The short answer

•     Opening the live production ranker is the least promising rung. Platforms are unlikely to do it, and code alone would not reveal the live data or objectives. The rungs that matter are the trace standard, the audit algorithms, reference rankers, and a viewer-owned data commons, and all four can be built without anyone’s permission.

•     Crowd-sourced data cannot replace platform logs, but it shows something logs cannot: what viewers actually experienced, across platforms, from the viewer’s side.

•     A contribution ledger can reward information value instead of viewing volume. Paying per view-minute would pay people to watch more, which is the opposite of the goal.

•     Suggested start: a small, consenting, language-stratified panel; viewer-side data only; three audit indices; the ledger running in shadow mode (recording standing, paying nothing) until value is actually realised.

O1 Objectives

The platform states what its ranking optimises

Yes

Medium–high

Useful, but self-reported and unverifiable alone

O2 Trace standard

Public schema and reason-code codebook that platforms, agents and viewer-side tools can emit or read

No

High

Enabling: makes every later rung comparable

O3 Audit algorithms

Public code that turns traces into divergence indices

No

High

Works on any trace source, including crowd-sourced

O4 Reference rankers

Transparent alternative rankers, used as baselines or as a viewer’s own choice

No

Medium–high

Gives a yardstick and a real alternative

O5 Audited access

Vetted third parties query platform logs under privacy controls

Yes, or law

Medium

High fidelity, but depends on law and goodwill

O6 Production ranker

Code and weights of the live system

Yes

Low

Limited alone: a snapshot goes stale; live data and objectives stay hidden

O7 Data commons

Viewer-contributed traces under tiered disclosure

No

Medium

The only rung showing the viewer’s side across platforms

L1 Capture

Records exposures and responses on the viewer’s device

Local-first: the raw trace stays on the device; only coarsened or aggregated forms leave. This mirrors the local-first, edge-when-useful, external-when-necessary ordering of the proximity network pilot.

L2 Schema and codebook

The Attention Trace Schema (ATS) and a shared set of reason codes

An open standard that platforms, agents and tools can emit

L3 Lineage graph

Links works, variants, exposures, rankers, objectives and responses

Lineage is the primary object, not the individual view. This parallels the Innovation Memory graph in the Problem Solver–Solution Seeker Exchange.

L4 Algorithms

Audit indices, attribution, reference rankers

Open code, versioned, reproducible

L5 Commons and ledger

Tiered data access, governance, incentives

Benefit follows contribution; control does not follow money

Exposure

Item ID, source-work ID, variant ID, surface (feed, search, share, agent), position, time bucket

Raw only on device or for the contributor; coarse time at public tier

Selection reason

Candidate source, ranker stage, active objective(s), signal families used (as codes, not values), ranker version hash

Codes may be public; underlying values never

Delegation

Who selected: viewer, subscription, platform ranker, social referral, or delegated agent (with agent version)

Public as aggregate

Response

Dwell bucket, completion, skip, share, return within a window

Raw only on device or for the contributor

Downstream

Whether the response changed the next selection; creator-side changes that followed, where the creator contributes

Vetted-researcher tier

Consent

Scope, purpose, expiry, consent language, withdrawal flag

Always attached to the record

Content lineage

Source work and its derivation chain (reel, full cut, dub, tutor); provenance marks where present

Public

A1

Trace minimiser (on device)

Reduces the raw trace to schema-allowed, coarsened fields before anything leaves

Over-coarsening destroys signal

A2

Private aggregator

Secure aggregation with differential-privacy noise and a minimum cohort size below which nothing is released

Noise budget exhausted by repeated queries

A3

Lineage linker

Joins exposures to source work and variant using content fingerprints and provenance marks

Fingerprints fail on AI-regenerated variants

A4

Commonality Index

For one source work, the average pairwise overlap of exposed segments across viewers (shared segments divided by all segments either saw). Low values mean the shared text has fragmented.

Segmenting long and short variants consistently

A5

Retention Pressure Index

Density of retention devices (hooks, cliffhangers, paywall gates) per unit time against a genre baseline, and how closely changes in that density track gains in exposure

Subjective labelling of “devices”; needs annotators

A6

Amplification Ratio

An item’s or class’s share of exposure in the audited feed divided by its share under a reference ranker

Reference ranker choice drives the answer

A7

Objective drift detector

Shift in the reason-code distribution for a stable cohort when a ranker version changes

Cohort not truly stable

A8

Decision-share decomposition

The share of selections made by the viewer, by subscription, by platform ranker, by social referral, or by a delegated agent. An empirical answer to “who decides what watching looks like”.

Attribution of ambiguous cases such as autoplay

A9

Attribution estimator

Coarse, interval-reported attribution of an exposure to factor families (prior history, popularity, social, paid promotion) using matched cohorts and reference-ranker comparison

Causal over-claiming; must report uncertainty

A10

Reference rankers

R0 chronological; R1 viewer-declared intent; R2 commonality-aware; R3 track-record-weighted. Viewers can compare them against their real feed, or adopt one.

