Company overview — app docs

company-overview — the big picture

Produces one-page, investor-facing company profile slides for a client firm (default La Caisse): analyst-grade research, honest sourcing, and slides composed from evidenced components — never copied templates.

This is an AOS application workspace: it inherits the kernel — the work loop, the commit gate, the state system, the derived wiki, the audits — from the AOS kernel at commit 2e72e07c933c321129e09a238c0cbd827cc5369d (the copy is recorded in working/state/kernel.lock; an audit watches it for silent drift). What the kernel is and why: the kernel repo's docs/12-app-template.md.

What this app actually does, its parts, and their pages: written as the app is built. Nothing here pretends to exist yet.

What is here — every file, in plain language

Every file and folder in this workspace. The commit gate blocks any commit that adds a file without a line here.

Inherited from the AOS kernel (locked — see working/state/kernel.lock)

These arrived by copy from the kernel at 2e72e07c933c321129e09a238c0cbd827cc5369d; changing one needs a reason in the lock's divergences section. What each does is documented in the kernel repo; one line each here:

• working/checks/audit_staleness.sh — inherited kernel file (locked).
• working/checks/build_dashboard.py — inherited kernel file (locked).
• working/checks/githooks/commit-msg — inherited kernel file (locked).
• working/checks/githooks/pre-commit — inherited kernel file (locked).
• working/checks/inventory_check.sh — inherited kernel file (locked).
• working/checks/kernel_tether.py — inherited kernel file (locked).
• working/checks/make_app_workspace.py — inherited kernel file (locked).
• working/checks/pre_commit_hook.py — inherited kernel file (locked).
• working/checks/run_completeness.py — inherited kernel file (locked).
• working/checks/stamp_timestamps.py — inherited kernel file (locked).
• working/checks/wiring_check.py — inherited kernel file (locked).
• working/functions/add-system.md — inherited kernel file (locked).
• working/functions/derive-wiki.md — inherited kernel file (locked).
• working/functions/evaluate-panel.md — inherited kernel file (locked).
• working/functions/index.md — inherited kernel file (locked).
• working/functions/loop.md — inherited kernel file (locked).
• working/functions/retrieve.md — inherited kernel file (locked).
• working/procedure.md — inherited kernel file (locked).

This workspace's own files

• CLAUDE.md — the boot file every fresh chat reads first: what

this workspace is, its boundaries, the read order.

• docs/ — these user-facing pages (README.md, 00-big-picture.md,

01-what-is-here.md — this page, 02-status.md, 03-decisions.md, 04-deferred.md, 05-the-slide.md, 06-component-catalogue.md, 07-how-analysts-build-it.md — the three slide-documentation pages written for the owner's review, 08-your-actions.md).

• wiki/ — AI-facing derived notes (README.md explains the node

format; index.md lists every node; starts empty).

• working/state/checkpoint.md — what is open right now; the first

thing a booting chat reads.

• working/state/audits.md — the audit receipts table with its

staleness alarms.

• working/state/kernel.lock — which kernel version this workspace

copied, file fingerprints, and any reasoned divergences.

• working/lessons/ — this app's own lessons, one per file

(.gitkeep holds the empty folder).

• working/writing-check.md — the two writing standards, checked

before every commit (usage log starts empty).

• .claude/settings.json — wires the commit gate into Claude Code

(hooks only; permission mode is set per workspace by the owner).

• working/references/ — the app's authoritative sources, imported

2026-08-02 from the old repo's company-overview-app-info folder (frozen evidence; the source of truth for what the slide product is):

• authoritative/ — the analyst specs (content model, then

presentation brief) and their README.md: tier 1 2026-06-29-analyst-company-overview-spec.md, tier 2 2026-06-28-analyst-worker-instruction-one-pager.md.

• benchmark-slides/ — the seven real analyst slides, split

Tier 1/Tier 2: dfm foods.JPG, sims lifecycle services.JPG, FineToday.png, Mader.png, Tynor Orthotics.png, UGL Transport.png, ValueLabs.png, with the folder README.mds carrying the owner's verbatim annotations (do not edit).

• recons/ — per-slide structured reconstructions from the previous

build: one folder per slide with recon.yaml (every element, machine-readable) and eval-reconciled.md (the adversarial-then- reconciled evaluation of that slide's strengths and defects).

• README-imported.md files — the old folder's own maps, kept as

provenance.

• working/assets/ — house image assets: la-caisse-logo.png

(the producer firm's emblem, 164x121, canonical copy — the asset whose loss was the predecessor's defining defect).

• working/app-settings.md — the owner-visible switches: producer

firm (the toggle the owner asked for), authority order, the defaults-with-declared-deviation rule, the hard floor.

• working/analysis/ — the app's own analysis workings:

benchmark-slide-reports.md holds the seven per-slide analyses (element-by-element evaluation, formatting facts, rebuilding analysis, quality verdicts — one section per benchmark, verbatim from the 2026-08-02 agent runs) that the human-facing slide documentation is written from.

Status

The date this page was last touched is the modified: date at its top. Newest first within each list.

Done

• This app's docs are on your phone now (2026-08-02, kernel D51).

