Information Bulletin Strategy

• Information Bulletin Strategy is a set of design patterns that coordinate selective disclosure under operational constraints like privacy, auditability, and delay tolerance.
• It integrates methods such as asynchronous publication, public verifiability, and controlled expressivity to shape downstream behavior and system performance.
• Its applications span opportunistic networks, bulletin-board search, verifiable voting, queue control, economic design, and multilingual alert systems.

Information bulletin strategy denotes a family of design patterns in which a bulletin, bulletin board, or scheduled disclosure mechanism is used to coordinate information release under operational constraints such as privacy, auditability, searchability, strategic interaction, delay tolerance, or multilingual publication. In the literature considered here, the term does not refer to a single canonical architecture. Rather, it appears in several technically distinct forms: a public bulletin board used as an asynchronous rendezvous point in opportunistic networks; a crawler-and-ranking pipeline for bulletin-board attachments; a peered Web Bulletin Board for verifiable elections; periodic queue-status broadcasts that shape reneging and jockeying; strategic disclosure rules in sender–receiver and competitive search models; auction mechanisms that separate internal information from external disclosure; and controlled-language phrase catalogues for time-critical multilingual warning bulletins (Farkhondeh, 2019, 0909.2489, Culnane et al., 2014, Kiggundu et al., 6 Aug 2025, Velicheti et al., 2024, Du et al., 2 Jun 2026, Tunuguntla et al., 14 Jan 2026, Winkler et al., 2015).

1. Core design idea and recurring dimensions

Across these works, a bulletin strategy is a mechanism for making selected information publicly or semi-publicly available in a way that changes downstream behavior. The released object may be a stored message, an election record, an indexed attachment repository, a queue-state descriptor, an advertisement-like signal, an auction-facing information bundle, or a pre-authored multilingual warning sentence. The shared structural issue is not merely publication, but publication under constraints: who may infer what, when disclosure occurs, what is hidden, and how the released information shapes later decisions.

Several recurring dimensions appear. One is the distinction between internal information use and external disclosure. In ad auctions, the seller should retain the finest information internally while disclosing it as coarsely as possible externally, because publisher revenue is increasing in information granularity and decreasing in disclosure granularity under the Information-Bundling Position Auction (Tunuguntla et al., 14 Jan 2026). A second is asynchrony: opportunistic-network bulletin boards operate through delayed WRITE and READ actions rather than direct sender-to-recipient exchange (Farkhondeh, 2019). A third is public verifiability: in election systems, the Web Bulletin Board is treated as a “broadcast channel with memory,” and the central problem is making publication append-only, globally visible, and robust against faulty or dishonest peers (Culnane et al., 2014). A fourth is controlled expressivity: the Swiss avalanche bulletin succeeds by limiting authors to a small sublanguage of predefined phrase templates, thereby making fully automatic translation possible without proofreading (Winkler et al., 2015).

A plausible implication is that “information bulletin strategy” is best understood as a systems-level design problem rather than a single algorithmic technique. The relevant optimization target changes by domain: metadata privacy in opportunistic networking, retrieval precision in bulletin-board search, robustness in verifiable voting, delay and impatience in queueing, revenue in markets, or publication latency and translation reliability in warning systems.

2. Privacy-preserving asynchronous bulletin boards in opportunistic networks

In opportunistic networking, the bulletin strategy is a public bulletin board (PBB) used as an asynchronous rendezvous point inside a disconnected, store-carry-forward environment, combined with mix-network routing so that senders, receivers, and their interaction pattern are hard to link (Farkhondeh, 2019). The architecture contains four entities: clients, normal relay nodes, mixer nodes, and one public bulletin board node. In the simulation, node 0 is the bulletin board, and the board is implemented as a Java component called BB, attached to a host object.

Internally, the board is modeled as $n$ indexed cells. Each cell stores a tag preimage and a value, operationalized as two arrays indexed by cell number. The board exposes two logical methods. In a WRITE, a sender provides an encrypted message, a preimage of a tag, and a chosen cell index; the board stores the ciphertext in the value array and the tag preimage in the tag array at that index. In a READ, a receiver sends an index and the preimage of the corresponding tag; if the supplied preimage matches the stored tag, the board returns the stored value, otherwise it returns null. The paper also states that only the intended recipients can delete the value of the cell by using the preimage of the corresponding tag, although deletion logic is not developed in implementation detail (Farkhondeh, 2019).

