Add network-society campaigns and indirect contribution breakdowns

TODO.md

@@ -16,6 +16,8 @@ Getting **testnet to a two-person shared lab** is also a standing plan, not a pi
 
 ----
 
+- We should have a UI for making it easy for a user to follow projects aligned with *similar* statements to the ones he prefers, and then if he wants to he can do the alignment-attestations to make those projects also show up on his own preferred cause board.
+
 - **(Ask)** Verify work-request contribution/refund links against actual chain events, and verify response providers and work items against the project's beneficiary/registry before presenting verified payment or provider authority. The first slice displays reported contribution/refund amounts and authenticates acceptance only against the provider named by the response author. See [work requests](specs/tech/subsystems/work-requests/README.md).
 - **(Ask)** Complete work-request UI revisions and withdrawals, version/history inspection for offers and responses, and a browser-tested playground demonstration. SDK versioned publication actions exist; the current UI primarily creates records.
 

continuity/2026-10-16-ai-pause-board-handoff.md

@@ -0,0 +1,35 @@
+# Handoff: AI-pause board, then the rest of the network-societies campaign
+
+Next session: author the families that are still missing. The AI-pause attester pass is done. Do not publish an implication the attester has not blessed. Do not write a new plan over `fake-data-generation/output/campaigns/network-societies-v1-b2/` or over `output/campaigns/network-societies-v1-b2-ai-pause/` if that directory is later executed.
+
+## Read first
+
+- [`fake-data-generation/campaigns/network-societies-v1-b2.md`](../fake-data-generation/campaigns/network-societies-v1-b2.md) — what batch 2 publishes, and the rule that the mined directory is the pre-pin plan.
+- [`fake-data-generation/campaigns/network-societies-v1-b2-statements.json`](../fake-data-generation/campaigns/network-societies-v1-b2-statements.json) group `ai-pause`, and the matching people in [`network-societies-v1-b2-decisions.json`](../fake-data-generation/campaigns/network-societies-v1-b2-decisions.json).
+- [`fake-data-generation/brainstorming/network-societies-campaign-plan.md`](../fake-data-generation/brainstorming/network-societies-campaign-plan.md) — which families are still not manifests, and the attester quality gate.
+- [`specs/product/statements-are-peculiar-for-good-reasons.md`](../specs/product/statements-are-peculiar-for-good-reasons.md) — containment versus nudge. The AI-pause modifieds were written to that rule. They have not been checked by the model.
+- Branch `campaign`, not pushed. This slice is `75705202` ("Add an AI-pause board with milder wordings and dissenters."). Parent is `dd3578dc`.
+
+## What is already true
+
+The cause is one plain board, `ai-pause`, so every sentence stays on it. It is not bridge-shaped: the five bridge roles are not all present. Five live-attester arrows are now in `implicationAllowances`. The planner allows a plain-board modified when the evidence label is `live-attester-modified-to-commonality`.
+
+The shared sentence is a treaty that pauses new research aimed at making AI systems more capable, for the length of that pause, while work on catastrophic harm continues, and that does not switch off current tools. Five modified statements contain that sentence and keep a reason the short sentence omits: small chance, jobs, industry warnings, not-a-Luddite, children. Naturals do not name the treaty. `pole-stop` (Hal) and `pole-accelerate` (Vic) reject the pause. Noor and Tess are nudge targets: they signed a natural and left the treaty unsigned.
+
+Seven spotlight users are `user-061` through `user-067` (Priya, Dean, Noor, Sam, Tess, Hal, Vic), persona `ai-pause-voice`. Five projects are in `network-societies-v1-b2-projects.json` under `ai-pause`. Scale is 16 causes, 117 statements, 67 users, 57 projects. Alignment attestations were raised from 40 to 57 so the new projects can be funded. `npx tsx --test test/campaignPlanner.test.ts` passed before the commit. The commit hook's fast suite passed.
+
+The mined run in `output/campaigns/network-societies-v1-b2/` is still the older 60-user, 100-statement, 52-project plan. Rewriting that directory would mark its wallets unprovisioned.
+
+## What is done (2026-10-16, later session)
+
+The eleven AI-pause pairs were sent to `evaluateImplicationWithLLM` on `deepseek/deepseek-v3.2`. All five designed-yes pairs blessed at high confidence, and all six designed-no pairs refused at high confidence, on the first wording. The rows are appended to `network-societies-v1-b2-implication-evaluations.json` (79 pairs, same prompt fingerprint). The five blesses are in `implicationAllowances`. The planner now accepts a live-attester modified whose statement id starts with `modified-` even when the board is plain, so the industry, not-a-Luddite, and children arrows are planned, not only `modified-left` and `modified-right`. `npx tsx --test test/campaignPlanner.test.ts` passed (15 tests). The fresh plan is `output/campaigns/network-societies-v1-b2-ai-pause/` (67 users, 117 statements, 57 projects, 491 writes). It was not executed. The mined directory was not rewritten.
+
+## What to do
+
+1. Author the next family batch. Still not manifests: low-carb, cancel-proof support, unwoke, Democracy 2.0, Islamic reform, and less regulation. US-friendly Canadians was expanded past the five-statement draft in a later session. Home learning and Grey County civic life stop at the pilot planks. Do not force a new triple onto an overlap that already shares a concrete want. The campaign plan's "Next task" paragraph lists those six families.
+2. Delegated spending and a recurring monthly pledge stay out. Do not start a bulk random run. Do not execute the AI-pause plan unless you mean to, and do not overwrite either existing output directory.
+
+## Suggested skills
+
+- `thorough-tester` — after an attester pass, check that Hal and Vic are not supporters of the pause through an arrow, and that Noor and Tess stay unsigned on the treaty.
+- `browser-driving` — only after a fresh local execute. Open the AI-pause board and confirm the milder sentences are visible next to the shared sentence.