The kernel's site builder renders this repo's docs/ folder — read-only, nothing here changes — as a page of the workspace site: https://aos-8-3-1-docs.pages.dev/company-overview.html (also via the site's Sites button). Every page carries its file number in the menu, and this repo's own commit history is the page's Turns view. Practically: your review of the slide documentation (pages 5, 6, 7) no longer needs the repo or attachments.

• **The kernel's record audit reached into this repo and found one

compliance gap — fixed (2026-08-02, kernel D48).** The twelve-agent audit you asked for verified this app's three spawn-day commits (attestations true), its imports, lock, emblem, and analyses (all verify on disk) — and caught that the seed-lessons file carried no "could-be-wrong" line, which this workspace's own filing rule demands of every filed fact. The line is now added: the eight rails come from one build's history as narrated by one audit, and are floors to keep, not proof that keeping them suffices.

• **The slide documentation is written — and it stops here for you

(2026-08-02, your D45 gate).** The three pages you asked for exist, written macro to micro from the persisted analyses and the frozen references, every material claim cited: docs/05-the-slide.md (what the product is, its anatomy, the authority order, the hard floor, and all seven benchmarks dated — none is current), docs/06-component-catalogue.md (thirteen components, each with its job, its evidence, its confirmed defects, and the rule that follows — a catalogue, never a template, per your ruling), and docs/07-how-analysts-build-it.md (the reconstructed research routes and selection judgement — marked as reconstruction — plus the six confirmed cross-slide findings, led by the missing-takeaway defect). Before reaching you, a fresh-context verifier checked the pages against the evidence and found 20 defects — overgeneralised "all seven" claims, one swapped attribution, a reversed claim about where the ask sits, unglossed jargon — 19 fixed, 1 overturned on recheck with the verbatim quote now cited in place. Nothing derives from these pages until your review: no app wiki, no procedures.

• The kernel folder's rename landed here (2026-08-02). You renamed

the kernel workspace to aos-8.3.1; this app's three references to the old path (the lock file's kernel-source line and two paths in CLAUDE.md) now point at the new name — declared edits, per the kernel's deferred row 19 checklist. The kernel tether verified all 18 locked files against the renamed kernel afterwards. Because CLAUDE.md is part of the boot contract, the change carries a visible waiver in working/state/audits.md (path strings only, no rule changed) instead of a fresh boot test.

• The app's sources and safeguards are in place (2026-08-02). The

authoritative analyst specs, the seven benchmark slides (with your verbatim annotations preserved), and the per-slide structured reconstructions were imported from the old repo as frozen reference evidence; the La Caisse emblem is a real image asset (the canonical copy, verified identical across the old lineage); the predecessor's paid-for failure lessons are seeded in working/lessons/ so they cannot be lost the way the old rebuild lost them; and working/app-settings.md holds your producer-identity toggle and the standing rules (authority order, defaults-with-declared- deviation, the hard floor).

• **First kernel upgrade adopted, by hand, hours after birth

(2026-08-02).** Importing the frozen reference evidence tripped this workspace's own gate (frozen files must not be edited, so they cannot carry our timestamp headers) — the kernel gained a narrow exemption for working/references/, and this workspace adopted the fixed gate script by re-copy and re-lock: the tether shows the change openly instead of a silent divergence. This was the kernel's deferred row 15 firing for real; the upgrade procedure gets extracted from this first hand-run.

• Workspace spawned from the AOS kernel (2e72e07c933c321129e09a238c0cbd827cc5369d, 2026-08-02T15:13:14+08:00)

by make_app_workspace.py; the initial commit passed this workspace's own gate.

In progress

Nothing — the build is at your review gate. After your review of the three documentation pages: the app wiki derives from them, then the macro research-loop procedure (kernel docs/13, build approach).

Waiting on you

One page owns the list — docs/08-your-actions.md. As of now it holds one open item: your review of the slide documentation.

Decisions

One entry per decision that mattered, newest at the bottom. Format: Concluded / Source (file:line, command + output, or fetched URL — or the words "nobody has checked this") / Argument. Decisions against something are recorded the same way.

No decisions yet. Kernel-level decisions live in the kernel repo's docs/03-decisions.md and are not duplicated here.

Deferred — tracked, with what would close each

What is not fixed now gets a row with its closing condition, so nothing depends on anyone remembering. This page is on the boot path.

The slide — what this app produces and what governs it

The first of three pages that document the product before any machinery is built (your stop point, kernel D45: you review these pages before anything derives from them). This page is the top of the macro-to-micro story: what the slide is, its anatomy at a glance, whose rules outrank whose, and the seven real slides the evidence comes from — each dated, because none of them is current.

How to read the citations. Evidence lives in three places, all inside this app: the seven benchmark analyses in working/analysis/benchmark-slide-reports.md (cited here as "analysis §1"–"§7"); the per-slide reconstructions and evaluations in working/references/recons/<slide>/ (the file eval-reconciled.md is each slide's verdict of record — defect codes like "W1" come from there); and the two authoritative documents in working/references/authoritative/ (cited as "tier-1 spec" and "tier-2 brief"). A claim without a citation is my synthesis and says so.