End-to-end communication assumes that users $A$ and $B$ previously shared a secret symmetric key, the first bulletin-board cell index, and the preimage or tag for the first message. Each encrypted message contains not only application payload but also the next random cell index and next tag. This creates a chained sequence of one-time hidden mailboxes rather than a fixed mailbox. The sender is advised not to save messages in continuous cells; instead, the sender picks a random index for the next message. Messages can traverse up to three mixer nodes. At each mixer, they are buffered, delayed, and reordered before forwarding; the paper states that messages become mixed without any change to their context (Farkhondeh, 2019).

The stated security objective is primarily metadata privacy rather than content confidentiality alone. The design aims at unlinkability between send and receive events, anonymity of sender and receiver identities with respect to observers and intermediaries, hiding of contact addresses, and protection of the social graph. The attacker model is informal: malicious users may observe traffic, analyze traces, reconstruct the social graph, or flood the bulletin board. There is no formal cryptographic adversary model, no proof of anonymity, and no quantified corruption threshold for compromised mixers (Farkhondeh, 2019).

Evaluation in the ONE simulator uses two scenarios with 41 nodes, node 0 as the bulletin board, board size 100 cells, radio range 25 m, transmission speed 100 MB/s, and at most three mixer nodes on a path. Scenario 1 uses a $1 \text{ km} \times 1 \text{ km}$ park with humans walking at 1–2 km/h; Scenario 2 uses a smaller area and speeds of 2–3 km/h. The metrics are total packet delivery ratio, total packet latency average, and total packet latency median. In Scenario 1, delivery ratios are high—0.97905 for WRITE and 0.985225 for READ—but average latency is about (1142.2516) s for WRITE and (1131.41987)5 s for READ, which the authors judge not acceptable. In Scenario 2, delivery remains high—0.982175 for WRITE and 0.976075 for READ—while average latency improves to 8.4144 s for WRITE and 8.76625 s for READ, which the paper considers acceptable performance (Farkhondeh, 2019).

This strategy is therefore a mix-assisted hidden mailbox chain. Its strongest contribution is the concrete integration of a public board with asynchronous unidirectional messaging in a DTN simulator; its main limitations are the absence of formal security analysis, the assumption of an initial secure meeting, vulnerability to flooding, and underspecified board-state management.

3. Bulletin boards as searchable repositories and classified attachment stores

A distinct bulletin strategy appears in bulletin-board search, where the central problem is that users often care less about the HTML notice page itself than about the attachments linked from it, such as doc, txt, and xls files (0909.2489). The paper on PrisCrawler states that almost 90% of bulletin-board information exists in attachments. Traditional crawlers rank webpages, whereas attachments typically contain no hyperlinks and therefore have no intrinsic hyperlink-based importance signal. This yields what the paper calls the “non-relevance” problem and the “chaos” problem: attachments are both weakly ranked and difficult to classify.

PrisCrawler is introduced as a subsystem of a bulletin board search engine. Its role is to collect bulletin-board pages and attachment files, classify attachments by type, transfer contextual relevance from page to attachment, and store the result as what the paper calls “clear classified and associated databases” (0909.2489). Its workflow extends the traditional crawler loop. The system retrieves and stores a bulletin-board page, scans its hyperlinks, determines whether each hyperlink points to an attachment, classifies attachments via suffix, computes relevance through Attachrank, writes relevance information at the head of the attachment record, stores the attachment, and enqueues webpage links for further crawling.

The architecture contains eight modules: Link Provider, Page Retriever, Link Scanner, Attachment Classifier, Link Storager, Attachrank Producer, Relevance Writer, and Attachment Storager. The core ranking idea is Attachrank, which is explicitly modeled after PageRank. If page $A$ has attachments $M_1,\dots,M_n$, and $PR(A)$ denotes the PageRank-style score of the page, then the attachment rank is defined as

$AR(M_i)=PR(A).$

The paper also states

$PR(A)=AR(M_1)=\cdots=AR(M_n).$

Attachrank is therefore a relevance-propagation scheme from webpages to files, not a separate recursive graph process over attachments (0909.2489).

The experiments use two BUPT bulletin boards, from which a normal crawler first collected about 60,000 webpages and attachments. The same source content was then organized either as a normal structure database or as the classified and associated database produced by PrisCrawler. Using the same pretreatment subsystem and retrieval subsystem, with Lemur as the retrieval system, the paper reports a normal database average pretreatment time of 74.7 minutes versus 63.5 minutes for the classified and associated database, interpreted as a 15% decrease in pretreatment complexity. Average search time falls from 0.102 seconds to 0.082 seconds, interpreted as a 20% decrease in retrieval subsystem complexity. Average precision rises from 68% before Attachrank to 85% after Attachrank (0909.2489).