Becoming a new default that is itself unaccountable

A11

Integrity checks

Duplicate and sybil detection, coordination and outlier detection, replay protection

False positives excluding legitimate minority viewers

Data donation

Exports their own watch history through a platform’s data-access feature and donates it

Lowest friction; legal right of access; long history

Retrospective; formats change; shows watched items, not what was skipped, and not why

Viewer-side capture

Runs an app or extension that records what appeared in their feed

Real-time; captures exposures, including skipped items

Needs high trust; may conflict with platform terms; device limits

Active annotation

Tags “why I think I saw this” or “this hook felt manipulative”

Ground truth for reason codes and for A5

Labour-heavy; subjective

Creator-side contribution

Shares variant tests and analytics

Shows how creators adapt to the feed (provocation 1)

Creators may not want methods exposed

Platform trace

Platform emits ATS records voluntarily or by mandate

Highest fidelity

Depends on consent or law

Institutional panels

A convening body recruits a stratified panel

Legitimacy and representativeness

Slow; needs a trusted convener

T0 Public

Aggregates and indices only

Released only above the minimum cohort size (a threshold gate on release)

T1 Vetted researcher

Query access to richer aggregates; no downloads of raw rows

Approved purpose, logged queries, privacy budget

T2 Contributor

The contributor’s own data in full, plus a personal “attention mirror”

Authenticated contributor only

T3 Device-only

Raw trace

Never leaves the device

Opportunity

A vantage point nobody else has: a viewer’s real feed, a creator’s variant tests, a platform’s logs

Viewers, creators, platforms

Marginal information: how under-represented the vantage is

Capability

Algorithms, indices, annotation, replication, legal and ethical review

Engineers, researchers, annotators, reviewers

Reproducibility, independent replication, reuse

Market

Recruitment, institutional adoption, reach to journalists and regulators, paying users of the indices

Conveners, universities, media bodies, distributors

Quality of what their recruiting or distribution produces, not head count

Data donation

Valid consent; schema-conformant; passes integrity checks; not a duplicate; cohort threshold met

Marginal information (under-represented language, region, platform, age band); persistence over time; reuse in findings. Volume is capped with diminishing returns.

Algorithm or index

Reproducible on a held-out test set; stays within the privacy budget

Reproducibility; independent replication; reuse; privacy cost (counts against)

Audit finding

Independently replicated; method and data lineage published

Replication; adoption by journalists, regulators or platforms

Annotation

Agreement with other annotators above a floor

Agreement; rarity of the case labelled

Recruitment and convening

Recruited contributors pass the data gate

Quality of resulting contributions

Platform or creator trace

Schema-conformant and verifiable

Fidelity; coverage; disclosure tier agreed

Contributor ID

Pseudonymous, non-transferable

Currency and type

Opportunity, Capability or Market; and the contribution type from the table above

Gate record

Which checks passed, when, under which version of the gate

Weights

Per-dimension scores with the method version used to compute them

Lineage pointers

Dataset versions, algorithm versions and findings this contribution fed into, or was fed by

Realisation log

Each realisation event that drew on this lineage and the share distributed

Algorithm Empire

Platforms keep selection and resist disclosure; machine-made variants tuned per ranker

Shadow measurement from the viewer’s side; evidence base for regulators and journalists. Exposure to platform-terms conflict is highest here.

Viewer’s Republic

Viewers or their agents choose form and ranker

Supplies the trace standard and reference rankers for viewer-chosen feeds. Agents must emit traces too, or the agent becomes the next unexamined medium (A8 tracks this).

Authored Revival

Fixed, human-made, signed works become scarce and valued

Records track record. A non-transferable contribution ledger is itself the kind of asset that cannot be rendered by AI, which is the point of provocation 7.

Gaming of indices

Published measures become targets and are optimised against

Publish methods; rotate probes; keep held-out audit sets; report ranges

Easier manipulation

Open ranker detail helps spammers

Open objectives and audit metrics rather than every parameter; delayed disclosure

Inference and re-identification

Traces reveal religion, politics, sexuality, health

Local-first capture, minimisation, differential privacy, release thresholds

Surveillance repurposing

The standard is used to monitor citizens rather than audit systems

No identity fields in the schema; device-only raw tier; published threat model; counsel

Poisoning and sybils

Fake or coordinated contributors move an index

Integrity checks; diversity-weighted standing; replication; deferred payout

Volume incentive

Rewards drift toward more watching

Weights on marginal information, with caps on volume

Commons capture

Platforms extract without contributing; or capital-style recentralisation

Non-transferable standing; no purchasable governance; reciprocal licence; defined council seats

Legal exposure

Platform terms, data protection, classification of shares

Counsel before the pilot; shadow-mode ledger first

Self-selection bias

Panel is not representative

Stratified recruitment; reweighting; declared limits

H1

Viewer-side traces from a modest stratified panel reproduce divergence patterns seen in any available platform-side or creator-side data, within a stated tolerance

The indices vary more with panel composition than with the feed itself

H2

Signed, localised reason codes are understood and used more than free-text explanations

Comprehension is no better in a two-language trial

H3

Ledger weights on marginal information produce a more diverse set of vantage points than flat per-contribution payment

A/B recruitment shows no difference in diversity or in data quality

H4

Published indices change behaviour of creators or platforms in useful ways, not only gaming

Observed change is explained by gaming alone

H5

Contributors keep participating without upfront payment

Attrition under shadow mode exceeds a threshold set in advance

0. Specify (months 0–3)

ATS v0.1; codebook draft in two languages; prior-art search; legal view; threat model

