IAM stands for Identity and Access Management. It is a complex domain which covers user accounts, authentication, authorization, roles, permissions and privacy. It is an essential pillar of the cloud stack, where users, products and security meets. The other pillar being billing & payments 💰.
This curated 
list expose all the technologies, protocols and jargon of the domain in a comprehensive and actionable manner.
The majority of the features making B2B users happy will be implemented by the IAM perimeter.
“Overly permissive AWS IAM policies that allowed `s3:GetObject` to `*` (all) resources”, led to \$80 million fine for Capital One. The only reason why you can't overlook IAM as a business owner.
A little *click-baity*, but author admit that “It depends on how much you trust them to 1. Stay in business; 2. Not jack up your prices; 3. Not deprecate services out from under you; 4. Provide more value to you in business acceleration than they take away in flexibility.”
In a Stanford class providing an overview of cloud computing, the software architecture of the platform is described as in the right diagram →
Here we set out the big picture: definition and strategic importance of the domain, its place in the larger ecosystem, plus some critical features.
Mozilla's security and access guidelines.
“This document divides cloud vulnerabilities into four classes (misconfiguration, poor access control, shared tenancy vulnerabilities, and supply chain vulnerabilities)”.
🆓 A Neo4J-based tool to map out dependencies and relationships between services and resources. Supports AWS, GCP, GSuite, Okta and GitHub.
Security is one of the most central pillar of IAM foundations. Here are some broad concepts.
A meta-critic of account management, in which features expected by the business clash with real user needs, in the form of user stories written by a fictional project manager.
In the same spirit as above, but broader: all the little things we overlook as developers but users really care about. In the top of that list lies account-centric features, diverse integration and import/export tools. I.e. all the enterprise customers needs to cover.
Sound advice to lay down the foundation of a future-proof IAM API.
On the concept of usernames as identifiers, and the complexities introduced when unicode characters meets uniqueness requirements.
💸 User login, user registration, 2FA and profile management.
🆓 Modern PHP user login and management framework.
The foundation of IAM: the definition and life-cycle of users, groups, roles and permissions.
An up to date set of recommendations for developers who are not cryptography engineers. There's even a [shorter summary](https://news.ycombinator.com/item?id=16749140) available.
Aims to bring together cryptography researchers with developers, focusing on uses in real-world environments such as the Internet, the cloud, and embedded devices.
“This paper has two major purposes. The first is to define some of the terms and concepts behind basic cryptographic methods, and to offer a way to compare the myriad cryptographic schemes in use today. The second is to provide some real examples of cryptography in use today.”
Foundational papers of cryptography.
“If you are using compare-by-hash to generate addresses for data that can be supplied by malicious users, you should have a plan to migrate to a new hash every few years”.
The whole authentication stack is based on cryptography primitives. This can't be overlooked.
“The phrase 'random number generator' should be parsed as follows: It is a random generator of numbers. It is not a generator of random numbers.”
“Do not assume that UUIDs are hard to guess; they should not be used as security capabilities (identifiers whose mere possession grants access)”. UUIDs are designed to be unique, not to be random or unpredictable: do not use UUIDs as a secret.
A benchmark of all identifier formats.
Funny take on the global aspect of unique identifiers.
Tokens, primary keys, UUIDs, … Whatever the end use, you'll have to generate these numbers with some randomness and uniqueness properties.
Quick overview of Google's Zero-trust Network initiative.
More companies add extra layers of VPNs, firewalls, restrictions and constraints, resulting in a terrible experience and a slight security gain. There's a better way.
💸 Identity & Access Proxy and Access Control Decision API that authenticates, authorizes, and mutates incoming HTTP requests. Inspired by the BeyondCorp / Zero Trust white paper.
💸 An identity-aware proxy that enables secure access to internal applications.
🆓 A cloud-native, identity-aware proxy and policy enforcement point that orchestrates authentication and authorization systems via versatile rules, supporting protocol-agnostic identity propagation.
Zero trust network security operates under the principle “never trust, always verify”.
🆓 A CNCF framework issuing short-lived, cryptographically-verifiable identities (SVIDs) to workloads across heterogeneous environments.
🆓 Minimalist Apple MDM server and library to enroll and manage the identity of Apple devices, inspired by MicroMDM.
