Coverage Tracking

Gives the origination effort a memory. Without it, the same three directories get worked four times and half the map never gets touched — and it all feels like progress.

Read first:

• references/market-map-method.md — the map is what coverage is measured against
• references/sourcing-sources.md — the source types being tracked

Step 0 — Load

1. The market map for the sector — segments, population estimates, discovery surface
2. coverage-log.md in the workspace, creating it from templates/coverage-log.md if absent
3. targets.csv

If there is no market map: coverage cannot be measured, because there is no denominator. Say that plainly and offer to run market-map. Do not invent a population estimate to fill the gap.

Step 1 — Compute what can be computed

From targets.csv, count and never estimate:

• Rows per segment
• Rows per region
• Rows per status: identified, classified, scored, briefed
• Rows per fit_score bucket
• Rows per ownership_type in the keep pile
• Rows per email_status and owner_route
• Rows per source

From the map, take the population estimates. Label them as estimates with the map's stated method every time you use them. A coverage percentage against an estimated denominator is direction, not measurement, and presenting it as measurement is the exact failure this skill exists to prevent.

Step 2 — Establish source status

For every source in the map's discovery surface, one status:

terms prohibit is permanent and carries a reason and a date, so nobody re-attempts a source that should not be worked. See references/research-ethics.md.

Any source in the map with no status is not started — surface it as a gap, not as coverage.

Step 3 — Name the gaps

The section that stops silent truncation from reading as coverage. For each gap: what it is, why it exists, what leaving it costs, and a priority.

Gap types to check for explicitly, every time:

• Segments in the mandate that are not started or under 30 percent
• Regions in scope with few or no rows
• Sources on the map never worked
• Rows stuck at a status — identified but never classified, scored but never briefed
• Anything capped, sampled, or truncated in a previous run, and where
• Not found clusters — a segment where ownership could not be established for most rows usually means the wrong source was used, not that the data does not exist

Step 4 — Build the rescore and refresh queue

What is stale, why, and when it is due:

Step 5 — Update the log and report

Write coverage-log.md with all sections updated and a dated session-log row.

🚧 HARD GATE — before writing

Show the computed counts, the source status table, and the gap list. Get approval. Then write. Never delete a previous session-log row — the history is the value.

Coverage — Commercial fire & life safety, Southeast US
Updated 2026-08-17

SEGMENT COVERAGE (against map estimates, method in market-map)
  Commercial contract-recurring   ~180 of ~240 est.   75%   worked
  Commercial project               ~40 of ~400 est.   10%   partial
  Residential                      out of scope
  Mixed                            ~22 of ~90 est.    24%   partial

SOURCE STATUS (9 sources on the map)
  worked            4
  in progress       1
  not started       3   ← the real gap
  terms prohibit    1   (logged 2026-08-02, never retry)

PIPELINE
  Identified 164 → Classified 139 → Scored 145 → Briefed 14
  Strong Fit 14 · Fit 52 · Near Miss 41 · Not a Fit 38
  Keep pile ownership: founder 61, family 38, partner 12, Not found 25
  Reachable now: 30 by email, 22 manual route only

GAPS, ranked
  1. Three Tier 1 sources never worked (2 licence rosters, 1 certification
     registry) — est. 60-100 companies, the highest-yield remaining work
  2. Commercial project segment at 10% — decide whether it is in mandate at all
     before spending more here
  3. 25 rows with ownership Not found — all from Tier 2 sources only; the state
     registry was not checked for these
  4. Georgia and Alabama: 6 rows between them vs. ~70 est. population

CAPPED OR TRUNCATED IN PREVIOUS RUNS
  - Map density sweep stopped at 3 of 7 metros (2026-08-10) — not coverage

RESCORE / REFRESH QUEUE
  - 23 rows blocked from Strong Fit by missing signal → due now
  - Sponsor re-screen on the 66-row keep pile → due 2026-10-15
  - Market map section 3 → due 2026-10-15

THE COVERAGE SENTENCE
  "We have enumerated ~180 of an estimated 240 companies in the commercial
  contract-recurring segment in the Southeast, from 5 of 9 sources on the map,
  and the biggest untouched gap is the two state licence rosters."

Notes

• Never present an estimated coverage percentage as a measurement. State the method or do not state the number
• A gap you did not name is a gap the user thinks does not exist. This skill's whole job is to make shortfalls visible
• terms prohibit is permanent. Record it once, honour it forever
• The coverage sentence is the test. If you cannot fill it in from the log, the log is out of date — and "it feels like we've covered it" is exactly the belief this file exists to disprove
• Run this at the end of every session. It takes two minutes and it is what makes the next session start in the right place