What the product is

A one-page, investor-facing company profile slide: a single page that tells an investor what a company is, what it earns, who owns it, and — when there is a deal — what is being asked for it. It is produced by a named client firm (the producer defaults to La Caisse; the toggle lives in working/app-settings.md) about a target company.

The tier-2 brief states the quality bar in one sentence: the slide "should be designed from an investor's perspective … It should convert research into a clean, structured, investor-style company overview", and "the goal is to produce a clean one-page company profile, not a raw research summary" (tier-2 brief, opening and closing sections).

Who actually reads it varies more than a template admits. Across the seven benchmarks the observed situations are: a live private-equity exit teaser (the short marketing page a seller circulates to potential buyers), a division profile built for a carve-out decision (selling a division out of its parent company), a live secondary buyout (one private-equity owner selling to another), a listed-company investor overview, buy-side screening for a control deal (work done for a potential buyer weighing a majority purchase), an in-flight carve-out screen, and a live buyout in exclusivity (one bidder granted sole negotiating rights) — analysis, cross-slide headline facts. The composition follows the situation — slides omit "standard" blocks and can be judged right to: UGL's missing management section is confirmed correct for its use-case, and Mader relocates its one-liner into its first bullet (analysis §6, §4). Not every absence passes — UGL's missing one-liner is a confirmed defect (eval C3). That finding shapes this whole documentation: components are chosen, not templated (your ruling, kernel D44).

Anatomy at a glance — macro before micro

Every one of the seven benchmarks is built from the same three bands (per-slide inventories, analysis §1–§7):

1. Header — the producing firm's emblem at top-left, the company name as the title, and on most slides the company's own logo at top-right with a rule (horizontal line) beneath (six of seven; UGL's recorded header is mark, name, and a page number — analysis §6). Five of seven carry a one-line summary of the company or transaction under the title. The tier-2 brief prescribes the title be the company name alone ("Do not write extra wording like 'Company Profile'") — both tier-1 benchmarks violate this ("Company Profile | DFM Foods", DFM eval W9; "Company Profile | Sims Lifecycle Services", Sims eval W8), which is part of why benchmarks sit below the specs in authority. 2. Body — two asymmetric columns in six of seven benchmarks; UGL runs two columns over a full-width bottom band (analysis §1–§7 inventories). The body holds some selection of: a business description, deal terms, a products/services block, key financials, revenue splits, shareholding, operating statistics, and management. The tier-2 brief prescribes a four-quadrant layout instead; the evaluations rule that "layout non-conformance is NEVER a defect" (verbatim from the DFM eval's binding layout note, working/references/recons/dfm-foods/eval-reconciled.md), and your ruling makes the brief a default a run may depart from with a logged reason — with two-column as the prior (kernel docs/13, authority section). The full component inventory is the next page, 06-component-catalogue.md. 3. Footer — notes (currency, fiscal-year-end, footnote definitions) and a named source strip, present on all seven slides (analysis §1–§7 inventories).

Authority — whose rules outrank whose

Three sources govern content and form, in this order (your ruling, recorded in working/app-settings.md):

1. The tier-1 spec (from the analyst, 2026-06-09, verbatim in working/references/authoritative/) — the content checklist: what a company description must cover (founding, headquarters, main business, positioning with market share where available, footprint, segments, customers, KPIs, shareholding, recent events), the 3-to-5-bullet / 100-to-150-word cap, the segments table, the operating-KPI examples, and the financials block (revenue, EBITDA — earnings before interest, tax, depreciation and amortisation, the profit measure investors compare companies on — margins, CAGR (compound annual growth rate), cash-flow conversion, and net debt to EBITDA, i.e. leverage). 2. The tier-2 brief (from the analyst, 2026-06-28, verbatim alongside it) — the presentation register: name-only header, a 15-to-25-word one-liner, "founded in" not "incorporated", honest visuals only ("Do not add visuals just for decoration"), takeaway commentary under charts ("The commentary should not just describe the chart. It should explain the key takeaway."), no management section, four quadrants, investment thesis deferred to a future version. Your ruling (kernel D44): this is experienced judgement, not law — encoded as defaults with declared deviation. A run follows the brief unless the company genuinely warrants departing, and every departure is logged with its reason and surfaced for review, never silent. 3. The benchmarks (seven real slides) — evidence of practice, known imperfect, and not up to date (your note, 2026-08-02; working/app-settings.md). They show how real analysts actually compose — including where they depart from the brief and where they are simply wrong (six of seven quantitative blocks carry no takeaway line; one chart is accounting-impossible — see the table below). This documentation treats them as evidence of practice at their date, never as today's standard.

Where the sources contradict each other — four-quadrant versus two-column, management-off versus six-of-seven-slides-have-it — the resolution is your D44 ruling: brief as default, benchmark practice as evidence for when a deviation is warranted, deviation always declared.

The hard floor — what never yields

A small set of rules is mechanical, not judgement (your ruling; working/app-settings.md). A slide that misses one FAILS regardless of how good everything else is:

• The strict header: producer emblem at left (the La Caisse mark

as a real image file — working/assets/la-caisse-logo.png, rendered as an image, never typeset, never substituted; if the asset is missing the run STOPS and ASKS you), company name, company logo at right.

