SC Execution History
Purpose
What Symphony Core actually did, derived from git commits rather than from what plans said would happen. Written to be read as context by future agent sessions before proposing work: the record here is the base rate against which any new plan should be judged.
This document describes the past. It does not set direction — that is sc-goals.md — and it does not track the present — that is sc-current-status.md and sc-weekly-scorecard.md.
Scope and window
Repository history begins 2025-10-23. The window is 9.8 months (297 days) to 2026-08-15, not twelve. Secondary figures are drawn from the personal coreOS repo where they change the reading. Regeneration commands are in Appendix A; unverifiable claims are in Appendix B.
Headline
| Metric | Value |
|---|---|
| Total commits | 893 (844 non-merge, 49 merges) |
| Days with at least one commit | 214 of 297 (72.1%) |
| Longest silence | 7 days |
Commits carrying Co-Authored-By: Claude | 800 (89.6%) |
| Distinct human contributors | 1 |
Presence is not the constraint. Something is committed on nearly three days in four, and the longest gap in the repository's life is a week.
Activity by month
| Month | Commits | Note |
|---|---|---|
| 2025-10 | 25 | Repo founded 10-23, 9 days |
| 2025-11 | 97 | Doc-framework build |
| 2025-12 | 102 | Folder reorganisation phases 1–3 |
| 2026-01 | 36 | Lowest full month |
| 2026-02 | 112 | People-ops MVP, client document kit |
| 2026-03 | 51 | Cleanup and migration tail |
| 2026-04 | 57 | Global-pivot strategy drafted |
| 2026-05 | 122 | SOP suite, pivot cutover |
| 2026-06 | 138 | Peak — Meta ads, ClickUp canon, PR workflow |
| 2026-07 | 92 | Help-centre insourcing, Banyan Tree |
| 2026-08 | 61 | Partial, 15 of 31 days |
Surge months (97–138) alternate with troughs (36–57) on a roughly three-month cycle.
The March–April 2026 trough is not a stall. coreOS took 54 and 83 commits in those months while this repo took 51 and 57. Combined output was flat to rising — attention relocated to the personal repo rather than stopping.
Where the work went
Commits touching each top-level directory. Rows exceed 100% because a commit can touch several.
| Directory | Commits | % |
|---|---|---|
04-operations/ | 276 | 30.9% |
08-reference/ | 217 | 24.3% |
09-clients/ | 189 | 21.2% |
docs/ | 149 | 16.7% |
05-platform/ | 116 | 13.0% |
03-sales/ | 111 | 12.4% |
10-people-operations/ | 83 | 9.3% |
CLAUDE.md (single file) | 77 | 8.6% |
02-marketing-brand/ | 71 | 8.0% |
06-team-training/ | 66 | 7.4% |
.claude/ | 53 | 5.9% |
01-strategy/ | 39 | 4.4% |
11-engineering/ | 31 | 3.5% |
07-customer-education/ | 20 | 2.2% |
The folder holding the company's goals, strategy, roadmap and status received less commit activity than CLAUDE.md alone.
Process versus delivery
The central measurement. Each commit is assigned a dominant category by counting files touched per bucket; the method is in Appendix A so it can be re-run.
- DELIVERY —
09-clients/,03-sales/,02-marketing-brand/,01-strategy/products|financial/ - PROCESS —
04-operations/,08-reference/,06-team-training/,docs/,CLAUDE.md,.claude/,scripts/ - OTHER — everything else
| Month | DELIVERY | PROCESS | Delivery % | Process % |
|---|---|---|---|---|
| 2025-10 | 2 | 14 | 8.0% | 56.0% |
| 2025-11 | 25 | 36 | 26.3% | 37.9% |
| 2025-12 | 20 | 65 | 20.0% | 65.0% |
| 2026-01 | 14 | 12 | 40.0% | 34.3% |
| 2026-02 | 40 | 57 | 36.0% | 51.4% |
| 2026-03 | 8 | 31 | 16.7% | 64.6% |
| 2026-04 | 23 | 15 | 44.2% | 28.8% |
| 2026-05 | 33 | 77 | 27.5% | 64.2% |
| 2026-06 | 23 | 81 | 20.0% | 70.4% |
| 2026-07 | 38 | 39 | 41.8% | 42.9% |
| 2026-08 | 17 | 29 | 33.3% | 56.9% |
| Total | 243 | 456 | 28.8% | 54.1% |
Confirmed independently two other ways: any-touch counting gives 350 vs 547; raw file-touches give 1,150 vs 2,136 (27.6% vs 51.2%).
