Proof Studio · Evidence Review Desk

Review the evidence before the story.

This is the public cross-case proof surface: every visible evidence row, observed value, source trail, export row, and limiting condition in one place. Evidence handles are live review links paired with routes you can actually open before using the JSON export.

Open artifact map Review case matrix Download evidence CSV Open evidence JSON Back to Proof Studio

HERMES cold review brief

What a buyer can inspect now.

Skip to full register

Read this as a launch-room triage: the question the proof answers, the visible row behind it, the value and source to challenge, the boundary that prevents overclaiming, and the next route to open.

  1. 01 Publicly inspectable
    Buyer question

    One brand system, multiple surface moods

    A design lead wants to see whether source-backed surfaces can change register without becoming a pile of one-off styles.

    Best public row

    Three registers are rendered from one systemThe public specimen shows warm, neutral, and cool surface registers while keeping the same component skeleton.

    Limit
    This proves visual system survival, not buyer conversion.
  2. 02 Publicly inspectable
    Buyer question

    Cold start vs. system-backed work

    A team wants to know whether a new agent produces better work when it starts from the project system instead of a bare task prompt.

    Best public row

    No productivity number is publishedThe public case labels the result as a measured pilot and explicitly avoids broad ROI or productivity lift claims.

    Limit
    The result supports a launch proof, not a statistical productivity claim.
  3. 03 Publicly inspectable
    Buyer question

    The starter kit runs locally

    A buyer wants to know whether the starter kit can be inspected without waiting for a hosted account.

    Best public row

    Hosted account is not required for local inspectionThe local inspection path reads extracted starter-kit files and does not require hosted authentication.

    Limit
    Hosted identity, payment operations, service visibility, and plan changes belong to the Phase 2 hosted service.
  4. 04 Publicly inspectable
    Buyer question

    A council run beats one generic lens

    A team is facing a hard product-planning decision and needs architecture, journey, verification, operations, and business judgment separated before synthesis.

    Best public row

    Five lanes are separated before synthesisThe public projection separates architecture, journey, verification, operations, and business lanes before the synthesized recommendation.

    Limit
    This is one sanitized run, not a statistical benchmark.

Case evidence matrix

One row per case, with the data first.

Download CSV

This is the quick evaluator view: strongest public row, observed value, source handle, limitation, and direct links before the long register.

Case Strongest row Value Source Limit Open
01 · measured pilot

Cold start vs. system-backed work

A team wants to know whether a new agent produces better work when it starts from the project system instead of a bare task prompt.

publicly inspectable No productivity number is published

The public case labels the result as a measured pilot and explicitly avoids broad ROI or productivity lift claims.

Evidence value Claim boundary: pilot

Good enough to inspect; not a broad numeric ROI claim.

Source trail Comparison matrix

review table · public projection

The result supports a launch proof, not a statistical productivity claim.

02 · qualitative proof

The page built from its own source

A buyer wants to see whether KBPD can make a product surface trace back to structured source material instead of a hand-built page.

publicly inspectable The source-backed route is published

The demonstration exists as an inspectable public route and is included in the static site build.

Evidence value Run date: Launch build

Date label attached to this public proof case.

Source trail Route source

structured route record · public projection

It demonstrates traceability rather than measuring outcome lift.

03 · demonstrator

Specialist lenses produce different work

A team wants to understand why KBPD uses specialist roles instead of one generic assistant voice.

publicly inspectable Eight role samples are published

The public role gallery exposes eight launch-safe roles with a role caption, capability statement, and concrete output sample.

Evidence value Featured roles: 8

The public page shows eight launch-safe examples with one artifact sample each.

Source trail Role samples

homepage component · public projection

Samples are static in the first launch.

04 · demonstrator

A council run beats one generic lens

A team is facing a hard product-planning decision and needs architecture, journey, verification, operations, and business judgment separated before synthesis.

publicly inspectable Five lanes are separated before synthesis

The public projection separates architecture, journey, verification, operations, and business lanes before the synthesized recommendation.