• House furniture present: header rule, one-liner position, source

strip — the always-on-every-slide set.

• Font floors: body text and reference text never shrink below the

floor to make content fit (your sizes: main text ~12pt, references ~10pt — kernel docs/13, build approach; the benchmarks' measured ratios are in the catalogue page).

• The source strip: named sources on the slide, always.
• Never invent — facts and graphics. A placeholder image or an

unsourced number at ship time is a hard FAIL or an ASK, never silently shipped (the previous build's central failure — its story is in the kernel's docs/13-company-overview-plan.md).

Why so small a floor: everything else is judgement the analyst (or a run) may exercise — the floor exists to make the non-negotiables mechanical so judgement can be free everywhere else (kernel D44).

The seven benchmarks, dated

None of these slides is current. Every one is stale against today (2026-08-02), and three of the situations they describe have since moved on (Sims' FY26 results were imminent at analysis time; Mader's FY25 has since reported; UGL's carve-out process has since resolved — analysis §2, §4, §6). Dates are established from evidence on the slides — a currency-rate date, a source date, a "price as of" note (each row cites its tell).

The pattern in that defect column is not an accident: the absence of takeaway lines is the number-one systematic defect across the whole set (confirmed on six of seven slides — analysis, headline facts), and quantitative-spine gaps (leverage, cash, trend) are number two. Both become standing rules in the rebuild; the full argument is on the process page, 07-how-analysts-build-it.md.

Next page: 06-component-catalogue.md — every component that appears on these slides, what job it does, and the evidence for when it earns its place.

The component catalogue — the parts a slide is composed from

The middle of the macro-to-micro story (05-the-slide.md is the macro). This page catalogues every component observed across the seven benchmarks: the job it does, who carries it, the forms it takes, the defects the evaluations confirmed, and the rule that follows. It is a catalogue, never a template — your D44 ruling: a new slide is composed the way an analyst composes, choosing components for the situation — and the evidence below includes omissions judged correct (UGL's missing management block and shareholding donut) as well as omissions graded as defects. Citations work as on the previous page ("analysis §n" = working/analysis/benchmark-slide-reports.md; "eval Wn/Cn" = that slide's eval-reconciled.md in working/references/recons/; "tier-1 spec" / "tier-2 brief" = working/references/authoritative/).

1. Header block

Job: identity and provenance in one glance — whose slide, about whom. Evidence: six of seven slides open with the same four-piece band: producing firm's mark top-left, company name as title, company logo top-right, rule beneath; UGL's recorded header is mark, name-only title, and a page number (analysis §1–§7 inventories). Forms: the producer mark varies (deal-team swirl on DFM, FineToday, Mader, UGL; an ellipse on Sims; concentric circles on Tynor; a lens emblem on ValueLabs) — which is exactly why the previous build lost it: hedged identity was demoted to "optional" (kernel docs/13, lesson 5). Here it is hard floor: the La Caisse mark as a real image asset. Defects: both tier-1 slides title themselves "Company Profile | <name>" against the brief's name-only rule (DFM eval W9, Sims eval W8). Rule that follows: name-only title; producer emblem, name, company logo, rule — mechanical presence check (hard floor, working/app-settings.md).

2. The one-liner

Job: place the company (or the deal) in one sentence before the reader commits to the page. Evidence: present on five of seven (DFM, Sims, FineToday, Tynor, ValueLabs); absent on UGL and Mader — with opposite verdicts: Mader's first bullet doubles as the relocated one-liner and its eval rates the miss minor ("failure is positional" — analysis §4, eval W1), while UGL's absence is a confirmed defect (eval C3). Forms: the brief's cap is 15–25 words; observed: FineToday 17 words (in range), ValueLabs 13 (slightly under, minor defect — eval), Tynor a 9-word strap-line judged tautological (eval, minor). Rule: default present, 15–25 words, bold italic under the title (DFM's one-liner "1.15x bold italic", analysis §1; Tynor's strap "1.3x bold italic", analysis §5); omission needs a declared reason — Mader's relocation passed, UGL's bare absence did not.

3. Business description bullets

Job: the company's story compressed to its salient facts — the tier-1 spec's checklist: founding year and headquarters, main business, what it is best known for, market positioning (with share figures if available), footprint, key segments, customers, KPIs, shareholding, recent events. Cap: "3 to 5 most salient bullet points with 0 to 2 sub-bullets each, 100 to 150 words in total" (tier-1 spec, verbatim). Evidence of craft: the bullets are sequenced as an argument, not listed — DFM walks history → concentration facts → ownership arc with entry multiple → the ask, deliberately last so it hooks the chart beside it (analysis §1); Tynor sequences leadership → quantified market (~30% share, 15%+ CAGR — "quantified, not adjectival") → moat → the ask (analysis §5); FineToday's second bullet is the page's only prose synthesis (two withdrawn IPOs — setting the exit as a third attempt, analysis §3). Defects: Mader runs 6 bullets past 150 words (eval W12); DFM writes "Incorporated" where the brief demands "founded in" (eval W12). Rule: 3–5 bullets, ≤150 words, sequenced toward the situation's question, "founded in", the ask last when there is one.

