Return to the full study

KEDDEH SYSTEMS · INTERACTIVE-CONTRACTS · R1 · 18 October 2026

Seven contextual interactive test contracts — R1

Reading edition · 18 October 2026. Complete current technical specification, including the session-local facility persistence clarification. One operational code location is shown by its file label. The exact current source and earlier reading edition are separately retained. This is an edited reading projection, not a byte-identical source capture.

Status: independent oracle specification for the central study; not an application or browser receipt. Original editions and code remain unchanged. Test IDs are stable. Every facility should show its task, input types and units, exact computation, result, and conclusion boundary next to the controls. It must provide an explicit Return to study route. The separate spatial-radix investigation is outside this file.

Shared admission and result rules

• Preserve submitted input text and identify the source contract/version used. Changing input creates a new attempt; it must not rewrite a retained earlier result. Show a rejection with its reason and retain the valid previous result as a previous result, never as the new attempt's answer. Facility inputs and result history are session-local unless explicitly exported. Reader checkpoint custody covers only its declared note, fragment, selected view, continuation and kernel profile; it does not restore the facility computation heap. A remount/reset must disclose this scope before users rely on retention.
• .1 is a literal origin designation, never a numeric operand. Native (-1+) superscripts and predecessor-dependent notation are not ordinary powers. These seven facilities do not evaluate that algebra. Do not use JavaScript eval or substitute a convenient scalar law.
• A supplied conventional numeric zero is a value. unavailable, unspecified, null, an empty field, false, and missing fields are not aliases for it. Do not infer absence from a zero value or zero coefficient.
• For the new conventional comparison profiles, rational inputs are JSON strings matching -?(0|[1-9][0-9]*)(\.[0-9]+|/[1-9][0-9]*)?, at most 64 characters. No leading/trailing whitespace, exponent, NaN, Infinity, numeric JSON values, native expressions, or implicit defaults. A UI may explicitly trim before capturing the submitted canonical input, but must not claim the trimmed bytes are the original bytes. Results are reduced p/q strings, or integer strings when denominator is one. Quantum sampling uses its separately declared finite-number contract below. This parser is an explicitly bounded UI profile, not a native Keddeh algebra rule.
• Domain rejection and a valid non-unique result are different outcomes. An underdetermined mixture and an ambiguous counterpart are informative admissible results, not successful unique deductions.
• Fixtures contain fixed expected answers. The independent harness calculates actual answers separately and compares them; it does not obtain expected answers by running the integrated implementation. The retained source tests are additionally replayed in isolated Python/Node contexts.
• Exact arithmetic certifies the declared finite calculation. It does not certify measurement precision, calibration, physical validity, cosmology, quantum hardware, or arbitrary-state reconstruction. No stochastic fit or numerical tolerance is hidden in these fixtures.

preservation — Preserve an observation without losing its context

Displayed task: “Preserve a supplied observation, reconstruct it in a fresh preserver, and compare its exact source bytes, ordered roles and explicit observation status.”

Input: existing execution-contract-r1.json carrier {sourceText, sourceSHA256, sourceBytes} and prior retained records, admitted by the exact existing observation-envelope schema. sourceBytes counts UTF-8 bytes. Source SHA-256 is the digest of those exact bytes. Numeric reported values here follow the retained schema's finite JSON-number type, not the new rational-string profile used by the arithmetic facilities. Units/reference/method/resolution/uncertainty are explicit supplied or unspecified slots. Origin must be literal .1.

Computation: run unchanged preservation-core-r4.js / createObservationPreserver; verify carrier digest and byte count; parse and validate; revalidate prior records; detect conflicting observation identity; retain and reconstruct the exact source plus parsed echo. Ordered roots and children remain in source order. A fresh replay of the same exact record is IDEMPOTENT. This is software source custody, not operator evaluation.

Independent expectations: baseline reported 0 → PRESERVED, source digest dc0e74e4690969cc8e002be875c4e266fd9bbd405f7cd532607e3e566cf1b0d3; unavailable and unspecified each → PRESERVED with their own status and no numeric invention. Same source replay → IDEMPOTENT. Equal scalar/new observation ID → a distinct retained record. Same ID/different observer or order with fresh valid pin → CONFLICTING_OBSERVATION_ID. Changed source/stale pin → REJECTED_SOURCE_IDENTITY. A missing role, numeric-string reported value, null reported value or origin alias → REJECTED_SCHEMA. A retained parsed-echo mutation → REJECTED_RETAINED_RECORD. Signed zero's source token must remain exact.

UI: offer the existing real fixture selector plus an explicit source-text/carrier inspector; do not manufacture a simplistic schema substitute. At least reported-zero, unavailable, same-ID conflict, stale-pin, order mutation and fresh replay must be available. The original acceptance suite and its 39 vectors are available as the wider test evidence.

