🧠What If an Exam System Could Separate Paper Assembly from Paper Visibility?
Resolve an admissible paper from approved structure. Release it only when the declared visibility conditions are met.
📝 SLANG-Exam
Examination security often focuses on protecting a final question paper after it has already been selected.
The paper may be stored on a server, printed, transported, sealed, copied, or passed through several authorized hands before the examination begins.
Every additional location and transfer can create another exposure surface.
But there is another question:
Does the final paper need to be selected and visible long before candidates are allowed to receive it?
SLANG-Exam explores a bounded structural approach.
It keeps the examination context, blueprint, approved question records, selector rules, authority conditions, and release conditions separate.
The governing relation is:
exam_result = resolve(submitted_input, versioned_profile, versioned_ruleset)
The idea is not to claim that examination security can be solved by one resolver.
The narrower idea is:
An admissible examination form can be selected from declared structure without treating one pre-stored final paper as the only source of paper availability.
📄 The Difference Between a Question Bank and a Final Paper
A question bank and a final assembled paper are not the same thing.
A question bank may contain many approved questions together with information such as:
- topic
- difficulty
- marks
- question type
- answer-key reference
- exposure group
- approval status
A blueprint describes what the paper must contain:
- total questions
- total marks
- topic distribution
- difficulty distribution
- question-type distribution
- forbidden question combinations
- exposure limits
The final paper is one selected set that satisfies all applicable constraints.
SLANG-Exam asks whether that final selection can remain unresolved or withheld until the declared conditions justify it.
The current reference operates on question metadata and content commitments. It does not contain question text.
🧩 What the Resolver Actually Checks
SLANG-Exam separates several questions that are often blended together:
Is the submitted input supported?
Is the examination context complete and internally consistent?
Can the approved question records satisfy the blueprint?
Which admissible paper should be selected under the declared selector?
Is assembly authorized for the declared audience?
Is the assembled paper admitted for visibility?
Can the result later be reconstructed and verified?
A simplified flow is:
submitted input -> admission -> canonicalization -> blueprint checks -> bounded selection -> release decision -> evidence
The reference may return:
RESOLVEDINCOMPLETECONFLICTFORBIDDENUNSUPPORTEDABSTAIN
A non-result is not automatically a system failure.
It may mean that the submitted structure does not justify forcing a paper.
🚦 A Paper Is Not Forced
Consider a few examples.
If the bank does not contain enough approved questions:
missing capacity -> INCOMPLETE
If the declared bank identity does not match the derived bank identity:
identity mismatch -> CONFLICT
If the input exceeds the supported schema, question-bank, marks, or resource boundary:
outside declared limits -> UNSUPPORTED
If the structure is valid but the exam window is closed:
assembly resolved + visibility withheld -> FORBIDDEN
If a selector requires uniqueness but more than one admissible paper exists:
multiple admissible papers -> ABSTAIN
The practical principle is:
Resolve when the declared contract supports a result. Refuse when it does not.
🖼️ SLANG-Exam Reference Diagram
SLANG-Exam separates input admission, blueprint feasibility, deterministic selection, authority, release visibility, and verifiable evidence.
👁️ Assembly and Visibility Are Different Decisions
One of the most important distinctions in SLANG-Exam is:
assembly != visibility
A paper may be structurally assembled while its selected question list remains withheld.
The current reference makes the paper visible only when:
- assembly resolves
release_authorized = trueexam_window_open = true
When visibility is withheld, the system may preserve the paper identity while returning:
paper_visible = false
and:
selected_questions = null
This means the system can preserve verifiable evidence that a bounded paper was resolved without automatically revealing its contents through the public result surface.
The distinction is:
A result can exist without being admitted for disclosure.
🎯 How Is the Paper Selected?
SLANG-Exam currently demonstrates several selector modes.
1. Canonical Ranking
CANONICAL_RANK
Eligible questions are placed in a deterministic canonical order.
The resolver selects the first admissible paper under that ordering.
This mode is reproducible:
same canonical input + same declared selector + same versioned profile + same versioned rules -> same selected paper
But it does not provide selection secrecy.
Anyone who possesses the same public selector inputs and question-bank metadata may be able to calculate the same result before release.
Determinism and secrecy are different properties.
