☰ Contents
AISA v2.0 / Technical documentation / Challenges Register (appendix)

Challenges Register (appendix)

F verified factP decided planC open challenge

Generated by pwsh scripts/wiki_challenges.ps1 from every C statement in the wiki; the triage columns come from challenges_triage.csv. Edit the challenge where it is made and the triage in the CSV — never this page.

Every row carries the investigation or proposed solution of 02.09.2026 where one exists — the measured evidence is in Measurements 2026-09-02 and the two challenge rounds in Challenge Rounds 2026-09-02 — matched by page and text, so the pointer survives renumbering. The CH-nn ids are assigned at generation time, in document order: they renumber whenever a C is added or removed, so cite a challenge by its page and subject, never by its id. Classes follow the decision register: A needs evidence that exists (a query, a file, a repo); B needs a decision by the owner; C is deferred by design — it needs data that does not exist yet and names the event that will produce it; D is a build item — settled by building it. ? means not yet triaged.

Class Count
A — evidence 6
B — decision 4
C — deferred by design 29
D — build item 1
Total 40

A — evidence

Id Page · section Challenge Owner Closes when Investigation / proposed solution
CH-10 Rating Files · Challenges Cache invalidation across replicas is asserted here but the invalidation call has not been exercised from the platform. First deployment to a multi-replica environment settles what the deployer must send. platform team first deployment to a multi-replica environment the invalidation call has not been exercised from the platform
CH-12 Stage Abacus · Challenges Confirm the QT obsoletion state on the target branch before a production-bound product whose process carries a quotation step (Stage IPAL remark); UI evidence so far is from serdica-ui master, backend from bulstrad-staging (KI-054 documents drift). Vladimir QT obsoletion landed on the target branch (Source work item) Challenge Rounds § CS: production simulation criteria (B — Vladimir + Bulstrad)
CH-17 Stage Application and Verification · UI-driven application LDAP provisioning of aisa.ui. (Bulstrad IT); whether „Enable login" runs under a non-administrator role. Vladimir + Bulstrad IT the aisa.ui.<env> accounts exist on QA (password in the directory; roles from the platform - D134) no platform identity exists in PROD IPAL today (Challenge Rounds § R2 § 3: seven EMPL accounts)
CH-21 Agents Memory · 7.3 The past-case entries P Past cases are not articles; their entry shape is § 5. Retrieval mechanics — title/grep index versus vector search (the estate retains Qdrant as a rebuildable index [F: docs/architecture/health-ai/00-program.md, inspected 31.08.2026]) — are decided by m… support author measure both retrieval mechanics on the migrated corpus (Agents Memory § 7.2) Repo evidence: retrieval mechanics and the two-independent-uses rule
CH-30 Memory Derived Capabilities · Challenges The four channel clients are not built where the workspace says they are. Source and a Release build exist for three of them, one has no build output at all, and every skill names a publish path none of them occupies. v2 must decide whether it inherits them… Vladimir dotnet publish of the four CLIs or corrected paths in CLAUDE.md and the v1 skills Memory-derived capabilities § Challenges: three CLIs have a non-publish Release build
CH-37 Decision Register · D20 — PL/SQL and git ✅ ANSWERED by research remaining: the polling cadence is uncalibrated — nobody has measured how often the developer saves, and that parameter drives the whole design. Needs one live measurement. Vladimir one live measurement of how often a developer saves during a session measure the save interval on a real session and set the watcher cadence from it; until then the design keeps the parameter named and unset

