Mediawiki Site Statistics

Files:	388	total
Pages:	1,857	total
Users:	1

metadata (normative)

Title:	Main_Page
Author:	Ralph B. Holland
version:	1.2.0
Affiliation:	Arising Technology Systems Pty Ltd
Contact:	ralph.b.holland [at] arising.com.au
Publication Date:	2025-12-23T13:00Z
Status:	Ongoing.

The preceding metadata is CM-defined and constitutes the authoritative provenance record for this artefact.

All fields in that table (including artefact, author, version, date and reason) MUST be treated as normative metadata.

The assisting system MUST NOT infer, normalise, reinterpret, duplicate, or rewrite these fields. If any field is missing, unclear, or later superseded, the change MUST be made explicitly by the human and recorded via version update, not inferred.

As curator and author, I apply the Apache License, Version 2.0, at publication to permit reuse and implementation while preventing enclosure or patent capture. This licensing action does not revise, reinterpret, or supersede any normative content herein.

Authority remains explicitly human; no implementation, system, or platform may assert epistemic authority by virtue of this license.

Overview

Frequent updates are made to this page.

For the time being this site only accepts verified white-listed agents. No anonymous access if permitted.

Geo blocking is also active: SG is ALWAYS denied due to weakened governance as a result of the enacted SG Copyright Act 2021 statutory exceptions for AI.

Singapore

Singapore has one of the clearest statutory exceptions for AI training and text/data mining in the world. The relevant law is the Copyright Act 2021, particularly Part 5, Division 8 (Computational Data Analysis) and the anti-contract-override provisions in Section 187.

Official statute

Singapore Statutes Online – Copyright Act 2021

1. Computational Data Analysis (CDA) Exception

Section 243: Definition of Computational Data Analysis

Section 243 defines "computational data analysis" broadly. It includes:

identifying, extracting, and analysing information using computer programs;
using copyrighted works as examples to improve computer programs.

The Act gives an explicit AI-related example:

using images to train a computer program to recognise images.

This is significant because the statute itself contemplates machine-learning training as a form of computational data analysis.

Section 244: Copying or Communicating for Computational Data Analysis

Section 244 creates the operative exception.

If certain conditions are met, a person may:

make copies of copyrighted works;
store or retain those copies;
communicate copies used for verification or collaborative research purposes.

The key conditions include:

The copying must be for computational data analysis or preparation for it.
The user must have lawful access to the source material.
The copies cannot be used for unrelated purposes.
Distribution of the copies is restricted to verification and collaborative research contexts.

The statute expressly says that "making a copy" includes storing or retaining the copy, which is important for AI training pipelines and dataset creation.

What counts as lawful access?

The Act gives examples:

Circumventing paywalls is not lawful access.
Accessing material in breach of database terms is not lawful access, subject to Section 187's contract-override rules.

In practice, this generally means:

publicly available webpages;
subscription content obtained through valid subscriptions;
licensed databases accessed according to lawful credentials;

are more likely to satisfy the access requirement than hacked, pirated, or bypassed content.

2. Prohibition on Contract Override

Section 187: Permitted Uses That May Not Be Excluded or Restricted

This is one of the most unusual provisions.

Section 187 states that contract terms are void to the extent they attempt to exclude or restrict statutory computational data analysis rights under Division 8.

The provision applies to terms that directly or indirectly prevent:

making copies;
supplying copies;
performing works;

where those acts would otherwise qualify as a permitted CDA use.

Why this matters for AI and web data Many websites publish terms such as:

"No AI training"

or

"No text and data mining"

Singapore's Section 187 potentially limits the effectiveness of such contractual restrictions where the activity otherwise falls within the statutory computational data analysis exception. The statute specifically identifies Division 8 (computational data analysis) as a category whose permitted uses cannot be contracted away.

The Act also contains an anti-evasion rule in Section 188, which can invalidate contractual choice-of-law clauses if they are used to circumvent Singapore's permitted-use protections.

3. Fair Use Singapore also has a general fair use provision.

Section 190: Fair Use

Section 190 provides that:

fair use of a copyrighted work is a permitted use;
fair use of protected performances is also a permitted use.

The fair-use analysis is flexible and considers factors such as:

purpose and character of the use;
nature of the work;
amount used;
effect on the market.

Unlike the CDA exception, fair use is not AI-specific. It functions as a broader safety valve for activities that may not fit neatly within another statutory exception.

4. Why Singapore Is Viewed as AI-Friendly

For AI developers, the combination of:

Section 243 (broad CDA definition),
Section 244 (copying for computational data analysis),
Section 187 (anti-contract-override protection),
Section 188 (anti-evasion through foreign-law clauses), and
Section 190 (general fair use),

creates one of the world's most permissive legal environments for training AI systems on lawfully accessed data. Commentators frequently compare it to Japan's similarly broad text-and-data-mining regime.

