Methodology v6.3 · Ontology v0.2.1
The P49 Ecosystem-Reconstruction Method
A reproducible way to reconstruct how a complex market or ecosystem is actually structured — as a typed, evidence-backed graph — and to extract analytical insight that survives scrutiny.
Section 1
What P49 is & why it exists
P49 is a reproducible method to reconstruct how a complex ecosystem or market is structured — capturing it as a typed, evidence-backed graph — and then to extract analytical insight from that structure. It turns "study a market" from an act of authorship into an act of engineering.
The core problem P49 was built to fix is a failure mode common to naive research pipelines: they collapse the journey from sources → synthesis into a single leap. In doing so they silently drop the three middle layers that actually create depth:
Dossiers
Triangulated, per-actor evidence — the raw, sourced foundation.
Tracks
Deep functional notes that give entities meaning before extraction.
Question-runs
Iterative analytical passes that interrogate the whole corpus.
Skip those layers and the result is predictable: flat, untriangulated output — assertions with no evidence trail, no cross-checking, no graph to query. P49's answer is structural rather than exhortative. It does not ask the analyst to be thorough; it makes skipping physically impossible by placing hard gates between layers. A downstream artifact simply cannot be created until the upstream gate has formally closed.
Depth comes from STRUCTURE — triangulated dossiers → deep tracks → ≥5 question-runs → 5-Whys synthesis — not from word count.
Section 2
The 7 Principles
Seven principles govern every P49 run. Together they make a reconstruction portable across domains, auditable in every claim, and resistant to the shortcuts that quietly destroy analytical quality.
Domain-agnostic
One canonical typology serves any domain — electricity, AI, biotech, finance. The method does not change when the subject does.
Evidence-attached
Every claim carries a confidence score, an evidence_url and a supporting quote. Provenance discipline is non-negotiable.
Graph-ready
Strict separation of type and instance keeps the corpus machine-buildable into a real, queryable graph.
Controlled expansion
Explicit inclusion, expansion and stop rules govern scope. The instruction is never "dig forever".
Versioned change control
New types enter only through an RFC — never inline. The ontology evolves deliberately, with a record.
Skill-first execution
Each phase is a named skill. Ad-hoc writing does not count as a phase being done — the skill must run.
Honest disclosure mandatory
Every deliverable states, above the fold, what is simulated, what is unverified and what was bypassed. Limits are disclosed, not buried.
Section 3
The 3-Factory Architecture
P49 organizes work into three sequential "factories", each with a single mandate and a hard gate guarding the exit. The metaphor is deliberate: each layer is an assembly line that produces a specific, inspectable artifact, and nothing passes downstream until quality control signs off.
L1 — Sources Factory (A-1, A0, A0.5, A1, A1.5, A2, A3): collect and canonicalize source material with strict provenance discipline and no interpretation. Output: dossiers plus sources.json.
L2 — Object Factory (A4–A7): turn deep functional tracks into typed entities and relations, then assemble a real, queryable graph in networkx.
L3 — Analytics Factory (A8–A13): assess coverage, audit contradictions, run analytical question passes, synthesize root causes and publish.
The next-layer artifact is not created until the previous gate closes — no shortcuts.
Section 4 · The sequential heart
★ The Full Phase Pipeline
This is where the method lives. Every reconstruction walks the same numbered spine, phase by phase, gate by gate. The stepper below shows the full run; the cards beneath give each phase its purpose, its input→output, and the skill that executes it.
L1 — Sources Factory
Domain Cartography
An inductive "map of the territory" (10 sub-phases) drawn before any formal ontology — used for large domains to understand the shape of the field first.
Reconnaissance
Three sketches of the field: an actors-map, a dependencies-map and an events / decisions timeline.
Provenance gating
Enforces the founding rule of the corpus: no entity without a source enters the master.
Parameters & paradigm
Fill parameters.yml — scope, research questions and boundaries that frame the whole run.
Ontology design
Map the domain onto the canonical 11 types / 19 relations via a bridge table. Missing concepts become RFC candidates — never invented inline.
Hubs roster
List the hub actors that warrant a dossier — the spine of the evidence base.
Dossiers
Raw, per-actor dossiers: ≥300 words, ≥3 independent sources (triangulation), full provenance.
- ≥80% of dossiers completed
- each dossier ≥3 sources & ≥300 words
- all 10+ actor categories covered
- ≥70% of entities carry a source_url
- F1 / F2 / F3 quality floors met
L2 — Object Factory
Research plan / tracks plan
Define the functional tracks to research, with non-overlapping scope.