Workloads, services and devices are principals too. They need identities to authenticate to one another and be granted access: the non-human counterpart to human user accounts.
An overview and comparison of all token-based authentication schemes for end-user APIs.
In the same spirit as above, but this time at the service level.
How-to in a nutshell: 1. Small root of trust; 2. TLS isn't enough; 3. Certificate-based tokens; 4. Crypto Auth Tokens (CATs). See the [slides](https://web.archive.org/web/20260306052223/https://rwc.iacr.org/2018/Slides/Lewi.pdf) for more details.
Protocols and technologies to verify that you are who you pretend to be.
A summary of [NIST Special Publication 800-63B](https://pages.nist.gov/800-63-3/sp800-63b.html) covering new password complexity guidelines.
The only way to slow down offline attacks is by carefully choosing hash algorithms that are as resource intensive as possible.
Recent scientific research calls into question the value of many long-standing password-security practices such as password expiration policies, and points instead to better alternatives such as enforcing banned-password lists and MFA.
This study recommend the association of: blocklist checks against commonly leaked passwords, password policies without character-class requirements, minimum-strength policies.
“Arbitrary low limits on length and character composition are bad. They look bad, they lead to negative speculation about security posture and they break tools like password managers.”
🆓 Shaming sites with dumb password rules.
🆓 A collection of password rules, change URLs and quirks by sites.
🆓 Specification defining site resource for password updates.
Good news: you're not stuck with a legacy password saving scheme. Here is a trick to transparently upgrade to stronger hashing algorithm.
The oldest scheme for auth.
The primary source of account hacks is password spraying (on legacy auth like SMTP, IMAP, POP, etc.), second is replay attack. Takeaway: password are insecure, use and enforce MFA.
An excellent walk-trough over all these technologies.
Key take-away: “schemes based on email and SMS are more usable. Mechanisms based on designated trustees and personal knowledge questions, on the other hand, fall short, both in terms of convenience and efficiency.”
“Our analysis confirms that secret questions generally offer a security level that is far lower than user-chosen passwords. (…) Surprisingly, we found that a significant cause of this insecurity is that users often don't answer truthfully. (…) On the usability side, we show that secret answers have surprisingly poor memorability”.
Google security team's data shows 2FA blocks 100% of automated bot hacks.
Same conclusion as above from Microsoft: “Based on our studies, your account is more than 99.9% less likely to be compromised if you use MFA.”
Probably on the verge of paranoia, but might be a reason to rate limit 2FA validation attempts.
Or why you should not rely on automated phone calls as a method to reach the user and reset passwords, 2FA or for any kind of verification. Not unlike SMS-based 2FA, it is currently insecure and can be compromised by the way of its weakest link: voicemail systems.
On the UX aspects of 2FA.
“When my 2FA code is entered incorrectly I'd like to know about it”.
Doesn't work because there are no cellphone towers at stations in Antarctica.
🆓 Open-source authentication and authorization server providing two-factor authentication and single sign-on (SSO) for your applications via a web portal.
🆓 Simple, secure and fast identity management platform.
Building upon password-only auth, users are requested in these schemes to present two or more pieces of evidence (or factors).
NIST has said that 2FA via SMS is bad and awful since 2016.
Definitive research project demonstrating successful attempts at SIM swapping.
TL;DR: don't. For details, see articles below.
Passwords are not the be-all and end-all of user authentication. This article tries to tell you why.
What are magic links, their origin, pros and cons.
Introduce WebAuthn as a standard supported by all major browsers, and allowing “servers to register and authenticate users using public key cryptography instead of a password”.
Or why passkeys are not worse than passwords.
Part of the FIDO2 project, and also known under the user-friendly name of passkeys.
Describe how authentication works with security keys, details the protocols, and how they articulates with WebAuthn. Key takeaway: “There is no way to create a U2F key with webauthn however. (…) So complete the transition to webauthn of your login process first, then transition registration.”
A practical guide to stay safe online and prevent phishing with FIDO2, WebAuthn and security keys.
🆓 Open-source implementation for security keys written in Rust that supports both FIDO U2F and FIDO2 standards.
Guide to using YubiKey as a SmartCard for storing GPG encryption, signing and authentication keys, which can also be used for SSH. Many of the principles in this document are applicable to other smart card devices.
Quick overview of the important stuff.
PKI lets you define a system cryptographically. It's universal and vendor neutral.