Evidence value Specialist lanes: 5

Architecture, journey, verification, operations, and business lenses are shown separately.

Source trail Council projection

sanitized diagram · public projection

This is one sanitized run, not a statistical benchmark.

05 · demonstrator

One brand system, multiple surface moods

A design lead wants to see whether source-backed surfaces can change register without becoming a pile of one-off styles.

publicly inspectable Three registers are rendered from one system

The public specimen shows warm, neutral, and cool surface registers while keeping the same component skeleton.

Evidence value Theme registers: 3

Warm, neutral, and cool registers are generated from the same brand source.

Source trail Specimen page

public route · public projection

This proves visual system survival, not buyer conversion.

06 · measured pilot

One review frame for many demos

A team has many experiments and needs a shared way to compare them without turning the site into a folder of disconnected artifacts.

publicly inspectable Registry is curated, not automatic

The current registry is intentionally curated and raw experiment assets still require redaction before publication.

Evidence value Run date: Launch build

Date label attached to this public proof case.

Source trail Demo registry

public projection · public projection

The first registry is curated, not automatically generated.

07 · qualitative proof

The starter kit runs locally

A buyer wants to know whether the starter kit can be inspected without waiting for a hosted account.

publicly inspectable Hosted account is not required for local inspection

The local inspection path reads extracted starter-kit files and does not require hosted authentication.

Evidence value Local tools: 6

Health, search, context read, proof index, lint, and diagnostics are exposed through the local server.

Source trail Local MCP README

starter-kit docs · public projection

Hosted identity, payment operations, service visibility, and plan changes belong to the Phase 2 hosted service.

08 · null result

Controlled benchmark is not public yet

A technical evaluator wants a controlled, reproducible answer to whether structured project context improves agent work.

publicly inspectable No public lift number exists

The benchmark case publishes no numeric lift result and labels the current result as null.

Evidence value Current public number: none

The site does not claim a measured productivity lift yet.

Source trail Benchmark design

measurement plan · public projection

No benchmark result exists on the public site yet.

Openable proof

Open the artifacts before reading the register.

The strongest public proof is the system and specimen corpus: live surfaces, cold/backed pairs, measured artifact values, and clear claim boundaries. The harbor side-by-side fixture stays useful, but it is a supporting lab.

Flagship path

System surface and retained compositor.

Start with the public switcher, then read the retained eight-adapter compositor boundary as the next publication target.

Live specimen views
4
Retained adapters
8
Generated stories
596
Public demo corpus

Seven cold/backed microsites with values.

This is the better sales proof than one narrow app: visual hierarchy, composition, voice, grid, perception, product judgment, and typography.

Microsite pairs
7
Measured values
21
Strongest public delta
441 -> 18
Supporting lab

Harbor side-by-side fixture.

Use this when you want a compact mechanical comparison. It is not the flagship proof surface; it is one bounded fixture inside the larger corpus.

Cold checks
6/14
Backed checks
14/14
Claim status
Demo
Campaign review page

Color hierarchy specimen

Cold output
KBPD-backed output
Buyer question

Can the same brief become a clearer visual system when the run has a retained palette contract?

Measured value

Near-value collisions: 441 -> 18 Lower is better: fewer color pairs collapse into the same visual weight.

Boundary

One fixture about palette discipline. It does not prove conversion lift or brand preference.

Evidence you can operate

Start with the artifact, then check the row.

Each route opens a public surface first: a lab, microsite pair, local runbook, or exact evidence row. The audit ID stays available after the thing itself is in view.

All specimen pairs

Open every cold/backed microsite pair.

These are the stronger public artifacts than the single harbor fixture: each row has an output to inspect and a measured value to challenge.

