A principle without a mechanism is a poster. Every principle below carries its enforcement.
Immanis Praeses Niralamba Learning Labs is an independent research organisation in Bengaluru, founded in July 2025 as Nexus Learning Labs and incorporated as Immanis Praeses Niralamba Learning Labs Private Limited on 2 September 2026. We work on neuromorphic computation and machine cognition without an institution behind us. Everything published under the earlier name is this organisation’s work and is bound by this charter.
“We” means the Labs — the organisation, not a person. Today that is one researcher, who founded the lab unfunded and built it from nothing. Tomorrow it is the Labs’ directors, its employees and its named collaborators. Everyone who works here is bound by this document from their first day, and it does not bend for anyone’s convenience, including the founder’s.
To ground well-defined computational mechanisms in careful, reproducible science, and to report what we find truthfully — whether it flatters us or not.
Serious laboratories bind themselves to mechanisms, not slogans. A small lab has fewer people; it does not have a lower standard. A principle without a mechanism is a poster. Every principle below carries its enforcement.
This is Tier One — how we do science. Tier Two, covering capital, deployment, people and stewardship, is in preparation and will be published as part of this charter.
Each principle states the tension it resolves — the easy impulse, and the discipline we choose instead — followed by the mechanism that enforces it and the consequence when it fails.
Speed looks like the only advantage a small lab has. It is not. A wrong result published first costs more than a right result published second — it costs every result that follows it.
A finding is not a finding until it has survived independent repetition inside the lab. Every headline number is reported across all random seeds run, with its spread — never a best run. Where a claim rests on one dataset or one cohort, that dataset is named in the same sentence as the claim, with its size.
This lab has no departmental review, no second principal investigator, no institutional sign-off. We do not pretend otherwise. In place of review we make the work observable: analysis plans are written before the run, commit history is timestamped, and every dataset, seed and script behind a published number is archived on Zenodo with a permanent DOI on the day of release. Anyone may check our arithmetic; nobody has to take our word for it.
The release checklist, which no paper leaves without.
Named consequence: if a released claim fails our own later repetition, we publish the correction ourselves, in the same venue, within 30 days — before anyone asks, and whether or not anyone noticed.
A null result costs the same months as a positive one and returns no headline. That is precisely why so few are published — and why the literature any small lab builds on is thinner and more optimistic than it appears. We pay that cost deliberately, because the alternative is a body of work whose positive results nobody has reason to believe.
Every experiment that was run is reported: the ones that worked, the ones that did not, and the ones that unsettled something we had already published. Negative results are written to the same standard as positive ones — same detail, same archive, same DOI. We do not describe a finding as promising, encouraging or suggestive. We give the number, its spread and the sample it came from, and let the reader judge.
Where our later work contradicts our earlier work, we say so plainly and name the earlier paper. We have already done this once: a constant we had published was disproved by our own subsequent experiment, and we published the disproof ourselves. That is not a confession. It is the mechanism working as designed.
A repeatable observation with no theory behind it is reported as exactly that — repeatable, unexplained, with the number of independent confirmations attached. We do not retrofit an explanation to make a result publishable.
A conclusion revised because better evidence arrived is science working. Concealment is the failure, not correction.
Nulls enter the publication queue at the same priority as positives, not a backlog.
Named consequence: a result withheld because it was inconvenient is treated as misconduct rather than discretion, and the correction is published under the name of whoever withheld it.
A funded timeline and an honest result pull in opposite directions. Science run to a deadline produces the result that fits the deadline.
We do not advance to the next stage until the current one is solid. Quality governs what we claim — never whether we communicate. Scope is what flexes. Dates do not slip silently: if a result will not be ready, we say so before the date with a revised date and the reason, not after it with an apology.
Every project carries stage gates, and a gate is passed on evidence rather than calendar.
Named consequence: a date missed without advance notice is written up as a process failure regardless of whether the science was correct. The science being right does not excuse the silence.
