12×2
A practical guide to deliberate digital choice

12×2

Build a deliberate information diet with feed audits, source maps, privacy controls, recommendation checks and repeatable browsing habits.

Verification notice. The recoverable archive preserved only the line “12x2 — Decide what you see on the internet” and an empty application shell. This new edition develops that verified idea as an independent educational resource; it does not claim that the former app or team still exists.

Orientation: turn a vague goal into a checkable decision

No person chooses every item in a feed. Platforms rank, recommend and repeat content using signals such as follows, clicks, watch time, hides, location, language and account history. The useful goal is therefore not total control. It is to make the system observable enough that deliberate choices can compete with automatic ranking.

A healthy information diet is built from sources, purposes, time boundaries and review rules. It distinguishes discovery from verification, interest from importance and privacy controls from content preferences. Clearing a history may change one signal while leaving follows, subscriptions, device identifiers or social connections intact.

12×2 treats digital choice as a small operating system. Twelve source slots and two review passes are a memorable model, not a magic number: first map what enters, then decide what deserves attention and what needs independent verification.

A reliable decision record names the entity being evaluated, the attribute that matters, the value or evidence observed, the date of that evidence and the action that follows. This sequence keeps an attractive page, familiar brand or confident recommendation from replacing verification. It also makes the process transferable: another person can inspect the same inputs and understand why the decision was made.

Definitions that keep the plan precise

Shared vocabulary is a control, not decoration. The definitions below separate concepts that are often collapsed in conversation. Use the final sentence in each card as an operational boundary.

Feed

A ranked sequence of items assembled for a person or context.

Decision use: It is an output of selection rules, not a neutral sample of everything available. Record the supporting evidence and its date instead of relying on a familiar label. If the evidence changes, revisit the downstream decision rather than preserving an obsolete conclusion.

Recommendation signal

An observed action or attribute used to estimate relevance.

Decision use: Clicks and watch time can mean curiosity, confusion or outrage rather than endorsement. Record the supporting evidence and its date instead of relying on a familiar label. If the evidence changes, revisit the downstream decision rather than preserving an obsolete conclusion.

Source

The identifiable origin responsible for a claim, report, dataset or creative work.

Decision use: A reposting account is a distribution channel, not automatically the original source. Record the supporting evidence and its date instead of relying on a familiar label. If the evidence changes, revisit the downstream decision rather than preserving an obsolete conclusion.

Information diet

The recurring mix of topics, sources, formats and attention allocated over time.

Decision use: Quality depends on diversity and purpose, not simply on volume. Record the supporting evidence and its date instead of relying on a familiar label. If the evidence changes, revisit the downstream decision rather than preserving an obsolete conclusion.

Verification

A separate process that checks identity, evidence, date, context and corroboration.

Decision use: Popularity and repetition cannot substitute for this process. Record the supporting evidence and its date instead of relying on a familiar label. If the evidence changes, revisit the downstream decision rather than preserving an obsolete conclusion.

Tracking

Collection or linkage of behavior across pages, apps, sessions or devices.

Decision use: Content controls and privacy controls operate on overlapping but different systems. Record the supporting evidence and its date instead of relying on a familiar label. If the evidence changes, revisit the downstream decision rather than preserving an obsolete conclusion.

Friction

A deliberate pause or extra step before opening, sharing or continuing.

Decision use: Small friction can interrupt automatic behavior without blocking legitimate access. Record the supporting evidence and its date instead of relying on a familiar label. If the evidence changes, revisit the downstream decision rather than preserving an obsolete conclusion.

Review pass

A scheduled inspection of inputs and outcomes.

Decision use: It turns one-time settings into a maintained system. Record the supporting evidence and its date instead of relying on a familiar label. If the evidence changes, revisit the downstream decision rather than preserving an obsolete conclusion.