Roughly 1.9 process commits for every delivery commit. Three findings follow:
- PROCESS exceeded DELIVERY in 9 of 11 months.
- The correlation is inverted. The two highest-output months in repo history were the two most process-heavy (2026-06 at 70.4%, 2026-05 at 64.2%). The only two months where delivery led were the two lowest-volume months. Additional energy has historically flowed into process, not delivery.
- The heaviest lifting has been self-maintenance. All fifteen of the largest commits by file count are reorganisations, renames, normalisations or migrations. Not one is content created for a client.
Client footprint
Excluding six bulk cross-client commits that inflate last-activity dates (renames, Drive migration, ClickUp ID population, doc-kit enforcement — hashes in Appendix A).
| Client | Commits | First | True last client-specific commit | Quiet |
|---|---|---|---|---|
| banyan-tree | 11 | 2026-07-29 | 2026-08-15 | 0d |
| iboost-online | 11 | 2026-05-14 | 2026-08-11 | 4d |
| rotary-sports (incl. rotary-ski) | 64 | 2025-11-19 | 2026-08-10 | 5d |
| moots-law (incl. mootslaw) | 26 | 2026-02-17 | 2026-08-10 | 5d |
| kraneworks | 27 | 2025-12-10 | 2026-07-10 | 36d |
| upscale-legal | 23 | 2025-11-03 | 2026-07-12 | 34d |
| ct-building-broker | 23 | 2025-11-07 | 2026-05-10 | 97d |
| bn-package-store | 18 | 2026-02-10 | 2026-05-11 | 96d |
| home-watch-203 | 16 | 2026-02-15 | 2026-05-11 | 96d |
| jp-floors | 9 | 2025-12-16 | 2026-04-20 | 117d |
| southbury-plaza | 9 | 2025-11-22 | 2026-04-20 | 117d |
| open-wize | 7 | 2026-02-17 | 2026-04-20 | 117d |
| ijazi-masonry | 1 | 2026-07-16 | 2026-07-16 | 30d |
The delivery portfolio is effectively four accounts. Six client folders have had no client-specific commit in 96–117 days, and for four of them the most recent touch was a repo-wide field-population pass rather than work for the client. ijazi-masonry has one commit in its entire life — the day it was onboarded.
Note the overlap with revenue: four of the six quiet accounts (southbury-plaza, ct-building-broker, jp-floors, bn-package-store) carry the largest single lines in the G1 gap.
Outbound sales has no footprint
03-sales/prospecting/ — 7 file-touches across 4 files, all templates or strategy, last touched 2026-03-02, silent 165 days. 03-sales/service-packages/ last touched 2026-02-12, silent 183 days.
Within 03-sales/lead-folders/, 74 of 93 file-touches (80%) are two names: lead-banyan-tree (47) and lead-iboost-online (27). All lead activity is inbound or referral. No outbound motion has ever run.
Strategy canon is resurrected, not maintained
| Document | Commits | Created | Last touched | Days since |
|---|---|---|---|---|
sc-goals.md | 2 | 2026-08-14 | 2026-08-14 | 1 |
sc-current-status.md | 5 | 2026-05-01 | 2026-08-10 | 5 |
sc-execution-roadmap.md | 3 | 2026-05-01 | 2026-08-10 | 5 |
sc-strategy.md | 3 | 2026-05-01 | 2026-05-31 | 76 |
Thirteen commits across four documents in 9.8 months, clustered into exactly three episodes: creation (2026-05-01), initiative-system import (2026-05-31), and consolidation (2026-07-16 → 2026-08-14). Between 31 May and 16 July — 46 days — no strategy document was edited at all, while the repo took 230 commits. The 2026-08-10 roadmap commit is titled "Q3 rollover"; its own changelog records the rollover as 41 days late.
Sprints and PRDs are write-once
Sprint backlogs live 0–3 days. Ten exist, 31 commits between them. Four have exactly one commit — created, never reopened. 2026-05-client-analytics-foundation was written on 15 May and never touched again.