4. Deal terms and recent news

Job: what is being asked, by whom, at what multiple — the reason the reader was handed the page. Evidence: carried as the closing bullets of the description block on every live-deal slide: DFM's EV (enterprise value — the whole-business price) of ~US$500–800M at 16–26x (analysis §1), FineToday's binding offer at ~14–15x (analysis §3), Tynor's 60% at ~$400m / 20x — where the arithmetic ties across the page: $20m EBITDA × 20x = $400m (analysis §5, "the slide's strongest holistic quality"), UGL's ~A$500–600M expectation plus a 2016 take-private precedent as anchor (analysis §6), ValueLabs' $1.2bn / 15x in exclusivity with the losing bidders footnoted (analysis §7). Craft detail: ValueLabs carries a re-underwritten EBITDA ("$80m vs indicated sell-side EBITDA of $100m") — the evaluation refuted the claim that this dual figure is a weakness: "it is a strength", diligence honesty on the page (analysis §7). Rule (brief): competing bids get a short bullet, not an automatic comparison table; a rejected bid's reason in 3–7 words (tier-2 brief).

5. Products / services block

Job: make what the company actually sells legible — "explain the company's main products, services, platforms, or business lines" (tier-2 brief). Forms observed (one per situation, no two alike): a 3-up image band — pack shots, channel logos, plants map (DFM, analysis §1); a four-step process flow with outputs and a monetisation legend — for a services business "the process IS the product" (Sims, analysis §2, rated "best-in-class" by its eval); value-chain steps R&D → manufacturing → sales with logo boxes mapping the geography split (FineToday, analysis §3); product-family cards with photos and SKU badges — photos "make an unfamiliar device category legible" (Tynor, analysis §5); a segments table whose logo grids double as the tier-1 spec's "Sample Customers (logos)" (UGL, analysis §6); a 2×2 services/case-study table — case studies as the right proof device for a services firm (ValueLabs, analysis §7). The brief's guard: visuals only if they explain — "Do not add visuals just for decoration." Rule: choose the form that makes the business legible; every visual earns its place.

6. Key financials block

Job: the quantitative spine — what the company earns, in which direction, with how much debt. The tier-1 spec's floor: declared fiscal-year-end, revenue, EBITDA, EBITDA margin, CAGR, cash-flow conversion, and net-debt-to-EBITDA "sometimes" (tier-1 spec §4). Forms: a 3-series grouped bar chart with margin ovals and growth brackets (DFM, analysis §1); a combo chart with an annual/half-year panel split that avoids splicing interim periods (Sims, analysis §2); grouped bars plus 9-month interim columns giving "the momentum read a live buyer needs" (FineToday, analysis §3); a 10-row table that is "nearly the whole §4 checklist INCLUDING the 'sometimes' leverage line" — net debt/EBITDA 0.6x→0.3x while growing at 30% CAGR (Mader, analysis §4); a single revenue chart with an EBITDA callout ribbon (Tynor, analysis §5); prose figures only (UGL, analysis §6); a single FY25A bar ("A" marks actual figures, as opposed to estimates) with stat tiles (ValueLabs, analysis §7). Defects — the set's second systematic gap: no leverage or cash on DFM (eval W4), FineToday (eval W2 — on a leveraged buyout), Tynor (eval W3 — "choices, not sourcing limits"); no trajectory at all on UGL (eval C1) and ValueLabs (eval W1); and the set's single worst defect is here: Sims plots EBIT margin above EBITDA margin in all seven periods — accounting-impossible, "a numerate reader loses trust in the entire financials block" (eval W1). Rules that follow: the spine is floor content (revenue, EBITDA, margin, trend, leverage — or a declared reason for absence); sanity checks are mechanical (EBITDA ≥ EBIT per period; every growth callout must reconcile — Sims' 9% oval computes to 7.57%, eval W6; every bar labelled — DFM eval W13); interim periods never spliced into annual series.

7. Takeaway / insight lines

Job: say what the numbers mean — "The commentary should not just describe the chart. It should explain the key takeaway." (tier-2 brief, with worked examples: "Around half of revenue is exposed to AI substitution risk."). Evidence: this is the number-one systematic defect of the whole benchmark set — confirmed missing on six of seven slides (DFM W1, Sims W7, FineToday W4 — "sharpest principle-level miss", Mader W2 — "the single most investor-relevant reading is left entirely to inference", UGL C6, ValueLabs W4 — "clearest violation of a core principle"; analysis, headline facts). The one nuance: UGL's segment-donut bullets are the takeaway — the evaluation refuted that instance, showing the function can be served by adjacent prose (analysis §6). Mader's eval states the rule best: takeaway lines are "a presentation discipline fully within the author's control" (analysis §4). Rule: every quantitative block carries an insight line (or adjacent prose demonstrably doing that job) — enforced everywhere in the rebuild.