The strategy is attachment-centric rather than page-centric. It treats the bulletin notice page as a semantic wrapper carrying title, time, and explanatory information, and the attachment as the substantive object to be indexed and retrieved. A plausible implication is that bulletin systems in document-heavy environments should regard files as first-class searchable entities rather than as mere endpoints behind notice pages. The reported gains are promising, but the paper also notes limitations: only two bulletin boards were used, both from the same institution; no significance tests are reported; and classification is shallow, relying mainly on suffixes rather than semantic content analysis (0909.2489).

4. Bulletin boards as public verifiability infrastructure in elections

In verifiable voting, the bulletin strategy is the design of a Web Bulletin Board (WBB) that serves as the public repository where election evidence is posted so that voters, auditors, and candidates can independently check that votes were recorded and tallied correctly (Culnane et al., 2014). The WBB is not merely a website. It is the cryptographic and procedural anchor for transparency, auditability, dispute resolution, and end-to-end verification. The paper treats the board as needing three properties often assumed informally in voting papers: once items are posted, they are not removed; the final published content is fixed; and all parties get the same view. This is why the WBB is described as a “broadcast channel with memory.”

To avoid a single point of failure and a single point of trust, the paper proposes a peered bulletin board implemented by $n$ peers using threshold cryptography. The adversarial model is strong: the network is under Dolev–Yao control; messages may be intercepted, blocked, delayed, duplicated, diverted, or spoofed; up to $A$0 peers may be dishonest; and honest peers may also be temporarily unavailable. The protocol separates operation into a posting phase and a publication phase, thereby avoiding Byzantine agreement on every posted item (Culnane et al., 2014).

For an item $A$1 in period $A$2, the posting protocol has three rounds. The user submits $A$3 to peers. Each peer that accepts it signs $A$4 with its ordinary signing key and forwards that signature to others. Once a peer has seen a threshold number of ordinary signatures on $A$5, it issues a threshold-signature share. The user collects at least $A$6 such shares and combines them into a bulletin-board receipt

$A$7

This receipt is evidence that the item has been accepted strongly enough that it must later appear on the published board. The paper explicitly shows that the peer-to-peer confirmation round is necessary; without it, a dishonest peer could help create a receipt for an item that does not later appear on the board (Culnane et al., 2014).

Publication is handled by an optimistic hash-based commit plus fallback reconciliation. Peers first try to agree on hashes of their local boards. If sufficiently many peers sign the same hash, they publish the board together with a threshold signature on its hash: $A$8 If hashes disagree, peers exchange databases, update by union, and retry the optimistic step. The design also handles conflicting posts through a symmetric irreflexive clash relation $A$9. Honest peers refuse to sign conflicting items. The critical threshold condition is

$B$0

which ensures quorum intersection through at least one honest peer. The paper proves that this threshold is sufficient to guarantee that receipted items appear on the published board and that clashing items do not both appear (Culnane et al., 2014).

A major contribution is formal verification in Event-B. The proof proceeds through refinements from an idealized specification of bulletin-board behavior to increasingly concrete protocol suites, including multiple periods, clash handling, and optimistic commit with fallback. The published board is required to satisfy $B$1, where $B$2 is the set of confirmed items and $B$3 the set of received items. The verification also establishes non-interference among the interacting subprotocols (Culnane et al., 2014).

This strategy turns the bulletin board from an informal assumption in voting protocols into a concrete distributed service. Its guarantees are strong under the stated threshold and storage assumptions, but liveness still requires additional communication assumptions; honest peers must not lose their databases; and the paper assumes that dishonest peers do not equivocate by sending different databases to different honest peers during publication. The strategy is motivated by the vVote system and by the need to publish daily over a multi-day election period.

5. Periodic information bulletins as control signals in queueing systems

In impatient queueing, information bulletin strategy refers to periodic broadcasting of queue descriptors to autonomous tenants in a decentralized multi-tenant system, so that tenants may decide whether to remain in a queue, switch to another queue, or process locally (Kiggundu et al., 6 Aug 2025). The setting is motivated by 6G-style decentralized control, where autonomy requires timely access to system status. The paper asks three questions: what queue status should be shared, how much should be communicated, and how often updates should be dispatched.