Entity–attribute–value evidence map

The table converts the topic into inspectable records. An entity is the thing being evaluated; an attribute is the property that affects the decision; the value is the current observation; and the final column states why the value changes action. Empty values should remain visibly unknown instead of being filled with assumptions.

EntityAttributeValue or evidence to recordDecision consequence
SourceOwnershipNamed publisher, author or organization with a traceable pageSupports accountability and conflict review.
ClaimEvidence typePrimary document, data, direct observation, expert analysis or unsupported assertionDetermines the verification route.
ItemPublication dateOriginal date plus material update datePrevents old context from being mistaken for current fact.
FeedRanking basisFollowing, engagement, chronology, location, topic or paid placementExplains why an item appeared.
AccountPreference signalsFollows, likes, hides, subscriptions, searches and watch historyIdentifies controls that may alter recommendations.
BrowserPrivacy controlCookie policy, storage, tracking protection and permission stateLimits data flow rather than choosing source quality.
SessionPurposeResearch, communication, entertainment, transaction or habitProvides a stopping condition.
Source setPerspective coveragePrimary, local, specialist, opposing and synthesis sourcesReduces dependence on one distribution path.
ShareVerification statusUnopened, read, source-checked, corroborated or correctedMakes forwarding an explicit decision.
ReviewOutcome measureUseful actions, corrected beliefs, time spent and repeated low-value patternsEvaluates the diet by effect, not engagement alone.

Source: Ownership

The working value is Named publisher, author or organization with a traceable page. Supports accountability and conflict review. Verify the value at the point of use, preserve the source and date, and mark uncertainty explicitly. A proxy value may be useful for planning, but it must never be presented as a confirmed current fact.

Claim: Evidence type

The working value is Primary document, data, direct observation, expert analysis or unsupported assertion. Determines the verification route. Verify the value at the point of use, preserve the source and date, and mark uncertainty explicitly. A proxy value may be useful for planning, but it must never be presented as a confirmed current fact.

Item: Publication date

The working value is Original date plus material update date. Prevents old context from being mistaken for current fact. Verify the value at the point of use, preserve the source and date, and mark uncertainty explicitly. A proxy value may be useful for planning, but it must never be presented as a confirmed current fact.

Feed: Ranking basis

The working value is Following, engagement, chronology, location, topic or paid placement. Explains why an item appeared. Verify the value at the point of use, preserve the source and date, and mark uncertainty explicitly. A proxy value may be useful for planning, but it must never be presented as a confirmed current fact.

Account: Preference signals

The working value is Follows, likes, hides, subscriptions, searches and watch history. Identifies controls that may alter recommendations. Verify the value at the point of use, preserve the source and date, and mark uncertainty explicitly. A proxy value may be useful for planning, but it must never be presented as a confirmed current fact.

Browser: Privacy control

The working value is Cookie policy, storage, tracking protection and permission state. Limits data flow rather than choosing source quality. Verify the value at the point of use, preserve the source and date, and mark uncertainty explicitly. A proxy value may be useful for planning, but it must never be presented as a confirmed current fact.

Session: Purpose

The working value is Research, communication, entertainment, transaction or habit. Provides a stopping condition. Verify the value at the point of use, preserve the source and date, and mark uncertainty explicitly. A proxy value may be useful for planning, but it must never be presented as a confirmed current fact.

Source set: Perspective coverage

The working value is Primary, local, specialist, opposing and synthesis sources. Reduces dependence on one distribution path. Verify the value at the point of use, preserve the source and date, and mark uncertainty explicitly. A proxy value may be useful for planning, but it must never be presented as a confirmed current fact.

Share: Verification status

The working value is Unopened, read, source-checked, corroborated or corrected. Makes forwarding an explicit decision. Verify the value at the point of use, preserve the source and date, and mark uncertainty explicitly. A proxy value may be useful for planning, but it must never be presented as a confirmed current fact.