8. Revenue splits

Job: expose concentration — where the money actually comes from. Forms: donut pairs (Mader: geography + sector, analysis §4; UGL: segment + geography, analysis §6), stacked bars (FineToday: geography 43/41/9/7 and product 49/26/23/2 — "two markets ≈84%, hair ≈ half", analysis §3), badge overlays (Sims' "% of Group EBIT" pills quantifying the carve-out question, analysis §2). Defects: FineToday's splits are a static average that suppresses the time dimension (eval W3); UGL's two segmentations never reconcile and its table gives no %-of-revenue (eval C2); UGL's 81% eastern-seaboard concentration sits uninterpreted (eval C6); Mader's sector donut is FY22-stale, third-party, and sums to 101% (eval W6). Rule: splits carry their as-of date, reconcile with each other, and get a takeaway (component 7).

9. Shareholding structure

Job: who owns it now — and, in a deal, who will. Forms: a pre/post donut pair with honest forward ranges for an unsigned deal (ValueLabs: founders 100% → 15–20% / CVC 80–85%, consistent colour semantics, analysis §7); prose where prose is better — FineToday's eval explicitly rates "prose-not-donut shareholding a design judgement" (analysis §3), DFM's "97%" rides in a bullet correctly (analysis §1), and Mader's founder-51% line under the share chart is a "first-order governance fact" whose prose form the analysis files under composition-follows (analysis §4); omitted where uninformative (UGL: 100% CIMIC would be "a single uninformative wedge", analysis §6). Rule (brief as default): donut with a takeaway line; prose or omission as declared deviations, which the evidence shows are often right.

10. Management block

Job: who runs it — when that answers the situation's question. The contradiction, stated openly: the tier-2 brief says drop it ("Do not include a management team section. It is not useful for this one-page company profile.") — yet six of seven benchmarks carry one (all but UGL; analysis §1–§7), and each time the evaluations found a situation-specific job it was doing: Advent's professionalisation thesis, with hire dates aligning to the recovery (DFM, analysis §1); who would run the separated business (Sims, analysis §2); all three "appointed by CVC" (FineToday, analysis §3); founder-led with an exit-ready CFO (Tynor, analysis §5); continuity in a founder-led buyout — though the Apr/May-25 role-start cluster goes unexplained (ValueLabs, analysis §7). UGL omits it and the eval confirms the omission correct (analysis §6). Resolution (your D44 ruling): default OFF per the brief; include as a declared deviation when the situation gives it a job, as these six did.

11. Operating statistics / KPI tiles

Job: the tier-1 spec's sector-adaptive metrics (§3 — utilisation, engineers, active clients, churn (the rate at which customers leave), and LTV/CAC (customer lifetime value against acquisition cost) as examples). Forms: five stat tiles — 7k+ engineers, 30+ offices, 12-yr average tenure, 300+ clients, NPS 80/90 (net promoter score, a customer-loyalty measure) (ValueLabs, analysis §7); a single operating KPI plus two flagship certifications (Sims, analysis §2); SKU badges giving breadth (Tynor — but breadth is not mix: SKU counts never become revenue mix, eval W1, "the single biggest content gap"; analysis §5). Notable absence: ValueLabs omits utilisation — the spec's first example for exactly this kind of firm (analysis §7). Rule: KPIs chosen per sector from the spec's menu; a breadth number never substitutes for a mix number.

12. Footer — notes and source strip

Job: honesty infrastructure: currency and fiscal-year-end declarations, footnote definitions, named sources. Evidence: all seven slides carry it (analysis §1–§7): DFM names five sources and footnotes its as-of dates — "Footnoted as-ofs let the reader judge freshness" (analysis §1, verbatim); UGL honestly footnotes that its geography donut is an analyst-built proxy from a project register (analysis §6); ValueLabs runs a three-line strip with exact split percentages and retention intent in the footnotes (analysis §7). Defects: UGL uses footnote ⁶ but never defines it (analysis §6); FineToday omits its FX date and fiscal-year-end entirely (confirmed gaps, analysis §3). Rule: source strip is hard floor; every footnote referenced is defined; currency, FX date, and fiscal-year-end always declared.

13. Charts and visual furniture — the formatting layer

The micro end of macro-to-micro: measured facts from the seven slides, usable as house style evidence. All from the FORMATTING sections of analysis §1–§7 (colour values are approximate where the benchmark is a photographed screen — noted per slide there).

• Layout: two asymmetric columns on six of seven; UGL adds a

full-width bottom band under its two columns. DFM is 1090×819 (~4:3); FineToday approximates 16:9.

• Title: serif, 2.2x–3.5x body size (Sims 2.2x, DFM 3.2x, Tynor

3.0x, ValueLabs 3.5x — the largest). One-liner ~1.15–1.3x, bold italic.

• Footers: 0.55x–0.75x body — the smallest text on every page.

Your floor keeps these honest: main text ~12pt, references ~10pt, titles free (kernel docs/13; hard floor).

• Colour systems: each slide runs one dark anchor (navy #14225A

DFM, #01205F Sims, #1B1E3D Tynor, #1B1B2E ValueLabs; charcoal #2b2f38 Mader, #2b2621 UGL; near-black olive FineToday) plus two to three accent colours for series and callouts. Sims shares one colour code between its process diagram's stars and its chart legend — "one visual grammar" (analysis §2).

