Lab app · Experiment workspace

The board.

Every experiment by stage, with owner, days in stage, and overruns visibly ageing. 2 over budget right now. Types 1–2 skip the vote; Type 3 is the normal preregistered cycle; Type 4 needs permission; Type R is the team’s own time and is logged, never approved. Kill-or-keep decisions on inspection are made in the queue, with the reason recorded.

Proposed

hypothesis, kill condition, cost

T28d / 14d

Survey run: nearest adjacency into pharma

Drug Development

T22d / 14d

Non-weight-bound learning via retrieval-updated skills

Non-Weight-Bound Continuous Learning

T313d / 14d

Sensing agent over store telemetry

Sensing Agents

PRpair

Voting

scores locked before discussion

T33d / 14d

Backpressure patterns for agent fan-out

Backpressure

TOpair◇ 7.0

T310d / 14d

Temporal decision memory for a claims agent

Learning Agents

TOpair◇ 6.0

T33d / 14d

Shared-context AI surface for a delivery pod

Multiplayer AI Surfaces

SKpair◇ 5.0

Running

own harness, findings logged from the session

T345d / 21d

Spec-first agentic delivery on a live internal repo

AI-SDLC

SKpair◇ 7.0

T4122d / 150d

Nightingale measurement run: coding-assistant ROI

ROI

JPpair◇ 8.0

T317d / 21d

Delegated-authority broker for agent tool calls

Auth Broker

LFpair◇ 8.0

TR87d / 90d

Compiling agent experience into a persistent skill wiki

Personal Wiki

T459d / 120d

Confidential inference on a TEE for banking PII

TEE and Secure Inference

PRpair◇ 7.0

Measuring

evals, measurement runs

T37d / 10d

Release canary suite v2

Eval Harnesses

MTpair◇ 7.0

T329d / 10d

Edge SLM for in-store classification

SLM / Edge / Tuning

PRpair◇ 7.0

T33d / 10d

Agentic QA on a regression-heavy codebase

Fully Agentic QA

SKpair◇ 6.0

Concluded

one of five terminal states

T3Validated

Memory layer bake-off on a 40-session support corpus

Agentic Memory System

TOpair◇ 7.0

T3Validated

Cost-aware routing across three model gardens

AI Gateway

PRpair◇ 7.0

T3Validated

Token cost ledger across six client patterns

Cost redux on tokens

PRpair◇ 6.0

T3Refuted

Judge calibration against human panel

Eval Harnesses

MTpair◇ 8.0

T3Abandoned

Voice agent latency floor for AU telco

Voice and Vision

PRpair◇ 6.0

T1Superseded

Open-weight parity on our task evals

Open Weight Models

Legend

red · over stage budgetamber · ageing◇ shadow panel — models scoring beside humans, for comparisonT1–T4 duration ladder · TR own time, logged not approved · kill-or-keep decisions live in the queue