The exception proves the rule: 2026-07-help-center-insourcing has 11 commits over 3 days with real execution telemetry in the trail — "WS1 progress", "WS1-3 autonomous authoring complete (10 articles)", "WS2 complete - intake form live on intake.symphonycore.com". It shipped. It is one of ten.
PRDs: 8 of 20 (40%) have exactly one commit. Median touches per PRD is 2. Only two carry implementation-status updates.
Cadence
There is none. Day-of-week distribution is flat Monday to Friday (135–162 commits, a 20% band), with 20% of commits on weekends. No Monday planning spike, no Friday review spike.
Every scheduled artifact has one or two commits total: operational-change-log.md (2 commits, both 2026-05-19, self-described as "append-only… to prevent silent drift"); session summaries (2, both 2025-12-21); the monthly KPI refresh (SOP written 2026-07-16, exactly one snapshot ever produced).
Only two rhythms have more than two observations, and both are under 30 days old: the rotary-sports client digest (v2.0 → v2.5 across 19 days, one client) and the ClickUp audit playbook (three runs, all on one client).
The rhythm that does exist is coarser: a repo-wide consolidation sprint roughly every three months — 2025-12, 2026-03, 2026-06-14, 2026-08-10. These are the four largest single-day file-churn events in the repo.
Abandonment signature
Second-level paths with ≥4 commits and ≥60 days of silence. Genuine abandonments, excluding content relocated to other repos by design:
| Path | Commits | Burst | Silent |
|---|---|---|---|
03-sales/prospecting | 4 | 2025-10 → 2026-03 | 165d |
03-sales/service-packages | 5 | 2025-11 → 2026-02 | 183d |
01-strategy/Business-Plans | 7 | 2025-10 → 2025-12 | 227d |
06-team-training/knowledge-sharing | 5 | 2025-10-29 → 11-01 | 286d |
10-people-operations/onboarding | 32 | 2025-11 → 2026-06 | 69d |
02-marketing-brand/campaigns | 18 | 2025-12 → 2026-06 | 71d |
10-people-operations/contractor-registry.md | 15 | 2026-02 → 2026-05 | 87d |
Relocations are not abandonment. 08-reference/platforms, 08-reference/tech, 08-reference/glossary and parts of 11-engineering/ show 162–165 days silent because they were moved to the reference-docs and sc-documents-archive repos in March 2026. Nine further paths went dark on a single day, 2026-06-14, in a deliberate mass-retirement merge (#74).
Named initiative outcomes visible in the trail: web-agents (15 commits in 2 weeks, Nov 2025, retired 2026-06-14); Extendly (cancelled 2026-07-03, replaced by the in-house help centre); SearchAtlas (drawn down 2026-07-24, $138 refunded); _meta/ and START-HERE.md (both retired). Deprecation is handled well. What is missing is a comparable record of things launched.
What this record implies for planning
- Effort is not the constraint. Direction is.
- Extra capacity has historically been absorbed by process work, not delivery. A plan that does not explicitly protect delivery time will lose it.
- Planning documents are written once and referenced, not maintained. A plan that requires periodic hand-updating will not get them.
- The four accounts with the largest revenue upside are among the six with no delivery activity in over three months.
- The only completion pattern with a good record is an external counterparty or a money date. Extendly's renewal, SearchAtlas's monthly bill, a GitHub issue, a coach waiting on pre-work — all completed. Internally-generated deadlines have a near-zero rate.
Related documents
- SC Goals — the two goals this record should be read against
- SC Weekly Scorecard — the live weekly loop
- SC Current Status — operating numbers
- SC Execution Roadmap — quarterly milestones
Appendix A — Regeneration
All commands read-only. Run from the repository root.
# Headline
git rev-list --count HEAD
git log --format='%b' | grep -ci 'Co-Authored-By: Claude'
git log --format='%ad' --date=short | sort -u | wc -l
# Activity by month / by ISO week
git log --format='%ad' --date=format:'%Y-%m' | sort | uniq -c | sort -k2
git log --format='%ad' --date=format:'%G-W%V' | sort | uniq -c | sort -k2 | tail -25
# Commits per top-level directory (deduplicated per commit)
git log --name-only --format='COMMIT:%h' | awk '
/^COMMIT:/{h=$0; next}
NF{ split($0,a,"/"); key=h"\t"a[1];
if(!(key in seen)){seen[key]=1; cnt[a[1]]++} }
END{ for(k in cnt) printf "%6d %s\n", cnt[k], k }' | sort -rn
# Process vs delivery — dominant category per commit, per month
git log --name-only --format='COMMIT|%h|%ad' --date=short | awk -F'|' '
function cls(l){
if(l ~ /^09-clients\//) return "DELIVERY";
if(l ~ /^03-sales\//) return "DELIVERY";
if(l ~ /^02-marketing-brand\//) return "DELIVERY";
if(l ~ /^01-strategy\/(products|financial)\//) return "DELIVERY";
if(l ~ /^(04-operations|08-reference|06-team-training)\//) return "PROCESS";
if(l ~ /^docs\//) return "PROCESS";
if(l ~ /^(CLAUDE\.md|README\.md|START-HERE\.md|\.claude\/|_meta\/|\.gitignore|scripts\/)/) return "PROCESS";
return "OTHER";
}
/^COMMIT\|/{ if(h!="") flush(); h=$2; d=$3; delete n; next }
{ if($0=="") next; n[cls($0)]++ }
function flush(){ best="EMPTY"; bv=0;
for(c in n){ if(n[c]>bv){bv=n[c]; best=c} }
print substr(d,1,7)"\t"best }
END{ if(h!="") flush() }' | sort | uniq -c
# Client footprint, excluding bulk cross-client commits
BULK='f1b8da6|66d133d|8680e14|2e3a743|cb71b81|cb27b3c|86af59a|fdba331|feaaa2e'
for c in $(git log --name-only --format='' | grep '^09-clients/' | cut -d/ -f2 | sort -u); do
printf "%-22s " "$c"
git log --format='%ad|%h|%s' --date=short -- "09-clients/$c" \
| grep -vE "\|($BULK)\|" | head -1
done
# Strategy-doc churn
for f in 01-strategy/sc-goals.md 01-strategy/sc-current-status.md \
01-strategy/sc-execution-roadmap.md 01-strategy/sc-strategy.md; do
echo "### $f"; git log --follow --format='%ad %h %s' --date=short -- "$f"
done
# Cadence: day-of-week distribution
git log --format='%ad' --date=format:'%u %a' | sort | uniq -c | sort -k2
# Largest commits by file count
git log --format='%h|%ad|%s' --date=short --no-merges | while IFS='|' read h d s; do
echo "$(git show --name-only --format='' "$h" | grep -c .)|$d|$h|$s"
done | sort -rn | head -20
Appendix B — What could not be verified
| Item | Issue |
|---|---|
| A twelve-month window | History begins 2025-10-23. Everything here covers 9.8 months. |
| Conventional-commit scope counts | 334 of 893 commits (37.4%) are free-form and carry no scope; scope counts undercount pre-2026-02 work. |
--follow across renames | The 2026-03-06 kebab-case migration breaks rename tracking. Client totals are merged by hand; commits touching both old and new paths may be double-counted (rotary-ski/rotary-sports, May 2026). |
| 49 merge commits | --name-only emits no file list for merges, so they are unclassified. 2026-06 has 23 merges — its split is computed on 115 of 138 commits. |
| The DELIVERY/PROCESS boundary | A judgment call. Reclassifying 05-platform/ (116 commits) as delivery moves the split to roughly 42%/54%. The stated figure is the conservative delivery estimate; the direction of the finding survives either way. |
| Commits as a unit of work | A one-line typo fix and a 77-file migration each count as one. Line-diff volume was not aggregated. |
| Revenue | No revenue data exists in git. "Touches delivery" is a path-proximity proxy, never a revenue measurement. Client activity here describes documentation, not billing. |
Revision History
| Date | Version | Change Description | Author |
|---|---|---|---|
| 2026-08-15 | 1.0 | Initial record, mined from 893 commits across 9.8 months. Establishes the process-versus-delivery ratio (1.9:1), the effective four-account delivery portfolio, the absence of any weekly cadence, and the write-once pattern in sprints and PRDs. Regeneration commands and unverifiable-claim register included so the document can be rebuilt rather than hand-maintained. | Symphony Core Systems Team |