Conclusion: exact custody and distinctions survived the exercised round trip. Not physical truth, numeric-origin conversion or native operator qualification.

krs-counterpart — Which counterpart follows from this readout?

Input: source-defined finite PREPARATIONS table, observed label, and optional frame/lineage/observed_time. Each supplied label is a nonempty string; an omitted context is JSON null. Empty strings are not wildcards. Labels have no physical units. Use the four exact retained W1–W4 records by default. The optional W5 conflict fixture is a declared added preparation, not an edit to the originals.

Computation: filter every preparation by the observed label and every supplied context. Deduplicate and sort counterpart labels. No matches → NO_ADMISSIBLE_PREPARATION with null counterpart; one distinct counterpart → DEDUCED_UNIQUE; multiple → AMBIGUOUS with null counterpart. Keep all matching preparation identities, including multiple preparations sharing one counterpart.

Expected: negative/no context → candidates [positive-A,positive-B], matches [W1,W2,W4], AMBIGUOUS. negative/F1 → positive-A, matches [W1,W4], DEDUCED_UNIQUE. negative/F2 → positive-B. negative/lineage L4 → positive-A. positive → negative-A. unknown → no admissible preparation, not zero. Add W5=(negative,positive-C,F1,L5,T3,synthetic:005), then negative/F1 → AMBIGUOUS [positive-A,positive-C].

Rejects: non-string label, blank context, non-record input, unknown extra schema field. An unknown nonempty label is a valid no-match result, not a parser error.

Conclusion: conditional identifiability on this complete finite table. Not native KEX/ZCG/QZL execution or a universal complement law.

line-origin — Inspect shared line identity and descend without replacement

Input: existing parent line L1–L12, integer child count 1–12, and an explicitly supplied ordered-count profile d=1–8, n=1–6, f=1–2. All count fields are JSON integers, excluding booleans; identities are literal strings. This is an explicit finite UI profile, not the source runtime's full unbounded domain.

Computation: use unchanged LineOriginRuntime. Inspect the selected parent and all frame memberships; descend creates parent.1 … parent.count, each with its own origin equal to its ID, without replacing the parent; repeat descent is idempotent for the same children. No frame membership is inherited implicitly by children. The cardinalities are ordered-product counts d^n and d^(n f), dimensionless, in this declared counting model.

Expected: L5 origin L5, frames [M_A,M_B,M_C], shared true. L1 origin L1, frames [M_A], shared false. Descend L1 by 5 → L1.1…L1.5, unchanged L1 origin, no duplicate children on repeat. d=5,n=3,f=2 → 125 and 15625; d=5,n=4,f=2 → 625 and 390625.

Rejects at wrapper boundary: L0/unknown parent, fractional/negative/zero/oversized child count, numeric strings or booleans for counts, missing fields. The legacy runtime itself does not validate all count inputs; the UI must do so before calling it. Do not claim its original self-test establishes admission safety.

Conclusion: shared identity, overlap, recursive naming and declared ordered-product arithmetic in the source reference model. Not coordinate dynamics, Hilbert-space dimension under unspecified assumptions, quantum advantage or hardware execution.

wp05-mixture — Can these two known views separate source and foreground?

Input: y1, y2 in a single declared common signal unit U, and known dimensionless second response a; rational strings. First response is fixed as y1=s+f; second is y2=s+a f. Unit field must be U for this synthetic profile. Signed conventional components are permitted. Unknown calibration is outside this profile.

Computation: if a≠1, f=(y2−y1)/(a−1), s=y1−f, rank=2, then substitute into both original equations. If a=1 and y2=y1, rank=1, UNDERDETERMINED; return no unique s/f. If a=1 and y2≠y1, rank=1, INCONSISTENT for this exact noiseless model; return no s/f. Do not divide by zero or manufacture a least-squares answer.

Expected: (5,8,2) → (s=2,f=3). (5,5,1) → UNDERDETERMINED. (5,8,1) → INCONSISTENT. (5,4,2) → (6,−1). (0,0,2) → numeric components 0,0, not absence. Nearly dependent: (5,5003001/1000000,1001/1000) → (1999/1000,3001/1000); compared with (s=2,f=3), errors are −1/1000,+1/1000. Nonzero determinant gives structural uniqueness but not useful measurement precision.

Rejects: missing/blank/nonfinite/malformed rational, zero denominator, unit mismatch, booleans, numeric JSON coercion, native expression tokens. a=1 is admitted and receives its meaningful outcome.