Schema reviewed by two external parties; legal view in hand

1. Pilot panel (months 3–9)

A few hundred contributors stratified by language and region; data donation plus annotation; one or two surfaces; A4–A6 computed; ledger in shadow mode

H1 and H2 results; shadow-mode retention measured (H5)

2. Open release (months 9–15)

Reference rankers; audit code under a reciprocal licence; researcher tier with query access

Independent replication of at least one finding

3. Ledger live and institutions (months 15–24)

First realisation events; council seated; engagement with platforms and regulators on O1 and O5

Review of ledger fairness and of capture risk

Findings

•     The three claims have different standing. Bypass (C1) is conditional: it works where the heavy charge sits in a layer that can be bypassed. Participation (C2) is partly supported: many modules can exist, but taking demand away from the integrated platform is the hard part. The cycle (C3) is untested: none of the cases examined shows a full differentiation-integration loop yet.

•     The “heavy charge” often sits outside the platform. On phones, app-store billing can add up to 30% on top of a platform’s own 8–12% fee (Patreon, per Sacra). Unbundling the platform without unbundling payment and distribution rules moves little money.

•     The chokepoint moves to discovery. A directory limited to awareness is still a ranker, so the ranking becomes the new medium. This is why audit and trace standards from the first paper matter here.

•     Impact is highest in loyal-audience niches, institutional and educational content, and regional-language long tails; lowest in mass video and live events. This is a judgment-based ordering, not a measurement.

•     P-AXIM space: 6 axes with 5 states each gives 15,625 configurations. The distribution is wide (11.9 of 13.9 possible bits of entropy). The most probable structure, the integrated platform persisting, holds only about 1.2%, and the five probable structures together hold about 2.1%. “Probable” here means relatively more likely, not likely.

•     The improbable five are mostly cycle breakers. They exit the market cycle through regulation, community trust, viewer-side agents, or a ledger-settled commons. That is why they are improbable and why they are worth cheap probes.

•     Correction to my earlier note. I said UPI makes direct payment between peers cheap. That holds for person-to-person transfers and small merchants, but a 0.4% merchant fee on payments above ₹2,000 has been notified and then postponed to after Diwali (secondary sources; see section 3.6). Rail cost is now a policy variable, not a given.

C1

Bypass. A consumer can connect to a provider peer to peer; the platform is restricted to awareness and a directory; the platform’s heavy charges are avoided.

The charge sits in a layer that cannot be bypassed (mobile OS billing, payment rails, trust and dispute handling), or the platform simply cuts its fee.

C2

Participation. Modularising the awareness-delivery value stack lets more players take part.

Players can join but cannot win demand; or modules cost more to coordinate than the platform bundle saves.

C3

Cycle. The unbundling forms an active gradient, with differentiation-integration cycles that align services with market segments.

No re-bundling occurs, or it occurs but follows no segment logic; or the gradient closes before a cycle completes.

Understanding

Separates what a platform fee pays for (awareness, delivery, trust, payment, ranking), so the bundle becomes legible and negotiable.

High

Low cost, and already visible in fee breakdowns such as Patreon’s 8–12% plus app-store charges.

Creation

New module businesses (directories, payment and trust modules, ranker authors) and new creator business models.

Medium

ONDC lists 22 buyer apps, 71 seller apps and 16 logistics providers (mid-2024), so module entry is real.

Transformation

Moves rent away from delivery toward discovery and trust.

Medium–high

Plausible, but the direction of migration is itself uncertain (section 3.6).

Communication

Open directory and trace standards let different players interoperate.

Medium

Depends on adoption of a shared schema.

Decision-making

Creators, viewers and regulators can compare stacks and choose modules.

Medium

Needs the audit indices from “Open Attention”.

AI augmentation

Viewer-side agents assemble modules on demand.

High potential, high risk

The agent can become the next unexamined platform (provocation 6).

Discovery

The gradient lens predicts where value migrates in each cycle phase.

Medium

Untested; see falsification below.

Cognitive

A shared vocabulary for unbundling: awareness, delivery, trust, payment, ranking, settlement.

Delivered by this paper

Generative

New structures that were not obvious before: the ten in Part B, including cycle breakers.

Delivered, unvalidated

Outcome

Lower creator fees, more participants, better discovery, measured in a real segment.

Not measured. No outcome evidence yet.

Creator margin

Lowest at scale; fees 8–12% plus OS billing on phones

Higher if payment and distribution rules allow

Higher still if settlement overhead stays low

Viewer convenience

Highest: one login, one bill

Lower: several modules, several failure points

Medium: the agent hides the modules but adds its own risk

Discovery quality

High for mass content, opaque

Variable; depends on directory scale

Potentially high and auditable, if rankers are open

Trust and disputes

Platform arbitrates

Weakest link: needs independent trust modules

Ledger and reputation help, but it is new

Delivery cost

Platform absorbs; priced into fee

Creator or CDN pays; peers can share load

Same as B

Participation breadth

Narrow: players are suppliers

Wide: modules compete

Widest, with credit for each contribution

Regulatory exposure

Platform bears it

Spread across modules; unclear liability

Highest uncertainty: ledger shares and agents

Resistance to capture

Low: the platform is the capturer

Medium: the directory can become the new platform

Depends on governance guardrails

Loyal-audience niche creators

4

4

4

4

3

0.25

Best fit: margin matters, audience follows the creator, direct billing is natural.