B — decision

Id Page · section Challenge Owner Closes when Investigation / proposed solution
CH-07 Premium Vocabulary · 6. What this leaves open — and who has to decide it The premium-algorithm choices have no owner in the twelve parts. Part 5 carries the tariff and the LD rows and part 8 the INSIS conditions and taxes (Configuration Module § 7.1), but nothing in the parts menu or in the ownership table names the write-off me… S1 author + Vladimir the next existing-product change that touches a cancellation or a mid-term annex (CF-04 / CF-06 / CF-18) a premium_rules section of the product specification carrying the four settable choices (write-off method, short-rate table, rounding target, loading/discount application order) with a value per product; quote verification asserts them on a cancellation and a mid-term change, not only on new business. The two code-side choices (tax application, post-annex plan) stay Development plugin items
CH-18 Support Module · 0. The agents P Waiting is a route, not a gap (D147). A request that can wait moves to a waiting state that names three things. Its trigger: a step only the customer can perform — the customer-question step type of Stage Solution Take § 1; a step only an external owner… Vladimir a decision whether a deferred wish pauses the contract clocks (clocks.pause_states) D147 defined the waiting state and left the deferred-wish pause state open
CH-23 Memory Derived Capabilities · Challenges The operational-recovery symptom domain has no owner of record. Its five operations are filed under the transfer investigator, and the argument above says they should not be. This needs a D-row either way, because it decides whether privileged host executio… Vladimir a D-row after the first ten incident cases (Stage Investigation § 2 measures reassignments and final leads) Memory-derived capabilities § Challenges: operational recovery stays with the transfers investigator until the cases show whether privileged host execution needs a profile of its own
CH-28 Memory Derived Capabilities · Challenges The health payment packet leaves the estate. It produces a file a bank consumes, which is not a target write, a message or a document — the write classes in Gating do not obviously contain it. Vladimir a Gating write class for files consumed outside the estate (D-row) Memory-derived capabilities § Challenges: the bank file is an external effect; it needs its class before the health payment skill publishes

