February 2026 — From bootstrap to pyramid

14 February – 28 February 2026 · 195 commits · pre-versioning era

The repo exists for fifteen calendar days in February 2026 — Saturday Feb 14 to Saturday Feb 28. Fifteen days from one commit to a four-layer analytical pyramid, a customer-facing Explorer, a Tauri desktop build, three reference taxonomies, a calibration harness, and four foundational concept papers. The product has no name beyond nuMetrix, the engine has no name at all, and the architectural vocabulary that will become the Senses is two months from being invented. None of that matters yet. The shapes get built first; the names come later.

This is the month before everything else.

The bootstrap

On Saturday Feb 14, d9ab3dbf lands at 12:47: Bootstrap nuMetrix hospital material flow lakehouse. A medallion-style dbt project, a Dagster definition, a synthetic-data generator, three taxonomies, a CLAUDE.md, a hospital-tenant skeleton — all in one commit. The shape is committed before any of the layers above it exist.

By Sunday night the project has Evidence dashboards, multi-tenant SAP+Navision dispatch, four synthetic tenants (alpha, beta, gamma, delta), the rebuild.sh script, and a user guide. Two days. Seven commits. The architecture that the next twelve weeks of journals will retrospect about is already legible.

The vocabulary arrives (W08)

The first full week — Mon Feb 16 through Sun Feb 22 — names the things that will live on the bootstrap shape. Five nouns in seven days:

1. The probe (53ec472a Feb 17) — “Add probe DSL framework v1: YAML-defined diagnostics compiling to dbt SQL”. The shape that all later signal/perspective/thesis/verdict work descends from. One type on Tuesday, six by Sunday.
2. The Explorer (6001f8b8 Feb 19) — “Add nuMetrix Explorer: customer-facing SvelteKit app”. The interactive analytical surface that will host every later Canvas-style page.
3. The hypothesis (ad0524bd Feb 21) — “Add hypothesis system — business questions evaluated against probe evidence”. The second layer of the analytical pyramid; eight commits the same Saturday flesh it out.
4. The calibration harness (df8aae0a Feb 21) — defect manifest, calibrate.py, Explorer integration. The discipline that makes “the system can be audited against itself” a real claim rather than a slogan.
5. The lineage layer (f345348f Feb 21) — “self-documentation lineage layer — YAML definitions, compiler, runtime metrics”. The framing that becomes the four-level lineage explainer in W12 and the Ledger paper (Sense 18) in April.

Three of these get renamed in W14 (probe → signal, hypothesis → thesis, the “diagnosis” that arrives in W09 → verdict). None of them get retired. The first vocabulary is not the right vocabulary, but it’s the right shape.

The 72-hour sprint and Diagnosis (W09 partial)

Sunday Feb 22 is day one of a 72-hour sprint that runs through Wednesday Feb 25. The output is captured in docs/journal/72h_in_february.md — the project’s first journal-style document, written before the journal cadence existed.

On Tuesday Feb 24, a2010044 ships the Diagnosis layer: eight starter rules, a validator, a compiler, a contract, platform views, an Explorer detail section in three languages. The fourth layer of the analytical pyramid lands. Probe → Assessment → Hypothesis → Diagnosis: the shape is complete.

The same afternoon, 0ff36502 commits strategic_directions.md — “five paths evaluated, Diagnosis + Temporal chosen”. The roadmap document arrives after the implementation. (This pattern will repeat for years.)

The remainder of W09 (Feb 26-28) does the polish: weekly time buckets (d08a683e), the Jump Bar (5496a49b), the interactive pipeline graph (67a83902), the cross-platform Tauri builds (3965f04f), the entity registry (3bb13793), the hypothesis-centric executive summary (b1e9aef3) — and the concept papers that prefigure four months of architectural work: SMEbits (5b6cefc8), authorization & journal (2b6a185a), cloud v2.0 (fe0be0c7).