Deep tracks
Functional notes ≥1500 words that answer 7 canonical questions and cite ≥5 dossiers — so entities gain functional context before extraction.
Extract & aggregate
Pull typed entities and relations from the tracks and aggregate them.
Graph
Build the queryable graph and compute real metrics in networkx.
- each track ≥1500 words + 7 canonical questions + ≥5 dossier references
- A8 coverage verdict = GO or CAUTION
- on LOOP → back to A5 (deepen the tracks)
L3 — Analytics Factory
Coverage assessment
Six-dimension verdict on how completely the domain is covered.
Contradictions
Audit five classes — low-confidence, low-validity, null-attributes, isolate-nodes, fact-conflicts — and either resolve or explicitly mark_open.
Question-runs
≥5 iterative analytical passes of the questions through the corpus. One pass is not an iteration.
Synthesis
≥5 root causes, each with a 5-Whys chain; a central paradigm; root-cause cards. Closed by a mandatory Council gate.
- synthesis reviewed and approved before publication may begin
Reflect / iterate
Step back, capture lessons, and decide whether to loop.
Publication suite
≥4 standalone artifacts with honest-disclosure above the fold. Closed by a mandatory Council gate.
- publication suite reviewed; disclosure verified above the fold
Optional & conditional phases
Iterative graph expansion (optional)
A 9-operation cycle to grow the graph: detect → classify → typify → link → ask-meaning → generate-questions → expand → verify → commit.
Merge phase (mandatory after N≥3 parallel batches)
Seven operations, M1–M7: dedupe, classify bridge edges, recompute coverage. Without it, A11 synthesis is blocked.
Section 5
The Meta-Cartography Toolkit v0.2.1
The toolkit is the shared vocabulary that makes P49 domain-agnostic. The same eleven node types and nineteen relations describe any ecosystem, which is exactly why a reconstruction is portable and a graph is buildable.
11 actor / node types
19 canonical relations
Plus, four supporting schemas
Because the typology is fixed, a concept that does not fit is treated as a signal — not an excuse to invent a new type inline, but a candidate for the RFC pathway (Principle 5).
Section 6
Quality discipline — F1–F8 and 8 guardrails
Two complementary control systems keep a run honest. The F-rules (F1–F8) govern the data — what may be committed to the graph. The guardrails govern the process — how the work is executed and audited.
F1–F8 · data-quality floors
| Rule | What it enforces |
|---|---|
| F1 | ≥2 alternative labels per entity (alt-labels for matching & dedupe) |
| F2 | Every relation ∈ the canonical 19 — strict, no inline new types |
| F3 | Confidence + evidence_url required at commit |
| F4 | A15 merge mandatory after N≥3 batches |
| F5 | Non-overlapping track scope |
| F6 | Batch-schema conformance |
| F7 | Mono-type — one primary actor_type per entity |
| F8 | Aggregate-entity avoidance — use a discrete division (not "Google", but "Gemini") |
8 process-control guardrails
| Guardrail | Guardrail |
|---|---|
| Hard gates between layers | Skill-first execution |
| Pre-flight checklist before each phase | Existing-work guard (don't redo) |
| Real audit logs — non-theatrical | Mandatory skill-reflection |
| Honest disclosure | TASK phase-tracking |
Section 7
Worked proof — the Electricity case
The method is not theoretical. This very project is a live, end-to-end application of P49, walking A0 → A13 on the formation of the electricity market. It demonstrates that the pipeline runs to completion and produces a queryable, evidence-backed result.
From A0 reconnaissance through A13 publication, the run produced 31 deep functional tracks built on roughly 850 sources, aggregated into a 298-node / 295-edge typed dependency graph, a normalized profit-pool model, and a BCG-style analytical report. The two artifacts below are the most direct evidence that the method runs end-to-end.
Section 8
One-glance summary — and when to LOOP
The whole method reduces to a single spine, punctuated by checkpoints. Read it left to right; whenever a gate's verdict fails, the run loops backward rather than pushing forward on weak foundations.
The spine: territory → ontology → dossiers → tracks → graph → coverage → contradictions → question-runs → synthesis → publication. The gates are checkpoints between the factories; on a failed verdict the run does not bluff forward — it loops back to strengthen the upstream layer (most often, back to A5 to deepen the tracks). That single discipline — proceed only on a clean verdict, loop otherwise — is what keeps P49 output structurally deep instead of merely long.
Gates make depth mandatory. You cannot skip a layer, so the output cannot be flat.