wf5-vs-qdash-delta.md
c16 · wf5 vs. qdash — Value & Feature Delta
A side-by-side of two takes on the AI-agent control plane:
- wf5 — bob-wire
c16wireframe, "persona-first synthesis." Route:/c16/wireframes/wf5/control-plane-v1/*· static lo-fi, placeholder data, 23 pages. - qdash / qcontrol — the standalone SPA at
http://127.0.0.1:7878/. Single-page, feature-first shell, backed by live/api/*telemetry (real instrumented agents, polled, with a live-sync clock). 12 views.
The goal here is the delta of value/features — what each one carries that the other doesn't, and where they cover the same surface at different depth.
TL;DR — the two axes of difference
The two prototypes diverge on two independent axes:
| Axis | wf5 | qdash |
|---|---|---|
| Organizing principle | Persona-first — 5 seats, shared pages re-led per seat (?seat lensing) | Feature-first — 4 functional sections, one flat app for everyone |
| Fidelity / data | Wireframe — static, placeholder, illustrative | Working app — real data polled from a live agent telemetry stream |
So the headline isn't "which has more features" — it's that wf5 is a wider product surface expressed as a wireframe, while qdash is a narrower surface that actually runs on real data. Each holds value the other lacks.
1. Shared surface (present in both)
Both ship these, so they're not where the value delta lives — but note the treatment differs (see §4):
Fleet Overview · Agent Inventory (master-detail) · Cost Center · Performance · Findings (CISO worklist, OWASP-LLM/MITRE-tagged) · Security Posture · Data Protection · Policy & Enforcement · Compliance & Frameworks · Teams & Users · Agent Identity & Access · Integrations · Agent Profile (drill target).
That's ~13 surfaces in common — the core control-plane spine is the same in both.
2. Only in wf5 (value qdash does not have)
These are net-new surfaces and structural ideas with no qdash equivalent:
| # | wf5 capability | The value it adds |
|---|---|---|
| 1 | Persona seats + per-seat lensing | One data model, five different first questions (Finance leads on spend-at-risk, Security on coverage, Exec on the governance index…). Feels like five tailored apps without being five apps. |
| 2 | Activity Log ("what happened?") | Reconstructed agent → action → target → outcome decision log; filter by family/flagged/tied-to-finding; per-agent rollup. The forensic / incident-replay trail. qdash has no event-history view. |
| 3 | Findings ↔ Activity pairing | Curated worklist (triage) links bidirectionally to the exhaustive record (forensics) — move from "what needs a human" to "what actually happened" in one click. |
| 4 | Effectiveness ("is it working?") | Task success / intervention rate / quality by workflow & agent — the outcome layer above Performance. qdash measures speed and cost but never whether the agent succeeded. |
| 5 | Usage Limits | Forward-looking spend ceilings/caps with cap-coverage %, an "uncapped" worklist, breach handling. qdash shows spend but offers no budget control surface. |
| 6 | Showback | Risk-weighted spend per identity, "unchargeable liability" for unattributable identities. (qdash's Teams view hints at chargeback but has no dedicated showback.) |
| 7 | Executive Governance & ROI Board | A packaged, exportable board: governance index trended to target, $-at-risk concentration, findings-backlog flow, "are the agents earning their keep?" ROI line ($/successful-task). qdash has no executive/board view at all. |
| 8 | Value-first page contract | Every page opens with purpose + computed bottom-line + recommended next action (WirePagePurpose). Articulates the "so what" before the table. |
| 9 | Lifecycle states on inventory | active / stale / forgotten / decommission / shadow — fleet hygiene & cleanup. qdash inventory is runtime-only. |
| 10 | Evidence pack (Compliance) | Previewable masthead, scope, per-framework why-trails, attestation — the artifact you hand an auditor. qdash exports findings JSON/CSV but has no assembled evidence pack. |
| 11 | Policy dry-run economics | "Cost of flipping to Enforce now," kill-switch / halt-agent. qdash shows enforcement mode but the Assist/Enforce path is labeled projected. |
Plus seat-home launchpads (5) — lensed KPI + "your surfaces" grids — which have no qdash analog.