The model uses a heterogeneous two-queue system with arrivals

$B$4

FCFS service, and queue-specific service rates $B$5 and $B$6. Each queue periodically sends status information every $B$7 seconds, with dispatch intervals fixed at

$B$8

Information staleness is modeled through a parameter $B$9, so effective degradation enters as $1 \text{ km} \times 1 \text{ km}$0 (Kiggundu et al., 6 Aug 2025).

Two information models are broadcast in round-robin fashion. The first is a Markov model of service-rate states, with average service rates

$1 \text{ km} \times 1 \text{ km}$1

Tenants compare service-rate distributions through first-order stochastic dominance. The second is a Markov model of queue-length inter-changing times (ICD). For queue $1 \text{ km} \times 1 \text{ km}$2, the rate of state change is simplified to

$1 \text{ km} \times 1 \text{ km}$3

so the expected time between successive changes in queue size is

$1 \text{ km} \times 1 \text{ km}$4

The service-rate bulletin is richer; the ICD bulletin is more aggregated and may induce overreaction (Kiggundu et al., 6 Aug 2025).

Tenant behavior is modeled through reneging and jockeying probabilities. Under service-rate information, waiting-time comparisons and first-order stochastic dominance drive switching and abandonment decisions. Under ICD information, the jockeying probability from $1 \text{ km} \times 1 \text{ km}$5 to $1 \text{ km} \times 1 \text{ km}$6 uses a sigmoid rule: $1 \text{ km} \times 1 \text{ km}$7 The queue adaptation problem is cast as choosing service rates to minimize delay and impatience: $1 \text{ km} \times 1 \text{ km}$8 subject to service-rate bounds and stability constraints. The analytical solution is intractable, so the paper uses a rule-based queue policy for predictive modeling rather than a fully specified RL algorithm (Kiggundu et al., 6 Aug 2025).

The numerical study uses 300 simulation runs per interval over arrival-rate settings

$1 \text{ km} \times 1 \text{ km}$9

The measured outcomes are reneging rate, jockeying rate, waiting time, and objective value. Without the policy, the paper reports “chaotic and variability” in reneging and jockeying profiles regardless of the information model, though this volatility reduces as dispatch intervals become larger. Even without policy, the service-rate Markov bulletin produces less impatience than the ICD bulletin. With the rule-based policy, impatience becomes more controllable, waiting-time medians drop, and dispatch intervals between 5 and 7 seconds appear near-optimal, especially when the service-rate model is shared (Kiggundu et al., 6 Aug 2025).

The paper reports optimized versus non-optimized objective values at each interval. For example, at $A$0 s, the optimized objective is 0.39 versus 1.62 without optimization; at $A$1 s, 0.77 versus 3.40; at $A$2 s, 0.749 versus 0.84; and at $A$3 s, 0.46 versus 1.27. Aggregate statistics are a mean optimized objective of 0.53, a mean non-optimized objective of 1.78, standard deviations 0.15 and 0.97 respectively, and average improvement 1.26 (Kiggundu et al., 6 Aug 2025).

This use of bulletin strategy is neither archival nor repository-based. It is an online control mechanism in which scheduled disclosures alter agent behavior and thereby feed back into system policy. A plausible implication is that, in decentralized systems, the informational content and timing of bulletins can themselves function as control variables.

6. Strategic disclosure, competitive search, and auction design

A further family of bulletin strategies arises in information design and market design, where the bulletin is a disclosure policy rather than a board. In "Learning How to Strategically Disclose Information," the sender observes a Gaussian state

$A$4

and commits to a signaling policy $A$5, while a receiver with unknown, possibly adversarially varying type chooses an action after updating beliefs (Velicheti et al., 2024). The sender’s problem is an online Stackelberg game, and the key technical reduction is from arbitrary signaling policies to posterior or error covariance choice. With entropy-based information cost, the sender solves

$A$6

Under full information feedback, Online Gradient Descent yields $A$7 regret in the regularized setting; without information cost, the rate is $A$8; with partial information, the paper establishes sublinear regret using a bandit method (Velicheti et al., 2024). The practical lesson stated in the paper is to treat disclosure as a learnable choice of posterior precision rather than a binary reveal-or-hide decision.