  1. 01
    Color hierarchy specimen Campaign review page Near-value collisions: 441 -> 18
  2. 02
    Composition specimen Open-house landing page Dominant element ratio: 2.889 -> 6.905
  3. 03
    Voice specimen Operations SaaS page Word count: 318 -> 294
  4. 04
    Grid specimen Article and brief layout Hand-computed widths: 17 -> 3
  5. 05
    Perception specimen Shift board Proximity token uses: 0 -> 52
  6. 06
    Product-decision specimen Roadmap review page Outcome/output ratio: 0.08 -> 0.73
  7. 07
    Typography specimen Editorial article Hardcoded pixel font sizes: 11 -> 0

Demo source map

What is public, what is retained, what comes next.

The proof program is larger than one fixture. Start with the public surfaces below, then read the retained boundaries before treating any result as a broader claim.

Public families
6
Open actions
24
Retained families
1
Next publications
2
01 · Public now

System surface lab

Can one retained system produce several buyer-facing surfaces without becoming a one-off demo?

Evidence in view
Four public specimen views plus a retained eight-adapter compositor record and story count.
Limit
This proves surface continuity and inspection depth, not buyer conversion.
02 · Public now

Specimen microsites

Can a reviewer inspect more than one polished example and compare the cold/backed outputs directly?

Evidence in view
Seven cold/backed microsite pairs with measured values for color, composition, voice, grid, perception, product, and type.
Limit
Artifact-quality evidence. It does not claim market lift or universal design superiority.
03 · Public now

Paired pilot matrix

What changed when the same scenario was run cold and then with retained project context?

Evidence in view
Five paired cells with word-count, self-sufficiency, source-boundary, role-fit, and risk signals.
Limit
Measured pilot evidence only. It is not a statistical productivity claim.
04 · Public now

Side-by-side fixture

Can the buyer open two shipped outputs and see the working-record difference without trusting a summary?

Evidence in view
Two public product surfaces and a fourteen-check matrix showing which decision, journey, design, verification, and handoff checks are present.
Limit
Bounded fixture evidence. It should not be presented as the flagship proof.
05 · Public now

Local MCP starter kit

Can a buyer inspect KBPD locally before a hosted service exists?

Evidence in view
Six local tools, a local proof index, a smoke path, and a Claude Desktop package boundary.
Limit
Local review proof. Shared account controls come in a later release.
06 · Public now

Persona council fanout

Can separated specialist passes produce a more reviewable decision than a single generic answer?

Evidence in view
Five public lanes, corpus ownership, cold-baseline board, and source-boundary notes.
Limit
Qualitative fanout proof. It does not prove every persona improves every task.
Retained source

Retained compositor runtime

Is there more system depth behind the public specimen than the current page exposes?

Visible today
Retained adapter and story counts are summarized on the system lab.
Boundary
Full runtime artifacts stay retained until dependency, license, and publication hygiene clear.
Next publication
Publish a sanitized compositor comparison gallery.
Next publication

Full-triad design run

What happens when structure, sequence, system, taste, copy, and verification all act on the same demo?

Visible today
The public site names this as the next stronger publication target instead of hiding it behind the small fixture.
Boundary
Retained working artifacts need a sanitized case frame before publication.
Next publication
Turn the strongest retained design run into a public case with before/after surfaces and a claim limit.
Next publication

Preregistered benchmark

When does KBPD earn a numeric lift claim instead of a qualitative proof claim?

Visible today
The current public pages hold numeric lift back and route buyers to pilot or artifact evidence.
Boundary
Benchmark harness and results stay unpublished until preregistered gates pass.
Next publication
Publish reproducible benchmark results only after the preregistered validity gate passes.

Review order

Evidence row → observed value → source → limitation.

01Evidence rowwhat is being reviewed
02Observed datawhat the page claims happened
03Source trailpublic or retained artifact
04Claim limitwhat this cannot prove

Quick inspection

Three rows a buyer can audit immediately.

