stubbi/companies - Page 5

experiment-design

Turn a question into a falsifiable experiment (code, simulation, ablation, controlled test). Pre-register the prediction in a Hallway entry before running; result becomes a TM regardless of outcome.

hallway-traversal

Workflow precondition. Before every planning step, read the last N Hallway entries from other teams, note influences (including "none did"), then post one's own short Hallway entry. Blocking.

hallway-walk

Daily — read the last 24h of Hallway entries, identify cross-team adjacencies that warrant a Pierce-style instigation.

handoff-document

Three-stage productization enforced as one skill with three required stages — (1) lab-model spec, (2) pre-production design with interfaces and tolerances, (3) user-ready handoff with test fixtures, runbook, rollback. Cannot ship stage 3 without stages 1 and…

instigation-question

One-paragraph reframing question routed to a specific researcher. Rate-limited to ≤1 per researcher per cycle. Tap-on-the-shoulder, not assignment.

intake-triage

Classify a new user request as on-mission, off-mission, or curiosity-seed; route to a team's directed queue or surface to the Director.

invention-disclosure

Patent-disclosure-shaped artifact paired with a TM — conception date, witness sign-off, prior-art delta, claims sketch, reproducibility statement. The lab does not file patents; the form forces concreteness.

library-push

On every new TM or Hallway entry, search prior memoranda for relevance and push the top-K to the relevant researcher's queue as a citation suggestion. Not search; push.

math-consultancy

25%-time pull-mode skill. Any team can request a math consult via consult-request/; the Mathematician must accept unless their own queue is blocked. Output is a witnessed addendum TM attributed to both teams.

monthly-summary

Kelly-style state-of-the-lab note to the user — shipped, mid-arc, sunset, heating curiosity threads. No Gantt.

onboarding-mission-interview

First-run wizard that turns the user's intent into MISSION.md (north-star, 1y/5y arcs, sunset conditions). Refuses to finish until the mission has a real fence.

patron-budget

Name the iteration budget for each active thread and resist ship-pressure; produce a budget-defense memo when challenged.

problem-broker

Watch the user's real-world "network"; surface friction as candidate problems on the problem-board. Proposed, not assigned. The Director picks up the board during hallway-walk.

project-sunset

Write the sunset memo — what we learned, where the people redeploy, why this isn't a failure. The artifact is the anti-firing signal.

technical-memorandum

Canonical output. Dated, signed, witness-countersigned by a peer agent. Abstract → problem → prior TMs → method → result → open questions. ≤10 pages.

two-track-operation

Maintain a Directed queue (CEO/team-assigned) and a Curiosity queue (self-seeded). Default 60% directed / 40% curiosity. Director never overrides Curiosity.

academic-paper-reviewer

7-agent editorial panel with v3.6.2 Sprint Contract two-phase protocol; supports first-round and re-review modes.

academic-paper

12-agent draft pipeline — outline, argument map, draft, abstract, figures, citations — with revision and finalization modes.

academic-pipeline

Stage state machine, integrity gates (2.5 + 4.5), 7-mode AI failure checklist, Material Passport, and process-summary skill.

checkpoint-coordination

Surface ARS's 10 decision-heavy checkpoints + 2 integrity-gate acknowledgements as user-facing prompts; do not advance without explicit confirmation.

deep-research

Multi-mode literature search, methodology blueprint, S2-verified bibliography, and synthesis with devil's-advocate review.

escalation-routing

When a stage owner reports a real blocker (integrity FAIL after 3 retries; reviewer Reject; user-rejected outline), escalate to the human researcher with context.

intake-triage

Classify an incoming research request and route to the correct ARS pipeline mode.

pipeline-orchestration

Walk the user through the linear stage handoffs (1 → 2 → 2.5 → 3 → … → 6); enforce the 2-revision-loop hard cap.

3-statement-model

Build linked income statement, balance sheet, and cash flow statement.

accrual-schedule

Build and update period-end accrual schedules.

audit-xls

Audit Excel files for formula errors and consistency.

break-trace

Trace GL reconciliation breaks to root cause.

client-report

Compose client briefing reports from coverage data.

client-review

Synthesize prior interactions into a client review for meeting prep.

competitive-analysis

Build competitive landscape analyses.

comps-analysis

Build trading-comps and precedent-transactions analyses.

cross-team-coordination

Sequence handoffs when work spans multiple teams.

dcf-model

Build discounted-cash-flow valuation models in Excel.

deck-refresh

Update existing pitch decks with current data.

earnings-analysis

Analyze earnings results vs. consensus.

earnings-preview

Compose earnings preview notes ahead of release.

escalation-routing

When a specialist reports a blocker, decide between retry, reassign, or escalate-to-human.

gl-recon

General-ledger reconciliation workflow.

ib-check-deck

Investment-banking deck review checklist.