Scoring formula

Goal: turn an arbitrary number of flags — positive and negative — into one -100–100 score plus a traffic-light band. 0 = neutral (no data). 100 = strongly positive. -100 = strongly negative.

Inputs (all data-driven, extendable)

data/categories.json — one entry per flag category:

{ "id": "boycott-conflict", "name": "Boycott / conflict complicity", "icon": "Megaphone", "defaultWeight": 3 }

Flag polarity — each flag is either a concern (negative, default) or a positive signal (positive).

Severity multiplier (per flag):

severity	multiplier
minor	1
moderate	2
severe	3
systemic	5

Confidence multiplier (derived from the flag's status):

status	multiplier
sourced	1.0
unverified	0.6

Decay multiplier — reserved extension point, defaults to 1.0. A future version could reduce the weight of old flags without changing the rest of the formula.

Step 1 — per-flag contribution

contribution(flag) = category.weight × severity.multiplier × confidence.multiplier × decay.multiplier

Step 2 — raw totals (per polarity)

rawNeg = Σ contribution(flag)  for all negative flags
rawPos = Σ contribution(flag)  for all positive flags

Step 3 — saturate each direction

A straight sum would let a company with many flags blow past any sensible scale. Apply diminishing returns separately to each direction so each component stays in [0, 100):

saturate(raw) = 100 × (1 − e^(−raw / K))

At raw = 0: saturate = 0 (no effect). As raw → ∞: saturate → 100 (component maxes out).

Step 4 — combine into -100–100 score

score = round(saturate(rawPos) − saturate(rawNeg))

A company with no flags has rawPos = rawNeg = 0, so score = 0 exactly — neutral/unknown, not perfect. Green flags push toward 100; red flags push toward -100.

K is a single tunable constant. Currently 12. Raising it makes flags hit more gently; lowering makes them hit harder. Calibrated against the first ~20 real seeded companies — revisit as more data is added.

Example at K = 12:

scenario	rawNeg	rawPos	score
no flags	0	0	0
one sourced, severe flag in a weight-3 category	9	0	-53
same, plus one sourced, severe green flag	9	9	0
one sourced, minor flag in a weight-2 category	2	0	-15
two sourced, minor red flags in weight-2 categories	4	0	-28

Step 5 — traffic-light band

score	band	meaning
34 to 100	green	low concern
-33 to 33	yellow	neutral / insufficient data
-100 to -34	red	high concern

Extending this later

• New category: add to categories.json with a defaultWeight. No formula change.
• New severity level: add a row to the severity table + its multiplier.
• Decay: implement the reserved decay.multiplier function; every flag already carries dateAdded.
• Per-user weighting: categories are already just weighted numbers — a future "mute category X" is a per-user override of category.weight at render time, no server-side change.

Scores do not flow along the ownership graph

A company is scored on its own flags only. Ownership is displayed for context, but a parent's conduct never lowers a subsidiary's score and vice versa.

If a parent's conduct genuinely implicates a subsidiary, record it as a flag on the subsidiary with its own source. That keeps every point on a company's score traceable to a sourced claim about that company, and keeps the score stable when an ownership edge is later corrected.

Editorial note

Default category weights reflect a starting judgment about relative severity. This is a defensible starting point, not a claim of moral authority — change via normal PRs.