Review: Outcome measure

The working value is Useful actions, corrected beliefs, time spent and repeated low-value patterns. Evaluates the diet by effect, not engagement alone. Verify the value at the point of use, preserve the source and date, and mark uncertainty explicitly. A proxy value may be useful for planning, but it must never be presented as a confirmed current fact.

Decision matrix: match the method to the situation

A decision matrix prevents one preferred solution from being forced onto every case. Read across the row: identify the situation, protect the priority, collect the minimum evidence, take a bounded action and respect the stop condition.

SituationPriorityEvidenceActionStop condition
Breaking eventTimeliness plus uncertaintyPrimary alerts and multiple credible reportersUse a short watch window and expect correctionDo not share an unverified screenshot.
Health or safety claimEvidence qualityPublic-health guidance and qualified sourcesTrace the claim to current primary guidanceDo not substitute influencer confidence for evidence.
Product decisionComparable criteriaOfficial specifications, independent testing and owner reportsSeparate facts, tests and anecdotesDisclose affiliate or sponsor incentives.
Local issueProximity and recordsLocal authority documents and on-the-ground reportingPrefer sources accountable to the locationCheck date and jurisdiction.
Long-form learningDepth and retentionBooks, papers, courses and deliberate notesRemove feed interruptionsStop when the source does not support the learning goal.
EntertainmentEnjoyment within a boundaryChosen creators and a time budgetUse subscriptions or listsEnd when autoplay replaces choice.
Privacy-sensitive taskData minimizationBrowser settings, permissions and provider policyUse the least data requiredDo not confuse incognito mode with anonymity.
Contested claimCorroborationIndependent sources with different evidence pathsMap agreements and disagreementsDo not count copied articles as independent confirmation.

Breaking event

Protect Timeliness plus uncertainty by collecting primary alerts and multiple credible reporters. The bounded action is to use a short watch window and expect correction. The plan must pause when this condition appears: Do not share an unverified screenshot. Recording the pause is a successful control, not a failed task.

Health or safety claim

Protect Evidence quality by collecting public-health guidance and qualified sources. The bounded action is to trace the claim to current primary guidance. The plan must pause when this condition appears: Do not substitute influencer confidence for evidence. Recording the pause is a successful control, not a failed task.

Product decision

Protect Comparable criteria by collecting official specifications, independent testing and owner reports. The bounded action is to separate facts, tests and anecdotes. The plan must pause when this condition appears: Disclose affiliate or sponsor incentives. Recording the pause is a successful control, not a failed task.

Local issue

Protect Proximity and records by collecting local authority documents and on-the-ground reporting. The bounded action is to prefer sources accountable to the location. The plan must pause when this condition appears: Check date and jurisdiction. Recording the pause is a successful control, not a failed task.

Long-form learning

Protect Depth and retention by collecting books, papers, courses and deliberate notes. The bounded action is to remove feed interruptions. The plan must pause when this condition appears: Stop when the source does not support the learning goal. Recording the pause is a successful control, not a failed task.

Entertainment

Protect Enjoyment within a boundary by collecting chosen creators and a time budget. The bounded action is to use subscriptions or lists. The plan must pause when this condition appears: End when autoplay replaces choice. Recording the pause is a successful control, not a failed task.

Privacy-sensitive task

Protect Data minimization by collecting browser settings, permissions and provider policy. The bounded action is to use the least data required. The plan must pause when this condition appears: Do not confuse incognito mode with anonymity. Recording the pause is a successful control, not a failed task.

Contested claim

Protect Corroboration by collecting independent sources with different evidence paths. The bounded action is to map agreements and disagreements. The plan must pause when this condition appears: Do not count copied articles as independent confirmation. Recording the pause is a successful control, not a failed task.

Two repeatable workflows

The first workflow builds a decision from evidence. The second protects execution and handoff. A step may be skipped only when its output is genuinely irrelevant and the reason is recorded.