🆓 Acts as a broker between CAs and environments, providing a central portal for developers to issue TLS certificates with 'sane' defaults.
🆓 A swiss army knife for PKI/TLS by CloudFlare. Command line tool and an HTTP API server for signing, verifying, and bundling TLS certificates.
💸 A suite of network fingerprinting methods to facilitate threat-hunting and analysis.
Certificate-based authentication.
Get up to speed on JWT with this article.
Learn how to use JWT to secure your web app.
Compared to API keys, JWTs offers granular security, homogeneous auth architecture, decentralized issuance, OAuth2 compliance, debuggability, expiration control, device management.
A good recap of all JWT pitfalls.
On token invalidation.
And [why your "solution" doesn't work](http://cryto.net/%7Ejoepie91/blog/2016/06/19/stop-using-jwt-for-sessions-part-2-why-your-solution-doesnt-work/), because [stateless JWT tokens cannot be invalidated or updated](https://news.ycombinator.com/item?id=18354141). They will introduce either size issues or security issues depending on where you store them. Stateful JWT tokens are functionally the same as session cookies, but without the battle-tested and well-reviewed implementations or client support.
A signed JWT is known as a JWS (JSON Web Signature). In fact a JWT does not exist itself — either it has to be a JWS or a JWE (JSON Web Encryption). Its like an abstract class — the JWS and JWE are the concrete implementations.
The standards are either completely broken or complex minefields hard to navigate.
Allows you to decode, verify and generate JWT.
JSON Web Token is a bearer's token.
Drag and drop your auth. Add authentication, user management, and authorization to your app with a few lines of code.
A curated list of providers, services, libraries, and resources for OpenID Connect.
Explain how these standards work using simplified illustrations.
A reference article describing the protocol in simplified format to help developers and service providers implement it.
Starts with an historical context on how these standards came to be, clears up the inaccuracies in the vocabulary, then details the protocols and its pitfalls to make it less intimidating.
A nice summary card.
Got multiple legacy systems to merge with their own login methods and accounts? Here is how to merge all that mess by the way of OIDC.
OAuth2-friendly adaptation of the Central Authentication Service (CAS) protocol. You'll find there good OAuth user flow diagrams.
“Updates and extends the OAuth 2.0 Security Threat Model to incorporate practical experiences gathered since OAuth 2.0 was published and covers new threats relevant due to the broader application”.
How to identify and exploit some of the key vulnerabilities found in OAuth 2.0 authentication mechanisms.
“PKCE is used to provide one more security layer to the authorization code flow in OAuth and OpenID Connect.”
💸 Open-source OIDC & OAuth2 Server Provider.
🆓 Open-source Identity and Access Management. Supports OIDC, OAuth 2 and SAML 2, LDAP and AD directories, password policies.
🆓 A UI-first centralized authentication / Single-Sign-On (SSO) platform based. Supports OIDC and OAuth 2, social logins, user management, 2FA based on Email and SMS.
💸 Open-source Identity Provider similar to Keycloak.
💸 An Open-Source solution built with Go and Angular to manage all your systems, users and service accounts together with their roles and external identities. ZITADEL provides you with OIDC, OAuth 2.0, login & register flows, passwordless and MFA authentication. All this is built on top of eventsourcing in combination with CQRS to provide a great audit trail.
🆓 Simple and opinionated OpenID Connect server designed for self-hosters. Single static binary with flat-file or SQLite storage.
OAuth 2.0 is a delegated authorization framework. OpenID Connect (OIDC) is an authentication layer on top of it.
The old OpenID is dead; the new OpenID Connect is very much not-dead.
“OAuth is a protocol for authorization: it ensures Bob goes to the right parking lot. In contrast, SAML is a protocol for authentication, or allowing Bob to get past the guardhouse.”
“Even though SAML was actually designed to be widely applicable, its contemporary usage is typically shifted towards enterprise SSO scenarios. On the other hand, OAuth was designed for use with applications on the Internet, especially for delegated authorisation.”
Identity is hard. Another take on the different protocol is always welcome to help makes sense of it all.
SAML is arcane at times. A another analogy might helps get more sense out of it.
Not only weird, SAML is also insecure by design, as it relies on signatures based on XML canonicalization, not XML byte stream. Which means you can exploit XML parser/encoder differences.
