luminarforge.
Pricing Compare Trust Log in Start free
The writing room — where the copy is made

A writing room, not a text box.

Five specialist benches take every post from angle to press — Strategist, Writer, Editor, Humaniser, Condenser — with a sixth on call for revisions. Every mark on this proof is one of them working.

Start free — 2,000 credits Read the marks ↓

A full post runs 250–500 credits — most land near 300 · on the rate card

Most tools show you a result. We show you the working.

Fig. 1 — galley, card №1847 · LinkedIn Illustrative proof — marks keyed below
① Strategist — the angleRepair-first routine — not another 10-step listicle. ②

Your skin barrier doesn’t fail in January. It fails in October — quietly — and spends three months sending you the bill.

④

In today’s fast-paced world, hydration matters more than ever. Most “winter skin” advice treats the symptom: more cream, more often.

③

The fix is sequence, not quantity. Repair, then seal — a study says 68% of adults layer wrong most routines layer these in the wrong order — and the barrier never gets a chance to close.

⑤

Start tonight: strip the routine to three steps and give it two weeks.

Condensed: 342 → 264 words · target 220–280 — fits the platform Gate: 84 — passed for press
③ = evidence check — unverified stat reframed ④ = AI pattern struck
The legend — six hands, read in order

Every mark has a bench behind it.

No single do-everything prompt. A specialist chain — each bench does one job, hands the copy on, and reports to the console as it works. The messages below are the production console’s own.

①

Strategist

Plans — never writes

Reads the brief and the client’s brand brain, then sets the thesis, narrative arc and hook — the angle note pinned to the proof. Strategy only; the copy comes later.

Analysing brief for “…” → Selected angle · 3 supporting points
②

Writer

Drafts in the client’s voice

Writes the post from the strategy — headline, body, hashtags, CTA — in the client’s Voice DNA. It may cite only from the verified-evidence list; anything unverified must be framed as a general observation.

Loading voice profile… → Draft complete — 342 words
③

Editor

Cuts — doesn’t rewrite

A surgical pass for depth, specificity and flow — targeted edits, not rewrites. It re-checks every cited statistic against the same evidence list the Writer used: the circled stat on the proof is this bench catching an unverified claim.

Reviewing for depth, specificity and flow… → Applied 6 editorial edits
④

Humaniser

Strikes the tells

The last creative pass — hunts the phrasing that makes AI copy feel like AI copy, against a maintained library of 205 banned patterns, and fixes rhythm, not just words. The struck line on the proof is its work.

Scanning for AI patterns… → 4 patterns addressed
⑤

Condenser

Makes it fit

Deterministic length enforcement against the platform’s real limits. It counts, and only cuts when the draft runs over — if the copy already fits, it touches nothing.

Condensing from 342 to 220–280 words… → 264 words
⑥

Improver

On call — the revise desk

Runs when you send a post back. It reads the score’s targeted brief and makes scalpel edits to the failing dimensions — under explicit orders not to degrade the ones that pass. “Think scalpel, not sledgehammer” — its own instructions.

Analysing quality gaps… → Applied 3 quality improvements

In a hurry? Quick mode runs the same chain with fewer hands — it skips the Strategist and Editor when you just need a fast draft. Full quality is the default.

Marks ①–⑤ on every draft · ⑥ on revisions
The console — no black box

You watch it happen, live.

Every bench streams its progress to the console as it works — what it’s doing, what it found, what it handed on. If a stage stumbles, you see where. If the copy is good, you saw it get that way.

Console — one runIllustrative transcript
StrategistSelected angle: the repair-first routine
WriterWriting… (247 words)
WriterDraft complete — 342 words
EditorApplied 6 editorial edits
HumaniserHumanisation complete — 4 patterns addressed
CondenserCondensed: 342 → 264 words (target: 220–280)
GateOverall 84 · lowest dimension 71 — passed for press
The gate — scored before it ships

We don’t ask the AI to rate itself.