continuity/2026-10-16-network-societies-b2-executed-handoff.md

@@ -0,0 +1,35 @@
+# Handoff: DeepSeek model for opencode-go, then the leftover campaign gaps
+
+Next session: figure out which DeepSeek model `opencode-go` should use. `deepseek-v4-flash` is a poor default for implication checks. Try whether 4.1 flash returns JSON. Do not treat that model hunt as a reason to rewrite the batch-2 statements again.
+
+## Read first
+
+- [`fake-data-generation/campaigns/network-societies-v1-b2.md`](../fake-data-generation/campaigns/network-societies-v1-b2.md) — what batch 2 published.
+- [`fake-data-generation/campaigns/network-societies-v1-b2-implication-evaluations.json`](../fake-data-generation/campaigns/network-societies-v1-b2-implication-evaluations.json) — 68 live pairs. Model `deepseek/deepseek-v3.2`. Prompt fingerprint `2f61d4d61235214eceee97e9a364da48b8a6833f509089941cb758728302e891`.
+- [`fake-data-generation/campaigns/network-societies-v1-b2-decisions.json`](../fake-data-generation/campaigns/network-societies-v1-b2-decisions.json) — allowed arrows and the five first-wording refusals.
+- [`verifier/llm-routing.json`](../verifier/llm-routing.json) — subscription pin. Every task kind is `xai/grok-4.6`. Allowed providers are `xai` and `opencode-go`. (NOTE FROM ADAM: Also might as well use grok-4.7 instead of grok-4.6.) Never OpenRouter for verifier leaves.
+- [`services/attester-core/src/llm-models.ts`](../services/attester-core/src/llm-models.ts) and [`services/attester-core/src/openrouter.ts`](../services/attester-core/src/openrouter.ts) — production OpenRouter default is `deepseek/deepseek-v4-flash-0731`, and `requestJsonCompletionWithUsage` caps `max_tokens` at 500.
+- Branch `campaign`. Do not commit on `dev` or `master`.
+
+## What is already true
+
+Batch 2 was executed locally into `output/campaigns/network-societies-v1-b2/` (gitignored). 446/446 mined and reconciled. The first pass failed 14 actions with insufficient ETH (`setBeliefsInBatch` and large publishes). Those wallets were topped up with `hardhat_setBalance` and the run resumed. Do not overwrite the pilot or `medium-realistic-v2` outputs.
+
+Twelve new modified-to-commonality arrows are planned with evidence `live-attester-modified-to-commonality`. Abortion and schools stay `accepted-bridge-role-pair`. The six clusters are not accepted seed. Employment modifieds and the Canada commonality were rewritten once so designed-no pairs refuse. Details are in the campaign note, not repeated here.
+
+`deepseek/deepseek-v4-flash-0731` returned an empty message body on the longer pairs. That is the same stall recorded for the abortion live check in [`statement-generation-exercises/02-compromise-abortion.json`](../fake-data-generation/statement-generation-exercises/02-compromise-abortion.json). v3.2 returned JSON. Heuristic `implies: true` was not used.
+
+The Grey County assimilation commonality board loaded at `http://commonality.localhost:8088/#/cause/0xEe6a18208ca061e2Eddd4A6fA7E4E4d4927eF352/network-societies-v1-b2-grey-assimilation-bridge`. The statement page shows "Arriving through other statements." Direct and indirect counts were 0 on that load.
+
+## What to do
+
+1. Decide the `opencode-go` DeepSeek model. `verifier/llm-routing.json` does not name one yet. An older note in [`verifier/verifier-checks-need-founder-level-understanding.md`](../verifier/verifier-checks-need-founder-level-understanding.md) pointed `clear-communication` at `deepseek-v4-pro`. Adam's guess: 4.1 flash may work better than v4-flash. Check the model ids `pi` / opencode-go actually accept, then run one short JSON implication pair and one longer modified-to-commonality pair. Record empty body versus a real decision. If v4-flash only fails because reasoning tokens eat the 500-token cap, say so. Do not switch production `PRODUCTION_OPENROUTER_MODEL` unless a model both returns JSON and matches the v3.2 designed-yes/no pattern on a few known pairs.
+2. The campaign gas estimate underfunded 14 wallets on this run. Look at `computeCampaignFundingNeeds` versus the gas viem actually requests (`setBeliefsInBatch` at 280000, publishes around 4e6). A resume should not need a manual `hardhat_setBalance`.
+3. Persona `set-belief` choices in the decisions file are still not applied. Eli can still be planned as believing the abortion commonality.
+4. Delegated spending and a recurring monthly pledge are still not campaign actions.
+5. The remaining brainstorm families (low-carb, unwoke as an umbrella, Democracy 2.0, Islamic reform, permitting, home-school beyond the pilot) are still unauthored. Do not force triples onto overlaps that already share a concrete want.
+
+## Suggested skills
+
+- `thorough-tester` — one short and one long implication pair per candidate model, plus the gas-estimate shortfall.
+- `browser-driving` — only if a later execute changes a board. This session already opened the assimilation commonality.