Run a feed audit

Run a feed audit decision flowFour-stage flow from evidence gathering to verification.Stage 1
Choose one feed and one representative seven-day per
Stage 2
Sample items at fixed points and record source, topi
Stage 3
Mark original sources separately from reposts, comme
Stage 4
Count concentration: repeated sources, repeated topi
Original decision diagram: each stage must leave evidence for the next stage.
  1. Choose one feed and one representative seven-day period; do not attempt to diagnose every platform at once. The step is complete when its evidence can be shown to the person responsible for the next decision.
  2. Sample items at fixed points and record source, topic, format, why it may have appeared and whether it served the session purpose. The step is complete when its evidence can be shown to the person responsible for the next decision.
  3. Mark original sources separately from reposts, commentary accounts and paid placements. The step is complete when its evidence can be shown to the person responsible for the next decision.
  4. Count concentration: repeated sources, repeated topics, repeated emotional triggers and items opened without a clear reason. The step is complete when its evidence can be shown to the person responsible for the next decision.
  5. Inspect available controls such as follow, unfollow, mute, hide, notification settings, chronology and history management. The step is complete when its evidence can be shown to the person responsible for the next decision.
  6. Change only a small set of signals so the effect can be observed rather than guessed. The step is complete when its evidence can be shown to the person responsible for the next decision.
  7. Repeat the sample after several days and compare source diversity, usefulness and unwanted repetition. The step is complete when its evidence can be shown to the person responsible for the next decision.
  8. Keep changes that improve the stated purpose; reverse those that merely make the feed quieter but less informative. The step is complete when its evidence can be shown to the person responsible for the next decision.

Verify before sharing

  1. State the exact claim in one sentence; a vague reaction cannot be verified. The step is complete when its evidence can be shown to the person responsible for the next decision.
  2. Open the source rather than relying on a screenshot, headline, excerpt or repost. The step is complete when its evidence can be shown to the person responsible for the next decision.
  3. Identify author, publisher, original date, update date and the evidence directly supporting the claim. The step is complete when its evidence can be shown to the person responsible for the next decision.
  4. Follow links to the primary document, dataset, statement or full context when available. The step is complete when its evidence can be shown to the person responsible for the next decision.
  5. Find at least one genuinely independent evidence path for consequential or contested claims. The step is complete when its evidence can be shown to the person responsible for the next decision.
  6. Check whether place, date, units, sample and definitions match the claim being made. The step is complete when its evidence can be shown to the person responsible for the next decision.
  7. Share with the appropriate uncertainty and source link, or do not share when evidence remains insufficient. The step is complete when its evidence can be shown to the person responsible for the next decision.
  8. If the claim later changes, correct the same audience rather than silently deleting context. The step is complete when its evidence can be shown to the person responsible for the next decision.

Worked scenarios: inputs, reasoning and failure checks

A dramatic clip dominates the feed

A short video appears repeatedly through several accounts. The audit first treats those accounts as distribution channels, then locates the earliest available upload and full-length source. The date and location do not match the current caption. Hiding one repost reduces duplication but does not resolve the claim; verification requires the original context and independent reporting. The user decides not to share, mutes two low-value aggregators and follows a local primary-information channel. The outcome is not ideological agreement. It is a cleaner evidence path and fewer duplicated signals.

Failure check: Ask which assumption, missing value or changed condition would reverse the decision. Then record the fallback before execution. This prevents a successful-looking result from hiding a broken premise.

Research becomes endless product browsing

The session begins with a need to compare three devices but drifts into dozens of recommendation videos. The planner writes five criteria, records official specifications, chooses two independent test sources and separates owner anecdotes into a different column. A thirty-minute boundary and a comparison table create a stopping rule. Affiliate links are noted as incentives rather than automatically rejected. At the second review pass, two models fail a non-negotiable criterion, so further content about them is removed from the queue. Attention returns to the decision instead of feeding the recommender.