C — deferred by design

Id Page · section Challenge Owner Closes when Investigation / proposed solution
CH-01 Delivery · 5. Work streams Open, from the component inventory: platform team the first plug-in's build the lead-in to the component-count row below, not a challenge of its own
CH-02 Delivery · 5. Work streams The component counts of § 2.1–§ 2.4 are what this wiki names today. The first plug-in's build finds what was still assumed; the inventory is amended in the same change, like the Whitelabel Catalogue. platform team the first plug-in's build amends the inventory in the same change D94 - the inventory is Delivery 1.4; amended like the Whitelabel Catalogue
CH-03 Source Details · DG-3 — Environment = branch set across the source's repositories Its internals are therefore sourced from the branch's retirement dossiers, not from reading its code; the source/intentgpt domain is built on the first case that touches it. source author first Source case touching intentgpt Challenge Rounds § CS: the source/intentgpt domain
CH-04 Value and ROI · 5. How the ROI number is produced P P The protocol for the two missing inputs (Challenge Rounds § R2 § 10): ten consecutive real interruptions across at least three developers and both channels, four timestamps each written as they happen — T0 noticed · T1 ticket work starts · T2 ends · T3 … development manager the ten-interruption CSV exists (two calendar weeks) Challenge Rounds § R2: § 10 switch-cost protocol + 25 % rule per type
CH-05 Components · Challenges The acceptance bar for a configured tariff — one reconciling quote, or a set across a factor matrix — is undecided and defines what verified means; settle it on the first slice (Product design § 11 DEC-6). S2 author the first abacus-only slice fixes the number of worked examples reconciliation 3 DEC-6; same question as Quote Verification
CH-06 Factor Sources · Challenges Whether the locality probe runs once per environment at connection time or per factor at plan time is open; the three queries are cheap enough for either, and the first case with a remote-sourced factor decides. S3 author the first case with a remote-sourced factor either shape is affordable; the case decides
CH-08 Quote Verification · Challenges How many worked examples make an acceptance — one reconciling quote or a factor matrix — is DEC-6 of the reconciliation, still open, and is fixed by the first slice. S2 author the first abacus-only slice reconciliation 3 DEC-6; same question as the Components acceptance bar
CH-09 Rating Files · Challenges The applyAlways and vocabulary checks above are assertions the estate has no tooling for. Whether they belong in the abacus configure skill or in a standing configuration.abacus.validator check is decided with the first abacus-only slice. S2 author the first abacus-only slice decides skill or standing check the assertions are the same either way
CH-11 Shared Vocabulary · Challenges Whether the tier distribution is measured once per customer at plug-in time or re-measured per case is open — it is the same question as the dated conventions of Product Configurator § A-19. S1 author the second customer plug-in reconciliation 1 A-19 - conventions are measured per estate, each count carrying its date
CH-13 Components · Challenges A source that is not in the workspace has unmapped seams; its node is built on the first case that touches it. source author the first Source case touching that source Challenge Rounds § CS-2.1: the source/intentgpt domain - stub now, summarizer on first case
CH-14 Source Module · 0.1 Its branch of the memory tree built on first case — repo absent here; known: Beth+AI components, health path being replaced source author first intentgpt case Challenge Rounds § CS: the source/intentgpt domain
CH-15 Data and Information Module · Challenges The HDesk and mail streams — counted 02.09.2026 (Challenge Rounds § R2 § 1) F: hdesk/ is the OTRS queue "Ablera", the customer's product-change and CR queue — 39 of its 52 tickets are Configuration-module intake, not support; the OTRS support stream rea… support author re-count on a full read-only export of the OTRS queue Challenge Rounds § R2: § 1 HDesk and mail streams
CH-16 Frequent Operations · Helper consolidation The seven operational-recovery rows — database-link backlog, broker channel, process halt, storage exhaustion, replica freeze, interrupted automation, business residue — are homed with support.investigator.transfers because their symptom usually arrives as … Support module owner (Vladimir) the first ten incident cases measured per Stage Investigation § 2 (reassignments and final leads) Frequent Operations § Helper consolidation: infrastructure-recovery rows stay with the transfers investigator until the incident cases show whether a seventh symptom domain pays for itself
CH-19 Agent Framework · Challenges Eval-set budget: a full re-run on every memory write is unaffordable; batching by domain, on a schedule and before publish, is the proposal — measured against the spend envelope (Metrics) once real numbers exist. platform team first eval cost measurement in the T1 experiment batching by domain and on a schedule; a full re-run only before publish (Agent Framework § 6.5)
CH-20 Agent Runtime · Challenges Compaction quality — what a summary must keep so that a resumed agent does not re-derive settled facts; measured on the T1 experiment's resumed cases. platform team resumed cases in the T1 experiment Challenge Rounds § CS: compaction quality
CH-22 Agents · 0.6 Auto-learning a missing domain How deep a summarizer goes on first creation, and what triggers a refresh of a node that already exists — a staleness signal, a schedule, or the next case that finds it wrong. platform team first auto-created source/serdica-backend/ node Challenge Rounds § CS: summarizer depth and refresh
CH-24 Memory Derived Capabilities · Challenges A tool owned by one module and bound by another module's skill has no stated precedent. The rating client is Configuration's, but every pricing investigation calls it; the print probe is Data's, but the configuration document-set change needs the same read.… Agents Memory owner (Vladimir) the first cross-module tool binding under the tool-package contract Memory-derived capabilities § Challenges: one owner per package; a foreign skill binds the published digest and proposes changes to the owner