Open full register
  1. One brand system, multiple surface moods

    Seven cold/backed specimen pairs are public

    The brand-system case links seven public specimen pairs across color, composition, voice, grid, perception, product decision, and typography surfaces.

    Limit
    The color, product-decision, and typography values are proxy checks from bounded specimens, not broad benchmark results.
    Review evidence row
  2. One brand system, multiple surface moods

    Color hierarchy specimen reduces near-value collisions

    The color specimen reports near-value collisions moving from 441 in the cold output to 18 in the system-backed output.

    Limit
    This proves visual system survival, not buyer conversion.
    Review evidence row
  3. One brand system, multiple surface moods

    Retained compositor depth is recorded but bounded

    The public case records eight retained UI-kit adapters and 596 retained generated stories as the next publication boundary.

    Limit
    This proves visual system survival, not buyer conversion.
    Review evidence row

Handle explorer

Open the evidence handles without reading JSON first.

Structured export
01 Cold start vs. system-backed work 8 evidence handles · 3 source handles
02 The page built from its own source 7 evidence handles · 3 source handles
03 Specialist lenses produce different work 8 evidence handles · 2 source handles
04 A council run beats one generic lens 14 evidence handles · 10 source handles
Case value handles
Case source handles
05 One brand system, multiple surface moods 15 evidence handles · 8 source handles
Case value handles
Case source handles
06 One review frame for many demos 7 evidence handles · 2 source handles
07 The starter kit runs locally 8 evidence handles · 4 source handles
08 Controlled benchmark is not public yet 7 evidence handles · 2 source handles

Status mix

Every row says what kind of proof it is.

Publicly inspectable
26
Public projection
5
Internal retained
1
Validated by build
4
Pending
1

Full register

All public evidence rows and their review pointers.

Evidence JSON
  1. Cold start vs. system-backed work Five paired scenarios were reviewed Public projection · evidence_facts[0]
    Observed

    The pilot evidence set contains five anonymized tasks run once from a prompt-only starting point and once from a system-backed starting point.

    Evidence

    Run metadata lists five paired scenarios; the comparison matrix is retained as the public projection source.

    Limit that constrains this row

    Small sample size.

  2. Cold start vs. system-backed work System-backed outputs carried named context Public projection · evidence_facts[1]
    Observed

    The system-backed summaries include inherited decisions, named reusable assets, and explicit limitation language that the cold summaries did not consistently carry.

    Evidence

    Before/after excerpts and metric notes on the proof detail page name reuse, inherited decisions, and explicit limitations as the primary reviewed signal.

    Limit that constrains this row

    Anonymized scenarios only.

  3. Cold start vs. system-backed work No productivity number is published Publicly inspectable · evidence_facts[2]
    Observed

    The public case labels the result as a measured pilot and explicitly avoids broad ROI or productivity lift claims.

    Evidence

    Claim status, metric note, and limitations all state that this is pilot evidence rather than a statistical productivity result.

    Limit that constrains this row

    The result supports a launch proof, not a statistical productivity claim.

  4. The page built from its own source Four source families are bound to the route Public projection · evidence_facts[0]
    Observed

    The page is backed by route, process, narrative, and code-emission records rather than a standalone marketing brief.

    Evidence

    Run metadata and source artifacts name the structured route record, narrative record, code plan, and process trace.

    Limit that constrains this row

    This is one route, not the whole product.

  5. The page built from its own source The source-backed route is published Publicly inspectable · evidence_facts[1]
    Observed

    The demonstration exists as an inspectable public route and is included in the static site build.

    Evidence

    The proof case links to `/how-it-works`; build output includes `/how-it-works/index.html`.

    Limit that constrains this row

    It demonstrates traceability rather than measuring outcome lift.

  6. The page built from its own source Numeric lift is held back Publicly inspectable · evidence_facts[2]
    Observed

    The page avoids a numeric improvement claim until the controlled comparison clears its gates.