In competitive sequential search, the bulletin becomes a seller’s credible pre-search disclosure, such as an advertisement, preview, or product snippet (Du et al., 2 Jun 2026). The receiver inspects products sequentially under costly inspection and follows Weitzman’s Index Algorithm. For seller $A$9, the relevant statistic is the amortized value

$M_1,\dots,M_n$0

where $M_1,\dots,M_n$1 is the Weitzman index. The paper shows that seller incentives are mediated by the induced distribution of $M_1,\dots,M_n$2, not directly by raw value distributions. With atomless priors, equilibrium exists. In symmetric environments with monotone increasing densities and sufficiently low search cost, full revelation is the unique symmetric equilibrium. When $M_1,\dots,M_n$3 exceeds the bottom of support, full revelation fails; for two senders the paper constructs a symmetric hinge-convex signaling equilibrium with a piecewise amortized-value distribution involving full revelation at low and high regions and pooling in the middle (Du et al., 2 Jun 2026). The strategic conclusion is explicit: in competitive search, optimal bulletins are designed to shape search indices, not merely posterior means.

In ad auctions, the bulletin strategy concerns how a publisher should use and disclose inventory-type information such as demographic, contextual, or behavioral audience segments (Tunuguntla et al., 14 Jan 2026). The proposed Information-Bundling Position Auction (IBPA) treats the ad inventory type as the publisher’s private information and lets advertisers submit multidimensional bids. The core theoretical conclusions are sharp. Publisher revenue is increasing in information granularity and decreasing in disclosure granularity. Moreover, full-information/null-disclosure IBPA weakly dominates GSP under any feasible information or disclosure regime (Tunuguntla et al., 14 Jan 2026). The allocation rule ranks advertisers by quality-adjusted marginal revenues derived from revenue curves: $M_1,\dots,M_n$4 The empirical study, using 9.6 million position auctions from a large retail media platform, reports that relative to GSP-FI-FD, IBPA-FI-ND yields revenue $M_1,\dots,M_n$5, allocation rate $M_1,\dots,M_n$6 percentage points, advertiser welfare $M_1,\dots,M_n$7, and total welfare $M_1,\dots,M_n$8 (Tunuguntla et al., 14 Jan 2026).

Taken together, these works show a consistent strategic pattern. A bulletin can be used to steer search order, infer preferences, preserve competition, or screen agents, and the key design variable is often not whether information exists but how much of it is revealed, at what precision, and under what commitment structure.

7. Controlled multilingual bulletin production and formal reasoning about information-sensitive strategy

In operational warning systems, bulletin strategy can mean constraining authorship so that translation becomes deterministic and publishable under severe deadlines. The Swiss avalanche bulletin is produced twice daily in four languages—German, French, Italian, and English—and the danger descriptions were too time-critical for manual translation or post-editing (Winkler et al., 2015). The solution is a catalogue of predefined phrases and predetermined combination rules, i.e., a controlled natural language for a restricted avalanche-warning sublanguage.

The catalogue contains 110 individual sentence templates, up to 10 segments per template with mean 4.3 segments, and 603 different lists of selectable options. It contains 7182 words with 757 unique words in German, and the authors state that the system can generate several trillion different sentences (Winkler et al., 2015). Forecasters author the source text in German by selecting a general sentence pattern and filling segments through pull-down menus. The system then maps segment selections directly to manually prepared target-language segments. Only limited runtime processing is used: segment order may differ by language but is fixed for that language, each segment may be split into two parts in the target language, and post-processing is restricted essentially to spaces and sentence-initial capitalization (Winkler et al., 2015).

The design is constrained by multilingual grammar. Articles are often bundled with nouns; prepositions may be stored with noun options; adjectives are permitted only when agreement remains stable across permitted contexts; pronoun references are restricted; and German case differences sometimes force separate templates. The authors explicitly note issues such as apostrophes, clitics, French split negation, elision in French and Italian, and the Italian impure “s,” all handled through phrase design rather than a grammar engine (Winkler et al., 2015).

Operationally, the system front-loads quality assurance. Before deployment, phrases were tested through random sentence generation, exhaustive option review across all languages, and fictitious bulletin writing by native-speaker avalanche experts. In the first two seasons, about 8000 danger descriptions were produced, and only 1 was not made available within 2 hours due to catalogue-related technical problems, giving catalogue technical availability of 99.999%. The time between the last possible source-text change and publication in all four languages fell from about one hour to a few minutes (Winkler et al., 2015).

