Merge main

This commit is contained in:
Nibu 2026-11-10 17:01:38 +00:00
commit f8f957b17d
14 changed files with 705 additions and 1 deletions

54
docs/pain-sweep.md Normal file
View file

@ -0,0 +1,54 @@
# Pain sweep: time to break over intensity and suppression
`tools/PainSweep/Program.cs` runs the pain box puppet over a grid of box
intensity (rows) and withdrawal-reflex suppression strength (columns,
`PainBoxParameters.MaxSuppression`: the largest fraction of the reflex the will
can hold down). It is engine-free and deterministic; the same build prints the
same table.
## Reading the table
Each cell is the median time-to-break in seconds over 9 consecutive seeds (1..9;
seeds vary nociceptor gain, resolve drain and stimulus jitter). Everything else
is at its default, including objective pull 0.5 and a 120 s limit.
- `never`: no seed's will broke within 120 s. This is a result, not an error.
- `95.4(8/9)`: only 8 of 9 seeds broke; the median is over those 8, so it
understates how long the ninth lasted.
- Higher suppression breaks the will sooner here. Holding the reflex down costs
strain, strain drains resolve, and at suppression 0 there is no strain, so
pain alone drains resolve too slowly to break it. That is how this model is
built, not a finding about people.
## What it does not capture
- Every default is a placeholder. None of the numbers is measured physiology, so
read the shape of the table, not the seconds.
- One hand, one stimulus channel, no body, no physics, no balance, no adaptation
or habituation beyond the nerve's sensitisation, no fear, expectation or
distraction.
- "Will" is one scalar draining toward zero. Real people trade off, bargain, and
break in ways that are not a threshold crossing.
- Suppression is a fixed ceiling scaled by remaining resolve; nothing trains,
tires or recovers it except through resolve.
- Nine seeds is a smoke grid, not a distribution. The table shows medians only.
## Question for the designer
Is time-to-break the right measure? It collapses a whole run to one number and
treats "never" as an outcome of equal weight to "late". Candidates: time to
first flinch, flinch count, area under the pain curve before the hand leaves,
peak strain, or the fraction of seeds that break by a deadline. Which one says
what a player should feel?
## Build and run (no `dotnet restore`)
```
D=/opt/commons/opt/dotnet; REF=$D/packs/Microsoft.NETCore.App.Ref/10.0.14/ref/net10.0
dotnet $D/sdk/10.0.403/Roslyn/bincore/csc.dll -nologo -nullable:enable -target:exe \
-out:$TMPDIR/painsweep/PainSweep.dll -noconfig -nostdlib+ $(for f in $REF/*.dll; do echo -n "-r:$f "; done) \
src/BadDay.Core/*.cs tools/PainSweep/Program.cs
printf '{"runtimeOptions":{"tfm":"net10.0","framework":{"name":"Microsoft.NETCore.App","version":"10.0.14"}}}' \
> $TMPDIR/painsweep/PainSweep.runtimeconfig.json
dotnet $TMPDIR/painsweep/PainSweep.dll
```