Slant · Brand Size-Up
The system map
Two systems, one seam, one direction of travel. The library holds the discipline and knows nothing about your brand. The engine knows your brand and borrows the discipline. What crosses between them — and what is forbidden from crossing — is the part worth reading first.
Automated, no human in the step
Human judgment required
Hard gate — failure halts the run
Artefact or store
01
The methodology library
Versioned, linted, regression-tested. Knows nothing about any particular brand.
SELF-MAINTAINING
SOURCE LAYER
~90
primary sources ingested —
books, papers, courses,
case studies
+ Slant's own project archives
GATED INGEST
plan → test batch
→ verify → scale
Deterministic post-checks:
links resolve, sources resolve,
raw layer untouched.
Two-agent verification.
THE CORPUS · 270 PAGES
Concepts 77
Methods 52
Strategy 15
Entities (who said it) 97
Synthesis (contradictions) 29
OPERATING DISCIPLINE
Every page states a confidence
band and its sources.
Pages are superseded,
never deleted.
Contradictions get a page,
not a shrug.
FIVE SCHEDULED ROUTINES — WHAT RUNS, HOW OFTEN, AND WHAT IT IS ALLOWED TO CHANGE
LINT
1st & 15th
13 hygiene checks
Broken links, orphans,
stale confidence,
unresolved contradictions,
index drift,
client-name scan.
READ & REPORT
SCOUT
weekly
Sources new material
Searches against current
gaps and proposes what
to read next.
Structurally incapable
of ingesting anything.
PROPOSE ONLY
EVAL
monthly
67 gold questions
A regression test for a
knowledge base. Re-run
after every ingest.
A falling pass rate means
an ingest broke something.
READ & REPORT
INGEST QUEUE
daily
Drains the backlog
One pre-approved item
per run. Added only after
the lint caught approved
work sitting undone.
THE ONLY ONE THAT WRITES
BACKLOG
quarterly
Re-aims the queue
Asks what the library
should learn next, given
what it now knows and
where it keeps coming
up short.
RECOMMEND ONLY
THE SEAM · ONE FILE, ONE DIRECTION
The diagnostic lens bank
59 diagnostic questions mapped to the six dimensions · 35+ named lenses,
each reduced to the one question it lets you ask · ~102 KB, rebuilt from
diagnostic-tagged canon · the lint verifies both copies byte-for-byte.
DIAGNOSTIC ONLY —
NEVER THE TREATMENT
Naming the problem ships.
Fixing it is a different engagement.
HONEST PROPORTION
28%
of the engine's loaded
instruction text is library-
derived. The other 72% is
engine-native, built from
real runs and what broke.
RETURN PATH
doctrine only —
abstracted method,
written brand-agnostically
✕ no brand data
✕ no client content
✕ no findings
02
The audit engine
Knows your brand in detail. Borrows its vocabulary from upstairs; gets its discipline from being wrong before.
WHAT IT READS
Intake — 4 fields
Company, site, name, email. The
substantive context comes after.
Outside-in reconnaissance
Your actual site and copy, not a
model's memory of it. Every social
platform navigated directly. Forum
and review mining against a hard
floor. Per-SKU data where it exists.
Where else the brand is encountered.
EVIDENCE STANDARD
Every claim carries a source tag
recording how it was obtained.
Inference is capped, never promoted.
HOW IT SCORES — COMPUTE FIRST, NARRATE SECOND
Six dimensions × 4–5 written criteria
CLARITY
CONSISTENCY
DIFFERENTIATION
RESONANCE
AVAILABILITY
CREDIBILITY
EACH CRITERION IS SCORED AGAINST A WRITTEN ANCHOR
100 fully · 75 mostly · 50 partial · 25 weak · 0 absent
…and tagged for what it rests on:
VERIFIED
INFERRED
ABSENT
Dimension = plain mean of its criteria, equal weight.
No weighting knob — a knob is a place to hide a judgment.
Confidence is derived from the tag mix, not graded by hand.
BANDS LEAD · STRONG 70–100 · SOLID 50–69 · EMERGING 0–49
WHO YOU GET COMPARED TO
Three sets, three selection rules.
Every comparison declares its lens.
1 · Shelf Set
Selected by revealed preference —
whoever pays to categorise you.
2 · Size & Structure Peers
Revenue, ownership, channel, scale.
The guard against unfair benchmarks.
3 · Same-Play Scan
Anyone running your move, any size.
Not-yet-threats included on purpose.
An unlabelled comparison is a hard
fail at the critique gate.
03
The gate stack
Nine checks between "the analysis is finished" and "the deck is right." No retries around, no warnings, no ship-with-caveats.
01
Assumption audit
Every load-bearing assumption surfaced and rated.
HUMAN WALKTHROUGH
keep · kill · reshape
02
Senior-strategist critique
Six lenses turned adversarially on the engine's own conclusions. Amendments and kills applied before anything advances.
03
Fact-check
Every claim traced to a primary source: verified, unverifiable, or wrong. A single false claim cannot ship.
04
AI-ism audit + empathy sweep
Generated-sounding prose fails. So does language that punches down — a diagnosis delivered with contempt is a worse diagnosis.
05
Asset acquisition
Logos and photography fetched and placed. A text-only deck is a failure, not a stylistic choice.
06
Image semantic verification
A vision pass actually looks at each candidate. A machine that never looked at the picture cannot know it's wrong.
07
Visual build gate
Enforced in code. Throws on a missing wordmark, on an unsourced attributed quote, and on any slide that had to shrink to fit.
08
Claim–source–scope trace
Run against the rendered deck, because rewriting a line to fit a slide can quietly break the source beneath it.
09
Regulatory sweep
Two-directional, primary sources only: catches recommending a violation, and inventing a rule that doesn't exist.
HALT → FIX THE CAUSE → REGENERATE → RE-RUN
THE ARTEFACT
22 slides + source appendix
HTML → PDF, typefaces embedded
HUMAN — LOAD-BEARING
Final read before anything leaves
Findings have been overruled here on every run
DELIVERED
Two business days
Compute time is ~30–45 minutes
THE GAP IS THE POINT
The difference between those
two numbers is a person reading it.
Inside the engine — the long-form walk through every part of this map. ·
How it works — the shorter version. ·
The Brand Size-Up — price, deliverable, turnaround.
Every count, threshold and rule shown here is drawn from the system's own specifications, logs and run records.
Where a claim could have been stated more impressively than the evidence supports — the 28% figure, the firewall's one recorded breach, the reproducibility gap — it is stated at the evidence instead.