Evaluation includes both forecaster and user studies. All seven forecasters reported high satisfaction, and none used the emergency “joker phrase” mechanism during the first two winters. In a blind comparison between old manually written and translated texts and new catalogue-generated texts, the mean recognition rate was 55%, so users could hardly distinguish the two origins. Aggregate language-quality ratings were very similar, with only a tiny overall decrease from 4.16 to 4.14 in the balanced dataset, while the authors conclude that the outputs are virtually equivalent to manually written and translated texts (Winkler et al., 2015).

A more abstract, formal perspective on information-sensitive strategy appears in work on Strategy Logic under imperfect information (Knight et al., 2019). There the central move is to avoid forcing strategies to be uniform at the semantic level. Instead, branching-time Strategy Logic and its epistemic extension make informational constraints expressible inside the logic itself. The paper introduces BSL and ESL, with quantification over strategies, explicit binding and unbinding, branching-time path quantification, and distributed knowledge operators over finite plays. Weak uniformity can be expressed by a formula such as

$M_1,\dots,M_n$9

possibly after unbinding other agents (Knight et al., 2019). This is not a bulletin-production system, but it provides a formal language for reasoning about strategic behavior when actions must remain consistent across indistinguishable information states. A plausible implication is that it supplies a general meta-logic for analyzing when and how bulletin-based information structures constrain admissible strategies.

Across these operational and formal works, the same structural principle reappears: effective bulletin strategies narrow the space of permitted information flows so that some downstream property becomes tractable—translation reliability, epistemic reasoning, or strategy consistency.

8. Synthesis, limitations, and recurring trade-offs

The surveyed literature presents information bulletin strategy as a domain-dependent response to information asymmetry and coordination constraints. In opportunistic networks, the bulletin is a hidden mailbox chain on a public board; in search, it is a mechanism for turning attachments into rankable objects; in elections, it is a threshold-signed public memory; in queueing, it is a periodic state descriptor; in economic design, it is a disclosure rule; and in avalanche warning, it is a controlled-language publication system (Farkhondeh, 2019, 0909.2489, Culnane et al., 2014, Kiggundu et al., 6 Aug 2025, Velicheti et al., 2024, Du et al., 2 Jun 2026, Tunuguntla et al., 14 Jan 2026, Winkler et al., 2015).

Several trade-offs are explicit. Privacy versus latency appears in opportunistic networking, where mix-induced delay is the cost of unlinkability (Farkhondeh, 2019). Structure versus expressivity appears in both bulletin-board search and phrase-catalogue translation: Attachrank treats attachments as first-class objects at the price of shallow attachment semantics, while the avalanche catalogue attains reliability by restricting language variety (0909.2489, Winkler et al., 2015). Robustness versus protocol complexity defines the peered Web Bulletin Board, whose distributed threshold design avoids trusting a single server but requires careful reasoning and formal verification (Culnane et al., 2014). Timeliness versus staleness-induced overreaction is central in impatient queueing, where richer service-rate bulletins outperform ICD bulletins and moderate update intervals outperform both overly frequent and overly infrequent updates (Kiggundu et al., 6 Aug 2025). Internal information use versus external disclosure dominates the economic papers: strategic disclosure, competitive search, and IBPA all emphasize that revealing less can be optimal even when learning more internally is beneficial (Velicheti et al., 2024, Du et al., 2 Jun 2026, Tunuguntla et al., 14 Jan 2026).

The limitations are equally recurrent. Several works remain simulation-based or heuristic rather than fully formalized. The opportunistic-network bulletin board lacks formal anonymity proofs and detailed cryptographic constructions (Farkhondeh, 2019). The queueing paper explicitly states that its analytical solution is intractable and that its controller is a rule-based heuristic without formal convergence proof (Kiggundu et al., 6 Aug 2025). The bulletin-board search paper relies on small-scale institutional experiments and shallow classification (0909.2489). The avalanche system depends on a narrow sublanguage and would be difficult to transfer to broader or more linguistically diverse domains (Winkler et al., 2015). In the economic papers, the strongest results rely on specific assumptions such as Gaussian priors and quadratic costs, atomless priors, independent private values, truthful direct revelation, or convex prior densities (Velicheti et al., 2024, Du et al., 2 Jun 2026, Tunuguntla et al., 14 Jan 2026).

What unifies these otherwise heterogeneous systems is the use of a bulletin mechanism to mediate information release in a way that is itself strategic. The bulletin is not a passive container. It is the locus where architectural goals—privacy, ranking, verifiability, control, persuasion, revenue extraction, or multilingual reliability—are encoded into what is published, withheld, indexed, or periodically broadcast.