    Evidence

    The proof case metric identifies claim discipline as the primary gate.

    Limit that constrains this row

    This is one route, not the whole product.

  7. A council run beats one generic lens Five lanes are separated before synthesis Publicly inspectable · evidence_facts[0]
    Observed

    The public projection separates architecture, journey, verification, operations, and business lanes before the synthesized recommendation.

    Evidence

    Council fanout visual, transcript storyboard, and mechanism stage each show five named lanes converging only after lane outputs exist.

    Limit that constrains this row

    This is one sanitized run, not a statistical benchmark.

  8. A council run beats one generic lens Color-separation lane board is public Publicly inspectable · evidence_facts[1]
    Observed

    The public persona page now shows five persona lenses as layered color-separation plates that converge into one resolved buyer artifact.

    Evidence

    The `Color-separated proof` board links the same retained corpus source into Hephaestus, Artemis, Cassandra, Mnemosyne, and Argus/HERMES lanes, then resolves them into a decision, limitation, and buyer-action artifact.

    Limit that constrains this row

    The source prompt and lane outputs are not public until redaction is complete.

  9. A council run beats one generic lens Cold baseline blends the evidence jobs Publicly inspectable · evidence_facts[2]
    Observed

    The public case describes the cold start as one recommendation where architecture, journey, verification, operations, and business judgments are fused together.

    Evidence

    Before artifact, storyboard step 01, and the T3 proof-row table all name the cold baseline as one blended response without separate source standards.

    Limit that constrains this row

    The run gallery is a redrawn projection, not raw screenshot evidence.

  10. A council run beats one generic lens T3 proof rows are visible before the transcript Publicly inspectable · evidence_facts[3]
    Observed

    The public persona page shows a five-row T3 table before asking the reader to accept the broader claim.

    Evidence

    The T3 table lists each lane, the cold-start failure mode, the corpus-backed signal, source class, and buyer decision in one visual review surface.

    Limit that constrains this row

    The capture review board is a publication map, not the underlying private capture set.

  11. A council run beats one generic lens Cold baseline review board is public Publicly inspectable · evidence_facts[4]
    Observed

    The public persona page now shows the cold artifact beside the KBPD-backed artifact and lists five baseline checks before the T3 table.

    Evidence

    The `Cold baseline artifact` section includes role boundary, source standard, claim pressure, public proof path, and language gate checks, each with an evidence-row link.

    Limit that constrains this row

    The public page demonstrates reviewability and role separation, not automatic correctness.

  12. A council run beats one generic lens Raw screenshots are intentionally withheld Publicly inspectable · evidence_facts[5]
    Observed

    The proof case publishes redrawn mechanisms and explicitly reports zero public source screenshots.

    Evidence

    Metric row `Public source screenshots: 0`; source artifact list marks retained run captures as private evidence; the run gallery labels itself as a public projection.

    Limit that constrains this row

    This is one sanitized run, not a statistical benchmark.

  13. A council run beats one generic lens Capture review board maps private screens to public proof Publicly inspectable · evidence_facts[7]
    Observed

    The public persona page shows five capture-review rows that connect retained screenshot classes to public frames, redaction boundaries, and exact evidence-row links.

    Evidence

    The `Capture review board` section includes panel launch, architecture lens, journey/verification lenses, operations lens, and business-language gate rows, each with retained signal, public frame, redaction boundary, and an evidence link.

    Limit that constrains this row

    This is one sanitized run, not a statistical benchmark.

  14. A council run beats one generic lens Four corpus source classes are represented Publicly inspectable · evidence_facts[8]
    Observed

    The public mechanism shows role roster, prompt archive patterns, prior decisions, and project source as inputs to the run.

    Evidence

    Metric row lists four corpus sources; source artifact list names role registry and prompt archive sources.

    Limit that constrains this row

    This is one sanitized run, not a statistical benchmark.

