Lightgap scorecard — unbounded usability, per use case

As of 2026-08-10 · fak v0.34.0 · lightgap_debt = 9

Every other scorecard in this repo tops out at 100. This one does not top out at all. Each cell asks a single question: relative to what you would do INSTEAD of fak, and relative to the best that is physically possible, how far does fak actually get you – and is that worth what it costs you to learn it? Being worse than the alternative is a negative score, not a low one. There is deliberately no overall number, because the answer differs by an order of magnitude across use cases and averaging them would hide exactly the thing you came here to find out.

The one line

Adopt iff you close more of the gap to physics than you consume of the adopter’s patience.

w_net = artanh(β) − artanh(load), where β = (fak − next_best) / (ceiling − next_best) is the fraction of the gap between the thing you would otherwise do and the best that is physically possible that fak closes, and load = (fak_hours − alt_hours) / tolerance_hours is the share of the buyer’s patience it costs. Because artanh is monotone, w_net > 0 exactly when β > load.

The score is unbounded in both directions: being worse than the alternative is a negative score, not a low one, and approaching the ceiling diverges — which is the honest description of a moat. Full model · ceilings and their derivations · every cell where fak loses · the comparisons nobody has run

The shape

Peak +1.01 — platform-team × observability (RELATIVISTIC)
Dent -4.35 — local-first × raw-speed (REGRESSIVE)
Eccentricity 5.36 nats; 16 of 45 scored cells are net-negative.

fak is not a sphere. It is a spike on a few axes attached to a body that dents inward on others — and the dents are not where the marketing is.

The sphere

Rows are use cases, columns are facets, each entry is w_eff (the claim-capped score decisions use).

use case	TOK	SPD	LONG	RUN	CTRL	OBS	PORT	OPS
solo dev, flat-rate plan	`-0.63`	`-0.28`	?	`+1.00`	?	?	?	`-4.06`
platform team	`+0.66`	`-0.56`	`+0.69`	`+1.00`	`+0.87`	`+1.01`	?	`-0.86`
regulated / audited	`+0.80`	`-0.42`	`+0.83`	`+1.00`	`+0.17`	`+0.72`	?	`-0.72`
local-first / offline	`+0.42`	`-4.35`	`+0.45`	`+1.00`	?	`+0.64`	`-1.20`	`-1.52`
fleet operator	`+0.72`	`-0.51`	`+0.74`	`+1.00`	`+0.55`	`+0.68`	–	`-0.68`
researcher	`+0.63`	`-0.60`	`+0.65`	`+1.00`	?	`+0.80`	?	–
framework builder	`+0.72`	`-0.51`	`+0.74`	`+1.00`	`+0.44`	`+0.68`	`-0.71`	`-1.06`

? = the buyer weights it and nobody has measured it. – = immaterial to that buyer (weight 0), not unscored.

Per use case

use case	verdict	pull (the reason)	drag (the price)	tolerance
Solo developer on a flat-rate plan	UNDECIDABLE	`+1.00` run-integrity	`-4.06` steady-state-ops	2 h
Platform team running agents for others	ADOPT	`+1.01` observability	`-0.86` steady-state-ops	40 h
Regulated buyer who must show their work	ADOPT	`+1.00` run-integrity	`-0.72` steady-state-ops	240 h
Local-first developer running models on their own hardware	BLOCKED	`+1.00` run-integrity	`-4.35` raw-speed	12 h
Operator running many agents unattended	ADOPT	`+1.00` run-integrity	`-0.68` steady-state-ops	80 h
Researcher measuring agent behaviour	ADOPT	`+1.00` run-integrity	`-0.60` raw-speed	24 h
Builder embedding it in their own product	ADOPT-WITH-SCARS	`+1.00` run-integrity	`-1.06` steady-state-ops	120 h

What each verdict means

verdict	means
ADOPT	one axis clears the switch bar and nothing material goes backwards
ADOPT-WITH-SCARS	the pull is real, but a material axis is `DRAG` or worse — go in knowing what you are eating
PILOT-ONLY	the best axis is positive but under the bar; worth a trial, not a migration
BLOCKED	a material axis is `REGRESSIVE` — no pull elsewhere buys it back
UNDECIDABLE	too much of what this buyer weights has never been measured against their actual alternative
HOLD	nothing clears rest

Verdict ladder

w_eff	verdict	means
≥ +3.80	AT-CEILING	at the limit — nothing better can exist on this axis
≥ +2.00	NEAR-C	category-defining; the alternative is not in the running
≥ +1.00	RELATIVISTIC	a real, large, net win worth restructuring for
≥ +0.50	CRUISE	a solid net win; adopt if this facet is what you came for
≥ +0.10	DRIFT	marginally better once effort is counted; easy to regret
≥ -0.10	REST	indistinguishable from doing nothing
≥ -1.00	DRAG	the alternative wins once you count the effort
≥ −∞	REGRESSIVE	actively worse than what you already have

Regenerate

fak score lightgap                      # the sphere
fak score lightgap --dents --unrun      # the honest parts
fak score lightgap --markdown-dir docs/lightgap-scorecard