Every post is scored across eight dimensions by deterministic NLP — real readability math, real entity counts, real sentence-variance measurement. Six weight the overall; two are informational.

The gate: overall 70 or better, no dimension below 50.

Fall short and the score names which dimension failed and why — then the revise desk fixes that, and only that.

DimensionWeightWhat it measures
Information density22%Named references and specific claims vs filler
Readability18%Grade level, sentence & paragraph variance
AI resistance18%Cliché density against the 205-pattern library
Engagement18%Hook strength and momentum
Platform SEO14%Keyword integration, platform conventions
E-E-A-T10%Experience and expertise signals, citation quality
AI Overview readinessinfoReported, not weighted
Snippet readinessinfoReported, not weighted
Pass gateOverall ≥ 70 and no dimension < 50
The room’s two enforcers
The spike — banned patterns

205 AI tells, banned by name

A maintained, versioned library of the phrases that make AI copy smell like AI copy — enforced twice: as never-use rules at write time, and scanned for at score time. If one slips through, the score drops and says why.

Let’s dive in… I’m thrilled to announce… It’s not just X — it’s Y.

Opening 26 · transitions 29 · LinkedIn tells 30 · formatting 31 · closing 25 · hollow intensifiers 31 · structural 33

Seven categories · maintained & versioned
The revise desk — bench ⑥

Fix what failed. Protect what passed.

One click sends a post back with its score attached. The Improver gets a targeted brief — what failed and why, and which passing dimensions it is forbidden to disturb. When two dimensions pull against each other, a resolution table settles the priority.

FIX — AI resistance (48): 2 clichés detected — use specific data, not curiosity-gap formulas.
PASSING — DO NOT DEGRADE: readability (88) · engagement (84) · platform SEO (82)
“Think scalpel, not sledgehammer” — its own instructions
What feeds the room
Feed 01

The brand brain

Every client’s knowledge base — positioning, products, personas — structured from their own site and documents, per workspace. The Strategist plans from it; the Writer writes from it.

Feed 02

Voice DNA

A measurable voice profile — eleven quantitative style metrics plus extracted traits, learned from real writing samples. The Writer drafts in it; the Humaniser polishes against it.

How Voice DNA works →
Feed 03

The verified-evidence list

Hallucination control that is architectural, not a policy promise: the prompt itself says “you may ONLY cite from this list.” No verified evidence? Then claims must be framed hypothetically — enforced at the Writer and re-checked by the Editor.

Feed 04

Research-informed briefs

Keyword gaps, competitor angles and fresh findings synthesised into the brief before a word is written — and verified stats flow into the evidence list, so research becomes citable, not decorative.

Said plainly — how we keep this honest
On AI detectors

We score resistance to AI tells against a maintained pattern library — heuristics, honestly applied. Nobody can promise “undetectable,” so we don’t.

On what a post costs

A full run through the room is 250–500 credits, most near 300 — measured from live production runs, published on the rate card. Failed runs aren’t charged.

The proof is ready

Put six specialists on your next post.

Start free with 2,000 credits — paste a client’s URL, brief the room, and read the marks on your first proof. Roughly a week of real production, on us.

Start with 2,000 free credits See the rate card

No card · No trial countdown · No per-client fees

luminarforge.
LinkedIn-native today · Instagram & Facebook rolling out
Features
AI Pipeline Voice DNA Campaigns Templates AI Imagery Publishing SEO Hub AI Suggestions
This page
The benches The gate The rate card Compare
Catalogue
The print room Pricing Trust & security Start free Log in
Company & legal
Contact Terms Privacy Cookies

The writing room’s proof and console transcripts on this page are illustrative; the stage messages are the production console’s own, and scores shown are examples, not guarantees. Quality scoring is deterministic NLP analysis — resistance to AI tells is a heuristic measure, not a detection guarantee. © 2026 Luminar Works.