2. Abstention When More Than One Paper Is Possible
ABSTAIN_ON_MULTIPLE
This mode requires the admissible paper to be unique.
Its bounded behavior is:
0 solutions + complete search -> INCOMPLETE
1 solution + complete search -> RESOLVED
2 solutions found -> ABSTAIN
uniqueness not established within limit -> ABSTAIN
Rather than silently choosing among multiple valid possibilities, the resolver can refuse to decide.
3. Single-Party Commit-Reveal
COMMIT_REVEAL_RANK
The submitted structure contains a selection commitment and a corresponding salt reveal.
The commitment is bound to the selection context:
selection_commitment = SHA256(domain, selection_context_id, salt)
This helps prevent one valid commitment from being reused with another bank, blueprint, audience, examination, or variant.
But the resolver verifies only the submitted structure.
It does not prove:
- when the commitment was created
- whether it was externally anchored
- whether the salt remained secret
- whether several salts were tested before commitment
Those remain operational and governance questions.
4. Multi-Party Commit-Reveal
MULTI_PARTY_COMMIT_REVEAL
Several declared participants contribute commitments and matching reveals.
The combined selector material depends on all accepted contributions.
The current reference supports between two and eight declared parties.
Its posture is described as:
CONDITIONAL_PRE_REVEAL_RESISTANCE
The word conditional matters.
The model does not prove that the participant roster was independently fixed, that commitments were timestamped, or that at least one contribution remained secret outside the resolver.
It verifies the final submitted participant, commitment, and reveal structure.
🔍 Why Canonicalization Matters
Some inputs may arrive in different array orders without representing different meaning.
For example, the reference normalizes question-bank order and supported commitment manifests.
Where ordering is declared irrelevant:
C(S) = C(pi(S))
therefore:
F_R(C(S)) = F_R(C(pi(S)))
But this does not mean that every ordering is irrelevant.
Ranked lists, execution traces, event histories, and other ordered structures may still preserve sequence.
The accurate principle is:
deterministic != universally order-independent
🧠 A Blueprint Must Work as One Whole Structure
It is not enough to check every constraint separately.
A bank may contain:
- enough algebra questions
- enough hard questions
- enough long-answer questions
- enough total marks
and still fail to contain one single paper satisfying all of those requirements together.
SLANG-Exam therefore performs bounded cross-constraint search.
It looks for one selected set that jointly satisfies the declared:
- question count
- total marks
- topic counts
- difficulty counts
- question-type counts
- exposure limits
- forbidden pairs
The search is limited.
search_nodes <= search_node_limit
If the required conclusion cannot be established within that boundary, the resolver does not pretend that a definitive answer was proved.
search exhausted -> conclusion not established within the declared bound
That does not mean no admissible paper exists outside the bound.
🧾 The Result Can Carry Evidence
SLANG-Exam produces deterministic identities for supported structures and results.
These may include:
- bank identity
- blueprint identity
- selection-context identity
- paper identity
- semantic result identity
- search-evidence identity
- bundle identity
- receipt identity
The project deliberately separates semantic meaning from operational search evidence.
result_id describes the declared semantic resolution.
search_evidence_id describes traversal-dependent evidence such as node counts and pruning information.
Two conforming implementations may agree on the semantic result without producing the same traversal statistics.
That leads to two verification levels:
Semantic conformance
same canonical input + same declared semantic contract -> same semantic result
Exact reference replay
same canonical input + same reference profile, ruleset, and traversal contract -> same complete reference bundle
🔄 Bundles and Receipts
A reconstruction bundle preserves the submitted input, normalized inspection surface, semantic result, and operational evidence required for exact reference replay.
A compact receipt provides a smaller public evidence object.
The receipt does not contain the complete submitted input or raw reveal salts.
It can still bind to the corresponding reconstruction bundle.
The broader idea is:
The system should not merely announce a result. It should preserve enough structure to inspect and verify what the result means.
🔐 What About the Question Bank?
SLANG-Exam does not remove the need to protect the question bank.
The following remain separate responsibilities:
- question creation and academic review
- secure content storage
- answer-key protection
- access control
- authentication
- digital signatures
- trusted time
- secure communication
- endpoint security
- printing security
- examination-center security
- invigilation
- incident response
The original SLANG-Exam concept focused on reducing reliance on an early final assembled paper while keeping question-bank protection as a separate security responsibility.
