C.26:8 - Common Anti-Patterns and How to Avoid Them
| Anti-pattern | Symptom | Repair |
|---|---|---|
| Quantum-like as prestige word | The case is only complex, uncertain, nonlinear, discrete or hard to measure. | Use the adequate subject account. A QL use requires the named obstruction, assumptions and consequence under :4. |
| Precautionary suppression | QL wording is rejected because it is unusual without testing the claimed contextual-model obstruction. | Recover the exact claim and apply its ordinary subject patterns. Retain QL-lite only if a named contextual-model obstruction survives and changes the admissible inference or action. If that comparison is unavailable, keep the proposed QL use unresolved; absence of a named ordinary pattern alone does not activate it. |
| Physical overread | The text sounds as if organizations, services, or teams are physically quantum systems. | Cite inherited QL-NQ; rewrite the claim as mathematical or representational. |
| Passive dashboard | A metric or score is used as a neutral fact after its publication or operational use changed behavior. | Use measurement and evidence patterns and, if needed, C.26.1. |
| Faithful-copy export | A survey, report, API response, or context map is treated as the live state itself. | Use bridge/export loss, C.26.2, or ordinary publication patterns. |
| Unsupported property or advantage | A shortcut is called linear, fast, cheaper or compressed without the grounds for the asserted property or comparison. | Establish a linearity claim from the operator, domain and linearity argument. For a comparative advantage, use the baseline, quantity, conditions and support required by :4.5; retain the losses, use limits, alternatives and return conditions relevant to that conclusion. |
| Hidden search problem | The option menu is frame-bound, but the text tries to solve it by naming QL. | Use the applicable search method to construct alternatives; examine a separate contextual-model obstruction only if it changes that work. |
| Cell-like service jump | A service or access bearer is called cell-like because it has a boundary or internal state. | Use A.6.P:4.11a to recover only the boundary, controlled-exchange, state, viability, behavior, coupling, resource, invariant, repair, or continuity claim the current decision needs, then use that claim’s subject pattern. Do not assemble the possibilities as one service bundle. Retain the analogy only for a residual QL issue that changes the decision. |
Near misses remain ordinary uses unless the separate :4 activation test establishes a contextual-model obstruction.
| Near miss | Why not QL by itself |
|---|---|
| Feedback loop | Use dynamics/control to recover the effect on reading, export or comparison before testing a further obstruction. |
| Metric gaming | Publication-induced behavior and its effect on a state reading can remain an ordinary metric/incentive problem. |
| Uncertainty | Ordinary epistemic uncertainty unless an exact probe/model frame, comparison frame, or effective reference scheme changes variable identity or comparison law and one named contextual-model obstruction remains. |
| Complexity | Complexity and a difficult shortcut, export or probe issue do not themselves identify a contextual-model obstruction. |
| Compression | A.6.3.CSC, A.6.3.RT, modeling, or implementation pattern first; QL only for state-representation residue. |
| DDD bounded-context cue | Direct boundary, local-sense, work, and model-use subject patterns first; the label does not identify a universal object or activate QL. Retain C.26 only when a named probe, order, comparison, or export obstruction changes the admissible inference. |
| Low-bit or quantized implementation | Engineering representation first; not QL because it is “quantized”. |
| Collective behavior | A.15, distributed cognition, routines and evidence patterns can support a bounded state reading without QL. |
C.26:8.1 - Cluster conformance scenarios
Use these as quick applicability tests. The ordinary result, including a result supplied by C.26.1–C.26.3, may be sufficient. Only a further named contextual-model obstruction activates C.26; the scenarios do not establish one merely by mentioning a child pattern.