CH-25 Memory Derived Capabilities · Challenges Does the twelve-part product build survive as one skill? It is the largest single procedure in the v1 workspace, and its distinctive act is the scope conversation — proposing the parts a specification implies and asking which are in scope. The five configur… Configuration module owner the first full-product case (CF-16) shows whether the scope conversation needs an owner beyond the normalizer Memory-derived capabilities § Challenges: the five stage skills absorb the emission; the normalizer's H1 review is the scope conversation until a case contradicts it
CH-26 Memory Derived Capabilities · Challenges Agent and office provisioning is catalogued under the access investigator (data.skill.agent-office) but half of it configures numbering counters and role audiences. Blank ranges (5 tickets asked) and vehicle catalogue entries (2) sit with master data; agent… Data & Manipulation module owner the first ten access cases Memory-derived capabilities § Challenges: catalogued as data.skill.agent-office under the access investigator; the numbering/role half may move to Configuration S5
CH-27 Memory Derived Capabilities · Challenges Is an outbound-text lint a tool or a profile rule? Ten recorded corrections make a rule list, but a check the model may overrule is not deterministic in the sense of Agents Memory § 19. It publishes as a tool only if its output is a typed finding the commun… communicator owner the third recorded correction after go-live Memory-derived capabilities § Challenges: a rule list in the communicator's method until its output is a typed finding the communicator must answer
CH-29 Memory Derived Capabilities · Challenges What seeds the past-case index — the archive miner or the migration pass? The five mining scripts produce per-shape aggregates; Agents Memory § 7 describes a distilled per-domain migration. They differ in granularity, and running both would produce two inde… Agents Memory § 7 migration owner migration wave 3 picks one index Memory-derived capabilities § Challenges: the distilled per-domain migration is the index; the archive miner's aggregates are evidence for it
CH-31 Memory Derived Capabilities · Challenges The health knowledge base carries five overlapping editions of one domain guide and the same data model in up to four places. The inventory files it as one article row; the migration has to pick the canonical pair — one platform-generic guide and one custom… health knowledge owner migration of the health node picks the canonical pair Memory-derived capabilities § Challenges: one platform-generic guide and one customer-values guide before the node is written
CH-32 Contracts · Challenges The drafts have been exercised against the eight Configuration, ten Support and eight Development written cases, not against a running stage. The first real case of each module either confirms a core field or moves it to the open body; the count of fields m… platform architect the first running stage produces and validates a draft instance (T2 slice) the drafts stay x-status: draft until then (D139)
CH-34 Gating · Challenges The first shape catalogue: which write shapes the five largest support families and the five configuration stages need, and how many of them a single template covers. Built on the first ten cases of each module. platform team + module authors the first ten cases of each module D72 - a shape is earned from observed instances, so the catalogue cannot precede the cases
CH-35 Gating · Challenges Whether check 2 can always re-derive the affected set cheaply — for a statement over a large table the re-derivation is a count query, but for a multi-statement block it is a set of them; the budget for the auditor is a per-profile ceiling (Agents § 0.1). write auditor author the first HW-approve on a shape over a large table a count query for the affected set; a sampled re-derivation where the count is not cheap
CH-36 Decision Register · D27 — the one-AI-runtime rule ✅ TAKEN, then softened by verification remaining: whether the health-AI owner wants the rewording, the rename, or neither. It no longer blocks anything here. health-AI owner the health-AI owner answers on the wording, or the scope sentences are reworded without them the decision stands: do not rename; reword the 16 scope sentences (1-2 hours against 1-2 days for the rename). Deferred by design - it blocks nothing in this wiki
CH-38 Decision Register · D11 — Investigator domain boundaries ✅ TAKEN remaining: most real cases span systems, so which investigator leads a cross-system symptom still needs a rule. It is set from the first ten cases rather than from a table drawn now. support.root profile owner the first ten cross-system cases the desk runs on the platform deliberately unset: which investigator leads a cross-system symptom is read off the first ten real cases instead of a table drawn now; the root routes by symptom in the meantime
CH-39 Measurements 2026-09-02 · RE-8 Retrieval mechanics and the two-independent-uses rule — the measurement and the v1 baseline — CH-50, CH-66, CH-69 The two-independent-uses promotion rule still cannot be measured: the corpus has no "use" events (which article an analysis actually relied on, versus merely linked). The held-out set above is the right sample; the missing data is a per-case "relied-on" mar… support author the ledger produces per-case relied-on marks the held-out set of the same page (section TB) is the right sample; the missing datum is which article a case relied on, not which it linked
CH-40 Product Configurator · Q. What the documents leave open, and the wiki already answers The configurator's seven open questions DEC-1 … DEC-7 (Product design § 11): DEC-1 (file output first) is the abacus-only slice in files mode (A-20); DEC-2 (single- or multi-tenant) is the plug-in model; DEC-3 (documents leaving the region) is C-9; DEC-4 an… S2 author the first slice fixes the number of worked examples (DEC-6) DEC-1 to DEC-5 and DEC-7 are answered by the wiki; only DEC-6 is open

D — build item

Id Page · section Challenge Owner Closes when Investigation / proposed solution
CH-33 Data Model · Challenges The scripts have not run: SRD_SUPPORT is not created (DB-7) and the QA backup path must be restored first (D45/D131). Closed by the first acceptance-install-clean.sql run on QA, whose output is pasted here. Vladimir + Bulstrad DBA SRD_SUPPORT created on QA after the backup path is restored (D45, D47, D131) install-clean and the acceptance scripts are the entry points; lint_ddl.py passes offline
PreviousRequirements Register (appendix)NextGlossary