Audio and text (podcasts, newsletters)

3

2

5

4

5

0.19

Already close to this structure, so the marginal gain is lower. A reference case, not a frontier.

Education and institutional content

4

4

3

3

3

0.14

Institutions can host and pay; trust and rights matter; ledger fit is strong.

Regional-language long tail

5

3

3

4

2

0.12

High importance, but adoption is held back by discovery and payment friction.

Mass entertainment video

5

2

5

2

1

0.03

Delivery cost and discovery scale favour integration.

Live sports and news events

4

1

4

2

1

0.01

Latency, rights and scale make peer delivery weakest here.

Patreon (Sacra)

Fee of 8–12%; blended take rate estimated to have risen from about 7% (2020) to about 8.5%; iOS in-app rules can add up to 30%; Apple’s billing requirement affects about 4% of creators by 1 November 2026.

One platform. The rising take rate suggests no fee-cut response yet, but also that creators pay the cost. The largest charge sits with the app store, not the platform.

ONDC (Inc42, mid-2024)

About 12 million transactions in July 2024; 22 buyer apps, 71 seller apps, 16 logistics providers; ONDC’s own analysis put it at about 3% of Swiggy and Zomato food volumes nationally.

Data is two years old and partly self-reported; no breakdown by buyer app, so concentration cannot be assessed. Participation is wide, share is small.

Bluesky (ContentGrip)

The protocol lets anyone build custom feeds; the 2026 roadmap leans into this; 42 million users since 2024; daily active usage fell 40% year over year by October 2025 (Similarweb via Forbes).

Openness at the ranker layer exists, but openness has not produced growth. No data on custom feed usage.

UPI fee regime (Inc42, CAclubindia, Finnovate)

Person-to-person transfers and small merchants stay at zero fee. A 0.4% merchant fee on payments above ₹2,000 (₹300 cap) was reported as notified in September 2026, set for 15 October, then postponed until after Diwali with no new date; the Supreme Court declined a stay on 29 September.

Secondary sources, and they disagree on status (Finnovate describes it as still proposed). Check NPCI or RBI notices before relying on it.

Podcast RSS and directories; PeerTube

General knowledge only: audio and text already work through open feeds and directories; federated video exists but is small.

Searches returned titles only; not verified in this session.

C1 Bypass

Conditional

Fee components are separable; web checkout avoids OS billing in some markets; UPI person-to-person is free.

Largest charge may be at the OS layer; platform can cut fees or down-rank outbound links; rail fees may rise.

C2 Participation

Partly supported

ONDC and Bluesky show many module players can enter.

Entry has not become share (about 3% of food volume; falling activity).

C3 Cycle

Untested

Conceptually consistent with unbundling and re-bundling in other industries (general knowledge, not tested here).

No full loop observed in the cases studied; each shows one phase only.

1. Intrinsic parameter discovery

Decompose the awareness-to-delivery cycle into parameters that can vary independently.

Six axes (4.2)

2. Probabilistic axis differentiation

Give each axis five states and an analyst prior.

State priors (4.3)

3. Soft cross-consistency filtering

Pairwise factors reward or penalise state combinations; nothing is hard-excluded.

Consistency field (4.4, Appendix A)

3.5 Empirical grounding gate

Classify each structure as Known/Established, Emerging/Contested, or Grounded White Space. Grounding is a hard constraint on what may be claimed, not a weighted score.

Grounding classes (4.7, 4.8, 4.9)

4. Probabilistic recombination

Enumerate 15,625 configurations; joint probability = product of priors × consistency factors, normalised; surprisal = −log2 p.

Ranked space (4.6)

5. Dual-track output

Track 1: probable structures to monitor. Track 2: improbable but coherent structures worth cheap probes.

Sections 4.7–4.11

A. Awareness locus

Who tells the viewer what exists?

A1 dominant platform directory; A2 plural competing directories; A3 viewer-side agent assembles awareness; A4 social or peer referral; A5 public-interest directory

B. Delivery path

How does content reach the viewer?

B1 platform-hosted streaming; B2 creator-hosted with CDN direct; B3 peer-assisted hybrid; B4 pure peer-to-peer transfer; B5 local proximity or offline mesh

C. Trust and payment

Who handles money, identity and disputes?

C1 platform-bundled; C2 independent payment and trust modules; C3 public payment rail plus reputation; C4 ledger or reputation-network settlement; C5 free or gift commons

D. Ranking governance

Who decides the order of what is shown?

D1 opaque proprietary; D2 open-trace audited; D3 viewer-selected ranker market; D4 regulation-mandated alternative; D5 community curation

E. Cycle orientation

What does the differentiation-integration dynamic do?

E1 stays integrated; E2 unbundles then re-bundles by segment; E3 durable modular niche; E4 unbundle sprint then collapse; E5 interoperable bundle-of-bundles

F. Segment alignment

What principle divides the market into segments?