On the technical and UX issues of single logout implementations.
A documented rant on the excessive pricing practiced by SaaS providers to activate SSO on their product. The author's point is, as a core security feature, SSO should be reasonably priced and not part of an exclusive tier.
Security Assertion Markup Language (SAML) 2.0 is a means to exchange authorization and authentication between services, like OAuth/OpenID protocols above.
Typical SAML identity provider is an institution or a big corporation's internal SSO, while the typical OIDC/OAuth provider is a tech company that runs a data silo.
Solution based on blind signatures. See the [slides](https://web.archive.org/web/20251203022343/https://rwc.iacr.org/2018/Slides/Mehta.pdf).
Not GCP's KMS, but the one at the core of their infrastructure. See the [slides](https://web.archive.org/web/20251203022343/https://rwc.iacr.org/2018/Slides/Kanagala.pdf).
💸 Secure, store and tightly control access to tokens, passwords, certificates, encryption keys.
💸 An alternative to HashiCorp Vault.
🆓 Editor of encrypted files that supports YAML, JSON, ENV, INI and BINARY formats and encrypts with AWS KMS, GCP KMS, Azure Key Vault, age, and PGP.
🆓 Audit Git repos for secrets.
💸 Searches through Git repositories for high entropy strings and secrets, digging deep into commit history.
Architectures, software and hardware allowing the storage and usage of secrets to allow for authentication and authorization, while maintaining the chain of trust.
Really basic overview of HSM usages.
AWS CloudHSM Classic is backed by SafeNet's Luna HSM, current CloudHSM rely on Cavium's Nitrox, which allows for partitionable "virtual HSMs".
🆓 Open-source project for building trusted execution environments (TEE) with secure hardware enclaves, based on the RISC-V architecture.
🆓 A specification and a reference implementation for the secure transfer, storage and processing of data.
A case study of vulnerability and exploitability of a HSM (in French, sorry).
HSMs are physical devices guaranteeing security of secret management at the hardware level.
A great introduction on the domain and its responsibilities.
A couple of real use-case to demonstrate the role of a TnS team.
Section dedicated to fraud management for billing and payment, from our sister repository.
Once you've got a significant user base, it is called a community. You'll then be responsible to protect it: the customer, people, the company, the business, and facilitate all interactions and transactions happening therein.
A critical intermediation complex driven by a policy and constraint by local laws, the Trust & Safety department is likely embodied by a cross-functional team of 24/7 operators and systems of highly advanced moderation and administration tools. You can see it as an extension of customer support services, specialized in edge-cases like manual identity checks, moderation of harmful content, stopping harassment, handling of warrants and copyright claims, data sequestration and other credit card disputes.
Is this paper aims at identity metasystem, its laws still provides great insights at smaller scale, especially the first law: to always allow user control and ask for consent to earn trust.
“To limit "friction" Uber allowed riders to sign up without requiring them to provide identity beyond an email — easily faked — or a phone number. (…) Vehicles were stolen and burned; drivers were assaulted, robbed and occasionally murdered. The company stuck with the low-friction sign-up system, even as violence increased.”
Customer name matching has lots of application, from account deduplication to fraud monitoring.
🆓 Wordlists for creating statistically likely usernames for use in username-enumeration, simulated password-attacks and other security testing tasks.
Some groups and content are illegal in some juridictions. This is an example of a blocklist.
💸 An open-source infrastructure for user identity and risk management.
🆓 Hunt down social media accounts by username across social networks.
Most businesses do not collect customer's identity to create user profiles to sell to third party, no. But you still have to: local laws require to keep track of contract relationships under the large Know You Customer (KYC) banner.
Why background check are sometimes necessary.
A really detailed analysis of suspicious contributor signups on OpenStreetMap. This beautiful and high-level report demonstrating an orchestrated and directed campaign might serve as a template for fraud reports.
🆓 A proposed method to “detects microcluster anomalies, or suddenly arriving groups of suspiciously similar edges, in edge streams, using constant time and memory.”
🆓 Open-source platform for visualizing and manipulating large graphs.
As an online service provider, you're exposed to fraud, crime and abuses. You'll be surprised by how much people gets clever when it comes to money. Expect any bug or discrepancies in your workflow to be exploited for financial gain.
“If you host an online community, where people can harm another person: you are on the hook. And if you can't afford to be on the hook, don't host an online community”.