| Scenario | Expected route | Avoid | Expected output |
|---|---|---|---|
| Teams prioritize the checks displayed on the readiness dashboard. | Use C.16 and A.15 to establish what was measured when. Use C.26.1 if the interaction affects the intended reading. | Taking a change in the displayed score as proof that the release condition improved. | Use existing evidence for the conditions it covers. Recover another observation when the release depends on an unresolved condition. |
| Workshop “discovers” a boundary but also creates alignment and local meaning. | A.6 / A.15 first; C.26.1 for false pre-probe discovery reading. | Exporting the workshop result as a timeless domain fact. | Record window, participants, carriers, unresolved rivals, and bridge/export limits. |
| API read warms cache and changes later timing. | Recover interface semantics and the sampling/write sequence; use C.26.1 while the output’s interpretation is unresolved. | Treating all timings as observations of the same cache regime. | Retain the supported pre-read or post-read quantity, or name the missing timing premise. |
| Two service health reports use different measurement frames. | C.16 / F.9 first; C.26 only if no admissible shared comparison frame remains. | Averaging the scores as one posterior-looking value. | Name the frame difference and either build an admissible comparison route or stop comparison. |
| Team survey says “aligned”, but incident behavior contradicts it. | Recover the observed work and use C.26.2 for the collective-state question. | Treating survey output as the team state. | State the bounded reading and any serious alternative that changes it. |
| Market “expects” a feature because many actors change behavior. | Identify the bearer or trace population and examine the domain, incentive and evidence account. | Inferring expectation or a collective System from parallel traces alone. | Retain the supported behavior or state claim, distinguishing it from the explanation of the movement. |
| Latency is green while support load and customer promise degrade. | C.25 / C.16 first; C.26.3 for multi-characteristic envelope regulation under disturbance when a candidate intervention, boundary condition, adaptation cost, or failure mode matters. Retain C.26 / QL only when the residual contextual-model obstruction condition holds. | Calling one green metric viability. | Add envelope variables, actuators, costs, and failure mode. |
| Summary compresses an architecture decision for executives. | Use an adequate coarsening account; retain QL only for a named residual obstruction. | Assuming the summary retains every decision-relevant distinction. | Use the summary where its contents suffice; recover omitted detail when the receiving decision depends on it. |
| Diagram translates the same system into graph form. | A.6.3.RT first; no QL unless incompatible representation, probe, or export cue remains. | Calling any diagram a QL state model. | State how the representation language or the reasoning medium changed and which source content must remain recoverable. |
| Low-bit model approximates expensive simulation. | Modeling, approximation, compression, or implementation pattern first; QL only if the shortcut claim depends on QL state, probe, or frame admissibility. | Treating low-bit or linear form as QL activation. | State the approximation and its relevant loss or error and use limits. If a performance advantage is claimed, supply its comparison under :4.5. |
| Assurance load is raised because a QL explanation is published, reused or written formally. | Identify the receiving conclusion and what support it needs under :12b. | Making a fuller record without resolving a consequential question. | Reuse the adequate account; add only the missing mathematical, empirical or assurance contribution needed for the new use. |
| Author claims QL is faster or better than a classical method. | Supply the named baseline, comparison quantity, conditions and grounds under :4.5. | Accepting superiority rhetoric. | State the supported comparison, or qualify or withdraw the unsupported advantage. |
The ordinary accounts and any supported QL finding can inform design options to test under the subject patterns:
| Problem | Design option to test |
|---|---|
| Dashboard changes behavior destructively. | Delay publication, split private and public metrics, add independent sampling, or publish confidence and loss boundaries. |
| Workshop creates alignment but is reported as discovery. | Record pre-workshop hypotheses, post-workshop commitments, and created boundary meaning separately. |
| API read disturbs state. | Add non-mutating read, shadow read, sampling window, idempotence declaration, or separate observation channel. |
| Metrics in two contexts are not comparable. | Build a bridge or coupling record or stop comparison. |
| Summary is overused as source. | Recover the omitted distinction needed by the receiving use, or retain the summary when its contents suffice. |
| Viability scalar hides damage. | Build envelope variables and actuators; add a failure-mode sensor. |
C.26:8.2 - AI and LLM work-cycle route examples
LLM-mediated work cycles can involve false passive reads, overused summaries, unsupported shared comparison frames and shortcuts without loss limits. The routes below recover the ordinary claim first. A C.26 child can supply its ordinary result; QL still requires the :4 activation test.
| AI case | Route |
|---|---|
| LLM summary of an architecture record | A.6.3.CSC, A.6.3.RT, and A.10 as applicable; overuse of a summary alone does not activate QL coarsening. |
| Prompted model evaluation changes model or prompt behavior | C.16 / B.3 first; C.26.1 only if the eval output is treated as a passive model-state read. |
| Agent work cycle “discovers” requirements | A.15 / A.10 / A.6 first; C.26.1 only if the interaction created the requirement framing. |
| Synthetic personas “represent market state” | A.10 / B.3 first; C.26.2 only if a low-recoverability state-reading claim is bounded with non-admissible use visible. |