F1 mass versus niche; F2 language or region; F3 content form (audio, text, video, live); F4 price tier; F5 institutional versus individual

A1 Dominant platform directory

30%

35.9%

A2 Plural competing directories

22%

18.1%

A3 Viewer-side agent

20%

17.1%

A4 Social or peer referral

20%

20.5%

A5 Public-interest directory

8%

8.3%

B1 Platform-hosted streaming

42%

31.1%

B2 Creator-hosted + CDN direct

28%

35.8%

B3 Peer-assisted hybrid

14%

19.8%

B4 Pure peer-to-peer transfer

6%

2.1%

B5 Local proximity or offline mesh

10%

11.3%

C1 Platform-bundled

38%

35.0%

C2 Independent payment and trust modules

28%

29.5%

C3 Public rail + reputation

22%

25.4%

C4 Ledger or reputation settlement

8%

7.6%

C5 Free or gift commons

4%

2.5%

D1 Opaque proprietary

46%

48.3%

D2 Open-trace audited

14%

14.1%

D3 Viewer-selected ranker market

20%

14.8%

D4 Mandated alternative

12%

15.9%

D5 Community curation

8%

7.0%

E1 Stays integrated

38%

35.1%

E2 Unbundle then segment re-bundle

34%

39.6%

E3 Durable modular niche

12%

11.2%

E4 Unbundle sprint then collapse

10%

8.6%

E5 Interoperable bundle-of-bundles

6%

5.5%

F1 Mass vs niche

30%

30.9%

F2 Language or region

22%

21.7%

F3 Content form

22%

22.6%

F4 Price tier

12%

11.3%

F5 Institutional vs individual

14%

13.4%

Configurations

15,625

5 states on each of 6 axes

Entropy

11.9 bits (maximum 13.9)

Flat distribution: about 3,900 structures are “effectively live”. No structure is individually likely.

Most probable structure

1.16%

The integrated platform persists (P1)

Five probable structures combined

2.1%

Selected as the top five that differ in at least three axes

Improbable selection rule

One per awareness locus

The most self-consistent structure (net factor at least 1) in the bottom decile of probability; for the dominant-platform locus, no structure qualified in the bottom decile, so the pool widened to the bottom quarter

Sensitivity (60 runs; each prior multiplied by lognormal noise, σ = 0.3)

P1–P2: top 1% in 100% of runs; P3: 92%; P4: 57%; P5: 55%

The first three probable structures are robust; P4 and P5 are borderline. The improbable structures stay in the bottom quarter in 95–100% of runs (I1, the dominant-platform variant, only 52%).

Structure

One platform runs the directory, hosting, payment, trust and ranking. Segmentation exists only as fee and feature tiers by audience size.

Path

Viewer opens the platform → platform ranker proposes → platform streams → platform bills and handles disputes → platform pays the creator minus its take.

Cycle dynamics

The cycle is suppressed. Differentiation happens inside the platform as product tiers, and the platform absorbs modular entrants before they scale (fee changes, features, link restrictions). The gradient stays open only in the tail, and the value locus is the ranker plus demand aggregation.

Segments

Mass entertainment, live events, anything where convenience and scale dominate.

Participants

Few. Third parties are suppliers and tools, not co-owners.

Ledger role

Negligible. A creator-side shadow ledger could record traced credit without settling anything.

Conditions

Network effects hold; OS rules keep favouring platform billing; fees remain tolerable.

Failure modes

Take rate rises past the point where creators with loyal audiences leave; regulation forces unbundling.

Grounding

Known/Established. Patreon’s 8–12% fee, blended take rate rising from about 7% to about 8.5%, and OS charges of up to 30% (Sacra).

Signals

Stable or rising take rates; outbound-link restrictions; no growth in independent payment modules.

Half-life call

Long, for as long as network effects hold.

Structure

The platform remains where people find things, but creators host content on their own sites or CDNs and take payment through independent modules. Ranking stays opaque. Unbundling begins in the middle of the audience-size range.

Path

Viewer finds the work on the platform → click-through to the creator’s own domain or app → independent payment → delivery from the creator’s host or CDN → platform sees only awareness metrics.

Cycle dynamics

(1) Differentiation: mid-size creators with loyal audiences move payment and delivery out. (2) Friction: several logins, failed payments, disputes. (3) Re-integration: creator-commerce suites bundle hosting, payment and a small directory for that tier. (4) Platform response: down-rank or restrict outbound links, or lower the fee for those who stay. The platform’s lever is the opaque ranker, which limits the gradient.

Segments

Mid-tier creators, courses, newsletters, premium communities.

Participants

Payment modules, hosting and CDN providers, creator-commerce suites, analytics providers.

Ledger role

Could credit the directory for awareness and the creator for conversion, as a referral share in place of a flat cut.

Conditions

Platform tolerates outbound links; OS billing rules do not force in-app purchase; payment rails stay cheap for the ticket size.

Failure modes

Outbound down-ranking; app-store rules; payment disputes; piracy of creator-hosted files.

Grounding

Emerging/Contested. Patreon’s web-checkout workaround against Apple’s billing rules and changes (Sacra).

Signals

Share of creators using off-platform checkout; platform policy on outbound links.

Half-life call

Medium. Depends on platform tolerance.

Structure

The platform directory dominates awareness; creators host; payment runs on a public rail with a reputation layer. The platform stays integrated at the top, and the unbundling is segmented by language and region where local creators and local rails dominate.