“You can think about the solution space for this problem by considering three dimensions: cost, accuracy and speed. And two approaches: human review and machine review. Humans are great in one of these dimensions: accuracy. The downside is that humans are expensive and slow. Machines, or robots, are great at the other two dimensions: cost and speed - they're much cheaper and faster. But the goal is to find a robot solution that is also sufficiently accurate for your needs.”
Moderation of huge social networks is performed by an army of outsourced subcontractors. These people are exposed to the worst and generally ends up with PTSD.
A documentary on these teams of underpaid people removing posts and deleting accounts.
Any online communities, not only those related to gaming and social networks, requires their operator to invest a lot of resource and energy to moderate it.
“A concise definition of Threat Intelligence: evidence-based knowledge, including context, mechanisms, indicators, implications and actionable advice, about an existing or emerging menace or hazard to assets that can be used to inform decisions regarding the subject's response to that menace or hazard.”
🆓 An open source intelligence (OSINT) automation tool. It integrates with just about every data source available and uses a range of methods for data analysis, making that data easy to navigate.
“A collection of several hundred online tools for OSINT”: domain, IP, email, username and social-network lookups useful for unmasking fraud and abuse.
🆓 “Collect a dossier on a person by username from 3000+ sites”, useful for account enumeration and unmasking fraud or abuse.
Open standards, tools and methodologies to support threat intelligence analysis.
Tags to organize information on “threat intelligence including cyber security indicators, financial fraud or counter-terrorism information.”
Fingerprints can be used as a source of signals to identify bots and fraudsters.
At one point you will let users upload files in your system. Here is a [corpus of suspicious media files](https://github.com/corkami/pocs) that can be leveraged by scammers =to bypass security or fool users.
🆓 Collection of multiple types of lists used during security assessments, collected in one place. List types include usernames, passwords, URLs, sensitive data patterns, fuzzing payloads, web shells, and many more.
🆓 CSV database of email addresses used by threat actor in phishing kits.
🆓 Tools to scan phone numbers using only free resources. The goal is to first gather standard information such as country, area, carrier and line type on any international phone numbers with a very good accuracy. Then search for footprints on search engines to try to find the VoIP provider or identify the owner.
🆓 Homoglyphs is a common phishing trick.
How to detect, unmask and classify offensive online activities. Most of the time these are monitored by security, networking and/or infrastructure engineering teams. Still, these are good resources for T&S and IAM people, who might be called upon for additional expertise for analysis and handling of threats.
Reference all open-source captcha libraries, integration, alternatives and cracking tools.
💸 Still an effective, economical and quick solution when your company can't afford to have a dedicated team to fight bots and spammers at internet scale.
Starts with a rant on how the service is a privacy nightmare and is tedious UI-wise, then list alternatives.
🆓 An open-source solution to protect upstream resources from scraper bots.
💸 Captchas solving service.
Another line of defense against spammers.
Perfect for this use-case, as bloom filters are designed to quickly check if an element is not in a (large) set. Variations of bloom filters exist for specific data types.
Radix trees might come handy to speed-up IP blocklists.
The first mechanical line of defense against abuses consist in plain and simple deny-listing. This is the low-hanging fruit of fraud fighting, but you'll be surprised how they're still effective.
🆓 Consolidates reputable hosts files, and merges them into a unified hosts file with duplicates removed.
💸 Extensive collection of list for security, privacy and parental control.
🆓 Mozilla's registry of public suffixes, under which Internet users can (or historically could) directly register names.
🆓 CIDR country-level IP data, straight from the Regional Internet Registries, updated hourly.
Top-5000 most common domain prefix/suffix list.
NSA's [XKeyscore](https://en.wikipedia.org/wiki/XKeyscore) matching rules for TOR and other anonymity preserving tools.
Official French denylist of money-related fraud sites.
Useful to identified clients, catch and block swarms of bots, and limit effects of dDOS.
🆓 A list of temporary email providers. And its [derivative Python module](https://github.com/martenson/disposable-email-domains).
💸 Cross-language temporary (disposable/throwaway) email detection library.
💸 Verify an email address's reachability over SMTP without sending anything, catching typos, disposable domains and role accounts at signup.
A list of domains for disposable and temporary email addresses. Useful for filtering your email list to increase open rates (sending email to these domains likely will not be opened).