• Callout furniture: CAGR ovals, margin ovals, growth brackets and

arrows, callout ribbons and speech bubbles (DFM, FineToday, Tynor, ValueLabs), badge pills (Sims). Defects to guard: unlabelled callout (Tynor's 33% badge), unreconciled callout (Sims' 9% vs 7.57% computed, eval W6), unlabelled bar (DFM FY25 EBITDA, eval W13), and one shipped editing artifact (UGL's stray red "s", eval C5).

Next page: 07-how-analysts-build-it.md — how these components get filled: the research routes, the selection judgement, and what holds across all seven slides.

How analysts build it — research routes, selection judgement, and what holds across all seven slides

The last of the three documentation pages (05-the-slide.md is the what, 06-component-catalogue.md the parts; this is the how). It answers your two design questions from the build plan: how did the analysts likely find each slide's information, and how did they decide what earned a place out of everything they found (kernel docs/13, build approach — your rebuilding analysis).

Honesty banner first: the per-slide rebuilding analyses are reconstructions — inferences from what is on each slide, marked inferential in the evidence file itself (working/analysis/ benchmark-slide-reports.md, header). Sections 2 and 3 below (the research routes and the selection judgement) are therefore not facts about what an analyst actually did; they are the best evidence-grounded account of what they must plausibly have done. Sections 1 and 4 (the situations, and the cross-slide commonalities) rest on confirmed observations — the evaluations' use-case verdicts and the analysis file's headline facts. Citations as on the previous pages.

1. Situation first — classification drives everything

The seven benchmarks serve seven different situations (the dated table in 05-the-slide.md), and every downstream choice — components, depth, even what counts as a defect — follows from that classification: a live-deal teaser builds its whole description toward the ask, placed deliberately last so it hooks the chart beside it (DFM, analysis §1); a carve-out screen quantifies the parent-contribution question (Sims' "% of Group EBIT" badges, analysis §2); a listed-company overview carries a share price chart and governance facts instead of deal terms (Mader, analysis §4); a screening page for a control deal makes the arithmetic tie out to the ask (Tynor, analysis §5). Two slides omit standard blocks and are judged right to (UGL, Mader — analysis, headline facts): composition follows situation. The previous build got this right too — "situation-first research that classifies the use-case early and re-prioritises everything" is on its worth-re-deriving list (kernel docs/13). Classifying the situation is therefore the first step of any future procedure — but procedures come after your review of these pages, not before (your gate, kernel D45).

2. The research routes — where each kind of fact comes from

Reconstructed per slide, then pooled. Each route names its evidence.

• Public filings and annual reports — the listed-era backbone:

founding, listing, segments, audited financials (DFM's listed-era facts, analysis §1; Mader's table, analysis §4; UGL's geography proxy is analyst-built from the parent's annual-report project register, honestly footnoted — analysis §6).

• Failed-IPO and deal filings are goldmines — FineToday's

financials, splits, management, and facilities all plausibly ride one document: the securities registration statement from its withdrawn IPO (analysis §3). A dead process leaves the best public paper trail on a private company.

• Investor days and company presentations — the process/product

story: Sims' four-step ITAD block (ITAD = IT asset disposition, the business of retiring and reselling corporate IT equipment) traces to its Investor Day (analysis §2).

• Broker research and data terminals — the scarcest, highest-value

content: a division inside a group has no standalone filings, so Sims' divisional breakout plausibly comes from a named broker note and estimates ("Nomura's Presentation (dated Mar-26)" is cited on the slide; analysis §2).

• Deal intelligence — press, market wires, banker channels — the

freshest and hardest facts: FineToday's days-old binding offer (analysis §3), UGL's mandate, range, and named likely bidder (analysis §6).

• Sponsor materials for private companies — when a private-equity

firm is in the deal, the hard facts (share, cost advantage, SKU counts, financials, terms) plausibly ride its deck or information memorandum: Tynor's footnote openly names the triad website / deck / market estimates (analysis §5); ValueLabs' internal KPIs (tenure, NPS), its dual EBITDA, and its losing-bidder list all imply information-memorandum or management access — "not public research" (analysis §7).

• LinkedIn and public bios — management blocks (DFM cites it,

analysis §1).

• Legacy collateral — heritage stats like UGL's "11,000+ wagons

built" (analysis §6).

• The advisor's own model — last resort, and it shows: Mader's

sector donut is a third-party estimate, FY22-stale, summing to 101% (analysis §4, eval W6) — the route exists, and its products need dating and flagging.

• What even good research cannot get — the honest-absence class:

market share in fragmented sectors (Sims, Mader — the soft "one of Australia's leading" is the correct move when share is unsourceable, analysis §4), named customers under NDA (Sims), nonpublic segment splits (Sims). The previous build's two closure states carry this: a fact is SOURCED or UNAVAILABLE-AFTER-SEARCH with sources tried recorded — never silently absent (kernel docs/13, worth-re-deriving list).

3. The selection judgement — what earns a place

The rebuilding analyses reconstruct not just where facts came from but why these facts survived when the analyst demonstrably held more. The pattern, slide by slide:

• Only goal-advancing facts survive. DFM surely held net debt,

net income, free cash flow, the FY25 growth driver, competitor names, full SKU history — all dropped: ugly, better handled in meetings, or an invitation to unfavourable comparison (analysis §1).