Path

As in P2, but payment is a rail-native intent or QR flow and trust leans on the rail’s own dispute mechanism and a reputation score.

Cycle dynamics

The gradient is controlled by rail pricing. Person-to-person transfers and small merchants at zero fee keep the gradient open for micro-creators. A 0.4% merchant fee above ₹2,000, if it takes effect, narrows it for larger tickets. The rail operator is effectively a valve on the cycle, which is unusual.

Segments

Regional-language creators, local services, small-ticket courses and memberships.

Participants

Rail operator, banks and payment apps, regional creators, local resellers and agents.

Ledger role

A natural settlement companion: shares settle over the rail.

Conditions

Rail policy stays favourable to small merchants; directory supports regional languages; payment flow is low-friction.

Failure modes

Fee regime change; fraud; rail concentration; platform blocking external payment links.

Grounding

Emerging/Contested. UPI fee status is disputed in the sources (notified and postponed, per CAclubindia; proposed, per Finnovate). Verify with NPCI or RBI.

Signals

Final status and date of the merchant fee; small-merchant threshold (reported as up to ₹1 lakh a month); creator uptake of rail-native checkout.

Half-life call

Medium, and policy-dependent.

Structure

Many directories compete, including a viewer’s own; viewers choose rankers; creators host; payment and trust are independent modules. Each content form (audio, text, video, live) settles into its own stack and re-bundles differently.

Path

Viewer’s chosen directory and ranker → pointer to creator’s host → payment module → delivery.

Cycle dynamics

Audio and text sit closest to this already (open feeds plus directories; general knowledge). Video lags because delivery cost and discovery scale favour integration. Re-bundling happens through “player apps” that bundle directory, payment and playback for one form.

Segments

By content form.

Participants

Directory operators, ranker authors, player apps, payment modules, hosts.

Ledger role

Settles shares among directory, ranker, host and payment; the trace standard from “Open Attention” lets rankers be compared.

Conditions

Open feed formats; interoperable payment; a market for rankers; a trace standard.

Failure modes

Directory fragmentation; one directory becomes the de facto platform; discovery quality below the integrated alternative; spam.

Grounding

Known/Established for audio and text (general knowledge, not re-verified); Emerging for video. Bluesky shows ranker-level openness exists, yet daily activity fell 40% year over year by October 2025: openness has not meant adoption.

Signals

Number of directories with real traffic; share of listening or reading through non-platform directories; ranker switching rates.

Half-life call

Medium to long for audio and text; uncertain for video.

Structure

Awareness moves out of platform feeds into messaging groups, communities and creator-to-viewer referral. Delivery and payment remain bundled in the platform. Only awareness unbundles. Segments divide by content form.

Path

Friend or group shares a link → platform plays it → platform bills.

Cycle dynamics

Awareness is the cheapest layer to move, so it differentiates first. The platform re-integrates by building messaging and community features or by treating shared links as ranking signals. Delivery stays integrated throughout.

Segments

Short video shared in groups; long-form shared by community recommendation.

Participants

Community admins, messaging apps, curators.

Ledger role

Referral lineage can be traced and curators credited; it is a natural first use of a shadow ledger.

Conditions

Messaging dominance; link previews that carry enough information; platform tolerance.

Failure modes

Misinformation; low quality; privacy leakage; the platform capturing the messaging layer.

Grounding

Known/Established in practice (messaging-driven discovery), not re-verified in this session.

Signals

Share of plays arriving from shared links versus feed recommendation.

Half-life call

Short to medium. Its probability comes from the priors, not from reinforcing factors.

Structure

The dominant directory is obliged by regulation to expose alternative rankers and interoperate. A low-price or free tier is delivered through local hubs; a premium tier stays on the platform. Settlement uses a ledger.

Path

Regulated directory announces an item → a local node (shop, campus, telecom edge) caches it → viewers fetch locally at low or zero data cost → ledger records node, directory and creator contributions → the premium tier is served by the platform.

Cycle dynamics

The cycle runs by price tier. The low tier is modular and the high tier integrated. Re-bundling is by hub operators such as telecom edges, which connects to the proximity network pilot.

Segments

Data-poor viewers; price-sensitive education and entertainment.

Participants

Telcos, local hub operators, regulators, creators.

Ledger role

Central: it settles among nodes, directory and creators.

Why improbable

Needs a strong mandate, a tolerant platform and ledger settlement at scale.

Failure modes

Piracy; moderation burden shifting to nodes; mandate watered down.

Grounding

Provisional white space (whole); components emerging. Regulation and local caching exist in parts; the whole was not found in this light search.

Stakeholder value; externalities

High for data-poor users. Externalities: rights leakage and moderation gaps at hubs.

Cheap probe

A campus or telecom-edge cache pilot for one regional-language education catalogue, with traces recorded in shadow ledger mode.

Half-life call

Long if mandated; fragile otherwise.

Structure

Community libraries, universities and volunteers share regional-language knowledge peer to peer, with no payment. Several directories compete; a shared interoperability layer links them. Segments divide by language.

Path

Viewer searches a regional directory → peers or institutional seeds transfer directly → no charge → trace recorded on the viewer’s device.