When a mechanism resists, the cheapest move is to change the question — swap in an easier benchmark, narrow the claim quietly until it becomes true, or report the configuration that happened to work. Each of these produces a publishable paper and no knowledge.
We go into the difficulty rather than round it. Benchmarks are chosen before the run, not after the result.
Persistence applies to the problem, never to a claim the evidence has stopped supporting. When three independent attempts fail to reproduce a result, the result is retired, not rescued.
An attempt log recording every reproduction attempt with its date and configuration, successful or not.
Named consequence: at the third failure retirement is automatic. Nobody has to win an argument to make it happen, including against the founder.
Verification feels like delay at exactly the moment momentum is highest.
Nothing carrying this lab’s name goes out unverified. Code runs from a clean checkout on a machine that has never seen the project. Figures regenerate from archived data. Every number in an abstract traces to the script that produced it.
The release checklist, and same-day archival of the artifact with a permanent DOI.
Named consequence: anything released without a completed checklist is withdrawn and re-released, never patched quietly.
A demonstration persuades faster than a proof, and the gap between the two is invisible to everyone except the person who built it.
Every capability we assert travels with the artifact that demonstrates it — running code, an archived dataset, or a benchmark reproducible on named hardware. “Runs on commodity hardware” is not a claim; “runs on a Raspberry Pi 5 at these numbers” is.
A claim register mapping every public claim to the artifact supporting it.
Named consequence: a claim whose artifact cannot be produced on request is withdrawn within seven days.
The pressure to generalise is strongest exactly where the evidence is thinnest. A result from one cohort wants to become a result about people; a result on one architecture wants to become a result about spiking networks.
A finding from 84 subjects is a finding from 84 subjects. Population, hardware, dataset and sample size travel with the claim wherever it goes.
Scope conditions appear in the deck, the website and the investor memorandum, not only in the paper where reviewers would catch their absence. A claim we would qualify for a referee is qualified for everyone.
The claim register carries each claim’s boundary conditions alongside its artifact.
Named consequence: overreach in any public-facing document is corrected in the same channel, at the same prominence, within seven days.
Principle 1 states that we have no peer floor. This principle is what we do about it. The temptation is to let the absence of an auditor become the absence of an audit — nobody is keeping score, and nobody will catch our weaknesses for us.
We publish our limitations as a standing, dated, public list — the current status of every result, sample sizes, what has not been validated, and what we cannot yet explain. Limitations belong where a reader finds them before they ask, not in the discussion section of a paper they may never open.
The limitations list is reviewed on a fixed quarterly cadence, versioned publicly, and every prior version stays downloadable.
Named consequence: if an outside reader identifies a limitation before we have published it, it goes on the list with the date and with credit to the person who found it.
This lab uses computational tools, including AI systems, across its research and engineering. Their use is disclosed by name in the methods section of every paper in which they were used. Tools do not hold authorship. Responsibility for every claim made here rests with the human being who signed it.
It is the why. The how lives in the process document (the four phases and three gates), and the who-does-what in the pipeline document. The invariants — hard rules, language, glossary — live in the universal standards, whose principles block points here.
When a principle and a deadline conflict, the principle wins — and the deadline is renegotiated openly, never missed in silence. That is the entire point of writing them down before the pressure arrives.
This charter is reviewed annually, and whenever a principle is found to be unenforceable in practice. Every version is dated; prior versions remain publicly available. All changes are recorded here.
The principles above are publishable and complete on their own. Beneath them, this lab holds one more commitment that the large labs do not name: the work is done as a discipline of devotion — as sadhana, not for applause. Being right is a form of truthfulness. Honest reporting is a form of non-harm to the record. Perseverance is a form of steadiness. Method, held this way, becomes practice.
(The private inner meaning of this layer — and of the lab’s mark — is held in private files only, and never appears in any public-facing document, including this one.)
This charter is published under Creative Commons Attribution 4.0 International (CC BY 4.0). Any laboratory may copy it, adapt it and bind itself to it, with attribution. We would rather the mechanisms spread than stay ours.