector works, or how a whole state works. The ceiling is bounded by the market it serves, not by the team. |
| **C** | substantive and replaceable | Does real work at real scale, and a close substitute exists. Removing it inconveniences its customers and moves nothing else. |
| **D** | local | The displacement is real to the people in the room and stops at the walls. Removing it would be noticed by its members and by nobody else. |
| **F** | inert | No estimable displacement. A directory listing, a ceremony, a duplicate, or a claim nothing corroborates. |
| **unranked** | not rankable | A page too thin to argue bounds from, and the honest escape hatch (P3) rather than a silent guess. Not a tier; not a placeholder for "low". |

Expected shape, as a drift guardrail and not a quota: `S` in the low single digits of a percent, `A`
under a tenth of the corpus, and the mass in `C`/`D`. **If `S` grows past roughly 4% of fact pages,
the rubric has drifted and the fix is in this file, not in the pages.**

## Rulings

These are the calls that actually decide placements. Each was settled against a real page.

1. **Rank the page's own atom, not the topic it touches.** `arpanet-fourth-node` is `A` and
   `golden-spike-transcontinental-railroad` is `S`, and the difference is not that the railroad
   matters more than the internet. The railroad page's subject is the railroad; the ARPANET page's
   subject is *Utah being the fourth node*, whose counterfactual is a different university getting a
   number. When a page is framed as "Utah's role in X", that role is what gets ranked — and a tier
   two steps below the topic's importance is a signal the atom may be framed too narrowly (P2).

   **This includes work Utah claims but did not host** (maintainer's ruling, 2026-08-11). Fletcher
   invented stereo at Bell Labs, Farnsworth did the decisive television work in San Francisco and
   Idaho, Hall's first synthesis was at GE — and `harvey-fletcher-stereophonic-sound` and
   `philo-farnsworth-television` still rank as the inventions they are named for, because that is
   their atom. `h-tracy-hall-diamond-presses` ranks lower *because its own page already scoped itself
   to the Utah press work* and says so in the second paragraph. That asymmetry is the mechanism, not a
   flaw in it: **the ladder is an audit of page framing.** A page that wants an honest tier scopes its
   atom honestly, and no page gets to keep a broad frame for prestige and a narrow one for scrutiny.

   **A multinational's atom in this corpus is its Utah operation**, unless the page argues that the
   whole company is the subject (added 2026-08-11; two raters hit this independently on `hexcel` and
   `sword-health`, both reasoning from the `adobe` anchor and reaching opposite conclusions about which
   thing they were ranking). `adobe` is the pattern to copy: the company is `A`-shaped, the page frames
   its atom as *Adobe's Utah engineering hub*, and so it is `B`. A page titled for the global
   corporation while existing here because of a plant is carrying two atoms and should pick one — until
   it does, rank the Utah operation, because that is why the page is in a Utah corpus at all.
2. **Magnitude, never sign.** `northrop-grumman-promontory` builds the boosters for every Artemis
   mission and the solid stages of the nuclear deterrent; `nsa-utah-data-center` is an intelligence
   capability of unbounded reach; `bingham-canyon-mine` electrified a country and left a permanent
   hole and a contaminated aquifer. All rank high, and the direction of the displacement is argued in
   prose on the page. A tier list that flinched here would rank Utah's most consequential work below
   its most agreeable work.
3. **An unverifiable claim is not a large claim; it is no claim.** A seed-stage company whose deck
   implies a reordered global industry, with nothing checkable behind it, is `F` or `unranked` — not
   `A` discounted to `C`. Bounds are estimated from evidence; where there is none there is nothing to
   estimate. `struvia.md` is the worked case: what it most needs is anything checkable. This is *not*
   a second Confidence field — a `High`-Confidence page can be `F` (a chamber of commerce, impeccably
   sourced) and a `Low`-Confidence page can be `A` (a live bet whose scale is legible and whose
   evidence is thin).
4. **An unproven bet keeps its ceiling; an implausible one does not.** Judge loosely — the world needs
   believers. A credible program whose success would be `S` sits at `A` while it is unfinished, one
   step down and not four: `fervo-energy` and `recursion-pharmaceuticals` are ranked on what the bet
   is worth if it lands, because that is the number a reader deciding where to spend five years
   actually needs. Name the bet with `**Bet:**` and let the reader discount it. Demote only when the
   premise itself stops being believable, not merely because it is unfinished.