Cycle dynamics

There is little cycle. The system is a stable modular commons that differentiates by language community. Integration occurs only at the interoperability layer (shared codebook or schema).

Segments

Regional-language long tail; public-domain and educational content; oral and folk archives.

Participants

Libraries, universities, community radio, volunteers.

Ledger role

Standing without money: recognition and access, not payout.

Why improbable

No money means weak incentives; pure peer transfer has little tooling for quality and rights.

Failure modes

Attrition; quality decay; no sustainable funding.

Grounding

Known in parts; whole not found. Peer swarms, libraries and community media exist separately.

Stakeholder value; externalities

High for language preservation; risk of unvetted content.

Cheap probe

One language, one institution, one content class (for example, public-domain textbooks), measuring seeding persistence over six months.

Half-life call

Long if anchored by an institution.

Structure

A viewer’s own agent assembles awareness from many directories, negotiates direct transfers with institutional providers (universities, public bodies) that fund the content, and applies the viewer’s chosen ranker.

Path

Viewer states intent → agent queries plural directories → agent selects a provider → peer transfer → agent logs a trace.

Cycle dynamics

The integrator moves to the viewer side. Directories differentiate; the agent re-integrates for each viewer. The cycle is therefore about the agent layer, which echoes provocation 6: the agent becomes the medium, but under the viewer’s control and with an audit trail.

Segments

Institutional versus individual: public knowledge, research and learning.

Participants

Agent developers, institutions, ranker authors.

Ledger role

Agents are a contributor class; their logs feed audit and settlement.

Why improbable

Needs trustworthy agents, institutions willing to fund, and a market for rankers.

Failure modes

The agent vendor becomes the new platform; prompt injection; provider spam.

Grounding

Provisional white space (whole); components emerging (agents are new; viewer-chosen rankers exist at the protocol level).

Stakeholder value; externalities

High for viewer agency. Externality: concentration of agent vendors.

Cheap probe

A prototype agent that queries two open directories and logs reason codes for each choice, tested with a small panel.

Half-life call

Short and uncertain.

Structure

Trusted communities (professional, alumni, cooperative, local institutional) curate and share content among themselves, with no payment and no platform. The community is the platform; trusted curators replace rankers.

Path

Curator posts → members fetch from each other or a local seed → no price.

Cycle dynamics

Durable. There is no re-bundling because community trust substitutes for platform services. The cycle appears only when a community outgrows the radius where members know the curators; then it splits (differentiates) instead of integrating.

Segments

Institutional and community content.

Participants

Community leaders, curators, local hosts.

Ledger role

Non-transferable standing for curators; no settlement.

Why improbable

Small reach, no monetisation, relies on voluntary curation.

Failure modes

Echo chambers; misinformation inside trusted circles; exclusion.

Grounding

Known/Established in analogue and in messaging-group forms; digital peer-transfer variant known in parts.

Stakeholder value; externalities

High trust, low reach. Externality: closed groups hide content from scrutiny.

Cheap probe

One cooperative or alumni body sharing a curated library by peer transfer, tracking participation and curator load.

Half-life call

Long, but small.

Structure

A public-interest body (library, broadcaster or university consortium) runs the directory. Content moves peer to peer. Peers earn standing for seeding and curation. A ledger settles contributions along lineage, and price tiers (free, low-cost, premium) are set by contributors.

Path

Contributor registers a work in the directory with its lineage → viewer discovers it → seeds transfer it → the ledger records exposures and contributions → realised value is shared along lineage.

Cycle dynamics

A slow cycle in a durable niche. The main risk is capture of the directory by capital, which recreates the recentralisation seen in many token-governed organisations. Guardrails from the Contribution Ledger Model apply: non-transferable standing, fixed share percentage, no secondary market, benefit share without control.

Segments

Price tiers across educational, cultural and public content.

Participants

Directory body, seeders, curators, creators, institutions.

Ledger role

Central. This is the closest of the ten to the “Open Attention” commons: the directory is the commons and the ledger is the settlement.

Why improbable

Needs a trusted convener, a working ledger, and legal clarity on shares and payments.

Failure modes

Cold start; classification of shares; bots; governance capture.

Grounding

Provisional white space (whole). Public directories and peer transfer exist; the ledger component is your own Contribution Ledger Model, with lineage-traced deferred compensation at Quirky as the cited precedent. Full prior-art search required before any claim.

Stakeholder value; externalities

High for contributors; externalities include legal exposure and uneven weighting across countries.

Cheap probe

Run the ledger in shadow mode inside the Open Attention pilot panel and test whether participation holds without payment (hypothesis H5 of the first paper).

Half-life call

Long if governance holds.