The current reference makes that boundary more explicit.
Reducing one exposure surface does not secure every other surface.
⚖️ Does Matching the Blueprint Prove Fairness?
No.
Two papers may contain the same number of questions, marks, topics, and declared difficulty labels while still differing in:
- actual difficulty
- ambiguity
- reading burden
- accessibility
- cultural loading
- time demand
- scoring behavior
- psychometric quality
SLANG-Exam enforces declared structural constraints.
It does not prove educational fairness or equivalence.
Academic review and independent psychometric validation remain necessary.
🌍 Where Could This Approach Be Useful?
The structural separation could be explored for:
- common examination papers
- center-specific variants
- candidate-specific variants
- certification assessments
- controlled practice examinations
- bounded question-set generation
- paper-release approval systems
- reconstruction and audit of selected forms
The useful question is not:
Can this resolver replace an examination system?
It cannot.
The useful question is:
Can an examination platform avoid treating one early, pre-stored final paper as the only way to know what candidates should receive?
SLANG-Exam provides one bounded reference for studying that question.
🧪 Current Reference Status
The current reference is SLANG-Exam v0.7.2.
Published checks include:
127/127 PASSreference self-test56/56semantic vectors reproduced56/56reference-evidence vectors reproduced10/10metamorphic relations reproduced3/3bounded-search probes reproduced
The repository also includes:
- the Python reference implementation
- frozen conformance vectors
- canonical and multi-party reconstruction bundles
- compact receipts
- receipt-to-bundle verification
- a multi-party commit-reveal profile
- permanent hostile-input and tamper coverage
- the reference diagram
- detailed documentation
These results apply only to the declared v0.7.2 implementation, schemas, profiles, rules, inputs, limits, and evidence boundary. They do not establish production qualification or third-party certification.
⚠️ What SLANG-Exam Does — and Does Not — Claim
SLANG-Exam demonstrates:
- bounded examination-form admission
- deterministic selection under declared selector rules
- explicit refusal states
- assembly and visibility separation
- canonical identities
- reconstruction bundles
- compact receipts
- semantic and exact-reference verification paths
It does not provide:
- a complete examination platform
- question-bank confidentiality
- authentication infrastructure
- institutional authority
- trusted timestamps
- digital signatures
- network or endpoint security
- printing security
- invigilation
- grading validity
- educational quality assessment
- legal or regulatory approval
- universal fairness
- guaranteed leak prevention
Its claim is narrower:
complete + consistent + admitted structure -> deterministic bounded resolution
incomplete, conflicting, forbidden, unsupported, or inconclusive structure -> explicit non-result state
🌐 Explore SLANG-Exam
The GitHub repository contains the reference implementation, frozen vectors, selector profiles, bundles, receipts, verification commands, architecture details, boundaries, Quickstart, FAQ, and Proof Sketch.
The SLANG-Observatory reference implementations and associated verification artifacts may be used, copied, modified, tested, studied, and redistributed without a license fee, subject to the repository LICENSE; documentation, architecture materials, specifications, diagrams, and explanatory content are governed by the separate terms stated there.
🌌 The Larger Question
Examination security is often framed as:
How do we protect the final paper after selecting it?
SLANG-Exam adds another question:
How late can final bounded selection and visibility remain unresolved without weakening the declared examination structure?
The question bank may still exist.
The blueprint may still exist.
Academic governance may still exist.
Security infrastructure may still exist.
But one fixed final paper does not have to serve as the only source of examination-form availability.
The principle is:
Preserve the approved question structure. Resolve within declared bounds. Release only when visibility is admitted. Keep evidence of what was resolved.
✍️ Authorship & Disclaimer
Created by the authors of the Shunyaya Framework.
SLANG-Exam is a bounded deterministic reference demonstration for examination-form admission, selection, visibility, and verifiable evidence.
It is not intended to replace examination authorities, academic governance, question-bank security, authentication, operational security, invigilation, psychometric validation, legal review, or institutional controls.
High-stakes use requires independent implementation review, security assessment, academic validation, appropriate authorization, and suitable operational safeguards.
OMP
Comments
Post a Comment