BOW Analytics · The Research Desk

The machine flags it. The desk assigns it. The piece pays it back.

Bow doesn’t keep a content calendar. The model mines its own output for tensions it can’t settle, the desk triages those signals into assignments, and every published piece must deposit residue— a detector, an embed, a data column — that widens what the machine can see next. Articles here aren’t posts. They’re moves in a research program.

24 open questions on the docket7 active desk files0/8 residue artifacts deliveredthis week on the ledger: 0 flips · 0 opened · 0 settled
Detector firesSignal triagedAssignment scopedInvestigated in the notebookPublished with live embedsResidue depositedDetectors see more
Live from the machine

Signals

Full docket (24) →

Nothing below was written by a person. These are the hottest tensions the detectors mined from the live model tonight, plus the most recent entries the ledger put on the record. Signals become assignments — or they get killed, on the record.

The model on the record

No ledger history on this instance yet — the nightly snapshot writes the first entry. How the ledger works →

The board

Desk assignments

Every file names its source signal, its lens angle, its format, its next physical action — and the residue it owes the machine before it counts as done. Killed files stay on the board: what the desk refuses to write is part of the record. The board is a typed file (lib/desk.ts) edited on triage, until the database migration lands.

Signal · 3

Method MemoSignalThis monthDesk · opened 2026-07-06

What Is a Win Worth? An Audit of Our Most Important Number

Signal · Knife edge Knife-edge detector: multiple verdicts flip within the plausible $-per-win range — the whole model leans on one number. live on the docket →

Lens angle Every lens prices a win differently; the audit shows how many verdicts each choice silently decides.

Needs data Hand-curated league-financials table (total salary vs. available wins) in data-seeds/

New data columnNew embed

Next action Curate the league-financials table; derive defensible $/win bounds from it.

Docket FileSignalBacklogDesk · opened 2026-07-06

Free Lunches: When a Trade Helps Both Teams

Signal · Free lunch The most counterintuitive standing output: the model says some trades create value on BOTH sides. live on the docket →

Lens angle The Skeptic's case that free lunches are model artifacts vs. The Consensus case that they're apron-relief transfers.

Needs data Agent sweep of the pairwise trade space to classify the mutual-gain population

New detector

Next action Scope the kill condition: if >80% of free lunches are pure apron-relief, the piece becomes a Method Memo.

Teaching NoteSignalBacklogDesk · opened 2026-07-06

Why a Dollar Doesn't Cost a Dollar (Track 101 companion)

Signal · Editorial call Education side needs a classroom on-ramp to the analytics side — derive it from the manifesto.

Lens angle One lens only (The Consensus), on purpose: students learn the dial exists before they learn to argue about it.

Teaching artifact

Next action Wait for the manifesto to publish, then derive — do not write this first.

Assigned · 2

Docket FileAssignedThis monthDesk · opened 2026-07-06

The Most Contested Player in the League

Signal · Worldviews disagree Lens-split detector: several verdicts sit ≥2 tiers apart across preset worldviews. live on the docket →

Lens angle All five lenses as characters — who they fight over most, and what kind of player he is.

New data columnNew detector

Next action Rank the league by buildLensSplit span; profile the top-decile cluster.

Ledger NoteAssignedThis weekDesk · opened 2026-07-06

The Week the Model Changed Its Mind (standing file)

Signal · Ledger event Nightly snapshot diffs: verdict flips and question turnover accumulate weekly.

Lens angle Which flips were stat drift vs. assumption sensitivity — annotate, don't just list.

Next action Ship the first recap from summarizeLastWeek; template the structure for weekly reuse.

In research · 1

Docket FileIn researchThis weekDesk · opened 2026-07-06

The Same Contract, Four Different Prices

Signal · Free lunch Trade detector keeps finding swaps where apron repricing makes both sides win. live on the docket →

Lens angle The Accountant vs. The Ring Chaser: identical production, opposite verdicts once the receiving team's apron is charged.

New embed

Next action Clip 3–4 case-study contracts through buildTradeAnalysis across all 30 team contexts.

Drafting · 1

Method MemoDraftingNowDesk · opened 2026-07-06

Show Your Work: Why Bow Publishes Its Assumptions

Signal · Editorial call Launch manifesto — the piece every future article links instead of re-justifying itself.

Lens angle All five lenses on one screen: the same player, five verdicts, and why that's the product.