continuity/2026-10-16-network-societies-draft-bridges-handoff.md

@@ -0,0 +1,37 @@
+# Handoff: evaluate the six draft bridges, then execute batch 2
+
+Next session: run the real implication-attester prompt on the six draft clusters, keep failed pairs unpublished, and only then execute `network-societies-v1-b2` locally. Do not treat bridge-shaped boards as attested implications.
+
+## Read first
+
+- [`fake-data-generation/campaigns/network-societies-v1-b2.md`](../fake-data-generation/campaigns/network-societies-v1-b2.md) — what the batch publishes and what it must not.
+- [`fake-data-generation/statement-generation-exercises/06-network-society-alliances.json`](../fake-data-generation/statement-generation-exercises/06-network-society-alliances.json) — draft wording. Not accepted seed.
+- [`fake-data-generation/statement-generation.md`](../fake-data-generation/statement-generation.md) and [`specs/product/statements-are-peculiar-for-good-reasons.md`](../specs/product/statements-are-peculiar-for-good-reasons.md) — attester plus routing. A bless is not enough.
+- [`fake-data-generation/PLAN.md`](../fake-data-generation/PLAN.md) section F.
+- [`fake-data-generation/brainstorming/network-societies-campaign-plan.md`](../fake-data-generation/brainstorming/network-societies-campaign-plan.md) — quality gate for new edges.
+- Branch `campaign`. This handoff's commit is the one that adds the six clusters. Do not commit on `dev` or `master`.
+
+## What is already true
+
+The four-family pilot was executed and reconciled locally (207/207). Details and board URLs: [`fake-data-generation/campaigns/network-societies-v1-pilot.md`](../fake-data-generation/campaigns/network-societies-v1-pilot.md). Do not overwrite `output/campaigns/network-societies-v1-pilot/` or `output/campaigns/medium-realistic-v2/`.
+
+Batch 2 is planned in source, not executed. Manifest: 60 users, 100 statements, 52 projects, 15 causes. Causes `abortion` and `schools` reuse accepted seed wording. Six new causes are bridge-shaped from the exercise file: `grey-assimilation`, `ranked-approval-poll`, `checkout-trial-bridge`, `minors-medicine`, `opinion-employment`, `canada-us-ties`. The older plain boards (crypto planks, civic-life planks, the voting trial, gender-conduct, Michigan authority) stay.
+
+`attest-implication` is capped at 4. Those four are the accepted abortion and schools modified-to-commonality pairs. The planner labels every such action `accepted-bridge-role-pair`. That label is false for the six new clusters, which is why their arrows are omitted. `campaignPlanner.test.ts` asserts the new cause ids have commonality boards and no implication actions.
+
+Persona `set-belief` choices are still not applied. Delegated spending and a recurring monthly pledge are still not campaign actions. The Canada–US commonality is the weakest draft ("I'd be okay with saying that…").
+
+A wiped local chain needs `node scripts/fund-local-service-wallets.mjs` before cause pages list projects. See the pilot note.
+
+## What to do
+
+1. Evaluate each new cluster with the real implication-attester system prompt (`cause-assist/src/implicationCheck.ts` / `IMPLICATION_EVALUATOR_SYSTEM_PROMPT`). Designed yes: each modified → commonality. Designed no: each natural → commonality, reverse arrows, and the assimilation cluster against the accepted immigration statements. Record hashes, prompt fingerprint, model, decision, and rationale. Heuristic `implies: true` is not evidence. If credentials are missing, stop and say so.
+2. Run the routing check from the statements spec. Rewrite the wrong role if a designed-yes fails or a designed-no passes. The Canada–US commonality is the first candidate to rewrite.
+3. Only after an affirmative model decision, raise the batch-2 `attest-implication` target so those pairs are planned, and store evidence that is not the accepted-role stamp. Leave refused pairs in the decisions file.
+4. Regenerate the plan into a fresh output directory. Run `npx tsx --test test/campaignPlanner.test.ts` from `fake-data-generation/`.
+5. Execute locally only after the plan matches the evidence. Walk the new bridge boards in the browser. Do not call the batch complete while a required journey is missing. The remaining brainstorm families (low-carb, unwoke as an umbrella, Democracy 2.0, Islamic reform, permitting, home-school beyond the pilot) are still unauthored. Do not force triples onto overlaps that already share a concrete want.
+
+## Suggested skills
+
+- `thorough-tester` — designed-negative pairs and the attest-implication cap.
+- `browser-driving` — after a local execute, open one new bridge (commonality plus one modified board) and confirm the new clusters have no implication chip until step 3.