Failure check: Ask which assumption, missing value or changed condition would reverse the decision. Then record the fallback before execution. This prevents a successful-looking result from hiding a broken premise.

Common failure modes and recoveries

Failure modes are most useful when paired with an observable signal and a small recovery. The goal is not to predict every problem; it is to detect a wrong path before it becomes expensive or irreversible.

Failure modeObservable signalRecovery
Following more sources without a purposeVolume rises but coverage gaps remain.Assign each source a role in a source map.
Treating engagement as preferenceOutrage clicks produce more unwanted items.Use explicit hide, mute and follow controls where available.
Clearing history as a complete resetOther subscriptions and identifiers still shape results.Map each signal and control separately.
Counting reposts as corroborationMany pages trace to one unsupported origin.Find independent evidence paths.
Using privacy mode as anonymityNetwork, account or site data may still be visible.Read the control’s actual scope and minimize data.
Opening every item to decide laterThe queue itself consumes the session.Use source and purpose filters before opening.
One permanent source listQuality and needs change over time.Schedule review passes and remove obsolete inputs.
Sharing before readingThe headline becomes a claim the source never made.Open, restate and verify first.

Following more sources without a purpose

The signal is: Volume rises but coverage gaps remain. Treat that observation as evidence that the current model is incomplete. Assign each source a role in a source map. After the correction, repeat the affected acceptance test; do not assume that changing the plan automatically repaired the outcome.

Treating engagement as preference

The signal is: Outrage clicks produce more unwanted items. Treat that observation as evidence that the current model is incomplete. Use explicit hide, mute and follow controls where available. After the correction, repeat the affected acceptance test; do not assume that changing the plan automatically repaired the outcome.

Clearing history as a complete reset

The signal is: Other subscriptions and identifiers still shape results. Treat that observation as evidence that the current model is incomplete. Map each signal and control separately. After the correction, repeat the affected acceptance test; do not assume that changing the plan automatically repaired the outcome.

Counting reposts as corroboration

The signal is: Many pages trace to one unsupported origin. Treat that observation as evidence that the current model is incomplete. Find independent evidence paths. After the correction, repeat the affected acceptance test; do not assume that changing the plan automatically repaired the outcome.

Using privacy mode as anonymity

The signal is: Network, account or site data may still be visible. Treat that observation as evidence that the current model is incomplete. Read the control’s actual scope and minimize data. After the correction, repeat the affected acceptance test; do not assume that changing the plan automatically repaired the outcome.

Opening every item to decide later

The signal is: The queue itself consumes the session. Treat that observation as evidence that the current model is incomplete. Use source and purpose filters before opening. After the correction, repeat the affected acceptance test; do not assume that changing the plan automatically repaired the outcome.

One permanent source list

The signal is: Quality and needs change over time. Treat that observation as evidence that the current model is incomplete. Schedule review passes and remove obsolete inputs. After the correction, repeat the affected acceptance test; do not assume that changing the plan automatically repaired the outcome.

Sharing before reading

The signal is: The headline becomes a claim the source never made. Treat that observation as evidence that the current model is incomplete. Open, restate and verify first. After the correction, repeat the affected acceptance test; do not assume that changing the plan automatically repaired the outcome.

Verification and handoff checklist

A checklist is evidence only when each item has a proof field. “Done” without a receipt, source, timestamp, comparison or visible test is a memory claim. The proof column below shows the smallest useful artifact.

CheckWhy it mattersMinimum proof
Every source has an assigned role.A list without purpose cannot reveal redundancy.Source map
Original and repost are separated.Distribution volume can imitate corroboration.Origin link
Dates and jurisdictions are visible.True information can be wrong for the current context.Context note
Paid placement is identified.Incentives affect selection and framing.Disclosure field
Recommendation controls are mapped.Different signals need different controls.Settings record
Privacy controls are scoped.A protection may cover tracking but not content ranking.Control description
Each session has a purpose.Without it, stopping is delegated to the feed.Session label
Consequential claims use primary evidence.Commentary cannot replace the underlying record.Evidence link
Sharing includes uncertainty.Confidence should follow evidence quality.Share note
A second review pass is scheduled.Settings decay and sources change.Review date

Every source has an assigned role.