The four-layer pyramid

By Feb 25 the analytical pyramid has the shape that will define the product for two months:

                  ╭───────────────╮
                  │   DIAGNOSIS   │   (Feb 24 — root-cause + recommendation)
                  ╰───────────────╯
                          ▲
                  ╭───────────────╮
                  │  HYPOTHESIS   │   (Feb 21 — business question)
                  ╰───────────────╯
                          ▲
                  ╭───────────────╮
                  │   ASSESSMENT  │   (Feb 20 — health score across probes)
                  ╰───────────────╯
                          ▲
                  ╭───────────────╮
                  │     PROBE     │   (Feb 17 — atomic diagnostic)
                  ╰───────────────╯
                          ▲
                  ╭───────────────╮
                  │   MEDALLION   │   (Feb 14 — bronze/silver/gold)
                  ╰───────────────╯

Four layers, all built in eleven days. The lower three layers (medallion, probe, assessment) live in dbt. The top layer (Diagnosis) lives in YAML compiled to dbt. The shape rhymes; the seams are clean. When Sense 14 renames everything in April, the shape will survive — only the labels change.

What’s still missing

The disciplines that the rest of the year will narrate as discoveries are absent in February — instructively absent:

• No artifact-store vocabulary. AFS / KLS / SIS — three weeks away.
• No pack / engine split. The project is one repo with one purpose. The pack abstraction is four weeks out.
• No CLI. Everything runs via just recipes. jinflow the binary doesn’t exist.
• No P2P2P / cloud architecture. Local-only. The cloud v2.0 concept paper at W09’s tail end is the first step.
• No Senses. The pattern of name the load-bearing idea before building it hasn’t been invented; the design docs that exist in February are post-hoc roadmaps, not prefiguring papers.
• No second pack. Millésime won’t appear for four more weeks. nuMetrix is the product, not a pack of the engine.
• No identity, no Cathedral, no Ledger, no Simulation, no Atelier, no Lens. Each of these will get a Sense in April or May.

What’s present in February that will outlive all of these: the medallion architecture, the probe-style DSL, the calibration discipline, the data-driven Explorer schema discovery, the trilingual documentation reflex, the Tauri distribution story. The bones, not the names.

Arc of the month

In one sentence: fifteen days went from one commit to a four-layer analytical pyramid with a customer-facing Explorer, three reference taxonomies, a calibration harness, and four concept papers — all without a name for any of it.

The journals from W10 onward retrospect renames and discoveries and sense papers. February has none of that. February has the working software. The vocabulary catches up next month.

Numbers across the month

Metric	Value
Commits	195
Calendar days	15 (Feb 14–28)
Working name	nuMetrix (until W10)
Source-system dispatchers shipped	3 (SAP, Navision, generic)
Synthetic tenants	4 (alpha, beta, gamma, delta)
Probe types in the DSL	6 (balance, mandatory_item, distribution_outlier, duplicate, ratio, assessment)
Probes shipped	11
Analytical-pyramid layers built	4 (medallion, probe+assessment, hypothesis, diagnosis)
Reference taxonomies	3 (ATC, MIGEL, BAG)
dbt models documented	68 → ~95
Concept papers committed	~6 (lineage, hypothesis, diagnosis, SMEbits, authz, cloud v2)
Acronyms invented	0 (AFS/KLS/SIS arrive in W10)
Sense papers	0 (Senses don’t exist as a pattern yet)
Releases	0 (the release pipeline is W11’s job)

The mood

The commit messages this month are doing-not-explaining. “Add this, fix this, document this, refactor this.” No essays in commit bodies, no architectural manifestos. The work is happening fast enough that naming it would slow it down. The Senses-style design discipline — name it before you build it — is a few months away from being invented; right now, the inversion holds: build it, then name it when you understand what you built.

By Sunday Feb 28, the project has fifteen days of working software with no vocabulary for any of it.

The vocabulary arrives next month.