• Exclusions can be justified — or judged errors. FineToday

dropping net income is agreed by its eval (irrelevant under a new capital structure); dropping the balance sheet on a leveraged buyout is "arguably an error" (eval W2; analysis §3). Tynor's revenue-deep-only financials for a 20x control deal: "choices, not sourcing limits" (eval W3). ValueLabs excluding the multi-year revenue trend when CAGR was derivable: "authorial, not sourcing" (eval W1). The selection judgement is real judgement — and it can be wrong; the evaluations grade it.

• Keep one clean series. Mader holds the FY25 half-year but keeps

a single audited FY20–24 series (analysis §4); Sims splits annual and half-year into separate panels rather than splicing (analysis §2).

• Names carry signal; comparisons invite trouble. Tynor keeps

backer names, drops funding-round details; hides competitor names behind "global peers" — the 3–5x price ratio argues the moat without opening a comp table (analysis §5). DFM drops competitor names entirely (analysis §1).

• Honesty survives selection. DFM states its concentration risk

twice (analysis §1); ValueLabs ships the conservative re-underwritten EBITDA next to the seller's number — refuted as a weakness, "it is a strength" (analysis §7); UGL footnotes its proxy-built donut (analysis §6).

4. What holds across all seven — the commonalities

The macro findings the whole set supports (analysis, cross-slide headline facts — these are confirmed observations, not reconstruction):

1. The missing takeaway line is the #1 systematic defect — six confirmed absences: DFM, Sims, FineToday, Mader, UGL, ValueLabs (analysis, headline facts); Tynor is the one slide not on the confirmed list. Even the partial exception proves the rule: UGL's segment-donut bullets do serve the takeaway function (that specific complaint was refuted), yet its 81% geographic concentration still sits uninterpreted (eval C6 — analysis §6). The brief demands takeaways; the benchmarks omit them; the evaluations grade the omission material every time it matters (Mader: "the single most investor-relevant reading is left entirely to inference"). Rule the rebuild enforces everywhere. 2. The quantitative spine gaps are the #2 defect class — leverage, cash, trend (five of seven slides; catalogue component 6). The spine becomes floor content with declared-absence as the only out. 3. Every slide is stale — dated in 05-the-slide.md; freshness is a property to be declared (as-of dates, FX dates, fiscal-year-ends), not assumed. 4. Composition follows situation — proven by absences judged correct (UGL, Mader). Hence a component catalogue, never a template (your D44 ruling). 5. Arithmetic that ties is a quality marker — Tynor's $20m × 20x = $400m ties the whole page together (analysis §5); Sims' impossible EBIT-above-EBITDA chart is the counterexample that destroys trust (eval W1). Mechanical sanity checks (EBITDA ≥ EBIT, callouts reconcile, bars labelled) cost nothing and guard exactly this. 6. Honest disclosure reads as strength, not weakness — the dual EBITDA, the stated-twice concentration, the footnoted proxy (section 3 above). The grounding rule (every claim sourced or marked unavailable-after-search) is the same instinct made mechanical (kernel docs/11 §6, this app's honesty floor).

What happens next — and what deliberately does not

These three pages are the stop point you set (kernel D45). Nothing derives from them until you have reviewed them: no app wiki nodes, no procedures. After your review, the build continues per the plan (kernel docs/13): the macro research-loop procedure first (research → evaluate → research more → until good enough for the slide's goal), then the micro procedures it calls — each derived from this evidence, entering as provisional pilots, ratified only by the first real company run.

Your actions — the one page listing everything only you can do

Obligations live only here; other pages point here instead of repeating them. This page is on the boot path.

Open now

Standing — always yours, never mine

Documentation — start here

Written for you, Josh: how company-overview works and where the work stands, without reading code or asking. Every docs page must appear in this list (the commit gate enforces it).

0. 00-big-picture.md — what this app is, where it sits in AOE, and how its parts fit. 1. 01-what-is-here.md — every file and folder, one plain sentence each. 2. 02-status.md — done / in progress / waiting on you, newest first. 3. 03-decisions.md — every decision with its evidence and the argument joining them. 4. 04-deferred.md — what is postponed, with the condition that closes each row. 5. 05-the-slide.md — what the product is, its anatomy, whose rules outrank whose, the hard floor, and the seven benchmarks dated. 6. 06-component-catalogue.md — every component a slide can carry: its job, the evidence, the defects, the rule. A catalogue, never a template. 7. 07-how-analysts-build-it.md — the reconstructed research routes and selection judgement, and what holds across all seven benchmarks. 8. 08-your-actions.md — the one page listing everything only you can do.

Files

Every file in the documentation folder, under its real name — tap one to open its page. Rows without a button exist in the repo but are not rendered on this site. The app's full inventory is its own 01 page.

Turns

One row per closed loop (newest first). The chips are the loop's own attestation, checkable against that commit's diff; rows marked pre-attestation predate the rule. This is the app repo's own history — it commits separately from the kernel.