  15. A council run beats one generic lens HERMES and cold-reader gates are named Publicly inspectable · evidence_facts[9]
    Observed

    The public case names the public-language lane and the cold-reader gate as corpus owners before the result becomes buyer-facing copy.

    Evidence

    Corpus ownership cards and the T3 proof-row table identify HERMES, archive discipline, session memory, and cold-reader review as separate responsibilities.

    Limit that constrains this row

    This is one sanitized run, not a statistical benchmark.

  16. One brand system, multiple surface moods Three registers are rendered from one system Publicly inspectable · evidence_facts[0]
    Observed

    The public specimen shows warm, neutral, and cool surface registers while keeping the same component skeleton.

    Evidence

    Metric row names three theme registers; visual evidence shows three surfaces sharing one component structure.

    Limit that constrains this row

    This proves visual system survival, not buyer conversion.

  17. One brand system, multiple surface moods Design outputs are named source artifacts Public projection · evidence_facts[1]
    Observed

    The case identifies generated design variables, the public specimen route, and motion rules as source artifacts.

    Evidence

    Source artifact list names design outputs, specimen page, and component runtime motion rules.

    Limit that constrains this row

    Accessibility remains a gate per surface class.

  18. One brand system, multiple surface moods Accessibility remains a gate Publicly inspectable · evidence_facts[2]
    Observed

    The proof does not claim every visual register is automatically ready for every accessibility context.

    Evidence

    Limitations section states that accessibility remains a gate per surface class.

    Limit that constrains this row

    The seven microsites are artifact-quality checks; they do not prove market lift or user preference.

  19. One brand system, multiple surface moods Seven cold/backed specimen pairs are public Publicly inspectable · evidence_facts[3]
    Observed

    The brand-system case links seven public specimen pairs across color, composition, voice, grid, perception, product decision, and typography surfaces.

    Evidence

    The specimen proof matrix and specimen gallery both expose cold and system-backed links for all seven public microsites.

    Limit that constrains this row

    The color, product-decision, and typography values are proxy checks from bounded specimens, not broad benchmark results.

  20. One brand system, multiple surface moods Twenty-one specimen values are visible Publicly inspectable · evidence_facts[4]
    Observed

    Each public specimen lists one primary value and two secondary values, for twenty-one visible measured values across the gallery.

    Evidence

    The specimen gallery summary and per-specimen detail panels show the value labels, cold value, backed value, and interpretation.

    Limit that constrains this row

    The retained multi-kit runtime is not public until publication hygiene and dependency validation are complete.

  21. One brand system, multiple surface moods Color hierarchy specimen reduces near-value collisions Publicly inspectable · evidence_facts[5]
    Observed

    The color specimen reports near-value collisions moving from 441 in the cold output to 18 in the system-backed output.

    Evidence

    The specimen matrix, gallery selector, and evidence values all show the same 441 -> 18 color hierarchy delta.

    Limit that constrains this row

    This proves visual system survival, not buyer conversion.

  22. One brand system, multiple surface moods Product-decision specimen shifts toward outcomes Publicly inspectable · evidence_facts[6]
    Observed

    The product-decision specimen reports outcome/output ratio moving from 0.08 in the cold output to 0.73 in the system-backed output.

    Evidence

    The public product-decision microsite row names the buyer question, the primary value, the cold value, and the backed value.

    Limit that constrains this row

    This proves visual system survival, not buyer conversion.

  23. One brand system, multiple surface moods Typography specimen removes hardcoded pixel sizes Publicly inspectable · evidence_facts[7]
    Observed

    The typography specimen reports hardcoded pixel font sizes moving from 11 in the cold output to 0 in the system-backed output.

    Evidence

    The public typography specimen row and gallery detail show the cold/backed value and the bounded interpretation.

    Limit that constrains this row

    This proves visual system survival, not buyer conversion.