3. Only in qdash (value wf5 does not have)
These are real, working capabilities — the value is mostly "it runs on live data," plus a few mechanisms wf5 never sketched:
| # | qdash capability | The value it adds |
|---|---|---|
| 1 | Live telemetry, end to end | Polls /api/{stats,inventory,agents,costs,security,egress,dlp,performance,activity,installations}; live-sync clock; "instrumented" filter separating real-signal agents from the rest. wf5 is entirely static placeholder data. |
| 2 | Current-vs-worst regression history | "Ever-observed risk — current vs worst" flags agents that regressed from their best posture. wf5's posture trend is projected/illustrative; this is computed from real history. |
| 3 | Content-free DLP movement, by class | Live, content-free counts of secrets-in-prompts, PII-in-responses, source-code egress — with the "legacy DLP can't see inside AI flows" framing. wf5 lists sensitive paths but doesn't break out these three live movement classes. |
| 4 | Weak-TLS / unallowlisted-egress detection | Real per-host egress checks (weak_tls, !allowlisted) surfaced on profile and posture. |
| 5 | Unattributed-access resolution gap | "Owner could not be resolved to a telemetry identity" — names the specific identities (duck.com, proxy-only) that break attribution, from real data. wf5 flags the concept; qdash shows the actual offenders. |
| 6 | Percentile-honest Performance | p95/p99 (not averages), out tok/s, explicit small-sample noise warnings tied to call count. More statistically careful than wf5's latency table. |
| 7 | est cost from a list-price table | Costs the provider didn't report are estimated from a list-price table and badged est per-row, live. |
| 8 | Master-detail depth | Inventory row → expand sessions → click session → full agent profile, all on real sessions. |
Note: qdash also practices "honesty labeling" (prev/proj/est badges, "Assist/Enforce
projected," "control mapping projected") — so that design law is present in both, not unique to wf5.
4. Same surface, different treatment
Where both have the surface, the depth/angle differs:
| Surface | wf5 | qdash |
|---|---|---|
| Cost Center | Spend by model/agent/user/provider, cache %, governance-tagged spend, est labels; + Usage Limits + Showback as separate control surfaces | Spend by agent/user/time, est rows, framed as OWASP LLM10 unbounded consumption; no caps, no showback |
| Security Posture | Score + "what to fix first" (Hardening-ROI), trend is projected | Score + factor deductions + real current-vs-worst regression |
| Data Protection | Sensitive paths/categories/users, high-severity rollup | Live content-free movement in 3 classes (secrets/PII/source), "legacy DLP is blind here" framing |
| Performance | Latency by model, slow-AND-expensive collisions, variance/red-rows | Percentiles + tok/s + sample-noise honesty |
| Teams & Users | Owner rollup: agents/spend/at-risk/findings/top-risk; "no owner to hold accountable" | Per-person, ranked by risk not spend, expandable to their agents, unattributed handling |
| Compliance | Auto-mapped frameworks + assembled evidence pack + attestation | Auto-tagged frameworks + labeled export preview (capability, not conformity) |
| Navigation | 5 persona seats → flat arc-ordered page list per seat | 4 feature sections (Assurance / Security / Identity & Access / Admin) |
5. The bottom line
wf5 is the broader product vision; qdash is the working proof.
- wf5 carries the strategy — persona-first IA, the value-first page contract, and five surfaces qdash simply doesn't have: Activity Log, Effectiveness, Usage Limits, Showback, and the Executive ROI Board. It answers "is it working?", "what happened?", and "are we on budget?" — questions qdash never reaches.
- qdash carries the proof — it runs on live agent telemetry and has the data-derived mechanisms wf5 could only sketch: regression history, content-free DLP movement classes, weak-TLS/egress detection, percentile-honest performance, and real attribution gaps.
The clean synthesis: qdash is, in effect, a live implementation of wf5's "Assurance + Security" spine (Fleet/Inventory/Cost/Performance + the five Security surfaces) for one persona — the CISO/operator — with no seats, no forward-control surfaces (caps/showback), no outcome layer (Effectiveness), no forensic layer (Activity Log), and no Executive board. If wf5's persona seats and its five net-new surfaces were layered onto qdash's live data plumbing, you'd have the full product.
Feature-count delta at a glance
- In both: ~13 core surfaces
- Only in wf5: Persona seats + lensing, seat-homes, Activity Log, Effectiveness, Usage Limits, Showback, Executive ROI Board, lifecycle states, evidence pack, value-first headers, policy dry-run economics
- Only in qdash: live telemetry/polling, regression history, content-free DLP classes, weak-TLS/egress detection, percentile performance, real attribution-gap resolution,
estpricing engine