Practical summary for AI use of web data

Under the Singapore regime, the strongest argument for permissive AI training is:

Publicly accessible or otherwise lawfully accessed web content may be copied and processed for computational data analysis.

AI training can fall within the statutory definition of computational data analysis.

Contract terms attempting to prohibit otherwise-permitted computational data analysis may be void under Section 187.

Separate questions may still arise regarding unlawful access, database rights, privacy law, confidential information, output infringement, or other non-copyright claims.

So the key statutory provisions are:

Section Subject Relevance to AI
243 Definition of Computational Data Analysis Explicitly includes machine-learning style training examples
244 Copying or Communicating for CDA Allows copying and retention for AI/data-mining purposes when conditions are met
187 Permitted Uses That May Not Be Contractually Restricted Prevents contractual override of CDA rights
188 Choice-of-Law Evasion Rule Prevents contractual circumvention of permitted-use protections
190 Fair Use Additional flexible defense for transformative uses

Official source:

Singapore Copyright Act 2021 (Singapore Statutes Online)

Breaking Telemetry Projections

The following stripes are projections across the timeline from 2025-12-23 to the time of projection.

Note the large wall of unattributed User Agent strings. This cohort is comprised of many bots that don't declare they are automata. Some are AI bots by the way they walk the corpus. Look at the massive volume - this should not be accepted as just background noise. The author has also enacted geo-fencing to exclude regions that exhibit unattributed abuse or have overly permissive laws - such as SG.

the large volume of bytes downloaded from the corpus mediawiki server striped by admitted agent across time

the large number of hits served by the corpus mediawiki server striped by admitted agent across time

Some telemetry projections

Synthetic records inserted on counter reset (and outages) cause the vertical lines.
The drop traffic is a result of the iptables chain match to ipsets.
The SYN rate and detection rate show that the server is being hit with traffic, and that traffic is rotating through IP address pools respectively.
Geo filtering and bot Verification were introduced to treat anonymous traffic and masquerading traffic respectively on 2026-04-17.
A new class of filter was introduced to drop anonymous access to metadata on 2026-04-23.
Geo filtering opened up on publications corpus 2026-04-28T26:00Z
Late pass spikes above baseline are due to local server traffic such as backup and archiving.

Only pages marked category:public are accessible due to Category:Access Control.

Introduction

(new) The corpus is now operating under the:

(2026-03-20T02:07Z) — CM Mandate.

This corpus was proudly developed with ChatGPT free and the author then enthusiastically jumped to the paid tier adoption to obtain reliable file upload and Project Context.

The development of the CM-2 corpus involved sustained experimentation with large language model platforms. These experiments exposed both stochastic behaviour (addressed normatively by CM-2) and systemic platform variations. The resulting observations informed the governance analysis and protocol design documented in this corpus.

These stochastic variations are well understood, but in colloquial language they are often aggregated by semantically summaries with terms and phrases such as:

Groundhog day,
Alice in Wonderland,
Rabbit holes,
Conversational Continuation,
Delusion,
Fabrication,
Drift,
Parroting,
Authority Inversion,
Normative Drift.

These Stochastic variations can be analysed with the newly released

(2026-02-15T08:38Z) — Governance Axes as a Multi-Dimensional Lens

semantic scaffolding providing a taxonomy and orthogonal multi-dimensional graduated pressure instrumentation. Perhaps, one of the significant contributions of the corpus - because this scaffold can be used to analyse pressures in other domains, institutions and systems - and are not constrained to LLM.

Axes telemetry instrumentation can be activated within LLM systems - see:

(2026-02-23T08:35Z) — Telemetry-Induced Constraint Salience: An Empirical Study in LLM Behavioural Compliance.

The entry point was the 2025-12-12 with the first publication on 2025-12-17.

The corpus is structured as a constructive demonstration of governed, corrigible knowledge infrastructure.

Licensing precedence for all artefacts is anchored to the timestamp of the first recorded Mediawiki version (UTC) and recorded in the metadata as Publication date. Subsequent revisions do not alter licensing precedence and are tracked separately as version provenance; not all version provenance is required to persist. Mediawiki versioning is used mainly for bot updates.

All papers in this category are published on a rolling basis. For licensing and precedence purposes, each paper’s publication date corresponds to its first recorded public revision (UTC). Subsequent edits do not alter publication status.

Leads

See:

Category:Cognitive Memoisation for the landing page.
RAG-minimisation for a strategy

Main Page