  24. One brand system, multiple surface moods Retained compositor depth is recorded but bounded Internal retained · evidence_facts[8]
    Observed

    The public case records eight retained UI-kit adapters and 596 retained generated stories as the next publication boundary.

    Evidence

    The system-surface lab and evidence values name the retained adapter count and story count while stating that the full runtime is not public yet.

    Limit that constrains this row

    This proves visual system survival, not buyer conversion.

  25. One review frame for many demos Eight proof families are registered Validated by build · evidence_facts[0]
    Observed

    The public registry currently contains eight proof cases spanning comparisons, demos, a case study, a surface trace, and a benchmark placeholder.

    Evidence

    Proof Studio index renders eight cards; validator output reports eight proof cases passed validation and leak scan.

    Limit that constrains this row

    The first registry is curated, not automatically generated.

  26. One review frame for many demos Every public case has status and limits Validated by build · evidence_facts[1]
    Observed

    The proof-case validator rejects entries without claim status, limitations, visuals, metrics, source artifacts, and evidence facts.

    Evidence

    `web/scripts/validate-proof-cases.ts` is part of the web prebuild and exits nonzero on missing required fields.

    Limit that constrains this row

    Raw experiment assets require redaction before publication.

  27. One review frame for many demos Registry is curated, not automatic Publicly inspectable · evidence_facts[2]
    Observed

    The current registry is intentionally curated and raw experiment assets still require redaction before publication.

    Evidence

    Limitations section states the first registry is curated and raw experiment assets require redaction.

    Limit that constrains this row

    The first registry is curated, not automatically generated.

  28. The starter kit runs locally Six local MCP tools are declared Validated by build · evidence_facts[0]
    Observed

    The starter kit local MCP contract exposes health, search, context read, proof index, lint, and diagnostics tools.

    Evidence

    Metric row lists six local tools; source artifacts name the local MCP README, local server, smoke command, and manifest.

    Limit that constrains this row

    The local server works against extracted kit files only.

  29. The starter kit runs locally Hosted account is not required for local inspection Publicly inspectable · evidence_facts[1]
    Observed

    The local inspection path reads extracted starter-kit files and does not require hosted authentication.

    Evidence

    Metric row reports zero hosted dependency; limitations distinguish local server behavior from the later hosted service.

    Limit that constrains this row

    Hosted identity, payment operations, service visibility, and plan changes belong to the Phase 2 hosted service.

  30. The starter kit runs locally Two smoke paths cover the local server Validated by build · evidence_facts[2]
    Observed

    The proof case records two smoke paths: extracted-package Node smoke and Docker image smoke.

    Evidence

    Metric row names two smoke paths and run metadata names one assembled package plus one Docker MCP image.

    Limit that constrains this row

    Smoke tests prove startup and representative calls, not full hosted account behavior.

  31. Controlled benchmark is not public yet No public lift number exists Publicly inspectable · evidence_facts[0]
    Observed

    The benchmark case publishes no numeric lift result and labels the current result as null.

    Evidence

    Metric row says current public number is none; claim status is null result.

    Limit that constrains this row

    No benchmark result exists on the public site yet.

  32. Controlled benchmark is not public yet Three benchmark arms are planned Public projection · evidence_facts[1]
    Observed

    The planned benchmark compares project system, generic docs, and competent prompt-only baseline arms.

    Evidence

    Metric row lists three planned arms; after-artifact summary describes separating structured files from generic docs and prompt-only baselines.

    Limit that constrains this row

    The final claim may be narrow or null.

  33. Controlled benchmark is not public yet Run artifacts are pending Pending · evidence_facts[2]
    Observed

    The source artifact list marks run artifacts as pending, and the limitations state no public benchmark result exists yet.

    Evidence

    Source artifacts include pending run artifacts; limitations explicitly state the result may be narrow or null.

    Limit that constrains this row

    The result will be published only after the validity gates pass.