   **The counterfactual applies to the bet, too — the ceiling is the prize divided by the number of
   credible pursuers** (added 2026-08-11, from the first corpus-wide run). Without this the ruling
   promotes every plausible preclinical company to `A`, because the *field's* prize is almost always
   enormous: a novel Gram-negative antibiotic would bend global antimicrobial-resistance mortality, and
   that is true of all several dozen groups chasing it. The prize belongs to the field; a page gets the
   share it would actually cause. So `A` for a bet needs a *distinctive* claim on the ceiling — being
   the only party doing the thing, or holding a checkable result nobody else has. `fervo-energy` (the
   only 500 MW enhanced-geothermal plant under construction) and `zanskar-geothermal` (an independently
   reported blind-system confirmation, the first in thirty years) clear it; `curza`, `halia-therapeutics`,
   and `rodatherm-energy` were adjudicated down to `B` for the opposite reason — real programs with real
   funding, pursuing prizes that many credible competitors are pursuing at the same time. This is the
   difference between "if this works, the world changes" and "if *we* are the ones who make it work."
5. **Enablers get the counterfactual difference, which is usually small.** This is charter metric 4
   doing its most unpopular work. An accelerator, chamber, association, event, or economic-development
   office is ranked on what would not have happened without it — nearly always a small acceleration of
   work that had other routes, however much throughput passes through. Almost the entire
   `Type: resource` shelf lands `D` or `F`, and that is the rubric working rather than failing. The
   exception is an enabler that is *itself* irreplaceable infrastructure: `utah-population-database` is
   high tier because BRCA was findable in Utah and not elsewhere, and `utah-forge` because there is no
   second US field laboratory de-risking enhanced geothermal.
6. **Depth substitutes for breadth.** Immense joy for a few outranks nothing (charter metric 1). A
   basic-science observatory touching a few hundred physicists, or a hospice with a dozen beds, is
   ranked on depth × permanence and is not capped for having a small audience. What caps it is
   replaceability, which is a different question.
7. **A tier is not a verdict on whether the page should exist.** The charter categorizes and
   prioritizes and never gatekeeps. `F` is the right answer for a page that is still the right answer
   to "who do I call in Cedar City". What the ladder changes is reading order, not membership.
8. **A person ranks on their own counterfactual, not on their best company's tier.** A `Type: person`
   page is not a trophy shelf for the ventures it links to, and inheriting a founder's company tier
   would double-count the same displacement on two pages. Rank what this person caused across a career
   that the field would not have produced without them — which sometimes exceeds any one of their
   companies, and much more often falls below the most famous one.
9. **Hype tier: something exceptionally funny or interesting may be bumped one tier, and the bump is
   marked with `*`.** Every tier list has the character who is objectively mid and top of everyone's
   watch list, and pretending otherwise makes for a worse list. `**Tier:** A*` means *ranked A, of
   which one step is charisma* — so stripping the asterisks recovers the pure displacement ladder and
   the drift guardrail above is measured unasterisked. Rules: **one step, never two**; it may reach
   `A*` but **never `S`**, because the top of the list is a claim about the world and an unfalsifiable
   `S` would cost more than it buys; and it must be earned by the page being *disproportionately worth
   reading* — funny, strange, instructive, or emblematic of how Utah actually works — never by
   affection for the subject, which ruling 3 already refuses. `pons-fleischmann-cold-fusion` is the
   worked case: what it displaced was a cautionary tale about how not to announce a discovery, and it
   is nonetheless the most interesting hour in the history of Utah science. The bump is most useful at
   the bottom, where it rescues a genuinely delightful page from the `F` pile and gets it read.
   (Maintainer's rule, 2026-08-11; the `A*` ceiling is the agent's call and is vetoable.)

## Hooks into the machinery

- `## Impact` sections argue depth, breadth, permanence, bounds, counterfactual — prose with
  reasoning, no scores. **A tier of `B` or above requires one**, whatever the page's `Type`; a resource
  that earns `B` has earned the argument, and the 200 that do not are left alone.
- `**Tier:**` is that argument's conclusion, and the two must agree. Where they disagree the prose is
  right, exactly as `## Evidence` beats `Confidence`.
- `**Bet:**` names the contrarian premise on pages that have one, and is what makes ruling 4 legible to
  a reader who would discount it differently.
- **Gems are `S` and `A`.** That is the whole point of the ladder: "surface all the gems" stops being a
  matter of an agent's taste and becomes a filter over [the tier list](../views/tier-list.md).
- Wanted-list gaps are ordered by the tier the missing page would plausibly hold.

## Provenance

First assigned across the whole corpus on 2026-08-11, 485 fact pages in one pass, by subagents working
from this file plus a hand-placed anchor set, with every `S` and `A` adjudicated by hand afterward.
Method, the anchor set, and the inter-rater divergence found on the way are in
`research/tier-list/`. This is charter probe E11 run at corpus scale: divergence between raters is a
gap in this document, and the fix goes here rather than into the pages.