A list without purpose cannot reveal redundancy. Preserve Source map with a date and owner. If the proof contradicts the plan, update the plan first; never rewrite the evidence to match the preferred conclusion.

Original and repost are separated.

Distribution volume can imitate corroboration. Preserve Origin link with a date and owner. If the proof contradicts the plan, update the plan first; never rewrite the evidence to match the preferred conclusion.

Dates and jurisdictions are visible.

True information can be wrong for the current context. Preserve Context note with a date and owner. If the proof contradicts the plan, update the plan first; never rewrite the evidence to match the preferred conclusion.

Paid placement is identified.

Incentives affect selection and framing. Preserve Disclosure field with a date and owner. If the proof contradicts the plan, update the plan first; never rewrite the evidence to match the preferred conclusion.

Recommendation controls are mapped.

Different signals need different controls. Preserve Settings record with a date and owner. If the proof contradicts the plan, update the plan first; never rewrite the evidence to match the preferred conclusion.

Privacy controls are scoped.

A protection may cover tracking but not content ranking. Preserve Control description with a date and owner. If the proof contradicts the plan, update the plan first; never rewrite the evidence to match the preferred conclusion.

Each session has a purpose.

Without it, stopping is delegated to the feed. Preserve Session label with a date and owner. If the proof contradicts the plan, update the plan first; never rewrite the evidence to match the preferred conclusion.

Consequential claims use primary evidence.

Commentary cannot replace the underlying record. Preserve Evidence link with a date and owner. If the proof contradicts the plan, update the plan first; never rewrite the evidence to match the preferred conclusion.

Sharing includes uncertainty.

Confidence should follow evidence quality. Preserve Share note with a date and owner. If the proof contradicts the plan, update the plan first; never rewrite the evidence to match the preferred conclusion.

A second review pass is scheduled.

Settings decay and sources change. Preserve Review date with a date and owner. If the proof contradicts the plan, update the plan first; never rewrite the evidence to match the preferred conclusion.

Frequently asked questions

Can I completely control a recommendation feed?

Usually not. You can influence inputs and use explicit controls, but ranking also depends on platform rules and signals you may not see. The practical aim is observability and better defaults.

Does clicking “not interested” erase tracking?

No. It may change a recommendation signal, while cookies, account history, device data or other tracking remain governed by separate controls and policies.

Is a chronological feed neutral?

It removes one ranking layer but still reflects whom you follow, what is published and platform moderation. Chronology is a useful control, not a complete representation of reality.

How many sources are enough?

Use roles rather than a magic count: primary evidence, accountable reporting, specialist interpretation, local context and a genuinely independent challenge where stakes justify it.

Are screenshots reliable evidence?

They can document what appeared, but they may omit source, date and context or be altered. Locate the original and verify the underlying claim.

What is the 12×2 method?

Use up to twelve intentional source slots for the current goal, then run two passes: one to map inputs and one to verify outcomes. Adjust the number to the task.

Does private browsing make me anonymous?

No. It mainly limits some local history persistence. Sites, networks, accounts and other systems may still observe activity depending on the setup.

Is this the legacy 12x2 application?

No. The archive did not preserve a usable product. This is a new educational edition based only on the recovered idea of deciding what you see online.

Sources and verification boundaries

These sources support general methods and public-record checks. They do not certify a private business, guarantee a current service or replace direct confirmation. Access dates and exact source pages should be preserved when a decision depends on them.

Editorial status: independently rebuilt on 2026-07-26. Material claims should be rechecked when laws, provider settings, public-health guidance, menus, business records or local disposal rules change.