🆓 “A Ruby gem to check if the owner of a given email address or website is working for THE MAN (a.k.a verifies government domains).” Good resource to hunt for potential government customers in your user base.
This is a general list of words you may want to consider reserving, in a system where users can pick any name.
List of all the names that should be restricted from registration in automated systems.
🆓 Profanity blocklist from Shutterstock.
🆓 Uses a linear SVM model trained on 200k human-labeled samples of clean and profane text strings.
A collection of scientific studies of schemes providing privacy by design.
Data breach index.
Most privacy breaches were allowed by known vulnerabilities in third-party dependencies. Here is how to detect them by the way of CI/CD.
🆓 As the world becomes increasingly connected, the email marketing regulation landscape becomes more and more complex.
As the guardian of user's data, the IAM stack is deeply bounded by the respect of privacy.
Hashing is not sufficient for anonymization no. But still it is good enough for pseudonymization (which is allowed by the GDPR).
“Hashed email addresses can be easily reversed and linked to an individual”.
Explain the intuition behind [differential privacy](https://en.wikipedia.org/wiki/Differential_privacy), a theoretical framework which allow sharing of aggregated data without compromising confidentiality. See follow-up articles with [more details](https://desfontain.es/privacy/differential-privacy-in-more-detail.html) and [practical aspects](https://desfontain.es/privacy/differential-privacy-in-practice.html).
🆓 Context aware, pluggable and customizable data protection and PII data anonymization service for text and images.
As a central repository of user data, the IAM stack stakeholders have to prevent any leakage of business and customer data. To allow for internal analytics, anonymization is required.
Europe's reference site.
Best practices for developers.
A one-page summary of the above.
This paper demonstrates that, because of the lack of GDPR law enforcements, dark patterns and implied consent are ubiquitous.
List of GDPR fines and penalties.
The well-known European privacy framework
Covers all the important facets of user onboarding.
A huge list of deconstructed first-time user signups.
From Leaked Screenshots & A/B Tests.
A collection of tactics to increase the chance of users finishing the account creation funnel.
Some basic tips on the login form.
Create login forms that are simple, linkable, predictable, and play nicely with password managers.
To support both SSO and password-based login. Now if breaking the login funnel in 2 steps is too infuriating to users, solve this as Dropbox does: [an AJAX request when you enter your username](https://news.ycombinator.com/item?id=19174355).
“In this post we will look at the humble `<input>` element and the HTML attributes that will help speed up our users' two factor authentication experience”.
Summarizes the results from an academic study investigating the impact removing password masking has on consumer trust.
Notifications are hard. Really hard.
As stakeholder of the IAM stack, you're going to implement in the backend the majority of the primitives required to build-up the sign-up tunnel and user onboarding. This is the first impression customers will get from your product, and can't be overlooked: you'll have to carefully design it with front-end experts. Here is a couple of guides to help you polish that experience.
Ranking, popularity and activity status of open-source digital identity projects.
The source of all new features added to the IAM perimeter.
Also of note: [Identity Platform](https://cloud.google.com/identity-platform/docs/release-notes), [Resource Manager](https://cloud.google.com/resource-manager/docs/release-notes), [Key Management Service/HSM](https://cloud.google.com/kms/docs/release-notes), [Access Context Manager](https://cloud.google.com/access-context-manager/docs/release-notes), [Identity-Aware Proxy](https://cloud.google.com/iap/docs/release-notes), [Data Loss Prevention](https://cloud.google.com/dlp/docs/release-notes) and [Security Scanner](https://cloud.google.com/security-scanner/docs/release-notes).
Relevant keywords: [`IAM`](https://www.gcpweekly.com/gcp-resources/tag/iam/) and [`Security`](https://www.gcpweekly.com/gcp-resources/tag/security/).
All the latest accounts updates on DO.
Help makes sense of their huge service catalog. In the same spirit: [AWS In Plain English](https://expeditedsecurity.com/aws-in-plain-english/).
Describe all GCP products in 4 words or less.
Keep track on the activity of open-source projects and companies operating in the domain.
Cypherpunks overlaps with security. This wiki compiles information about the movement, its history and the people/events of note.
The header image is based on a modified photo by Ben Sweet.
[1]: Poison Study (Mira, 2007). [↑]