P1

Integrated Platform Persists

Known

×4.41

High for viewers; low for creators

Rent extraction; opacity

Long

P2

Platform Directory, Creator-Hosted Delivery

Emerging

×4.56

High for mid-tier creators

Piracy; payment disputes

Medium

P3

Rail-Native Direct Payment, by Region

Emerging

×3.15

High for regional creators

Fraud; policy dependence

Medium

P4

Plural Directories, Form-Specific Stacks

Known / emerging

×4.26

High for viewers who choose

Spam; fragmentation

Medium–long

P5

Social-Referral Awareness

Known

×0.96

Medium

Misinformation; privacy

Short–medium

I1

Directory as Public Utility, Mesh Delivery

Provisional white space

×1.82

High for data-poor viewers

Rights leakage; moderation gaps

Long if mandated

I2

Regional Gift Commons

Known in parts

×1.56

High for language preservation

Unvetted content

Long if anchored

I3

Agent-Brokered Peer Exchange

Provisional white space

×1.50

High for viewer agency

Agent vendor concentration

Short, uncertain

I4

Congregation Networks

Known

×1.95

High trust, low reach

Closed groups hide content

Long, small

I5

Ledger-Settled Public Directory

Provisional white space

×2.73

High for contributors

Legal exposure; capture

Long if governed

0. Integrated

Nothing; convenience and network effects dominate

P1

Take rate exceeds the cost of leaving; OS or rail rules open a gap

1. Awareness unbundles

Discovery moves into messaging and communities

P5

Creators see that delivery and payment can follow awareness

2. Payment and delivery unbundle for the middle tier

Hosting and billing move to creators and independent modules

P2, P3

Friction of several modules; disputes; credential fatigue

3. Plural directories and ranker choice

Directories compete; viewers choose rankers

P4

Coordination cost; discovery quality falls short of the integrated alternative

4a. Segment re-bundling

Player apps and creator suites bundle modules for one segment

New P1-like bundles inside a segment

The new bundler’s take rate rises; the next unbundling starts

4b. Exit through regulation

Mandated alternative rankers and interoperability

I1

Mandate takes effect or lapses

4c. Exit through the viewer’s agent

Integration moves to the viewer side

I3

Agent vendors concentrate or open standards win

4d. Exit through community or ledger

Trust substitutes for platform services

I2, I4, I5

Community outgrows its trust radius, or ledger governance fails

P1

Platform take rate and outbound-link policy

Stable or rising rates plus link restrictions confirm; falling rates or relaxed links suggest P2 is forming

P2/P3

Share of creators using off-platform checkout; status and date of the UPI merchant fee

Growth in off-platform checkout confirms; the fee taking effect weakens P3 for tickets above ₹2,000

P4

Traffic through non-platform directories by content form; ranker switching

Audio and text moving toward plural directories while video does not would confirm the form-based segmentation

P5

Share of plays arriving via shared links

A rising share with platform delivery unchanged confirms the awareness-only unbundling

I1

Campus or telecom-edge cache pilot for one regional-language education catalogue

Cache hit share and cost per viewer below the platform tier; a hub operator willing to run it

I2

One language, one institution, one content class

Seeders persist after six months without payment

I3

Prototype agent querying two open directories and logging reason codes

Panel members prefer agent-assembled results and can read the logs

I4

One cooperative or alumni body sharing a curated library by peer transfer

Participation holds; curator workload stays manageable

I5

Shadow-mode ledger inside the Open Attention pilot panel

Attrition below the pre-set threshold without payment

Pure peer transfer without a way to charge or enforce

B4 with C1; B4 with A1; B4 with D1; B4 with E1

×0.1–0.3

A platform has no mechanism to bill for content it does not deliver, and no reason to direct viewers to it.

Platform hosting with modular partners

B1 with A2; B1 with C2; B1 with C3; B1 with D3

×0.4–0.5

A platform that hosts the content would not give up directory, payment or ranking control.

Dominant directory with open rankers

A1 with D3; A1 with D5; A1 with D2

×0.3–0.5

The directory owner controls ranking unless forced.

Free commons with platform bundling

C5 with B1; C5 with E1; B5 with C1

×0.3–0.4

Platform economics assume payment.

Creator-hosted delivery with outside payment

B2 with C2; B2 with C3; B2 with C1

×1.5 (and ×0.4 for C1)

Creator hosting needs a payment and trust provider outside the platform.

Peer-assisted delivery with modular payment

B3 with C2; B3 with C3

×1.4

Hybrid delivery needs billing from somewhere other than the host.

Integration with bundled stack

E1 with B1; E1 with A1; E1 with C1; D1 with A1

×1.4–1.5

An integrated structure is internally consistent when each layer is bundled.

Plural directories with auditing or mandate

A2 with D2; A2 with D4; E5 with D2; C4 with D2

×1.2–1.4

Competing directories need a way to compare rankers.

Agents with ranker markets

A3 with D3; A3 with D1

×1.5 and ×0.5

An agent that picks for the viewer pairs naturally with viewer-chosen rankers.

Cycle states with their natural partners

E2 with A2, C2, B2, F1, F3; E3 with B4, B5, C4, A4; E4 with B3, B4; E5 with A2, D2, C4

×1.2–1.5

Each cycle orientation is easier to sustain with particular module choices.

Ledger settlement

C4 with D2; C4 with A2; C4 with B1

×1.3–1.4 and ×0.3

A ledger needs traceable, comparable rankers and cannot settle for a platform that hosts everything.

Local mesh and language communities

B5 with A4; B5 with F2; B5 with C3

×1.2–1.5

Local transfer fits community referral and regional segmentation.

Public interest

A5 with D5; A5 with C5; D4 with A1; D4 with E1

×1.2–1.4

A public directory pairs with community curation and non-commercial payment; mandates bite on dominant directories.

Price tier and free commons

F4 with C5

×0.5

Price tiers imply payment.