Conclusion: exact rank/reconstruction for known conventional responses; not a general inverse of source T(S,M_o,F,H,R), and not evidence that a physically absent foreground follows from a fitted zero.

wp17-residue — Propagate the declared retained state and test the null

Input: λ dimensionless per declared discrete step; κ in U_r/U_u; B in U_y/U_r; r0 and every η in U_r; every driver u in U_u; baseline/foreground/observer contributions in U_y. All are rational strings. Arrays drivers and eta have equal length 1–32. Output includes r0 through rN and y0 through yN. Unit labels are fixed as {observation:U_y,residue:U_r,driver:U_u}. No automatic inferred noise, initial state, cadence or fitted coefficient.

Computation: r[t+1]=λ r[t]+κ u[t]+η[t]; y[t]=g+foreground+observer+B r[t], with declared measurement error fixed to exact zero for this finite profile. Show mapped residue B r[t] separately. A finite output is not a stability proof. λ is not silently restricted to [0,1]; any admitted rational is computed, and growth is shown honestly.

Expected fixture: λ=1/2,κ=1,B=3,r0=0,u=[2,0,1,0],η=[0,0,0,0],g=10,foreground=1,observer=−1/2 → r=[0,2,1,3/2,3/4], y=[21/2,33/2,27/2,15,51/4]. B=0 leaves the same r but y is constant 21/2. κ=0,r0=2,u=[2,0,1],η=0 → mapped residue [6,3,3/2,3/4], so κ=0 alone is not the observational null. κ=0,r0=0,η=[1],u=[0] → next residue state 1 U_r and mapped observation contribution B r1=3 U_y. Scale κ,r0,η by 2 and divide B by 2 → same observations with doubled latent states.

Rejects: length mismatch, empty/over-limit arrays, missing initial state/noise, unit mismatch, malformed/nonfinite rational, booleans, native-power strings. Zero κ/B/λ/r0/driver/noise are legitimate numeric parameters with different consequences; none means the record is absent.

Conclusion: declared conventional finite recurrence and its conditional null/scale ambiguity. Not cosmological data, a Planck likelihood, physical residue, general stability, posterior fit or a Bayes factor.

celsius-scale — Convert a typed thermal reading without changing its status

Input: quantity temperature or temperature-interval; state reported, unavailable or unspecified; unit degC or K; literal origin .1; source provenance/roles retained outside the numeric result. A reported value is a rational string. Missing states instead carry a nonempty reason and no value. Units are not guessed. This standard scale comparison is conventional metrology, not native arithmetic.

Computation: for temperature, K = degC + 5463/20 and degC = K − 5463/20; for temperature-interval the numerical size is unchanged under this unit change. Return the exact converted rational and round-trip rational. Non-reported state is preserved with null numeric outputs and no conversion attempt. For this ordinary thermodynamic-temperature profile reject reported T<0 K (equivalently t<−5463/20 °C); intervals may be negative. The profile does not address special negative-temperature statistical systems. Original precision, uncertainty and source token remain unchanged; exact conversion cannot add calibration or measurement information.

Expected: 0 degC temperature → 5463/20 K → 0 degC; 100 degC → 7463/20 K → 100; −40 degC → 4663/20 K; 0 K → −5463/20 degC; interval 10 degC → 10 K; interval −10 K → −10 degC. unavailable/unspecified → same state and no numeric value. .1 stays .1. No ratio of Celsius readings is labelled a ratio of thermodynamic temperature or heat.

Rejects: missing state, blank/unknown unit, value on a missing state, missing reason, numeric zero substituted for missingness, origin alias, temperature below admitted bound, null/boolean/malformed rational. Example −273.16 degC rejects; −273.15 degC is admitted. Do not infer a water phase or boiling/freezing outcome without pressure/composition and a separate physical model.

Conclusion: exact typed affine/interval transformation and missingness preservation, not thermometer accuracy, thermal energy or phase-transition simulation.

Integration review protocol

1. Author exports exact test-view HTML/scripts and the mapping from study section to test ID. No prose may imply a facility that is absent.
2. Independent tests run exported computation entry points against fixtures.json; inspect actual admission checks, source module identities, and printed units/scope.
3. Browser actor separately verifies live input→result, rejection, Return to study, and reconstruction/route custody. An isolated oracle pass is never described as a browser pass.
4. Retain final package hashes and browser receipt hashes; match each reported input/output to the contract. Preserve earlier failed artifacts and annotate fixes.

Source custody: see source-manifest.json. Preserved source vectors: preservation-vectors.json. Fixed independent arithmetic/reference fixtures: fixtures.json. Rerun: python3 harness.py, then node source-replay.cjs.

quantum-bell — Prepare an ordered Bell state and inspect a valid measurement

Status: reviewed candidate, not active merely because a local oracle passes. The browser author must install the candidate through the authorized KEX authoring route, export the active module bytes and its source identity, and obtain a fresh target-runtime readback before presenting it as active.

