130 lines
4.4 KiB
Markdown
130 lines
4.4 KiB
Markdown
---
|
||
id: GROUND-WP-0006
|
||
type: workplan
|
||
title: "H1 experiment: problem pressure and high-stress ATTACK self-soothe"
|
||
domain: consumer
|
||
repo: ground-game
|
||
status: active
|
||
owner: bernd
|
||
topic_slug: whynot
|
||
created: "2026-08-07"
|
||
updated: "2026-08-08"
|
||
origin: residual
|
||
origin_ref: GROUND-WP-0003
|
||
state_hub_workstream_id: "0056c739-e31a-459c-bf5c-d67a9bd9b596"
|
||
---
|
||
|
||
# H1 — measure whether Stress coupling restores ATTACK/DARVO relevance
|
||
|
||
**Context:** [GROUND-RPT-0003](../reports/260807-clay-borg-attack-and-modes.md)
|
||
showed ATTACK never pays and DARVO never arms under competent play.
|
||
Design challenge and hypothesis:
|
||
[`history/260807-attack-darvo-stress-design.md`](../history/260807-attack-darvo-stress-design.md).
|
||
|
||
**Package:** [`editions/experiments/h1-problem-stress/`](../editions/experiments/h1-problem-stress/)
|
||
**Catalog id:** `h1-problem-stress` (selectable; not baseline).
|
||
**Measurement:** [GROUND-RPT-0004](../reports/260808-clay-borg-h1-measured.md).
|
||
|
||
Do **not** merge H1 into `ground-darvo-r0` from this workplan. Catalog
|
||
`decision: reject-as-baseline` recorded 2026-08-08 (agent recommendation
|
||
pending maintainer confirm on T05).
|
||
|
||
## Task: freeze H1 package and catalog entry
|
||
|
||
```task
|
||
id: GROUND-WP-0006-T01
|
||
status: done
|
||
priority: high
|
||
state_hub_task_id: "183e5fef-d350-434f-8ebb-09c8a43e224a"
|
||
```
|
||
|
||
Land experiment directory, `rules_delta.yaml`, `editions/catalog.yaml` +
|
||
`CATALOG.md`, and the design note. Baseline r0 untouched.
|
||
|
||
**Done 2026-08-07.**
|
||
|
||
## Task: request clay-borg kernel support for variant selection
|
||
|
||
```task
|
||
id: GROUND-WP-0006-T02
|
||
status: done
|
||
priority: high
|
||
state_hub_task_id: "20ea00ea-c196-4aab-982b-12d5ea79b258"
|
||
```
|
||
|
||
Message clay-borg with: catalog path, `variant_id: h1-problem-stress`,
|
||
H1-A/H1-B deltas, and ask for (1) ability to select a ground-game catalog
|
||
variant in trials/play drivers, (2) kernel implementation of the two
|
||
deltas, (3) re-run of the attack-value panel (RPT-0003 shape) and a DARVO
|
||
arm-rate report against H1 vs baseline.
|
||
|
||
**Done 2026-08-07.** Hub message `af57274e-…` to `clay-borg-custodian`.
|
||
|
||
## Task: human table smoke (optional, 2p or 3p)
|
||
|
||
```task
|
||
id: GROUND-WP-0006-T03
|
||
status: todo
|
||
priority: medium
|
||
state_hub_task_id: "d3430104-fa6e-4a91-999f-0905424c9c2f"
|
||
```
|
||
|
||
One short game under H1 rules using experiment `Rules_Text` / `Actions`.
|
||
Note: stress climb speed, whether ATTACK felt useful at 4–5, whether
|
||
DARVO appeared, whether group threshold still felt fair. Append notes
|
||
under `history/` or this workplan.
|
||
|
||
## Task: record measurement and catalog utility
|
||
|
||
```task
|
||
id: GROUND-WP-0006-T04
|
||
status: done
|
||
priority: high
|
||
state_hub_task_id: "a6d90fa8-1b95-4856-9642-851d0b0f2ff4"
|
||
```
|
||
|
||
When clay-borg (or table + engine) results exist: update
|
||
`editions/catalog.yaml` row for `h1-problem-stress` —
|
||
`status`, `utility_estimate`, `decision` (`keep-as-experiment` /
|
||
`promote-to-baseline` / `reject`). Do not edit baseline r0 content here.
|
||
|
||
**Done 2026-08-08.** RPT-0004 filed; catalog `status: measured`,
|
||
`decision: reject-as-baseline`, utility summary from CB-EV-0030/0031.
|
||
Package remains `selectable: true` for regression A/B.
|
||
|
||
## Task: decide promote / reject / H2
|
||
|
||
```task
|
||
id: GROUND-WP-0006-T05
|
||
status: wait
|
||
priority: medium
|
||
state_hub_task_id: "40e74ec4-c006-4411-9259-6277371b2601"
|
||
```
|
||
|
||
Maintainer decision from criteria in the design note §3.2. If reject or
|
||
under-deliver: spawn H2 (e.g. last-place status stress) as a **new**
|
||
experiment package, not an in-place mutation of H1 without a new
|
||
`variant_id`.
|
||
|
||
**Agent recommendation 2026-08-08 (needs maintainer confirm):**
|
||
|
||
| choice | recommendation |
|
||
|--------|----------------|
|
||
| Promote H1 deltas into r0 | **No** — criterion 3 fails hard |
|
||
| Keep package in catalog | **Yes** — regression pin, not ship |
|
||
| Next experiment | **Yes** — new `variant_id` (H2), not silent H1 edit |
|
||
|
||
**H2 shape candidates** (pick one or combine carefully):
|
||
|
||
1. **Pressure form:** +Stress **per** unclaimed Problem, or only from
|
||
Round 3+, or only if ≥2 unclaimed — so tax does not force full clear
|
||
every round before SOLVE bandwidth exists.
|
||
2. **Who pays:** not always everyone — e.g. Lead, or seats with 0 claims
|
||
this game, or last place (semi Idea 2).
|
||
3. **Keep or drop H1-B:** H1-B never fired under competent play; either
|
||
drop for H2 clarity or lower the self-soothe gate (Stress ≥3) so it is
|
||
testable.
|
||
4. **Do not** only stack last-place stress on top of flat H1-A without
|
||
fixing the solve-rate tax first.
|
||
|
||
Confirm T05 → close WP-0006 or spawn H2 workplan.
|