continuity/2026-10-16-network-societies-next-batch-handoff.md

@@ -0,0 +1,31 @@
+# Handoff: next network-societies batch
+
+Next session: author the first expansion batch of three or four families. Do not publish implication edges that have not been accepted by the attester. Do not wipe the local chain.
+
+## Read first
+
+- [`fake-data-generation/brainstorming/network-societies-campaign-plan.md`](../fake-data-generation/brainstorming/network-societies-campaign-plan.md) — sequence step 4, the fifteen-family table, and the statement quality gate.
+- [`fake-data-generation/campaigns/network-societies-v1-pilot.md`](../fake-data-generation/campaigns/network-societies-v1-pilot.md) — what the four-family pilot can and cannot show.
+- [`fake-data-generation/PLAN.md`](../fake-data-generation/PLAN.md) section F.
+- Branch `campaign`, not pushed. Latest commit `99fd0301` ("Fund the local alignment-trust bootstrap wallet."). Earlier on the same branch: `976eef50`, `c7648c02`.
+
+## What is already true
+
+The four-family pilot (Owen Sound checkout, Grey County civic life, Christian teaching, home learning) planned 40 users, 40 statements, 20 projects, and 207 writes. Execution and reconciliation were 207/207. Artifacts stay gitignored under `fake-data-generation/output/campaigns/network-societies-v1-pilot/`.
+
+Cause pages list a project only after the alignment-trust bootstrap has trusted the attester. A wiped chain leaves the generated bootstrap key with no ETH, the worker crash-loops, and every board says there are no aligned projects. `scripts/fund-local-service-wallets.mjs` now includes `ALIGNMENT_TRUST_BOOTSTRAP_PRIVATE_KEY`. On this chain that wallet was topped up, the worker admitted 8 attesters, and the class-host directory (`project-015`) shows under Fundable Projects on Christian teaching, civic life, and home learning. Project URL and board URLs are in the pilot note.
+
+The statement row label "Projects" means the count is zero. A real count reads "1 project".
+
+Implication candidates in the decisions file are still unpublished. Persona `set-belief` choices are still not wired. Delegated spending and a real monthly pledge are still not campaign actions.
+
+## What to do
+
+Add the next three or four families from the brainstorming table. Start with the accepted abortion bridge as a spotlight story, and prefer families that can reuse accepted statements where the wording already means the thing being said. Keep the four pilot families. One shared project on several boards is one contract and several alignments, which this pilot already proved.
+
+Do not plan over `output/campaigns/medium-realistic-v2/` or `output/campaigns/network-societies-v1-pilot/`. If the local stack is still up, leave it. A new execute needs a fresh output directory. If you wipe, run `node scripts/fund-local-service-wallets.mjs` before expecting cause pages to list projects.
+
+## Suggested skills
+
+- `thorough-tester` — when the new batch gets a plan, check the shared-project case and a board that only includes the statement through `alsoOnCauseIds`.
+- `browser-driving` — after a local execute, open the new boards and one pilot board that already lists the directory.