Candidate: candidate-p15-runtime.js, SHA-256 3a7ac468654c82809a4327d91ef95b596be4f81fd7bd8ace7e0a70a8727d129e. Original retained P15 SHA-256 7c00d68f35135504e8a0348611c877ff47208e6d579e0d02160e5cfaf8a2eeca remains historical and unchanged. The candidate's 62 grouped admission checks preserve the original nine built-in checks; these are not 71 independent physical experiments. All seven UI facilities must report which concrete code version actually executed.

Displayed task: “Apply H to the left qubit, tensor the right |0⟩, apply left-control/right-target CNOT, then inspect a supplied Born-interval draw and its normalized successor.” A second labelled phase control may compare H→S on the left with S→H; arrows denote execution order, not textual matrix multiplication order.

Input: mode bell, phase-after-H, phase-before-H, conditional-map, or gate-reject; explicitly supplied measurement sample is a finite primitive JavaScript number with 0≤sample<1. The UI must reject blank input before conversion and must not pass undefined to the module's RNG default when claiming reproducibility. Numeric zero and negative zero are valid draws. A string such as "0.25", null or false is not a numeric draw at the module API. All states, amplitudes, probabilities and samples are dimensionless; displayed standard basis labels are not claims of absent states.

Basis/order contract: K_NEG maps to |0⟩=[1,0], K_POS to |1⟩=[0,1]. Tensor nesting is left outer/right inner; labels are [00,01,10,11]. CNOT maps 00→00, 01→01, 10→11, 11→10. Default sequence produces Φ+=(|00⟩+|11⟩)/√2, amplitudes [(1/√2,0),(0,0),(0,0),(1/√2,0)], probabilities [1/2,0,0,1/2], and a nonseparable state (the predicate separable2 returns false). The complex pair is (real,imaginary).

Phase control: left H then S then tensor |0⟩ then CNOT gives (|00⟩+i|11⟩)/√2. Left S then H instead gives Φ+ because S|0⟩=|0⟩. Equal computational-basis probabilities do not erase the difference in relative phase. Display amplitudes and the operation order, not only probabilities. This is a conventional finite complex state-vector reference, not evaluation of native superscripts.

Independent expected measurements: default Bell sample 0, 0.25 → 00 with successor probabilities [1,0,0,0]; sample 0.5, 0.75, or 0.9999999999999999 → 11 with successor [0,0,0,1]. Use the candidate's actual computed probabilities and cumulative ordering for boundaries: its 1/√2 scalar gives p00=0.4999999999999999, not bit-exact decimal 0.5. At a cumulative boundary move to the next positive-support entry. A valid maximum draw must never fall through to a trailing zero-probability state. No claim of an empirical frequency distribution follows from one deterministic draw.

Gate/map distinction: H, S and CNOT must use applyUnitary. The map diag(2,1) is not unitary and applyUnitary rejects it with E_NONUNITARY_GATE. Under the separately labelled apply conditional-map API, applying it to |+⟩ gives normalized probabilities [4/5,1/5]; this API alone does not provide a physical success probability, a complete instrument, or CPTP-channel proof. Projection diag(1,0) applied to |+⟩ yields |0⟩; its zero-output branch on |1⟩ rejects with E_ZERO_VECTOR. Measurement is a conditional successor operation and is not required to be unitary.

Rejects: sample below zero, at/above one, NaN, either infinity, strings/null/booleans; nonfinite matrix coordinates; invalid/ragged dimensions; all-zero/near-zero state; invalid exposed predecessor norm. The retained candidate's constructor deliberately keeps finite Number-coercion compatibility for amplitudes, while matrix and sampling admission is stricter. The UI does not expose arbitrary amplitude expressions and must not claim the candidate removed that compatibility.

Numerics: ideal expected amplitudes/probabilities use absolute comparison tolerance 1e-12 in the independent teaching oracle; exact source/candidate regression tests retain their original tolerances and exact nominal comparisons. Candidate EPS remains 1e-10, unitarity predicate tolerance 1e-9, separability squared-determinant-modulus threshold |det|²<1e-16 (equivalently |det|<1e-8). Do not loosen these to make a failing input pass. quantum-fixtures.json retains fixed ideal values and actual rejection codes; original portable vectors retain explicit IEEE nonfinite tags.

Conclusion: correct ordered reference-state preparation, finite input admission, valid sampled outcomes and normalized successors under the pinned candidate. Not native quantum substrate, hardware execution, physical equivalence, Bell-inequality violation, quantum advantage or a universal demand that every quantum operation be unitary.