An experiment should be able to make the programme change course. These earlier labs keep their measured outcomes, declared success gates and the observations still needed.
The record below is sourced to the earlier Observatory at its reviewed September 2026 commit. Its runnable packages remain in that repository. The current IDA Live instrument and claim atlas sit here in the Bounded Observer.
Open the 41-paper research library · Inspect the dated laboratory source
Access names the observable and source. State asks what it forgets. Action declares the intervention. Law and Geometry specify the model's additional assumptions. Evidence compares a frozen prediction with a strong baseline on untouched outcomes. The laboratory cards below report where that sequence reached and where it stopped.
These independent laboratory packages remain in their original repositories with code, protocols and attribution intact. Their status is stated at the dated source; upstream and online projects may have changed since then.
In 214 older female mice, 3,445 of 4,104 outcomes were resolvable (83.94%), below the 90% prerequisite. Added recovery history cut error by 0.84%, below the declared 5% gate, and won in three of five folds. The result does not demonstrate longer life or causal rejuvenation.
Two held-out flights tested whether an added control profile improves prediction. Flight 02 covered 50 cell lines and 92 exact drug/dose identities; retained error improved only 0.13% against a 10% gate, winning one of five folds. Both candidates failed their declared success gates.
A finite-passage and rescue-boundary model separates illustrative dynamics from independent source measurements and proposes a discriminating experiment. It is not a validated cancer treatment or evidence of clinical efficacy.
An earlier comparison used 360 published metabolic-assay values. The bounded adaptive variant improved held-out-dose error against tested simple baselines under its assumptions; hormesis and redox mechanisms remained unresolved. The earlier hosted exhibit may require owner access.
The continuation tool has a frozen diagnostic and source checks. Its first diagnostic tied conventional invalidation on 30 decisions and 25 checks; it did not establish a performance advantage.
Frozen contracts, official forecast captures, later outcomes, missing runs and revisions are kept as an evidence trail. Collection and integrity checks do not themselves prove an improved model.
Earlier measurement and feedback prototypes connect to IDA Live here. Synthetic demonstrations and existing-data analyses do not yet establish physiological validity or useful added prediction.
The earlier proof and certificate atlas preserves mathematical foundations, corrections and counterexamples. The current capstone and claim atlas carry the new programme's status and next checks.
The earlier station organised eleven rooms and permanent instruments. These are source and status pointers to that work, not live migrations into this static site. Browser simulations can run from the earlier app or their standalone packages; the central desk and saved records require its server and storage services. Read the room and admission contract ↗
Disturb a seeded reactor and compare recovery controllers with conventional baselines. A playable control model is not a general control advantage.
Source ↗A consent-based, 24-trial timing task records local response behaviour. It is a measurement interface, not an EEG or awareness inference.
Source ↗A six-state simulated beacon and a dimensionless reference reaction-pathway calculation make established quantum behaviour inspectable. Neither is evidence of a new physical law.
Source ↗The expedition interface connects public Oslo observation capture/replay, AI investigation access where configured, and a genome request planner. It does not supply a live Atlas score or autonomous discovery.
Expedition source ↗Experiments can be inspected and exported from the browser. A local record is a record of a session, not admitted public scientific evidence.
Source ↗The shared inquiry engine, optional AI interpreter, private saved printouts and source-linked encyclopedia need the earlier server, D1 storage and configuration. The API contracts constrain what calculations may run.
Operating contract ↗A synthetic world offers a separate protocol, evaluations and a browser explorer for testing what a reduced observation misses. The controlled world cannot establish a biological law.
Protocol and results ↗The perspective map links research lines and lets a declared simulation choose a next question from observations. Learning a better policy across projects remains a future test.
Source ↗Version 0.2.1 finds states compatible with one observation but different targets, then compares additional measurements. It was accepted for local conditional integration; physical access and empirical calibration remain open.
Reproduction package ↗Recovery: establish the future outcome and a resolvable cohort before evaluating added history again; the earlier 83.94% outcome resolution missed its own admission gate. Virtual Cell: choose an independent response set and compare a predeclared representation with stronger baselines before crediting added control-profile information. Cancer Lab: measure the proposed below-threshold state and the later passage in the same experimental preparation before claiming a finite rescue mechanism. IDA: require an independently measured return dynamic and matched feedback comparison.
These are next-test directions, not claims that the tests have run. Each source package retains its full protocol and revisions.
The earlier Observatory also recorded candidate fixes in Ray Serve, NeuroGym, Pertpy and Graphiti (timestamp companion). The upstream PRs are the source for current review and merge status; a local test result does not imply upstream acceptance. Read the earlier evidence summary ↗
These results are scoped to the stated datasets, interventions and software versions. The failed gates remain failures. A synthetic demonstration tests code inside its model; a source hash identifies bytes; a prospective mechanism needs its own calibration and independent comparison. The research library retains the wider paper and prediction record without promoting historical author claims into new validation.