New embed

Next action Write the lede around the live verdict-flip demonstration; publish before any other piece.

Killed · 1

Desk BriefKilledDesk · opened 2026-07-06

Daily reaction column on trade rumors

Signal · Editorial call Tempting cadence-filler: react to every rumor cycle for traffic.

Lens angle None — and that's the tell. No detector fired.

Killed because Fails the desk's admission test: no detector fired, no falsifiable question, no residue. Bow reacts to news only when the model has something non-obvious to say (research/09 §4).

The portfolio

What the desk publishes

Seven formats, chosen so one editor can run all of them (research/09 §7). A format that requires residue cannot ship as prose alone — it must leave a detector, embed, data column, or model revision behind.

Ledger Note

Two paragraphs of human judgment on what moved on the ledger — the model's diary, annotated.

Audience
Returning readers; the weekly heartbeat.
Time
30–60 min
Cadence
Weekly in season
Rigor bar
Every claim is a ledger event or a live embed; no new analysis required.
Model tie
Built directly from diffSnapshots output; embeds re-run the live engines.

e.g. The Week the Model Changed Its Mind · Three Verdicts That Flipped on One Slider

Docket File

residue required

The workhorse: take one Open Docket question, investigate it, settle or sharpen it on the record.

Audience
Serious fans and analytics-literate readers.
Time
6–15 hrs
Cadence
2/month in season
Rigor bar
Full assumption-bounds sweep; the piece must show where (or whether) the conclusion flips.
Model tie
Starts from a machine-generated question with a stable id; publishes at its lineage.

e.g. The Same Contract, Four Different Prices · Free Lunches: When a Trade Helps Both Teams

Desk Brief

A fast, structured read on a live situation the detectors flagged — assignment-to-publish in a day.

Audience
Anyone following the news; Bow's answer to the hot take.
Time
2–4 hrs
Cadence
As signals warrant
Rigor bar
Only ships if a detector actually fired; conditional claims with stated assumptions only.
Model tie
Anchored to a specific signal (docket question or ledger event) cited in the piece.

e.g. Why the Model Calls This Trade a Free Lunch · The Docket Just Opened a File on Phoenix

Flagship Investigation

residue required

Quarterly original research on a territory question — the pieces a front office forwards.

Audience
All audiences, layered: story up top, methods appendix below.
Time
30–60 hrs over 4–8 weeks
Cadence
Quarterly
Rigor bar
New dataset or model extension; external-validity check; adversarial review pass.
Model tie
Must change the model — a new lens, detector, or assumption dimension exists afterward.

e.g. Do the Lenses Know Something the Market Doesn't?

Method Memo

residue required

Model changes, assumption audits, and decision retrospectives — being wrong, in public, on purpose.

Audience
The analytics community; low traffic, maximal credibility per reader.
Time
3–6 hrs
Cadence
On every model change + quarterly retrospective
Rigor bar
Auditable by a team analyst; every change versioned against the ledger record.
Model tie
Documents a diff to lib/aasv, lenses, or detectors; the ledger grades it later.

e.g. What Is a Win Worth? An Audit of Our Most Important Number · The Report Card

Teaching Note

residue required

A classroom-ready explainer that turns one desk finding into a Track 101/201 artifact.

Audience
Students and instructors on the education side of the house.
Time
2–4 hrs, derived from an already-published piece
Cadence
1/month, trailing the publication
Rigor bar
Plain language, glossary-linked; every number is a live embed a student can re-run.
Model tie
Reuses the source piece's embeds; links the lesson to the live model.

e.g. Why a Dollar Doesn't Cost a Dollar (Track 101 companion)

Partner Memo

A front-office-style brief for schools and partners — the desk's work in decision-memo form.

Audience
Partners, program directors, front-office-adjacent readers.
Time
2–3 hrs, derived from existing findings
Cadence
As partnerships require
Rigor bar
Memo discipline: recommendation, evidence, risk, dissent — no claim without a citation.
Model tie
Compiled from buildPlayerMemo/buildTeamBrief outputs and published findings.

e.g. State of the Apron: A Briefing for Partners

Nothing ships as prose alone

What a piece leaves behind

A desk file doesn’t end when the article publishes. Its embeds stay live (they re-run the real engines on every read), its interactive core can be promoted to a permanent tool, its findings can revise the model itself, and its leftovers become detectors and teaching artifacts. One investigation, six exits: