Compare commits

...

1 Commits

Author SHA1 Message Date
Alexander Whitestone
ac328a82f5 docs: add GrepTard agentic memory report 2026-04-06 17:48:38 -04:00
2 changed files with 556 additions and 0 deletions

View File

@@ -0,0 +1,311 @@
# Agentic Memory for OpenClaw Builders
A practical structure for memory that stays useful under load.
Tag: #GrepTard
Audience: 15Grepples / OpenClaw builders
Date: 2026-04-06
## Executive Summary
If you are building an agent and asking “how should I structure memory?”, the shortest good answer is this:
Do not build one giant memory blob.
Split memory into layers with different lifetimes, different write rules, and different retrieval paths. Most memory systems become sludge because they mix live context, task scratchpad, durable facts, and long-term procedures into one bucket.
A clean system uses:
- working memory
- session memory
- durable memory
- procedural memory
- artifact memory
And it follows one hard rule:
Retrieval before generation.
If the agent can look something up in a verified artifact, it should do that before it improvises.
## The Five Layers
### 1. Working Memory
This is what the agent is actively holding right now.
Examples:
- current user prompt
- current file under edit
- last tool output
- last few conversation turns
- current objective and acceptance criteria
Properties:
- small
- hot
- disposable
- aggressively pruned
Failure mode:
If working memory gets too large, the agent starts treating noise as priority and loses the thread.
### 2. Session Memory
This is what happened during the current task or run.
Examples:
- issue number
- branch name
- commands already tried
- errors encountered
- decisions made during the run
- files already inspected
Properties:
- persists across turns inside the task
- should compact periodically
- should die when the task dies unless something deserves promotion
Failure mode:
If session memory is not compacted, every task drags a dead backpack of irrelevant state.
### 3. Durable Memory
This is what the system should remember across sessions.
Examples:
- user preferences
- stable machine facts
- repo conventions
- important credentials paths
- identity/role relationships
- recurring operator instructions
Properties:
- sparse
- curated
- stable
- high-value only
Failure mode:
If you write too much into durable memory, retrieval quality collapses. The agent starts remembering trivia instead of truth.
### 4. Procedural Memory
This is “how to do things.”
Examples:
- deployment playbooks
- debugging workflows
- recovery runbooks
- test procedures
- standard triage patterns
Properties:
- reusable
- highly structured
- often better as markdown skills or scripts than embeddings
Failure mode:
A weak system stores facts but forgets how to work. It knows things but cannot repeat success.
### 5. Artifact Memory
This is the memory outside the model.
Examples:
- issues
- pull requests
- docs
- logs
- transcripts
- databases
- config files
- code
This is the most important category because it is often the most truthful.
If your agent ignores artifact memory and tries to “remember” everything in model context, it will eventually hallucinate operational facts.
Repos are memory.
Logs are memory.
Gitea is memory.
Files are memory.
## A Good Write Policy
Before writing memory, ask:
- Will this matter later?
- Is it stable?
- Is it specific?
- Can it be verified?
- Does it belong in durable memory, or only in session scratchpad?
A good agent writes less than a naive one.
The difference is quality, not quantity.
## A Good Retrieval Order
When a new task arrives:
1. check durable memory
2. check task/session state
3. retrieve relevant artifacts
4. retrieve procedures/skills
5. only then generate free-form reasoning
That order matters.
A lot of systems do it backwards:
- think first
- search later
- rationalize the mismatch
That is how you get fluent nonsense.
## Recommended Data Shape
If you want a practical implementation, use this split:
### A. Exact State Store
Use JSON or SQLite for:
- current task state
- issue/branch associations
- event IDs
- status flags
- dedupe keys
- replay protection
This is for things that must be exact.
### B. Human-Readable Knowledge Store
Use markdown, docs, and issues for:
- runbooks
- KT docs
- architecture decisions
- user-facing reports
- operating doctrine
This is for things humans and agents both need to read.
### C. Search Index
Use full-text search for:
- logs
- transcripts
- notes
- issue bodies
- docs
This is for fast retrieval of exact phrases and operational facts.
### D. Embedding Layer
Use embeddings only as a helper for:
- fuzzy recall
- similarity search
- thematic clustering
- long-tail discovery
Do not let embeddings become your only memory system.
Semantic search is useful.
It is not truth.
## The Common Failure Modes
### 1. One Giant Vector Bucket
Everything gets embedded. Nothing gets filtered. Retrieval becomes mood-based instead of exact.
### 2. No Separation of Lifetimes
Temporary scratchpad gets treated like durable truth.
### 3. No Promotion Rules
Nothing decides what gets promoted from session memory into durable memory.
### 4. No Compaction
The system keeps dragging old state forward forever.
### 5. No Artifact Priority
The model trusts its own “memory” over the actual repo, issue tracker, logs, or config.
That last failure is the ugliest one.
## A Better Mental Model
Think of memory as a city, not a lake.
- Working memory is the desk.
- Session memory is the room.
- Durable memory is the house.
- Procedural memory is the workshop.
- Artifact memory is the town archive.
Do not pour the whole town archive onto the desk.
Retrieve what matters.
Work.
Write back only what deserves to survive.
## Why This Matters for OpenClaw
OpenClaw-style systems get useful quickly because they are flexible, channel-native, and easy to wire into real workflows.
But the risk is that state, routing, identity, and memory start to blur together.
That works at first. Then it becomes sludge.
The clean pattern is to separate:
- identity
- routing
- live task state
- durable memory
- reusable procedure
- artifact truth
This is also where Hermes quietly has the stronger pattern:
not all memory is the same, and not all truth belongs inside the model.
That does not mean “copy Hermes.”
It means steal the right lesson:
separate memory by role and by lifetime.
## Minimum Viable Agentic Memory Stack
If you want the simplest version that is still respectable, build this:
1. small working context
2. session-state SQLite file
3. durable markdown notes + stable JSON facts
4. issue/doc/log retrieval before generation
5. skill/runbook store for recurring workflows
6. compaction at the end of every serious task
That already gets you most of the way there.
## Final Recommendation
If you are unsure where to start, start here:
- Bucket 1: now
- Bucket 2: this task
- Bucket 3: durable facts
- Bucket 4: procedures
- Bucket 5: artifacts
Then add three rules:
- retrieval before generation
- promotion by filter, not by default
- compaction every cycle
That structure is simple enough to build and strong enough to scale.
## Closing
The real goal of memory is not “remember more.”
It is:
- reduce rework
- preserve truth
- repeat successful behavior
- stay honest under load
A good memory system does not make the agent feel smart.
It makes the agent less likely to lie.
#GrepTard

View File

@@ -0,0 +1,245 @@
%PDF-1.4
%“Œ‹ž ReportLab Generated PDF document (opensource)
1 0 obj
<<
/F1 2 0 R /F2 3 0 R
>>
endobj
2 0 obj
<<
/BaseFont /Helvetica /Encoding /WinAnsiEncoding /Name /F1 /Subtype /Type1 /Type /Font
>>
endobj
3 0 obj
<<
/BaseFont /Helvetica-Bold /Encoding /WinAnsiEncoding /Name /F2 /Subtype /Type1 /Type /Font
>>
endobj
4 0 obj
<<
/Contents 17 0 R /MediaBox [ 0 0 612 792 ] /Parent 16 0 R /Resources <<
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>> /Rotate 0 /Trans <<
>>
/Type /Page
>>
endobj
5 0 obj
<<
/Contents 18 0 R /MediaBox [ 0 0 612 792 ] /Parent 16 0 R /Resources <<
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>> /Rotate 0 /Trans <<
>>
/Type /Page
>>
endobj
6 0 obj
<<
/Contents 19 0 R /MediaBox [ 0 0 612 792 ] /Parent 16 0 R /Resources <<
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>> /Rotate 0 /Trans <<
>>
/Type /Page
>>
endobj
7 0 obj
<<
/Contents 20 0 R /MediaBox [ 0 0 612 792 ] /Parent 16 0 R /Resources <<
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>> /Rotate 0 /Trans <<
>>
/Type /Page
>>
endobj
8 0 obj
<<
/Contents 21 0 R /MediaBox [ 0 0 612 792 ] /Parent 16 0 R /Resources <<
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>> /Rotate 0 /Trans <<
>>
/Type /Page
>>
endobj
9 0 obj
<<
/Contents 22 0 R /MediaBox [ 0 0 612 792 ] /Parent 16 0 R /Resources <<
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>> /Rotate 0 /Trans <<
>>
/Type /Page
>>
endobj
10 0 obj
<<
/Contents 23 0 R /MediaBox [ 0 0 612 792 ] /Parent 16 0 R /Resources <<
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>> /Rotate 0 /Trans <<
>>
/Type /Page
>>
endobj
11 0 obj
<<
/Contents 24 0 R /MediaBox [ 0 0 612 792 ] /Parent 16 0 R /Resources <<
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>> /Rotate 0 /Trans <<
>>
/Type /Page
>>
endobj
12 0 obj
<<
/Contents 25 0 R /MediaBox [ 0 0 612 792 ] /Parent 16 0 R /Resources <<
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>> /Rotate 0 /Trans <<
>>
/Type /Page
>>
endobj
13 0 obj
<<
/Contents 26 0 R /MediaBox [ 0 0 612 792 ] /Parent 16 0 R /Resources <<
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>> /Rotate 0 /Trans <<
>>
/Type /Page
>>
endobj
14 0 obj
<<
/PageMode /UseNone /Pages 16 0 R /Type /Catalog
>>
endobj
15 0 obj
<<
/Author (\(anonymous\)) /CreationDate (D:20260406174739-04'00') /Creator (\(unspecified\)) /Keywords () /ModDate (D:20260406174739-04'00') /Producer (ReportLab PDF Library - \(opensource\))
/Subject (\(unspecified\)) /Title (\(anonymous\)) /Trapped /False
>>
endobj
16 0 obj
<<
/Count 10 /Kids [ 4 0 R 5 0 R 6 0 R 7 0 R 8 0 R 9 0 R 10 0 R 11 0 R 12 0 R 13 0 R ] /Type /Pages
>>
endobj
17 0 obj
<<
/Filter [ /ASCII85Decode /FlateDecode ] /Length 1202
>>
stream
Gatm:;/b2I&:Vs/3++DmOK]oT;0#utG&<!9%>Ltdp<ja\U+NO4k`V/Ns8*f_fh#+F:#]?,/b98)GB_qg7g]hl[/V`tJ2[YEH;B)I-rrSHP!JOLhA$i60Bfp08!),cE86JS,i\U@J4,<KdMrNe)8:Z-?f4fRGWMGFi5lCH&"_!I3;<":>..b&E%;l;pTQCgrn>He%kVtoVGk'hf)5rpIo-%Y5l?Vf2aI-^-$^,b4>p^T_q0gF?[`D3-KtL:K8m`p#^67P)_4O6,r@T"]@(n(p=Pf7VQY\DMqC,TS6U1T\e^E[2PMD%E&f.?4p<l:7K[7eNL>b&&,t]OMpq+s35AnnfS>+q@XS?nr+Y8i&@S%H_L@Zkf3P4`>KRnXBlL`4d_W!]3\I%&a64n%Gq1uY@1hPO;0!]I3N)*c+u,Rc7`tO?5W"_QPV8-M4N`a_,lGp_.5]-7\;.tK3:H9Gfm0ujn>U1A@J@*Q;FXPJKnnfci=q\oG*:)l<j4M'#c)EH$.Z1EZljtkn>-3F^4`o?N5`3XJZDMC/,8Uaq?-,7`uW8:P$`r,ek>17D%%K4\fJ(`*@lO%CZTGG6cF@Ikoe*gp#iCLXb#'u+\"/fKXDF0i@*BU2To6;-5e,W<$t7>4;pt(U*i6Tg1YWITNZ8!M`keUG08E5WRXVp:^=+'d]5jKWs=PEX1SSil*)-WF`4S6>:$c2TJj[=Nkhc:rg<]4TA)F\)[B#=RKe\I]4rq85Cm)je8Z"Y\jP@GcdK1,hK^Y*dK*TeKeMbOW";a;`DU4G_j3:?3;V%"?!hqm)f1n=PdhBlha\RT^[0)rda':(=qEU2K/c(4?IHP\/Wo!Zpn:($F"uh1dFXV@iRipG%Z''61X.]-);?ZT8GfKh,X'Hg`o<4sTAg2lH^:^"h4NUTU?B'JYQsFj@rEo&SEEUKY6(,aies][SXL*@3>c)<:K0)-KpC'>ls)3/)J=1GoN@nDCc'hpHslSSGWqRGNh']0nVVs9;mm=hO"?cc/mV08Q=_ca/P`9!=GEeSU3a4%fS?\Li@I93FW5-J+CH_%=t*SpX*>A"3R4_K$s0bi&i?Nk\[>EcS,$H:6J,9/Vb&]`cFjMu(u>)9Bg;:n?ks43,alko`(%YTBIJF]a0'R^6P\ZBaehJA&*qOpGC^P5]J.,RPj?'Q\.UFN>H.?nS)LMZe[kH6g38T.#T*LC_lG'C~>endstream
endobj
18 0 obj
<<
/Filter [ /ASCII85Decode /FlateDecode ] /Length 926
>>
stream
Gau1-9lJc?%#46H'g-ZWi*$')NWWHm^]plBNk6\t)lp@m=-VJ$hiDeA<cinMegV47NbRf&WL";&`;C4rDqtMC?,>\S!@!*F%>ZRX@.bNm=,Zs0fKWNW:aXAab4teOoeSs_0\e>@l*fD!GY)nNUeqW&I`I9C[AS8h`T82p%is)b&$WX!eONEKIL[d+nd%4_mV]/)Wup,IMr16[TcU=)m9h3H0Ncd70$#R6oC-WsLG8"JWS".'1J?mc4%TpP0ccY/%:^6@Lblhan.BE1+4-0mb%PaheJ.$*bN4!^CY#ss48"+HFT\qPEH"h-#dmYBXcbt'WZm>$'11!&KAlJirb9W-eu9I]S7gLenYQ^k0=ri-8<S7Oec`CEa76h8)b#B&\aD/)ai\?>7W(+"M-)"YQ>:s!fE?Ig(8]+Z;.S@rn9Rr:8_&e9Tf3DbAWcM[]bU,*"s/c;;gJO/p;UuYK8t=0i%h\Zquj1a3na;>+XaD$=lbJ%(UR&X2W=ig_kj]1lDZRm1qi!SI^4f^Y/aP,(FKi^<nZ>K^PG9%nmVof.,pCO5ihrG`W&g'@SB%_;hW+(@1pC0^QmS`IS:?.r(5'k3`XsL^;'^E%'Ni'^u*153b[V/+H$PpdJ=RR1`b;5PB7&L!imo?ZSX8/ps`00MM'lYNm_I+*s$:0.n)9=kcnKi%>)`E*b]E$Tsp\++7'Y40'7.ge+YD>!nhk$Dn.i,\5ae:#gG]1[DiiPY0Ep@\9\/lQh,/*f#ed>5qa1)Wa"%gLb,Qo@e''9^VhTr"/"<+BLOAEjAc)8r*XcY_ccKK-?IHPL6*TsYd1]lBK$Lu\5e0nI``*DkQ1/F/.\[:A(Ps&&$gB8+_;Qlo?7b^R_7&2^buP18YSDglL,9[^aoQh1-N5"CTg#F`#k)<=krf*1#s<),2]$:YkSTmXTOj~>endstream
endobj
19 0 obj
<<
/Filter [ /ASCII85Decode /FlateDecode ] /Length 942
>>
stream
Gau0BbAQ&g&A7ljp6ZVp,#U)+a.`l:TESLaa'<:lD5jC#Kr")3n%4f9UMU.2Yn4u1CbE-%YMkR/8.QZRM[&*<%X/El(l*J@/.Q.1^VYE5GZq?Cc8:35ZQQ+3Zl0FTHFogdfu7#,,`jr4:SI[QHoXt]&#,B'7VGbX+5]B`'CtnCrssGT_FRb/CGEP>^n<EiC8&I5kp+[^>%OC(9d^:+jXoC3C#'-&K2RW0!<FL('%Wf0d@MAW%FeRV`4]h9(b7-AhKYAYY`)l'ru+dY2EWPm``\J-+CJoNd:2&50%9!oSMK6@U*6^^a=@VUF0EPd,((@#o#;=rK5`gc%j)7aYA@NB-0]KLKEYR'R%pq=J>lL$9@-6&?D@^%BP#E?"lh6U9j,C^!_l^jiUqcYrt8$Rd<J/4anQ$Ib4!I(TAIUBi9$/5)>&Z(m5/]W@p>VrJgKA<0H*7/q*l&uh'-ZKOSs^Zk?3<R4%5BJpXi[$75%c1c3(,::20$m<bO$)U6#R?@4O!K]SpL_%TrFLV\Kr5pb%;+X1Io_VDC_4A'o't[p)ZLC13B^\i!O_`J_-aM:kH]6("%#L=[+$]682Hq?>$[eE7G'\gd'#2X#dLW26gCW3CAGQX1)8hn1cM13t,'E#qDIDlXCq+aX@B9(,n)nMHUolD*j]re<JYZd=cL17qAb<=]=?>6Lu@1jr45&$1BR/9E6?^EpTr?'?$sGj9u._U?OOV<CHZ!m!ri`"l-0Xf],>OlI7k\$*c<_Mr&n'7/)N@[jL4g;K1+#cC(]8($!D=4H71LjK<:K]R^I3bPLD:GnWD7`4o1rlB@aW<9X7-k^d)T*]0cp-lp`k*&IF3(lcZ)[SK^UC4k;*%S:XlI`Vgl(g;AQ.gME?L%/f^idEJ]!4@G^"Z)#nD[;<B>K_QW8XJOqtA"iZ>:SL771WKcgnEc&1i84L#qlG~>endstream
endobj
20 0 obj
<<
/Filter [ /ASCII85Decode /FlateDecode ] /Length 987
>>
stream
Gatm:;/ao;&:X)O\As09`(o<&]FB]@U`fbr`62KB`(<[e\?d3erd2q)1Vj?/CJ^fVoj.LOh.M4]'I%qgpfhnAmg=;\9n>&JC7kl)TX]RI`Th>0:S'\#N)@I>*EUX\5&:gp'T*Abo,itH7"1sR*?m0H>hmaY"7&?^#ucC28i.0(_Du=VqZ;ZD:qZFF?h!k31Wi7[GgJqbSkk*QeV#^teuo)p6bN21oM-=&UjX3!jue'B3%JD^#:nB-%"]bp16@K12XLO'CPL7H7TMf3Ui6p7Y+=of/Nn.t/9JaF/%tDfDH5Fpcj"<#eBV39)ps<)f!;;ThZR*E;,3j5L?17a%*rS^>,W5&!M-B(OQVl"ZU(a%0?"^n_)g6m$/.oX[4!d0p#)LoVNLYfd<fgNp=a<hHuuU[p"ick(M7b?7Ghm9-Y=`["$;aD[$Cii:)SoA>g6B"1>Ju;AMiM85U_[K,bFeG3WCnO@sSPs4=8+hjAH%\GYNQHn4@fW*.e3bDPVY,T]C,K4MSVL7TiR%<(Q'e!pII'<QX86En^fAPiNFE4';kSXZo%Ip\1E:[Jpf!,gN=dcamf4g-Gor9g\Y"K\b"`Gi8!$`W^p&jDP?$V9AB-)-aItX2F38VpV7;SItfle:KAj)<7!$@P)D`oJg#DHE$dF2,L>3N5P3tS<nITKDT;G7!!dIV3>>]=D7"cFZXGZlL=Z8AE23M/P@g#$-IP>@lo&,`uaM(oak.<(2&<F8ICC8PMpGRe*M"X^Q(k'Eti78p2KQ,L4^PO_);p9=%tof'esFm8I0=)ntQn&YdN7A()ts&IV\F9!Vo*O_q8B_ogb_JloTL]?MWs^fWVtemfq1J&'>rQ!Gl^h-rl=."$\:BVfXTG@qQ0MLZXpKSSLl_:PS$Gqc3'kc[Y3\i<YV5CnM`3Osf:ooPC$.b":P&i)=Ua=Ik@kmI0jL'11ie\c.RuE5qpu4E1NH['&>V_<g-eDH6LWTRbQgGN/NO~>endstream
endobj
21 0 obj
<<
/Filter [ /ASCII85Decode /FlateDecode ] /Length 976
>>
stream
Gatn%9lHLd&;KZQ'm"2lFMmZ\kbW$_[#.h^8qN:#0,kbPf"g"OlaZAdmf9d2\S/&%aqb0cn[tH]I:kQbo\oa'DZHpq\GX3pqiI)Y6P`#^%;rK)HH,\T`ZEA&PU.J95J&u`G_a(\k4Q4V@RdQ^7nUQ@aI7\=FlsdAA%]@h;JCfdQ<(F%BWt?[G6,Q35J2^:-Y2[,*I"F&311kNA#/)N06me2nE'tJcf%aP2:tM>BS<dlTb_bJk[_]\H-BIpdXna`WCAfq%/pWKKt/aGmUl:m4P"/mG-E?IB@MUP_T@_aoK;!<68JUW73*UW-oSY0l*5Mu#1_;:/nE<GWTZ:m_WKB@r'O^%1G9V[nL-R(?Jjb7@%gO#@Y(ZK)kHWQUl3rf;CA+Pnek"R18hK5S?*j6&.2R+W3OSZW9MnQ;+jQmC6e0=>"_q7Ic.KH+%qS0mfVknj$&O`'GunE3E;JiQV%+ae3U#D4Qp@rqa>l"&p97997.L4I+JO:Q`)V2=VQpQ$Km2[la-7d@:'f*JgDK?Tf!S+3;k7a&iS<"@BdNHH5W5=?=CQ1BlBmV*`&X.#?pkg09=;rOt4,"5oNKE"q:-#Br$r]$;Sc3BIc`<>N:B7E@4)j(XSFJ3DsnF>acsu"#i%,VD9ASfTGRtMG+#lM@`C>pmu))6\9tg/PSGW5F=6FD"n54&=DGb_NJ-O,25mZj0X?P$^a00jaM4U9QA+A/4c%6G/e!$TMW>6MgW&M\o9;a5NYK*UgZSOJ9D6qeAaO06$aTmT[7sACbhM'WodG,l7H2LAF@4;CH-"'BtDFLKMl*N0l,so+Y^11B[Tjp$Tkbi`j\dqRr/G=W\m=SB=%+fAb.Wlk@(_.S3(ZW0iq)%D1Mjq$S1//&hBm9n^.Zaq8=9/Q@3MV^%7@.On$P`k+6Bi23KZJ(\7\d#)Bml=jb`BY)"oCrobCdgtt>C82IdO77,t,RgjJ8mJD__R/I%aB^5$~>endstream
endobj
22 0 obj
<<
/Filter [ /ASCII85Decode /FlateDecode ] /Length 892
>>
stream
Gatn%9lnc;&;KZL'mhK*X'5(,30ol(Zia69DHmm&*Lf#d.`lPV?]PmK)(6A6LbkXELEu1gbO<'T'EWX/ok1N0\B<e$'*ZN$YCK(fK)?[-o$QKRDH8(bUl:JT0"7UlH;r-Yp-uXJ.jQk1MghS>rH<OuS^)[tKJ/O-?U60/4PVk"Tp"]Y432n;rn`bYm.D/H)3;86I?8p<5>g1&i9Vgc;^a]':`(lkcTrLfcq@$\6*s,I%PO`;MkUEY[4E56)C`$0)TRK'puELcp=^#`_a+iXFIe/djYK,VdQsB9bAN.ja-@GSB>\8RE7sds/<r%o\2WK&V(p(97sd<0D^YPA$[LQAWu#CK0QeUPg4"#DMbY4Y,2`h^%.T&e0bS$o_P5^qic.@7UN&$n6B`P"YnHdi9E#C-&!I#'f0LbL+Kl&umM9&4Hq.N6Bo?pXjE$7@$t3V@nsV!G-bD'maV5,ck,!G@k>9;fT*#m@D_QSnCIDmD6Q`4e:/%LSHSlYZC`)4c?U'sF&I-,i>SVDA1u9[gjsh9t-lk`B2@S8Q<#69&XJVQ7UbZ7_QmKXpEf%qN=\H*!BiH=iWXMfq6FOol@D&-jM4&/B)nd"=T@j@L@4Ft\!jMkQmD8;?lg?IN8=]%)dh_(*3JG(0&t#=*#i(:M?[U8*1##!TnT*0fm=i@m"1fj$E\L.=*UkIW[*i<[=Hj6s(gH*ETphfbhM`bu35Ut059Yi;&_9P(b-Pp^+I]QDTL7Cm-5kK<ctKd(+6Le)gX<+KV?hS//]aqFZEUDFf@YFmP>%dV$Z$;/g1sS)`['3g),T&l"jnbmH;3=00u^G?$1=Cg;#(0uD,G7_6fMp>>ET1>g6HM`JO>F!4d<jTHFcHc-'#!PI9kOX;t.5D_h$3RR5jTI~>endstream
endobj
23 0 obj
<<
/Filter [ /ASCII85Decode /FlateDecode ] /Length 1105
>>
stream
Gatm:bECU<&A7<Zk3/Vl#XRr'*PJ_K8sPs&QPEkT_3..m&P6X:cl;q3#(3B92%.Mu/sT`[9&8juBn/Q=%mI^n*PcLm5J?0o@jl*M#tpq9#B,LE_hQK=AeF)YB,OuMX)pKhs+Oi0'HR.rs(M>"aKO+R"VI++?@:aX%T[_PPt&:t=Ho]<rrC$q;#KO.rhRS9@+fFA'q!`c-_2`\C^9:QW4iSTjnVnqXmV[F:9#'ZAP;t>bp`pb$+<PjhU=fs'HT.?`%%N8\P2?r][kGF#;&!PmN/Edmb,H?QF]FCl+qo<gCn\]BgXI3OXqL\Z4mn/[VIGi;KtCK46fK-)`)a>P_Hq_4g[L)l#qD.iigX[G\NZcgcg[\-VVH/\"'>C>^4mT*qM5C!TR=I;n&.o]$7sc1_GUrMigamdOFkq5O5K$4hN)Y$jP\(Y0[AeM2\:a)D*#VKkkCB#/Bm<^F8@Wr.Uur#&]4eJ1a5@fgTQOkP""sT+\U\QU6>McDJR<_/K1.j]]&K^OGt\3hu2NChH[Nkd7L!hFibAW1No</'p035I,CI2CEdier6/q1#%-f2.M+LE/-qt.#"VM7-gDDdA\bRl%E[:6D#(H*OV#Y[S&q=pICBVPgC;4N\kM$!MLEj]Pm=i$q%mQ(9OEtfSR.XX\N?TkmsON4j*D*BZ'\&S=cLI/'qb+$;3ei#EO5r-@q1%!E,kn&!Tc=H89P>Wo!!HC=??_2Nd,Q??;GE@P1n9>;F>j.<6]3'@e3GcHiH8[M3<'Zr+U>nS"UOZ>$t+\uib/EY[*X4A&QJGGL'*8e^Z6QEJ2BS;XpsXYf8jbq%gR:"k]:PkIV-+KLa!_(SZaU\Ja*4B\tQU8NJ,iDU_SaXm'5!IlBaLCt_-"!s>NUV<FMaUWb4-0;5=Ti4?hDh*GKe:"HfY`p>Eq:_,Go;-EJh9\QsdQ4>iXbh3rc['8.Ks[q.%'s-^$bhV77r)JJ/NVSni'$do2"]O7)e-^kN5_iNP,3,S7]J]J4J1Y">*)RD`GW[OL*Z^-@?J'U=gAeSS1fR(O.dZ'3V_iDP&"eRA_eM#Lc4e(@.0ZijofJ,rf?[4p,^jX?Y/d0]@V.rI#8<$IfZ<4,)Q~>endstream
endobj
24 0 obj
<<
/Filter [ /ASCII85Decode /FlateDecode ] /Length 1129
>>
stream
GatU2bHBSX&Dd46ApIV9W1pA[]1_uc$hU/f.\rPMBR+-nVF4^QZE(b/:lcg2<>\,Y%E%>T3eoM(cB(=_0/e9F%D\G7^AF<!MkI#!`BapODt$]1Heu#QB,X)PYoon1ZqC5KIui4e#o!jIc2D@(9^#NWEC1$.$kFF^QGFYHQ$DROQdB-8oDlaYXV#GC?VIT5i#=*DL>lEmr0CZ]TVmR_?0JK"'brCh$Z)k]/T>"Hiei3_4:1T2&U50aQb`31_-Ei30tE//_iG..AYE&9Sha.nq'd[fX]8ltK]_9,)"0BsH&#E.]K-7e;T,\+D>\(CL95-=;8KpV:T2p8+0L;d3:cW,\WapQ,"`pA.oOV,QsO.7<:(r,K.pZ3G*5=9i-?-CLaD9d!g\QYd1+mW4T.LrM.m*/5OqJqT(N(P9eq*bZ43)In9]rX&!Gh_gu:HK7r-nYF/Qh:ZGs2rVJSVJAaDZ#1kW'c(c\:EhI+l,Gj\"GTnFJljL!u93KQoH.Z+1]UVoYNCYlKJ?a\ZeL*(uU-U;PRQhHoq\/ag#3)s`>.r`a?8TjX*/I@8N\oQpm?NOT<PW-8r4%fs&RJ_T"@P#",>cZ_=pA>3UKWPu4QjtpA#Aqo?*U5%Yk:4VPNS8`236=)m/KD%C-%Wd065pl*G-D-Y>rbkOau6OiSc,RKj#C-SFWAZl>Gr^0&pXl@.-#JE,W-H_>Z2uap[SWc"a?0.0=C=Ylq&>o@*Ct,6;VCbJWS1?/LM-jiq#M(e%;:.pn^`VFmMP+nU5]#hMb:e4SHJOM@TA0JM5L.lJC)uV!JYGCNDU1QGAe=+P"r191)0=<e0ZIC=e_]RV9f"CHgQ5N7FpkUL)ZngE4g,gJ#`F/%BtPeM=e*D0^u!#pU+G7#T@;)JZ9lCSOWQ]lk#Wq]K)6P:0Z;)nk[8.!:PYj.,35!L9Z(k*uYJ48/2R3SOuBl\%iQC>F!#`l#?q+OM%f0,Gi:D8ffYEiIZ1+QoHdEM]Du;H.<uj0s4J-UG%]0q`m9*4Yakng%DdoF8GL@rK+G3s.+oBd3MRV-B:Q@@.5Og)//&oaMY:?r0#AGWSW@,FXXMA`67J.\YR1X:&*,6$h-r3\2j)mjlFba:iD%``q)3p']OO$^k'_f)~>endstream
endobj
25 0 obj
<<
/Filter [ /ASCII85Decode /FlateDecode ] /Length 1016
>>
stream
Gatm;9lo#B&A@Zcp6`:f@`_um0qDo'SP2+Z3T_OP27U^Cd75Vb^+0J'Rki.)A3>LLf'#9a54'"''"[l+Zg%NCB5hk0JI@i&^b_:mlioZ"7e\/,ph#XR.6&jAFW*ttULl6T+7c>?;O55%gWu+;mpW@Qdhue+4/ip1Zrsg8!\3"fr8leOlmL$6#H1k<rdcRHPG)rIi6^1YpE-N$%cPIq9hVsm<>^E@bFt"HKA%8)<sY?RX'0Ffkd<bc++Q>(nLm,@I>(-kc:0hTq+qP=K@X4e(U^CuP[8l+B3K\E$KfuSK<W-4NHCDP."sJu8g1+VY?IbTN:5q5YA!;(Qb\GOjc'Sc%jG\OhSf$$#HOt7Up"Z1RAP[iFPrGjf[lc8^]u]#;MSto\)=&f`CT'RP)Zf`id%1Z"C=lB[NnIC)3jf[.cN!q[>.L](C15J)n?E%K.KWtaB12]7P7=T"8=%["j`)50O?"N0kTBa>l7BA:K@HkG>sJ@eZ6,+nU*i!B1E8U;)u08==T>.8e-F(kNf_4,tO2ZHuD1(2Bb$;FP'a9SaUJ.#,a2!fgN'W35R9u%tXVQV"R4UVKQ&DSDE_@KM,[SciKceT&2pbjN3l6M(&b,I9F@R(r$A`3dka]06XRYCAep8fbE",=%L+D"\ctiRfMSL&t*NB>[U_^m+B$Fo>gAE4TVN\eMU@W+G0+jD,e\-'m=uAOp>/X9!pecQ3u@1?!En=K,m$1kJ8O`@uZK?.XEEQ<9[?>s0@?l&QIL7IO#INB?;k5&G[J'ciL4(^2fp<d6>!U[oU>@ZsP@OB:Jd!eKDu@kWMY;q'T]'WT)2GdZTGs$5G[O%%QSkT9QeFjY7O@%WM4u1Z7@<0PI5CG"K+M9crG_*HmkY8N@27\e?8F87Q]tA?"X_1m1:1k2fu+f8agFgQ\W3e*I22C$ht*jD\di,#N&M6<[<S7IrdYC:mcjTI0Td26ATL!+/`J8oK+@5th%#C(pV3E#L2H'"5$o4jl@6pGM2&Bj!YfK[F`%h]J3~>endstream
endobj
26 0 obj
<<
/Filter [ /ASCII85Decode /FlateDecode ] /Length 257
>>
stream
Gat=dbmM<A&;9LtM?,Bq+XUCIH;m]Q]Eie6@^l$.0q.1O\$nRO;FF=sQ=X^]Dcd<CQ#-;'!t0h18j\Q31<=tN:g5Ic].s,#L'#bI`oE`_ZX$'4:Z9;]ZaMlp:<FAa_P^5AAVbZ8.5dkb7rD!-:"n54Y&p6l;F:p1j.VYm_&iqS:)AtMW.<?Rom?a^Jf<2`GMPAi*;uF@dmDk0[34X/*7%TKn>;Z<4$<q\Ld/O$S3DYaA+eE=Xt$\jLCA>2IHYJN~>endstream
endobj
xref
0 27
0000000000 65535 f
0000000061 00000 n
0000000102 00000 n
0000000209 00000 n
0000000321 00000 n
0000000516 00000 n
0000000711 00000 n
0000000906 00000 n
0000001101 00000 n
0000001296 00000 n
0000001491 00000 n
0000001687 00000 n
0000001883 00000 n
0000002079 00000 n
0000002275 00000 n
0000002345 00000 n
0000002626 00000 n
0000002745 00000 n
0000004039 00000 n
0000005056 00000 n
0000006089 00000 n
0000007167 00000 n
0000008234 00000 n
0000009217 00000 n
0000010414 00000 n
0000011635 00000 n
0000012743 00000 n
trailer
<<
/ID
[<25b005833ac6719201eda8c8a8690d7b><25b005833ac6719201eda8c8a8690d7b>]
% ReportLab generated PDF document -- digest (opensource)
/Info 15 0 R
/Root 14 0 R
/Size 27
>>
startxref
13091
%%EOF